FEBRUARY 2012 VOLUME 60 NUMBER 2 IETPAK (ISSN 0018-926X)

PART I OF TWO PARTS

SPECIAL ISSUE ON MULTIPLE-INPUT MULTIPLE-OUTPUT (MIMO) TECHNOLOGY

Guest Editorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J. W. Wallace, J. B. Andersen, B. K. Lau, B. Daneshrad, and J. Takada 434

Antenna Design, Modeling, and Analysis

Design of a MIMO Dielectric Resonator Antenna for LTE Femtocell Base Stations . . . . . . . . J.-B. Yan and J. T. Bernhard 438
A Compact Eighteen-Port Antenna Cube for MIMO Systems . . . . . . . . . . . . . . . . . . . J. Zheng, X. Gao, Z. Zhang, and Z. Feng 445
Printed MIMO-Antenna System Using Neutralization-Line Technique for Wireless USB-Dongle Applications . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S.-W. Su, C.-T. Lee, and F.-S. Chang 456
Simple and Efficient Decoupling of Compact Arrays With Parasitic Scatterers . . . . . . . . . . . . . B. K. Lau and J. B. Andersen 464
Reducing Mutual Coupling of MIMO Antennas With Parasitic Elements for Mobile Terminals . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Z. Li, Z. Du, M. Takahashi, K. Saito, and K. Ito 473
A Compact Wideband MIMO Antenna With Two Novel Bent Slits . . . . . . . . .. . . . . . . . J.-F. Li, Q.-X. Chu, and T.-G. Huang 482
Characteristic Mode Based Tradeoff Analysis of Antenna-Chassis Interactions for Multiple Antenna Terminals . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H. Li, Y. Tan, B. K. Lau, Z. Ying, and S. He 490
Multiple Antenna Systems With Inherently Decoupled Radiators . . . . . . . . . M. Pelosi, M. B. Knudsen, and G. F. Pedersen 503
A Pattern Reconfigurable U-Slot Antenna and Its Applications in MIMO Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P.-Y. Qin, Y. J. Guo, A. R. Weily, and C.-H. Liang 516
Multiple Element Antenna Efficiency and its Impact on Diversity and Capacity . . . . . . . . . . . . . J. X. Yun and R. G. Vaughan 529
On the Accuracy of Equivalent Circuit Models for Multi-Antenna Systems . . . . . . . . . . . . . . J. W. Wallace and R. Mehmood 540
Channel Sounding and Modeling
A Low-Cost MIMO Channel Sounder Architecture Without Phase Synchronization . .. . D. Pinchera and M. D. Migliore 548
Impact of Incomplete and Inaccurate Data Models on High Resolution Parameter Estimation in Multidimensional Channel
Sounding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Landmann, M. Käske, and R. S. Thomä 557
A General Coupling-Based Model Framework for Wideband MIMO Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Y. Zhang, O. Edfors, P. Hammarberg, T. Hult, X. Chen, S. Zhou, L. Xiao, and J. Wang 574

(Contents Continued on p. 433)

(Contents Continued from Front Cover)

Multi-Link MIMO Channel Modeling Using Geometry-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J. Poutanen, F. Tufvesson, K. Haneda, V.-M. Kolmonen, and P. Vainikainen 587
Land Mobile Satellite Dual Polarized MIMO Channel Along Roadside Trees: Modeling and Performance Evaluation . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Cheffena, F. P. Fontán, F. Lacoste, E. Corbel, H.-J. Mametsa, and G. Carrie 597
Empirical-Stochastic LMS-MIMO Channel Model Implementation and Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P. R. King, T. W. C. Brown, A. Kyrgiazos, and B. G. Evans 606
System Performance Evaluation
Effectiveness of Relay MIMO Transmission by Measured Outdoor Channel State Information . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K. Nishimori, N. Honma, T. Murakami, and T. Hiraguri 615
Single and Multi-User Cooperative MIMO in a Measured Urban Macrocellular Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. K. Lau, M. A. Jensen, J. Medbo, and J. Furuskog 624
User Influence on MIMO Channel Capacity for Handsets in Data Mode Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J. Ø. Nielsen, B. Yanakiev, I. B. Bonev, M. Christensen, and G. F. Pedersen 633
Exposure Compliance Methodologies for Multiple Input Multiple Output (MIMO) Enabled Networks and Terminals . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N. Perentos, S. Iskra, A. Faraone, R. J. McKenzie, G. Bit-Babik, and V. Anderson 644
MIMO Transmission Using a Single RF Source: Theory and Antenna Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . O. N. Alrabadi, J. Perruisseau-Carrier, and A. Kalis 654
MIMO Capacity Enhancement Using Parasitic Reconfigurable Aperture Antennas (RECAPs) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R. Mehmood and J. W. Wallace 665
Eigen-Coherence and Link Performance of Closed-Loop 4G Wireless in Measured Outdoor MIMO Channels . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Webb, M. Hunukumbure, and M. Beach 674
Multipath Simulator Measurements of Handset Dual Antenna Performance With Limited Number of Signal Paths . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P. Hallbjörner, J. D. Sánchez-Heredia, P. Lindberg, A. M. Martínez-González, and T. Bolin 682
On Small Terminal Antenna Correlation and Impact on MIMO Channel Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Yanakiev, J. Ø. Nielsen, M. Christensen, and G. F. Pedersen 689
Compensating for Non-Linear Amplifiers in MIMO Communications Systems . . . . . . . . . S. A. Banani and R. G. Vaughan 700

CALL FOR PAPERS

Call for Papers: Special Issue on Antennas and Propagation at Millimeter and Sub-millimeter Waves . . . . . . . . . .. . . . . . . . . . 715
 IEEE ANTENNAS AND PROPAGATION SOCIETY
All members of the IEEE are eligible for membership in the Antennas and Propagation Society and will receive on-line access to this TRANSACTIONS through IEEE Xplore upon payment of the annual Society
membership fee of $24.00. Print subscriptions to this TRANSACTIONS are available to Society members for an additional fee of $36.00. For information on joining, write to the IEEE at the address below.
Member copies of Transactions/Journals are for personal use only.
ADMINISTRATIVE COMMITTEE
S. R. BEST, President T. S. BIRD, President Elect J. S. TYO, Secretary-Treasurer
2012 2013 2014 2015
Y. M. M. ANTAR M. ANDO* R. D. NEVELS* M. SALAZAR PALMA*
J. T. BERNHARD* D. B. DAVIDSON D. ERRICOLO
S. MACI M. EL-SHENAWEE L. JOFRE
M. MARTINEZ-VÁZQUEZ S. K. RAO C. PICHOT
D. F. SIEVENPIPER M. W. SHIELDS N. RILEY
Honorary Life Members: R. C. HANSEN, W. R. STONE
*Past President
Committee Chairs and Representatives
Antenna Measurements (AMTA): S. SCHNEIDER EuRAAP Representative: W. ROSS STONE Publications: T. S. BIRD
Antennas & Wireless Propagation Letters Editor-in-Chief: Fellows Nominations Committee: L. C. KEMPEL Region 8 Representative: B. ARBESSER-RASTBURG
G. LAZZI Finance: J. S. TYO Region 9 Representative: S. BARBIN
Applied Computational EM Society (ACES): A. F. PETERSON Gold Representative: Region 10 Representative: J. L-W. LI
AP-S/URSI Joint Meetings Committee: J. L. YOUNG Historian: K. D. STEPHAN
Awards: C. G. CHRISTODOULOU IEEE Press Liaison: R. J. MAILLOUX Sensor Council: A. I. ZAGHOUL, T. S. BIRD, M. W. SHIELDS
Awards Coordinator: C. A. BALANIS IEEE Public Relations Representative: W. R. STONE Standards Committee—Antennas: M. H. FRANCIS
Best Paper Awards H. NAKANO IEEE Social Implications of Technology: R. L. HAUPT Standards Committee—Propagation: D. V. THIEL
Chapter Activities: Z. SHEN Institutional Listings: T. S. BIRD TABARC Correspondent: C. A. BALANIS
CCIR: P. MCKENNA Joint Committee on High-Power Electromagnetics: TAB Magazines Committee: W. R. STONE
Committee on Man and Radiation: K. ITO Long-Range Planning: J. L. VOLAKIS TAB New Technology Directions Committee: S. C. HAGNESS
Constitution and Bylaws: O. KILIC Magazine Editor-in-Chief: W. R. STONE TAB Transactions Committee: M. A. JENSEN
Digital Archive Editor-in-Chief: A. Q. MARTIN Meetings Coordination: S. A. LONG Transactions Editor-in-Chief: M. A. JENSEN
Distinguished Lecturers: D. R. JACKSON Membership:
Transnational Committee:
Education: D. F. KELLY Nano Technology Council: G. W. HANSON & A. AKYURTLU
EAB Continuing Education: S. R. RENGARAJAN New Technology Directions: S. C. HAGNESS USAB Committee on Information Policy: S. WEIN
Electronic Design Automation Council: M. VOUVAKIS Nominations: M. SALAZAR PALMA USAB R&D Committee: A. C. SCHELL
Electronic Publications Editor-in-Chief: S. R. BEST PACE: J. M. JOHNSON USNC/URSI: J. T. BERNHARD
EuCap Representative: J. L. VOLAKIS Women in Engineering Representative: P. F. WAHID
AP Transactions website: http://ieeeaps.org/aps_trans/ AP Transactions Manuscript Central website: http://tap-ieee.manuscriptcentral.com
Chapter Chairs
Albuquerque: MICHAEL HARRISON Florida West Coast: JING WANG Milwaukee: SHRINIVAS G. JOSHI Seoul: JAEHOON CHOI
Argentina: GUSTAVO FANO Foothill Section: FRANK FREYNE Mohawk Valley: E. PAUL RATAZZI SE Michigan: TAYFUN OZDEMIR
Atlanta: KRISHNA NAISHADHAM Fort Worth: SAIBUN TJUATJA Montreal: KE WU Siberia Section Tomsk: ROMAN V. MESCHERIAKOV
Australian Capital Territory: DAVID MURRAY France: MHAMED DRISSI Morocco: MOHAMED ESSAAIDI Singapore: LING CHUEN ONG
Baltimore: JACK SCHOFIELD FSU Georgia: GIORGI GHVEDASHVILI Moscow: DIMITRY M. SAZONOV South Africa: RIANA GESCHKE
Bangalore, India: KALARICKAPARAMBIL VINOY Fukuoka: KAZUNORI UCHIDA Nagoya, Japan: NOBUYOSHI KIKUMA South Australia: CHRISTOPHE FUMEAUX
Beijing: DONGLIN SU Germany: GEORG BOCK Nanjing: WEI HONG Southern Alberta: ELISE FEAR
Benelux: GUY VANDENBOSCH Gujarat, India: SHRI RAJEEV JYOTI New Jersey Coast: FILOMENA CITARELLA Spain: JOSE I. ALONSO
Boston: GREGORY CHARVAT Harbin, China: QUN WU New South Wales: KARU ESSELLE Spokane DON MINFORD
Bulgaria: KATYA ASPARUHOVA Hong Kong: KWOK CHUNG Nizhny Novgorod: GEORGE L. PAKHOMOV Springfield: PAUL R. SIQUEIRA
Calcutta: DEBATOSH GUHA Houston: STUART A. LONG North Italy: GIUSEPPE VECCHI St Petersburg: SVETLANA P. ZUBKO
Central & South Italy: LUCIANO TARRICONE Hungary: LAJOS NAGY North Jersey: HAR DAYAL, KIRIT DIXIT Sweden: ANDERS RYDBERG
Central Texas: JEREMY PRUITT Huntsville: ERIC R. GRIGORIAN Norway: KARL MARTIN GJERTSEN Switzerland: MICHAEL MATTES
Chicago: DANILO ERRICOLO Hyderabad, India: LAKSHMI MERUGU Orlando: SIAMAK EBADI Syracuse: HAKAN PARTAL
Cleveland: MAX SCARDELLETTI Indonesia: FITRI YULI ZULKIFLI Ottawa: QIUBO YE Tainan: WEN-SHAN CHEN
College Station: GREG HUFF Israel: SHAMUEL AUSTER Philadelphia: JACK NACHAMKIN Taipei: HSI-TSENG CHOU
Columbus: DIMITRIS PSYCHOUDAKIS Japan Council: KEIZO CHO Phoenix: STEVE ROCKWELL Thailand: TUPTIM ANGKAEW
Connecticut: CHARLOTTE BLAIR Kansai: TAKESHI MANABE Poland: JERZY PIOTROWSKI Toronto: GEORGE ELEFTHERIADES
Croatia: RADOVAN ZENTNER Kharkov: OLEKSANDR M. DUMIN Portugal: NUNO BORGES DE CARVALHO Turkey: IBRAHIM TEKIN
Czech-Slovakia: JIRI DRINOVSKY Lebanon: ELIAS NASSAR Princeton-Central Jersey: ALLEN KATZ Tucson: HAO XIN
Dallas: MURALI IYENGAR Lithuania: VYTAUTAS URBANAVICIUS Queensland: ASHLEY ROBINSON UK/RI: ALI REZAZADEH
Dayton: ANDREW TERZUOLI Long Island: BRYAN TROPPER Rio de Janeiro: JOSE RICARDO BERGMANN Vancouver: ALON NEWTON
Denver-Boulder: MICHAEL JANEZIC Los Angeles: JOHN GIANVITTORIO San Diego: JERRY TWOMEY Victoria: KAMRAN GHORBANI
East Ukraine: KONSTYANTYN V. ILYENKO Macau: KAM WENG TAM Santa Clara Valley - San Francisco: PAUL HARMS Washington, DC: BRIAN RIELY
Eastern North Carolina: TODD NICHOLS Malaysia: MOHAMAD A. RAHIM Saratov/Penza: NIKITA M. RYSKIN West Ukraine: IRYNA IVASENKO
Egypt: HADIA EL-HENNAWY Melbourne, FL: RICK BOTSFORD Seattle: LIXIN CAI Winnipeg: VLADIMIR OKHMATOVSKI
Finland: ARTTU LUUKANEN
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION
Is the leading international engineering journal on the general topics of electromagnetics, antennas and wave propagation. The journal is devoted to antennas, including analysis, design, development, measure-
ment, and testing; radiation, propagation, and the interaction of electromagnetic waves with discrete and continuous media; and applications and systems pertinent to antennas, propagation, and sensing, such
as applied optics, millimeter- and sub-millimeter-wave techniques, antenna signal processing and control, radio astronomy, and propagation and radiation aspects of terrestrial and space-based communication,
including wireless, mobile, satellite, and telecommunications. Author contributions of relevant full length papers and shorter Communications are welcomed.
See inside back cover for Editorial Board.
IEEE Officers
GORDON W. DAY, President MICHAEL R. LIGHTNER, Vice President, Educational Activities
PETER W. STAECKER, President-Elect DAVID A. HODGES, Vice President, Publication Services and Products
CELIA L. DESMOND, Secretary HOWARD E. MICHEL, Vice President, Member and Geographic Activities
HAROLD L. FLESCHER, Treasurer STEVE M. MILLS, President, Standards Association
MOSHE KAM, Past President FREDERICK C. MINTZER, Vice President, Technical Activities
JAMES M. HOWARD, President, IEEE-USA
PETER N. CLOUT, Director, Division IV—Electromagnetics and Radiation
IEEE Executive Staff
DR. E. JAMES PRENDERGAST, Executive Director & Chief Operating Officer
THOMAS SIEGERT, Business Administration ALEXANDER PASIK, Information Technology
MATTHEW LOEB, Corporate Activities PATRICK MAHONEY, Marketing
DOUGLAS GORHAM, Educational Activities CECELIA JANKOWSKI, Member and Geographic Activities
EILEEN LACH, General Counsel & Corporate Compliance ANTHONY DURNIAK, Publications Activities
BETSY DAVIS, SPHR, Human Resources JUDITH GORMAN, Standards Activities
CHRIS BRANTLEY, IEEE-USA MARY WARD-CALLAN, Technical Activities
IEEE Periodicals
Transactions/Journals Department
Staff Director: FRAN ZAPPULLA
Editorial Director: DAWN MELLEY Production Director: PETER M. TUOHY
Senior Managing Editor: WILLIAM A. COLACCHIO Associate Editor: DAWN L. MENENDEZ
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION (ISSN 0018-926X) is published monthly by the Institute of Electrical and Electronics Engineers, Inc. Responsibility for the contents rests upon the
authors and not upon the IEEE, the Society/Council, or its members. IEEE Corporate Office: 3 Park Avenue, 17th Floor, New York, NY 10016-5997. IEEE Operations Center: 445 Hoes Lane, Piscataway,
NJ 08854-4141. NJ Telephone: +1 732 981 0060. Price/Publication Information: Individual copies: IEEE Members $20.00 (first copy only), nonmembers $126.00. (Note: Postage and handling charge not
included.) Member and nonmember subscription prices available upon request. Copyright and Reprint Permissions: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy
for private use of patrons, provided the per-copy fee indicated in the code at the bottom of the first page is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For all
other copying, reprint, or republication permission, write to Copyrights and Permissions Department, IEEE Publications Administration, 445 Hoes Lane, Piscataway, NJ 08854-4141. Copyright © 2012 by The
Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Periodicals Postage Paid at New York, NY and at additional mailing offices. Postmaster: Send address changes to IEEE TRANSACTIONS
ON ANTENNAS AND PROPAGATION, IEEE, 445 Hoes Lane, Piscataway, NJ 08854-4141. GST Registration No. 125634188. CPC Sales Agreement #40013087. Return undeliverable Canada addresses to: Pitney
Bowes IMEX, P.O. Box 4332, Stanton Rd., Toronto, ON M5W 3J4, Canada. IEEE prohibits discrimination, harassment, and bullying. For more information visit http://www.ieee.org/nondiscrimination. Printed
in U.S.A.

Digital Object Identifier 10.1109/TAP.2012.2186030

434 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Guest Editorial for the Special Issue on Multiple-Input
Multiple-Output (MIMO)
W E are pleased to present this special issue on mul-
tiple-input multiple-output (MIMO), which represents
a breakthrough in the use of antenna arrays in wireless trans-
mission. Unlike traditional phased array or diversity techniques
that enhance one signal of interest, MIMO systems employ
antenna arrays jointly at the transmitter and receiver to spatially
multiplex signals, providing tremendous capacity gains.
Although there has already been intense research in MIMO
wireless communications, and many obstacles in signal pro-
cessing, modulation, and coding for MIMO systems have been
overcome, outstanding questions in the areas of antennas and
propagation remain, making MIMO a timely topic for our
community. The need for research in this area becomes even
more apparent as new standards such as IEEE 802.11n, LTE
Advanced, and WiMAX that include MIMO operation are
implemented, revealing that physical devices, antennas, and
channels can no longer be oversimplified or neglected.
This special issue is organized into three main sections: 1) an-
tenna design, modeling, and analysis, 2) channel sounding and
modeling, and 3) system performance evaluation.
A. Antenna Design, Modeling, and Analysis
Although signal processing treatments of MIMO may treat
antennas as isotropic elements that are not affected by nearby
antennas or scatterers, real antennas exhibit non-isotropic pat-
terns and inter-element coupling. This section contains papers
that consider the challenges of designing compact MIMO an-
tennas with good performance, as well as novel and rigorous
ways to model and analyze such antenna systems.
Exploiting multiple polarizations is a possible method of
achieving a compact MIMO design with low coupling. Yan
and Bernhard present a clever design allowing two orthogonal
resonant modes of a compact dielectric resonator antenna
(DRA) for LTE700 femtocell applications, achieving polariza-
tion and angle diversities and 30 dB isolation. A low profile
tri-polarized antenna consisting of a dual-polarized ring patch
and a disk-loaded monopole is explored in Zheng et al. to build
an 18-port antenna cube, exhibiting lower mutual coupling and
simpler feeding than a dipole MIMO cube.
Several papers address the challenge of MIMO antenna de-
sign for compact user terminals exhibiting higher mutual cou-
pling and correlation. Su et al. implement a printed neutral-
ization line along one ground plane edge to decouple a two-
monopole array for a USB dongle application at 2.4 GHz, re-
quiring little modification of the ground plane. The use of par-
asitic structures for coupling mitigation is explored in several
contributions. Lau and Bach Andersen introduce the theory of
parasitic decoupling, whereby two arbitrary antennas of a given
antenna spacing can be perfectly decoupled with a reactively
loaded parasitic element acting as a reflector. Experiments re-
veal that decoupling is achieved with only a small penalty in
total efficiency. Z. Li et al. introduce a complementary perspec-
tive that the parasitic elements create a second path for cou-
Digital Object Identifier 10.1109/TAP.2012.2183909
0018-926X/$31.00 © 2012 IEEE
pling cancellation, demonstrating the principle by decoupling
two closely-coupled slot antennas using two monopoles as par-
asitic elements. J. Li et al. design an efficient wideband MIMO
antenna by combining a parasitic decoupling strip with right-an-
gled slits in the ground plane to obtain 2.4 GHz–6.55 GHz op-
eration and 18 dB isolation.
The ground plane of compact user terminals can play a major
role in the radiation of MIMO antennas at low frequency where
the chassis is excited. The theory of characteristic mode is ex-
plored by H. Li et al. in the context of designing efficient MIMO
antennas by placing the elements to avoid simultaneous ex-
citation of the chassis by more than one antenna element. A
tradeoff analysis shows that MIMO performance is significantly
improved by the increased isolation. Pelosi et al. carry out a
comprehensive study on the performance of small narrowband
antennas with and without a user in either MIMO mode or trans-
ceiver separation mode (TSM). This approach can relax the
duplex filter requirement in TSM, although user effects may
largely influence the antenna performance.
In order to further improve MIMO antenna performance in
a time-varying propagation channel, reconfigurable antenna el-
ements may be employed to optimize the antenna-channel in-
teraction. Qin et al. show that two pattern reconfigurable U-slot
antenna elements can provide capacity gain in measured line-of-
sight (LOS) and non-LOS channels, relative to two omnidirec-
tional reference antennas.
Metrics and models for MIMO antennas are considered by
two contributions. Yun and Vaughan isolate the role of antenna
efficiencies from correlation in the diversity and capacity per-
formance of a given MIMO antenna. Thereafter, the MIMO an-
tenna can be represented with an equivalent number of ideal an-
tenna branches that are called diversity order and capacity order,
respectively. The question of the validity and accuracy of equiv-
alent circuit models for MIMO arrays is addressed by Wallace
and Mehmood, where a method-of-moments analysis based on
first principles reveals that such models are exact under normal
circumstances, and that transmit and receive modes can be ana-
lyzed with a single unified model.
B. Channel Sounding and Modeling
This section focuses on accurate channel characterization
through sounding and modeling, which is vital to correctly
assess the benefits of MIMO transmission, allowing critical
tradeoffs and design decisions to be made.
Two papers in this section directly consider the topic of
channel sounding. Pinchera and Migliore present an interesting
measurement approach using a parasitic array instead of a
switched array. Using low cost switched parasitic elements
instead of a large multiport microwave switch dramatically
reduces the cost of MIMO channel sounding with only modest
reduction in accuracy. The impact of an imperfect underlying
model on the accuracy of high-resolution double-directional
MIMO channel estimation is studied by Landmann et al. It is
shown that modeling this uncertainty allows multipath to be
correctly classified as discrete or diffuse, and that imperfect
calibration can lead to large error in multipath estimates.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Sounder-based channel modeling is considered in two pa-
pers. Zhang et al. extend tensor-based MIMO modeling ap-
proaches to the case of wide bandwidth, which is required for
today’s wireless standards. The model is assessed using mea-
sured indoor channels, indicating a tradeoff between complexity
and accuracy when generating synthetic MIMO channel data.
Poutanen et al. propose a method for extending geometry-based
stochastic channel models to the case of multiple links, which
is important to analyze MIMO systems using coordinated trans-
mission or relays. This model is accomplished by having certain
clusters that are shared by the links, creating dependence in the
statistics of the MIMO channels.
Finally, this section includes two papers that present measure-
ment of land mobile satellite (LMS) channels. Cheffena et al.
consider the effect of signal shadowing by trees in MIMO-LMS
links, proposing a multipath model for trees based on multiple
scattering theory. The model is compared with direct FDTD
simulation, indicating that good accuracy can be obtained with
modest complexity. King et al. investigate the use of multiple
antennas to increase the capacity of LMS networks, where a
Markov chain is employed to characterize the time-variant na-
ture of shadowing and depolarization effects. The utility of the
proposed technique is illustrated through direct measurements
with an artificial LMS platform.
C. System Performance Evaluation
The final section deals principally with system-level aspects,
indicating how detailed characteristics of the propagation
channel, antennas, and devices affect the performance of the
overall MIMO system or network.
Two papers consider the emerging topic of relays and co-
ordinated MIMO transmission. Nishimori et al. evaluate the
capacity of relay-enhanced multi-antenna transmission in a
cellular environment through direct propagation measurements
taken in Yokkaichi City, Japan. This study shows that charac-
terizing path-loss differences is critical and that relay-enhanced
MIMO can provide a 50% improvement in capacity. Lau et
al. analyze urban propagation measurements involving three
coherent base stations and a mobile unit equipped with four
antennas. Capacity for cooperative transmission from the base
stations is analyzed, revealing dramatic sum-rate capacity gains
compared to non-cooperative methods.
User influence and exposure limits are considered in two con-
tributed papers. Nielsen et al. provide a detailed study of user
influence on the outage capacity for mobile devices in the data
mode operation. Six different handsets at two bands are char-
acterized for twelve different users, showing that handset de-
sign and hand position critically impact body loss, mean effec-
tive gain, and outage capacity. Perentos et al. consider compli-
ance and exposure testing of MIMO devices, which is important
as multi-antenna technology is increasingly incorporated into
advanced devices. The developed methodologies allow such
testing to be performed with scalar field probes, avoiding ex-
pensive upgrades of existing test equipment.
The use of parasitic arrays for MIMO transmission are con-
sidered in two papers, providing reduced complexity or capacity
enhancement compared to classical MIMO systems. Alrabadi
et al. develop the methodology of using a switched parasitic
array (SPA) with only a single active RF source to replace a
full MIMO transmitter with reduced cost and complexity. The
generalized method for forming the required orthogonal bases is
demonstrated through simulation and direct measurement with
435
a prototype SPA. Mehmood and Wallace propose flexible re-
configurable aperture (RECAP) antennas to increase MIMO ca-
pacity in interference-limited scenarios. Multi-user simulations
with a detailed noise model suggest that high reconfigurability
can lead to many-fold capacity increase.
Finally, four papers are included that extend or verify as-
sumptions made in existing modeling approaches for MIMO
systems. Webb et al. consider the coherence time and bandwidth
of channel state information in measured time-varying urban
channels, indicating how sensitive feedback methods are to time
and frequency offsets. The study shows that controlling the feed-
back rate can lead to significant improvements in mobile MIMO
systems. Hallbjörner et al. explore the impact of sparse multipath
on antenna correlation and diversity, in contrast to classical
treatments where infinite and uniform arrivals are assumed. Mul-
tipath channels are simulated using antenna arrays in an anechoic
chamber, showing that sparse multipath can lead to high vari-
ability or spread of channel statistics like correlation. Yanakiev
et al. study the use of correlation as a metric in the design stage to
predict handset performance in terms of MIMO capacity in real
scenarios. The surprising result is that correlation may have little
bearing on capacity, indicating correlation may be a misleading
figure of merit. Finally, Banani and Vaughan investigate the ef-
fect of non-linear amplifiers in practical MIMO systems and how
to compensate the resulting degradations to channel capacity. A
model for non-linear MIMO systems is introduced, and a blind
channel-estimation technique is developed to estimate and
track the channel in the presence of non-linearities.
To conclude this guest editorial, we would like to thank the
former Editor-in-Chief Dr. Trevor S. Bird and his successor
Prof. Michael A. Jensen, for providing us with the opportunity
to coordinate and organize this special issue and for their con-
tinued support throughout the process. We are also grateful to
the many anonymous reviewers who helped make the special
issue possible. We believe that the issue provides a true snap-
shot of the state-of-the-art in antennas and propagation research
in MIMO systems, serving as interesting reading as well as a
useful reference for years to come.
JON W. WALLACE, Guest Editor
School of Engineering and Science
Jacobs University
Bremen, Germany
JØRGEN BACH ANDERSEN, Guest Editor
Department of Electronic Systems
Aalborg University
Aalborg, Denmark
BUON KIONG LAU, Guest Editor
Department of Electrical and Information Technology
Lund University
Lund, Sweden
BABAK DANESHRAD, Guest Editor
Department of Electrical Engineering
University of California, Los Angeles
Los Angeles, CA
JUN-ICHI TAKADA, Guest Editor
Graduate School of Engineering
Tokyo Institute of Technology
Tokyo, Japan
436 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Jon W. Wallace (S’99–M’03) received the B.S. (summa cum laude) and Ph.D. degrees in electrical
engineering from Brigham Young University (BYU), Provo, UT, in 1997 and 2002, respectively.
From 1995 to 1997, he worked as an Associate of Novell, Inc., Provo, and during 1997 he was a
Member of Technical Staff for Lucent Technologies, Denver, CO. He received the National Science
Foundation Graduate Fellowship in 1998 and worked as a Graduate Research Assistant at BYU
until 2002. From 2002 to 2003, he was with the Mobile Communications Group, Vienna University
of Technology, Vienna, Austria. From 2003 to 2006, he was a Research Associate with the BYU
Wireless Communications Laboratory. Since 2006, he has been Assistant Professor of electrical
engineering at Jacobs University, Bremen, Germany. His current research interests include wireless
channel sounding and modeling, physical-layer security, MIMO communications, cognitive radio,
and ultrawideband (UWB) systems.
Dr. Wallace currently serves as an Associate Editor of the IEEE TRANSACTIONS ON ANTENNAS
AND PROPAGATION. He was awarded the H. A. Wheeler paper award in the IEEE TRANSACTIONS
ON ANTENNAS AND PROPAGATION in 2002.

Jørgen Bach Andersen (M’68–SM’78–F’92–LF’02) received the M.Sc. and Dr.Techn. degrees
from the Technical University of Denmark (DTU), Lyngby, Denmark, in 1961 and 1971, respec-
tively. In 2003 he was awarded an honorary degree from Lund University, Sweden.
From 1961 to 1973, he was with the Electromagnetics Institute, DTU and since 1973 he has been
with Aalborg University, Aalborg, Denmark, where he is now a Professor Emeritus and Consultant.
He was head of a research center, Center for Personal Communications, CPK, from 1993–2003.
He has been a Visiting Professor in Tucson, AZ; Christchurch, New Zealand; Vienna, Austria; and
Lund, Sweden. He has published widely on antennas, radio wave propagation, and communica-
tions, and has also worked on biological effects of electromagnetic systems. He has coauthored a
book, Channels, Propagation and Antennas for Mobile Communications (IEE, 2003). He was on
the management committee for COST 231 and 259, a collaborative European program on mobile
communications.
Prof. Andersen is a former Vice President of the International Union of Radio Science (URSI)
from which he was awarded the John Howard Dellinger Gold Medal in 2005.

Babak Daneshrad received the B.Eng. and M.Eng. degrees with emphasis in communications
from McGill University, Montreal, Quebec, Canada, in 1986 and 1988, respectively, and the Ph.D.
degree with emphasis in integrated circuits and systems from the University of California, Los
Angeles (UCLA), in 1993.
In January 2001, he co-founded Innovics Wireless, a company focused on developing 3G cellular
mobile terminal antenna diversity solutions and in 2004 he co-founded Silvus Communications.
From 1993 to 1996, he was a member of technical staff with the Wireless Communications Sys-
tems Research Department, AT&T Bell Laboratories, where he was involved in the design and
implementation of systems for high-speed wireless packet communications. Currently, he is a Pro-
fessor with the Electrical Engineering Department, UCLA. His research interests are in the areas
of wireless communication system design, experimental wireless systems, and VLSI for commu-
nications. His current research interests are cross disciplinary in nature and deal with addressing
practical issues associated with the realization of advanced wireless systems. The work is focused
on low power MIMO wireless systems, as well as cognitive radio communications.
Prof. Daneshrad is the recipient of the 2005 Okawa Foundation award, a coauthor of the best paper award at PADS 2004, and
was awarded first prize in the DAC 2003 design contest. He is the beneficiary of the endowment for “UCLA-Industry Partnership
for Wireless Communications and Integrated Systems.”
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012 437

Buon Kiong Lau (S’00–M’03–SM’07) received the B.E. degree (with honors) from the University
of Western Australia, Perth, Australia and the Ph.D. degree from Curtin University of Technology,
Perth, in 1998 and 2003, respectively, both in electrical engineering.
During 2000 to 2001, he worked as a Research Engineer with Ericsson Research, Kista, Sweden.
From 2003 to 2004, he was a Guest Research Fellow at the Department of Signal Processing,
Blekinge Institute of Technology, Sweden. Since 2004, he has been at the Department of Electrical
and Information Technology, Lund University, where he is now an Associate Professor. He has
been a Visiting Researcher at the Department of Applied Mathematics, Hong Kong Polytechnic
University, China, Laboratory for Information and Decision Systems, Massachusetts Institute of
Technology, and Takada Laboratory, Tokyo Institute of Technology, Japan. His primary research
interests are in various aspects of multiple antenna systems, particularly the interplay between an-
tennas, propagation channels and signal processing.
Dr. Lau is an Associate Editor for the IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION.
From 2007 to 2010, he was a Co-Chair of Subworking Group 2.2 on “Compact Antenna Systems for Terminals” (CAST) within
EU COST Action 2100. Since 2011, he is a Swedish national delegate and the Chair of Subworking Group 1.1 on “Antenna System
Aspects” within COST IC1004.

Jun-ichi Takada (SM’11) received B.E. and D.E. degrees from Tokyo Institute of Technology
(Tokyo Tech), Japan, in 1987 and 1992, respectively.
He was a Research Associate at Chiba University from 1992 to 1994, and an Associate Professor
at Tokyo Tech from 1994 to 2006 where he has been a Professor since 2006. From 2003 to 2007,
he was also a Researcher at the National Institute of Information and Communications Technology
(NICT), Japan. His current interests include the radiowave propagation and channel modeling for
various wireless systems, and regulatory issues of spectrum sharing.
438 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Design of a MIMO Dielectric Resonator Antenna for
LTE Femtocell Base Stations
Jie-Bang Yan, Member, IEEE, and Jennifer T. Bernhard, Fellow, IEEE
Abstract—A novel multiple-input multiple-output (MIMO) di-
electric resonator antenna (DRA) for long term evolution (LTE)
femtocell base stations is described. The proposed antenna is able
to transmit and receive information independently using TE and
HE modes in the LTE bands 12 (698–716 MHz, 728–746 MHz)
and 17 (704–716 MHz, 734–746 MHz). A systematic design method
based on perturbation theory is proposed to induce mode degen-
eration for MIMO operation. Through perturbing the boundary
of the DRA, the amount of energy stored by a specific mode is
changed as well as the resonant frequency of that mode. Hence, by
introducing an adequate boundary perturbation, the TE and HE
modes of the DRA will resonate at the same frequency and share
a common impedance bandwidth. The simulated mutual coupling
between the modes was as low as . It was estimated that in
a rich scattering environment with an Signal-to-Noise Ratio (SNR)
of 20 dB per receiver branch, the proposed MIMO DRA was able
to achieve a channel capacity of 11.1 b/s/Hz (as compared to theo-
retical maximum 2 2 capacity of 13.4 b/s/Hz). Our experimental
measurements successfully demonstrated the design methodology
proposed in this work.
Index Terms—Dielectric resonator antenna (DRA), long term
evolution (LTE), multiple-input multiple-output (MIMO) antenna,
mutual coupling, perturbation method.
I. INTRODUCTION
T HE Federal Communications Commission (FCC) re-
cently released the 700 MHz spectrum which was
previously used for analog television broadcasting [1]. A new
nationwide wireless broadband network based on long term
evolution (LTE) technology has been proposed to operate in
the 700 MHz spectrum [2], [3]. In the LTE Evolved UMTS
terrestrial radio access (E-UTRA) air interface, multiple-input
multiple-output (MIMO) technology plays an important role
in increasing the system’s spectral efficiency [4], [5]. Given
the lower operating frequency of the LTE system, as compared
to the WiFi and cellular standards, the antenna in handheld
devices such as a smartphone or a netbook must be electrically
Manuscript received May 27, 2010; revised December 14, 2010; accepted
February 05, 2011. Date of publication October 28, 2011; date of current version
February 03, 2012. This work was supported by the Motorola Center for Com-
munications at the University of Illinois at Urbana-Champaign and a Croucher
Foundation Scholarship.
J.-B. Yan was with the Electromagnetics Laboratory, Department of Elec-
trical and Computer Engineering, University of Illinois at Urbana-Champaign,
Urbana, IL 61801 USA. He is now with the Center for Remote Sensing of Ice
Sheets (CReSIS), University of Kansas, Lawrence, KS 66045 USA.
J. T. Bernhard is with the Electromagnetics Laboratory, Department of Elec-
trical and Computer Engineering, University of Illinois at Urbana-Champaign,
Urbana, IL 61801 USA (e-mail: jbernhar@illinois.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2174021
0018-926X/$26.00 © 2011 IEEE
small. This implies the mobile antennas are likely to be ineffi-
cient and the coverage of the system is therefore limited. This
is especially true if MIMO operation is needed at both mobile
and base station since the antenna efficiency would be further
reduced due to strong mutual coupling between closely-packed
mobile antennas. In view of this, LTE architecture includes a
femtocell solution for coverage extension [6]. Femtocells can
be considered as low-power access points serving indoor areas.
To exploit the richness in multipath propagation in indoor
scenarios, it is desired to employ MIMO antennas with a very
low mutual coupling as the base station antenna in a femtocell.
One possible solution would be the orthogonally polarized
MIMO antennas proposed in [7]. However the problem is that
such antennas would be oversized when scaled to operate at
700 MHz. Hence, a new MIMO antenna solution for LTE’s
femtocell base station is necessary.
In this work, a 700 MHz dual-mode MIMO dielectric res-
onator antenna (DRA) that is suitable for the new wireless
system is proposed. Although the cost of DRAs may be high
as compared to traditional PIFAs or microstrip antennas, they
have the advantages of compact size, high radiation efficiency,
and wide impedance bandwidth [8]. Another important feature
of DRAs is that the three dimensional structure offers more
degrees of freedom in exciting various orthogonal resonant
modes, and each mode can be utilized to transmit and receive
information independently. This makes the DRA an ideal
candidate for application in MIMO communication systems.
Indeed, a multi-mode usage of a single dielectric resonator
has been suggested in [9], but the emphasis is not on MIMO
applications.
The concept of a MIMO DRA was first described and
demonstrated by Ishimiya et al. in [10], [11]. It was experi-
mentally shown that a cubic MIMO DRA is able to achieve a
diversity gain of about 10 dB and has comparable performance
to traditional MIMO dipole arrays in practical IEEE 802.11n
systems. Nevertheless, in Ishimiya’s papers, no explicit design
method has been described. The major difficulty of applying
DRAs in MIMO systems is to make various modes to resonate
at the same frequency while maintaining low coupling between
the modes. Here, we introduce a systematic design method for
MIMO DRAs. The key in MIMO DRA design is to induce
degenerate modes (i.e., modes that have the same resonant
frequency). Conventionally, only DRAs that exhibit symmetry
can support degenerate modes [12] and this limits any further
size reduction of MIMO DRAs. Hence, a novel mode degener-
ation method based on boundary perturbation is proposed and
demonstrated in this work.
Section II describes the base design for the proposed MIMO
DRA, then, Section III introduces the boundary perturbation for
YAN AND BERNHARD: DESIGN OF A MIMO DIELECTRIC RESONATOR ANTENNA FOR LTE FEMTOCELL BASE STATIONS 439

Fig. 1. Perspective view of the split-cylindrical DRA ( ,

, , , and ).

mode degeneration. In Section IV, we evaluate the performance

of the perturbed antenna structure. Simulated results including
those for MIMO capacity are provided. Following that, some
experimental results are given in Section V as a validation to
the developed design methodology. Finally a conclusion and a
discussion of future work are given in Section VI.
Fig. 2. Theoretical magnetic field distributions for the (a) mode, and
II. BASE DESIGN (b) mode.

Consider a split-cylindrical DRA , with a radius

of 44 mm and a length of 80 mm residing on a ground plane
with dimensions as shown in Fig. 1. The mode and the
mode can be excited simultaneously using appropriate
excitation methods, such as probe feeds, aperture coupling or
microstrip feeds. The value of the subscript ranges between
zero and one, depending on the method of feeding [12]. Here,
a 50 microstrip-fed rectangular slot and a probe feed were
chosen to excite the and modes, respectively
(see Fig. 1). FR-4 epoxy board with thickness of 1.6
mm is used as the substrate of the microstrip line. The dimen-
sions of the slot are 50 mm 4 mm and the probe that excites
the mode has a length of 27 mm.
Fig. 2 shows the plots of the theoretical magnetic field distri-
butions for the two modes inside the DRA computed using Wol-
fram Research Mathematica [13]. It can be seen that
mode behaves as a magnetic dipole on the -axis while
mode radiates as a short magnetic dipole oriented along the Fig. 3. Simulated -parameters of the unperturbed cylindrical DRA.
-axis. The two modes are therefore orthogonal to each other
and should exhibit low mutual coupling. The resonant frequen-
cies of the mode and mode can be derived from simulation was performed using Ansys HFSS [14] and the sim-
the separation equation [8] and are found to be 653 MHz and ulated -parameters of the antenna are shown in Fig. 3. The
520 MHz, respectively, theoretically predicted and simulated operating frequencies of
the modes agree very well with each other. It can also be seen
that the coupling between the two modes is very low as
expected.
(1)
III. DESIGN OF MIMO DRA
(2) A. Boundary Perturbation
In order to work in a MIMO system, the two modes should
where is the speed of light in free space, and and are have the same resonant frequency and have a shared impedance
the first zeros of the zero-order Bessel function and the deriva- bandwidth. To accomplish this, we propose a mode degenera-
tive of the first-order Bessel function, respectively. A full-wave tion method based on boundary perturbation. For an arbitrarily
440 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

shaped dielectric resonator, the change in resonant frequency

due to a change of the cavity wall can be determined using per-
turbation theory [15], and is given by,

(3) Fig. 4. Boundary perturbation from the base of the split-cylindrical DRA
(Cross-sectional ( -plane) view).

where and are the permittivity and the permeability of the

dielectric resonator respectively, and are the resonant ra-
dian frequencies of the perturbed and unperturbed resonator, re-
spectively, and are the volume perturbed and the original
volume of the resonator, and and are the unperturbed
fields. Equation (3) indicates that the change in resonant fre-
quency is equal to the electric and magnetic energies removed
by the perturbation divided by the total energy stored [15], i.e.,

(4)

where and are time-averaged electric and magnetic

energies originally contained in the volume perturbed and
is the total energy stored in the unperturbed cavity. Fig. 5. Plot of change of resonant frequency against the perturbation param-
Now consider a boundary perturbation from the base of the eter .
split-cylindrical DRA as depicted in Fig. 4. The changes in reso-
nant frequencies of the mode and mode can be
computed using (3), and the result is shown in Fig. 5. It can be
observed that as the electric boundary is moved up, the resonant
frequency of the mode increases more rapidly than that
of the mode. Hence, at a certain perturbation value, the
two resonant frequencies should overlap and thus fulfill the pri-
mary requirement for MIMO antenna design. According to (4),
the difference in the rate of change of resonant frequency can
Fig. 6. Elliptical approximation of the perturbed cylindrical boundary (Cross-
be explained by the difference in the energy stored by the two sectional ( -plane) view).
modes in the perturbation volume . To verify the boundary
perturbation method, an HFSS simulation was carried out and
the result is also shown in Fig. 5. It can be seen that the result
predicted by the boundary perturbation method starts to deviate
from the result obtained from HFSS when the perturbation, ,
increases. This is due to the substitution of the original fields into
the perturbed fields during the derivation of (3). The difference
between the original fields and the perturbed fields would be in-
tolerable when the perturbation is too large. Thus, the deviation
at large perturbations is inherent in the perturbation analysis.
Nonetheless, the boundary perturbation method gives a good
initial guess on how much perturbation is required to make the
two modes resonate at the same frequency. According to the
HFSS simulation result, the two modes both resonate at 700
MHz when the perturbation, , is 13 mm.
In (3), there is no specific constraint on the geometry of the
cavity, hence, the proposed boundary perturbation method can
be applied to DRAs of any other shapes with arbitrary pertur-
bations. However, the difficulty of analysis of such structures Fig. 7. Simulated -parameters of the perturbed cylindrical DRA.
might be the evaluation of the integrals in (3).
YAN AND BERNHARD: DESIGN OF A MIMO DIELECTRIC RESONATOR ANTENNA FOR LTE FEMTOCELL BASE STATIONS 441

Fig. 8. Comparison of the simulated and measured radiation patterns of Port 1 (HE mode).

Fig. 9. Comparison of the simulated and measured radiation patterns of Port 2 (TE mode).

B. Boundary Approximation the perturbation, , is 13 mm. The dimensions of the perturbed

While the above described boundary perturbation method ob-
tains the solutions by approximating the fields, boundary ap-
proximation estimates the change in resonant frequency by ap-
proximating the structure of the resonator. According to Fig. 6,
a perturbed cylindrical boundary can be modeled by a half el-
lipse. The accuracy of this method depends on how well the
perturbed circular arc is approximated by an elliptic arc. The
resonant frequencies of various modes in an elliptical DRA can
be found by expanding the fields inside the cavity in Mathieu
functions and applying the technique of separation of variables.
A detailed analysis of an elliptical DRA can be found in [16].
The resonant frequencies of a series of split-elliptical DRAs of
various minor axes, which corresponded to the previously de-
scribed set of perturbed cylindrical DRAs, are computed, and
the change in resonant frequency estimated by this boundary
approximation method is plotted in Fig. 5. The results obtained
agree very well with those calculated by both the boundary per-
turbation method and full-wave simulations.
IV. SIMULATED ANTENNA PERFORMANCE
A. Antenna Characteristics
From Section III-A, the mode and the mode
of a split-cylindrical DRA will both resonate at 700 MHz when
DRA are 80 mm 84 mm 31 mm. Given the same resonant
frequency and a half-wavelength antenna separation, the dimen-
sions of a two-element MIMO PIFA would be 107 mm 214
mm 5 mm. Since coupling between antenna ports is another
important parameter to characterize MIMO antennas, the pro-
posed antenna structure was simulated in HFSS. The simulated
-parameters of the antenna were obtained with 50 termina-
tions at both ports and are given in Fig. 7. It can be observed
that the mutual coupling between the two modes is insensitive
to the perturbation (see Fig. 3) and is less than . This
is significantly lower than the mutual coupling in conventional
MIMO antennas that are based on dipole antennas, patch an-
tennas or PIFAs [17]–[20]. The impedance bandwidths (defined
as ) of the mode and the mode
are 10 MHz and 35 MHz respectively. The mode has a
relatively narrow bandwidth and limits the overall bandwidth of
the antenna. Nevertheless, the bandwidth of the mode
can be improved using well known bandwidth broadening tech-
niques, such as inserting an air gap between the ground plane
and the DRA [21], [22], or adding a matching stub at the end
of the microstrip line [23]. The simulated gains of the
and modes are 3.96 and 3.19 dBi, respectively. The sim-
ulated radiation patterns of the two modes, which are orthogonal
to each other, are given in Figs. 8 and 9. Hence, the antenna is
442
Fig. 10. The floor plan (16 m 19 m) of the office environment that was used
to estimate the MIMO channel capacity (notice that there is no line-of-sight
(LOS) path between the transmitter and any of the receivers).
able to exploit polarization diversity and the pattern orthogo-
nality leads to low mutual coupling between the ports.
B. MIMO Performance Evaluation
The channel capacity gain by using the proposed antenna
was evaluated with the aid of Remcom Wireless Insite [24].
In the simulation setup, a single transmitter and 1000 identical
receivers were placed in an office environment as shown in
Fig. 10. The office environment was constructed to resemble
a rich scattering environment (i.e., the channel statistics are ap-
proximately Rayleigh distributed). In order to resemble a time-
varying MIMO channel, the receivers were randomly spread
across the designated area of the office such that a 1000 non-
line-of-sight (NLOS) communication links were established. In
all the simulations, there were 80 paths for each channel realiza-
tion. The simulated complex radiation patterns (including both
polarizations, and ) of the proposed antennas were used at
the transmitter and all the receivers. 1000 samples of the unnor-
malized channel matrix were then obtained from the simulation
[25],
where there are communication links, and
is the unnormalized channel matrix of the -th link. Here,
represents the -th sample of the complex channel gain
between port of the transmitter and port of the receiver,
where subscripts , 2 and , 2:
Here is the number of path in the -th link; is the re-
ceived power contributed by the -th path in the -th link;
is the phase of the -th path in the -th link. From the simu-
lated channel data, it was found that the coherence bandwidth of
the wireless channel was much larger than the bandwidth of the
proposed antenna. Hence, for equal power distributed among
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 11. Estimated channel capacity of the proposed MIMO DRA.
the transmit antennas, the flat-fading MIMO channel capacity
of the -th link, , can be calculated by [25]–[27]
(7)
where and denote the number of transmit and receiver
antennas, respectively; is an identity matrix with dimen-
sion ; is the mean signal-to-noise ratio (SNR) per receive
branch; represents a complex conjugate transpose; and
is the -th normalized channel matrix
(8)
where denotes a Frobenius norm. The mean capacity and
the maximum achievable capacity obtained by using the pro-
posed antenna are plotted in Fig. 11. Fig. 11 also gives the theo-
retical channel capacities for single-input single-output (SISO),
(5) 2 2 and 3 3 channels with zero mean, unity variance, i.i.d.
complex Gaussian distributed channel elements for comparison.
The results indicate that the estimated mean channel capacity is
11.1 b/s/Hz at an SNR of 20 dB per receiver branch. The max-
imum achievable capacity is very close to the theoretical max-
imum 2 2 MIMO capacity of 13.4 b/s/Hz. The small discrep-
ancy between the theoretical and simulated capacities may be
due to the non-ideal scattering environment and finite mutual
coupling between the modes. Nevertheless, the simulation re-
sults reflect the utility of the antenna design, and a prototype
(6) antenna is presented in Section V.
V. MEASUREMENT RESULTS
The perturbed cylindrical DRA was built and tested in
the Electromagnetics Laboratory at University of Illinois
at Urbana-Champaign. The dielectric material ( and
) was supplied by Countis Laboratories
[28]. The dielectric block was bonded onto the ground plane
YAN AND BERNHARD: DESIGN OF A MIMO DIELECTRIC RESONATOR ANTENNA FOR LTE FEMTOCELL BASE STATIONS
Fig. 12. Measured -parameters of the perturbed cylindrical DRA.
with silver epoxy so as to prevent any air gaps between the di-
electric and ground plane. This is important because for DRAs
with high permittivities, air gaps of less than 0.05 mm can
be enough to significantly alter the expected input impedance
[12]. The -parameters of the perturbed cylindrical DRA were
measured using Agilent’s two-port Network Analyzer E8363B
(with 50 reference impedance). The measured results are
given in Fig. 12, which are very close to the simulated results
given in Fig. 7. Both modes are well matched at 717 MHz.
The coupling between the ports is less than at
the operating frequency. The measured impedance bandwidths
of the mode and the mode are 13.5 MHz and
35 MHz, respectively. The measured radiation patterns along
the three principal cuts are given in Figs. 8 and 9. Despite a
small distortion of the pattern at some angles, the measured
patterns agreed reasonably well with the simulated ones. The
complementary nature of the two orthogonal modes can still be
observed clearly.
VI. CONCLUSION
A 2-port MIMO antenna based on a split-cylindrical DRA is
described in this work. A mode degeneration method derived
from perturbation theory is proposed to make the TE and HE
modes of the split-cylindrical DRA resonate at the same fre-
quency. The proposed method has been verified by both full-
wave simulations and the boundary (elliptical) approximation
method, and can be applied to DRAs of any shape. The fabri-
cated MIMO DRA was tested and the experimental results show
very good agreement with the simulated results. Indeed, given
that the same operating frequency and the same dielectric mate-
rial, the antenna described in this paper is smaller in volume, has
lower profile, has a smaller ground plane and has much lower
mutual coupling as compared to the work in [10], [11]. The pro-
posed antenna is potentially suitable as the femtocell base sta-
tion antenna in the forthcoming nationwide mobile broadband
system based on LTE technology. Future work related to this
paper will be a frequency reconfigurable MIMO DRA which
can easily be adapted to other LTE bands and other wireless
standards.
443
REFERENCES
[1] Federal Communications Commission, “700 MHz band,” Auction 73
Feb. 2009.
[2] News Archives AT&T Inc., 2008 [Online]. Available: http://www.att.
com/
[3] News Archives Verizon Wireless, 2009 [Online]. Available:
http://news.vzw.com/
[4] 3GPP TS36.300, “Evolved Universal Terrestrial Radio Access
(E-UTRA) and Evolved Universal Terrestrial Radio Access Network
(E-URRAN): Overall Description,”.
[5] D. Astely, E. Dahlman, A. Furuskar, Y. Jading, M. Lindstrom, and
S. Parkvall, “LTE: The evolution of mobile broadband—[LTE part II:
3GPP release 8],” IEEE Commun. Mag., vol. 47, no. 4, pp. 44–51, Apr.
2009.
[6] V. Chandrasekhar and J. Andrews, “Femtocell networks: A survey,”
IEEE Commun. Mag., vol. 46, no. 9, pp. 59–67, Sep. 2008.
[7] C.-Y. Chiu, J.-B. Yan, and R. D. Murch, “Compact three-port orthog-
onally polarized MIMO antennas,” IEEE Antennas Wireless Propag.
Lett., vol. 6, pp. 619–622, 2007.
[8] K.-M. Luk and K.-W. Leung, Dielectric Resonator Antennas. Hert-
fordshire, England: Research Studies Press Ltd., 2003.
[9] L. K. Hady, D. Kajfez, and A. A. Kishk, “Triple mode use of a single
dielectric resonator,” IEEE Trans. Antennas Propag., vol. 57, no. 5, pp.
1328–1335, May 2009.
[10] K. Ishimiya, J. Langbacka, Z. Ying, and J.-I. Takada, “A compact
MIMO DRA antenna,” in Proc. IEEE Int. Workshop on Antenna
Technology: Small Antennas and Novel Metamaterials (IWAT ’08),
Chiba, Japan, Mar. 2008, pp. 286–289.
[11] K. Ishimiya, Z. Ying, and J.-I. Takada, “A compact MIMO DRA for
802.11n application,” presented at the IEEE Antennas and Propagation
Society Int. Symp., San Diego, CA, Jul. 2008.
[12] A. Petosa, Dielectric Resonator Antenna Handbook. Norwood, MA:
Artech House, 2007.
[13] Mathematica, Wolfram Research Inc., 2010.
[14] HFSS Ansys, Inc., 2010.
[15] R. F. Harrington, Time-Harmonic Electromagnetic Fields. New
York: IEEE Press, 2001.
[16] A. Tadjalli and A. Sebak, “Resonance frequencies and far field pat-
terns of elliptical dielectric resonator antenna: Analytical approach,”
in Progress in Electromagnetic Research, PIER 64, 2006, pp. 81–98.
[17] C.-C. Hsu, K. H. Lin, H.-L. Su, H.-H. Lin, and C.-Y. Wu, “Design
of MIMO antennas with strong isolation for portable applications,”
presented at the IEEE Antennas and Propagation Society Int. Symp.,
Charleston, SC, Jun. 2009.
[18] H. Zhang, Z. Wang, J. Yu, and J. Huang, “A compact MIMO antenna
for wireless communication,” IEEE Antennas Propag. Mag., vol. 50,
no. 6, pp. 104–107, Dec. 2008.
[19] K.-S. Min, D.-J. Kim, and M.-S. Kim, “Multi-channel MIMO antenna
design for WiBro/PCS band,” in Proc. IEEE Antennas and Propagation
Society Int. Symp., Hawaii, Jun. 2007, pp. 1225–1228.
[20] K. Chung and J. H. Yoon, “Integrated MIMO antenna with high isola-
tion characteristics,” Electron. Lett., vol. 43, no. 4, pp. 199–201, Feb.
2007.
[21] M. Cooper, “Investigation of Current and Novel Rectangular Dielectric
Resonator Antennas for Broadband Applications at L-band Frequen-
cies,” M.Sc. thesis, Carleton University, Ottawa, ON, Canada, 1997.
[22] S.-M. Deng, C.-L. Tsai, S.-F. Chang, and S.-S. Bor, “A CPW-fed sus-
pended, low profile rectangular dielectric resonator antenna for wide-
band operation,” in Proc. IEEE Antennas and Propagation Society Int.
Symp., Washington, D.C., Jul. 2005, vol. 4B, pp. 242–245.
[23] P. V. Bijumon, S. K. Menon, M. N. Suma, M. T. Sebastian, and P. Mo-
hanan, “Broadband cylindrical dielectric resonator antenna excited by
modified microstrip line,” Electron. Lett., vol. 41, no. 7, pp. 385–387,
Mar. 2005.
[24] Wireless Insite, Remcom Inc., 2006.
[25] J. D. Boerman and J. T. Bernhard, “Performance study of pattern recon-
figurable antennas in MIMO communication systems,” IEEE Trans.
Antennas Propag., vol. 56, no. 1, pp. 231–236, Jan. 2008.
[26] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” in Wireless
Personal Commun.. New York: Kluwer Academic Press, 1998, pp.
311–335.
[27] Z. Tang and A. S. Mohan, “Experimental investigation of indoor
MIMO Ricean channel capacity,” IEEE Antennas Wireless Propag.
Lett., vol. 4, pp. 55–58, 2005.
[28] Countis Laboratories [Online]. Available: http://www.countis.com/
444 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Jie-Bang Yan (S’09–M’11) received the B.Eng. de-
gree (first class honors) in electronic and communica-
tions engineering from the University of Hong Kong,
in 2006, the M.Phil. degree in electronic and com-
puter engineering from the Hong Kong University of
Science and Technology, in 2008, and the Ph.D. de-
gree in electrical and computer engineering from the
University of Illinois at Urbana-Champaign, in 2011.
He was a Croucher Scholar from 2009 to 2011
while he did his Ph.D. at the University of Illinois
at Urbana-Champaign. Upon graduation, he joined
the Center for Remote Sensing of Ice Sheets (CReSIS), University of Kansas,
where he is currently an Assistant Research Professor. His research inter-
ests include design and analysis of MIMO and reconfigurable antennas, RF
propagation, radar antenna designs, and fabrication of on-chip antennas. He
holds two U.S. patents and a U.S. patent application related to novel antenna
technologies.
Dr. Yan was the recipient of the Best Paper Award at the 2007 IEEE (HK)
AP/MTT Postgraduate Conference and the 2011 Raj Mittra Outstanding Re-
search Award at Illinois. He serves as a Reviewer for several journals and con-
ferences on antennas and electromagnetics.
Jennifer T. Bernhard (S’89–M’95–SM’01–F’10)
was born on May 1, 1966, in New Hartford, NY.
She received the B.S.E.E. degree from Cornell
University, Ithaca, NY, in 1988 and the M.S. and
Ph.D. degrees in electrical engineering from Duke
University, Durham, NC, in 1990 and 1994, re-
spectively, with support from a National Science
Foundation Graduate Fellowship.
While at Cornell, she was a McMullen Dean’s
Scholar and participated in the Engineering Co-op
Program, working at IBM Federal Systems Division
in Owego, New York. During the 1994–95 academic year she held the position
of Postdoctoral Research Associate with the Departments of Radiation On-
cology and Electrical Engineering at Duke University, where she developed RF
and microwave circuitry for simultaneous hyperthermia (treatment of cancer
with microwaves) and MRI (magnetic resonance imaging) thermometry. From
1995–1999, she was an Assistant Professor in the Department of Electrical and
Computer Engineering, University of New Hampshire, where she held the Class
of 1944 Professorship. Since 1999, she has been with the Electromagnetics
Laboratory, Department of Electrical and Computer Engineering, University
of Illinois at Urbana-Champaign, where she is now a Professor. Her industrial
experience includes work as a Research Engineer with Avnet Development
Labs and, more recently, as a private consultant for members of the wireless
communication and sensors community. Her research interests include recon-
figurable and wideband microwave antennas and circuits, wireless sensors and
sensor networks, high speed wireless data communication, electromagnetic
compatibility, and electromagnetics for industrial, agricultural, and medical
applications, and has four patents on technology in these areas.
Prof. Bernhard is a member of URSI Commissions B and D, Tau Beta Pi,
Eta Kappa Nu, Sigma Xi, and ASEE. She is a Fellow of the IEEE. She was an
organizing member of the Women in Science and Engineering (WISE) Project
at Duke, a graduate student-run organization designed to improve the climate
for graduate women in engineering and the sciences. In 1999 and 2000, she was
a NASA-ASEE Summer Faculty Fellow at the NASA Glenn Research Center,
Cleveland, OH. She received the NSF CAREER Award in 2000. She is also an
Illinois College of Engineering Willett Faculty Scholar and a Research Professor
in Illinois’ Coordinated Science Laboratory, and the Information Trust Institute.
She and her students received the 2004 H. A. Wheeler Applications Prize Paper
Award from the IEEE Antenna and Propagation Society for their paper pub-
lished in the March 2003 issue of the IEEE TRANSACTIONS ON ANTENNAS AND
PROPAGATION. She served as an Associate Editor for the IEEE TRANSACTIONS
ON ANTENNAS AND PROPAGATION from 2001–2007 and for IEEE Antennas and
Wireless Propagation Letters from 2001–2005. She is also a member of the ed-
itorial board of Smart Structures and Systems. She served as an elected member
of the IEEE Antennas and Propagation Society’s Administrative Committee
from 2004–2006. She was President of the IEEE Antennas and Propagation So-
ciety in 2008.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
A Compact Eighteen-Port Antenna Cube
for MIMO Systems
Jianfeng Zheng, Xu Gao, Zhijun Zhang, Senior Member, IEEE, and Zhenghe Feng, Senior Member, IEEE
Abstract—An 18-port compact antenna cube is proposed in this
paper. The cube, which has a volume of 0.76 0.76 0.76 , dates up to 12 electrical dipole antennas on all its 12 edges. The
provides 18 individual channels and is ideal for multiple-input
multiple-output (MIMO) wireless communications. On each of
the total six faces of the cube, a three-port tri-polarization antenna
is installed. All antennas adopt a metal backing configuration,
so the ground of all antennas forms a well shield Faraday cage,
in which other functional circuits can be installed. Experimental
measurements were carried out to evaluate the performance of
the antenna cube in different MIMO scenarios. The results show
that MIMO systems with the proposed compact antenna cube
outperform those with dipole antennas which occupy the same
number of RF channels but with much larger space. When a
vertical 3-dipole array, a horizontal 3-dipole array and a dual
polarization antenna are used in the user end (UE), respectively,
the capacity of the global selected MIMO systems with antenna
cube is about 2.7, 4.6, and 2.9 bits/s/Hz more than the full MIMO
systems with a vertical 3-dipole array as the access point (AP)
antennas. It is 1.9, 3.9, and 2.0 bits/s/Hz more than the full MIMO
systems with a vertical 5-dipole array as AP antennas. The perfor-
mance differences between the MIMO systems using global and
simplified selection circuits are small.
Index Terms—Antenna cube, antenna selection, multiple-input
multiple-output (MIMO), polarization.
I. INTRODUCTION
A PPLYING multiple-input multiple-output (MIMO) tech-
nology especially with antenna selection in access points
(AP) can improve the overall system capacity. However, to con-
struct enough antennas within a small volume is always a chal-
lenge.
In previous works, a number of compact MIMO antennas
have been proposed consisting of up to four ports, compact an-
tenna designs with more than 10 ports are less common and
mainly consist of a flat panel approach and are used in large size
base station. Recently an interesting approach, the antenna cube,
emerges. An antenna cube takes advantage of spatial and polar-
ization orthogonality to implement a large amount of antennas
Manuscript received December 20, 2010; revised March 28, 2011; accepted
August 15, 2011. Date of publication October 25, 2011; date of current ver-
sion February 03, 2012. This work was supported in part by the National Basic
Research Program of China under Contract 2007CB310605, in part by the Na-
tional Science and Technology Major Project of the Ministry of Science and
Technology of China 2010ZX03007-001-01, in part by Qualcomm Inc., and in
part by the Chuanxin Foundation of Tsinghua University.
The authors are with the State Key Lab of Microwave and Communications,
Tsinghua National Laboratory for Information Science and Technology,
Tsinghua University, Beijing 100084, China (e-mail: zjzh@tsinghua.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173449
0018-926X/$26.00 © 2011 IEEE
445
within a constrained volume. In [1]–[3], MIMO cube accommo-
24-port and 36-port antenna cubes suitable for MIMO wireless
communications are presented in [4].
However, existing cubes [1]–[4] demand a completely ded-
icated space for antennas. As the antenna elements in those
cubes are omni-directional, the inner space must be kept empty
to avoid performance degradation, i.e., other circuits cannot
be installed in the space. To resolve the problem, a compact
18-port planar tri-polarization antenna cube for MIMO systems
is proposed in this paper. A tri-polarization antenna makes full
use of the promising polarization domain, which is considered
an important resource for constructing compact antenna arrays
and enhancing system performance [6]–[8]. The antenna cube
employs tri-polarization antennas [9] as the basic elements.
To form a compact antenna cube, six tri-polarization antennas
are distributed on separate faces of a cube. This arrangement
achieves low mutual coupling and wide coverage within a
small volume mm with an operating frequency
band of 2.40–2.48 GHz.
In a real communication system, it is difficult to implement
a large amount of RF channels even at AP. Thus some sorts
of antenna switching must be involved for antenna-abundant
MIMO systems [10]–[12]. Accompanying with the antenna
cube, two simplified antenna switching schemes are proposed
in this paper. Measurement results demonstrate that in an indoor
environment, performance achieved by simplified switching
schemes is almost as good as that of a fully switching system.
Antenna design, measurement results and experimental ver-
ifications of the proposed compact planar tri-polarization an-
tenna cube are described in Sections II–V. Specifically, the tri-
polarization antenna is briefly introduced in Section I. Mea-
surement results of the 18-port antenna cube are presented and
discussed in Section II. Measurement procedure and analysis
framework are explained in Section III. Experimental results of
MIMO systems between the antenna cube and various terminal
antennas are discussed in Section IV. Conclusion is drawn in
Section V.
II. ANTENNA CUBE DESIGN
The conformal and low-profile tri-polarization antenna which
was proposed in [9] is a fundamental building block in the planar
MIMO cube and is briefly introduced here. The configuration
of the tri-polarization antenna is shown in Fig. 1. A ring patch,
which functions as two independent orthogonal polarized an-
tennas, and a disk-loaded monopole compose the tri-polariza-
tion antenna, and the operating frequency band is chosen to be
2.4–2.48 GHz.
446
Fig. 1. Geometry of the tri-polarization antenna: (a) top view and (b) side view.
Different from antennas used in other MIMO cubes proposed
in the literature, the planar tri-polarization antenna has a very
low profile, and the total height of the tri-polarization antenna is
5.8 mm. Furthermore, the full 74 74 mm sized ground plane
of the tri-polarization antenna makes it particularly suitable for
antenna mounted on the equipments. The advantage of low pro-
file together with the easiness of its conformal integration on a
cube surface makes the tri-polarization antenna a good candi-
date to construct the MIMO cube.
Apart from the above advantage of the tri-polarization
antenna, most importantly, the patch antenna mode and the
monopole antenna mode of the tri-polarization have the or-
thogonal polarization property to each other. With three ports
of this antenna working independently, the far field of this
antenna has three orthogonal linear polarizations. Specifically,
when the tri-polarization antenna is placed as in Fig. 1, i.e., the
monopole is along the -axis while the feed lines P1 and P2
are along the - and -axes in horizontal plane, respectively,
the E-field radiated by the ring patch is parallel to the ground
plane and can provide two orthogonal polarizations excited
through P1 and P2, while the monopole provides the vertical
polarization component and has an omni-directional radiation
in the azimuth plane. Fig. 2 shows the measured radiation
patterns of the tri-polarization antenna at 2.42 GHz. As shown,
the radiation pattern of monopole mode (port M3) and patterns
of the patch mode (port P1 and port P2) have orthogonal
polarizations to each other. The gains of the directional slot-fed
antennas at 2.42 GHz are 7.5 dBi for P1 and P2, while the
gain of the omni-directional coaxial-fed disk-loaded monopole
fed by M3 is 2.5 dBi. The main reason for the lower gain of
M3 compared with the gains of other two ports is the different
radiation properties between monopole and patch antennas.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
The omni-directional radiation property gives the monopole
mode lower gain compared to the directional patch mode.
In real communication applications, the position of mobile
terminals may rotate due to different communication scenarios
and the arbitrariness of user’s behavior. For the fact that the
three ports of this antenna radiate three polarized fields that
are orthogonal to one another, this antenna could receive elec-
tromagnetic wave with any kind of polarization by switching
among the ports of the antenna cube, thus avoid situations of
the polarization mismatch.
The tri-polarization antenna has a low planar profile and the
complete common ground, thus it is easy to construct the planar
tri-polarization antenna cube by embedding one tri-polarization
antenna on each face of a cube. The structure of the planar tri-po-
larization antenna cube is shown in Fig. 3. As shown, the six
planar tri-polarization antennas are fixed on the six faces of
the cube. Each antenna has 3 ports, and the antenna cube has
18 ports, which can provide up to 18 individual communica-
tion channels. The antenna cube operates at 2.4–2.48 GHz, and
the volume is 94 94 94 mm , about 0.76 0.76 0.76
where is the wavelength in vacuum. For convenience of de-
scription, the faces of the cube are numbered as shown in Fig. 3,
the up face is #1, the front, right, back and left faces are num-
bered as #2, #3, #4, and #5 respectively, and the bottom face
is #6. The three ports in a face are noted as P1, P2, and M3.
Each port in the antenna cube is denoted with the numbers of
faces and ports, for example, F#1-P1 represents the P1 port of
the tri-polarization antenna in the #1 face of the cube.
For the three ports of each tri-polarization antenna in the face
have three orthogonal polarizations, it is easy to obtain the full
radiation coverage in the whole sphere. Therefore, the MIMO
cube can provide good convergence for user terminals with any
rotation and position.
An important aspect to construct the antenna cube is to main-
tain relative low mutual coupling between any individual ports,
as mutual coupling will deteriorate the performance of MIMO
wireless communication systems. For the compact tri-polariza-
tion antenna cube, relatively low mutual coupling between an-
tennas of the proposed MIMO cube is mainly due to the choices
of antenna types, positions and orientations. As the three an-
tennas in a tri-polarization antenna employ orthogonal polariza-
tions, the mutual coupling between each port is relatively low.
The tri-polarization antenna has a ground backing, so the tri-po-
larization antennas in different faces radiate toward different di-
rections and inherently have low mutual coupling.
To verify the performance of the planar tri-polarization an-
tenna cube, a prototype antenna cube was fabricated, and the
photo of the cube is shown in Fig. 3. Due to the symmetric char-
acteristic of the antenna cube, only the tri-polarization antenna
#1 is measured. The measured reflection and transmission coef-
ficients are shown in Fig. 4. The results are pretty much identical
to the results reported in [9].
Between any two tri-polarization antennas in adjacent faces,
there are nine sets of transmission coefficients. As shown in
Fig. 5, there are three most significant results between antennas
in face #1 and #3. The isolations at 2.4–2.48 GHz band are all
better than 20 dB. The isolations between ports in opposite
ZHENG et al.: COMPACT EIGHTEEN-PORT ANTENNA CUBE FOR MIMO SYSTEMS 447

Fig. 2. Measured electrical patterns of the tri-polarization antenna: (a) , plane; (b) , plane; (c) , plane; (d) , – plane; (e) ,
– plane; (f) , – plane.

Fig. 3. Structure and photograph of the planar tri-polarization antenna cube.

antennas are all better than 25 dB, which is not illustrated here
for the reason of concision.
Overall, these results show that the proposed planar tri-polar-
ization antenna cube has good isolation among the individual
ports, which satisfies the requirement of MIMO systems.

III. MIMO SYSTEMS WITH THE ANTENNA CUBE

In prior works, the antennas presented for MIMO systems
were often validated by examining the channel capacity of the
full MIMO systems between antenna cubes in a narrow fre-
quency band. However, the full MIMO systems which support
more than 10 individual channels are too expensive and compli-
cated to use in personal wireless communication systems nowa-
days, such as WLAN equipments, and the communication sys-
tems mostly are wideband. To overcome these shortcomings,
the performance of the MIMO systems employing the antenna Fig. 4. (a) Measured return loss of the tri-polarization antenna #1 in the cube.
cube is examined in typical indoor scenarios with antenna se- (b) Measured isolation between ports in antenna #1.
lection among the whole WLAN frequency bands.
The measurements were carried out in Room 1010 on the 10th typical laboratory room as schemed in Fig. 6. The framework
floor of Weiqing Building in Tsinghua University, which is a of the room is reinforced concrete, the walls are mainly built
448
Fig. 5. Measured isolation between ports in adjacent tri-polarization antennas.
Fig. 6. Structure of the measured office.
by brick and plaster, and the ceiling is made with plaster plates
with aluminium alloy framework.
The scheme of the test-bench is shown in Fig. 7. The measure-
ment system consists of an Agilent E5071B network analyzer,
AP antennas, user equipment (UE) antennas, RF switches, a
computer, an auxiliary amplifier, and RF cables. The AP and the
UE are connected to a 16-to-1 RF switch and a 4-to-1 RF switch
respectively, and the switches are then connected to the network
analyzer. The auxiliary amplifier is between the transmit an-
tenna and the network analyzer to amplify the transmit signal.
The computer controls the measurement procedure and records
the data.
In the measurement, the transmit power of network analyzer
is set to 10 dBm, IFBW is 10 kHz, and sweep averaging is set on
with sweep averaging factor as 16, the noise floor of the network
analyzer is below than 90 dB when measuring S21. The loss
of the cable is less than 15 dB, the insertion loss of the switch
is about 4 dB and the power gain of the amplifier is about 10
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 7. Schematic of test-bench for MIMO system.
Fig. 8. Antennas used in measurement besides antenna cube: (a) vertical
5-dipole array; (b) vertical 3-dipole array; (c) horizontal 3-dipole array; (d)
dual-polarization antenna.
dB. With the SNR limitation of 15 dB, the dynamic range of the
measurement system is above 66 dB.
For the conveniences of measurement and installation, the
tri-polarization antenna in the bottom face of cube was removed,
thus only 15 ports of the cube were used.
The configurations of the measurements are listed in Table I.
The measurement campaign was carried out for twelve repre-
sentative MIMO systems, and the measured channel responses
are noted as , here is the type number of AP antennas and
is the one of UE antennas. On the AP side, four different arrays
were used respectively. They are a vertical three-dipole array, a
vertical five-dipole array, and an antenna cube with three/five
selected branches. The separation between adjacent antennas
of the dipole array is one wavelength, so the three/five-dipole
array’s size is two/four wavelengths. On the UE side, three dif-
ferent arrays were used alternatively. They are a vertical three-
dipole array, a horizontal three-dipole array and a compact dual
polarization antenna [13]. The size of three-dipole array is two
wavelengths and the dual polarization antenna is 0.8 wavelength
in size. The schemes of all dipole arrays and the dual-polariza-
tion antenna are shown in Fig. 8.
In each measurement, the AP antenna was fixed in the
center of the office room with a height of 1.2 m, and UE antennas
were placed in the 10 locales around the room sequentially with
a height of about 0.8 m. The locales UE2, 5, 8, 10 had Line-of-
sight (LOS) paths and UE1, 3, 4, 6, 7, 9 only had non-line-of-
sight (NLOS) paths, as illustrated in Fig. 6, where the locales
UE5, 10 and the locales UE2, 8 were in the broad-sight and
ZHENG et al.: COMPACT EIGHTEEN-PORT ANTENNA CUBE FOR MIMO SYSTEMS
TABLE I
MEASUREMENT CONFIGURATIONS
TABLE II
CONFIGURATIONS OF THE STUDIED MIMO SYSTEMS
end-fire directions of the referenced dipole array at AP. In each
place, channel matrices at 4 points separated by 6.5 cm, i.e.,
half-wavelength, were measured in order to obtain independent
fading, and denoted as , where represents the serial
number of the locale and is the serial number of measurement
point in the locale place. For each , the responses over
the whole WLAN band were measured.
As we measured the channel responses after midnight and
before dawn, the channels were supposed to be static, so the
elements in the channel matrix were measured
in sequence and the switching of the was completed by
using RF switches.
In the measurements of referenced MIMO system with a ver-
tical 5-dipole at AP and a 3-dipole array at UE, the maximum
measured S21 is 45 dB, and the average measured S21 is 52
dB. That is, the average SNR of the measurement is about 52
+ 90 dB 38 dB with referenced linear dipole arrays, here 90
dB is the S21 noise floor of the proposed measurement system.
The measured channel responses are assorted to construct the
wideband channel of referenced dipole-array MIMO systems
and MIMO systems with antenna cube. The frequency bands are
divided following the IEEE 802.11 specifications as shown in
Table II. That means, when studying the channel capacity of any
MIMO system with specified antennas and places, 14 wideband
channels are adopted based on the frequency partition of IEEE
802.11 specifications.
449
A. Transmitter Power Constraints
As the received power and richness of scattering are quite dif-
ferent at different UE locations, the measured channel response
matrices must be appropriately normalized. For the rich scat-
tering required to achieve low correlations for MIMO commu-
nications often produces low SNR, which in turn decreases the
channel capacity [14], we adopt the MIMO system with vertical
5-dipole at AP and vertical 3-dipole at UE as reference to nor-
malize the channel responses with average transmitted power
as discussed in [15]. This normalization considers not only
richness of the multipath but also the power gain.
Obviously, the can be expressed as in
(1)
where , which is different with the average SNR of the
measurements, is the assumed average SNR of the referenced
MIMO systems with channel response , is the number
of the locales and is the number of the measured points in
each locale, denotes the number of the channel bands,
and are the numbers of transmit and receive antennas
of channel responses, respectively, and is the number of
the measured frequency bins in the th band. means the
average received noise per frequency bin, and is the trace
operation.
In the following, the assumed average SNR is set to 15 dB
in analyzing the channel capacity.
B. Channel Capacity of MIMO Systems With Antenna
Selections Over Wide Bands
Though prior proposed cubes were demonstrated in full
MIMO systems, the transceivers of a full MIMO system using
antenna cubes might be too expensive to accommodate in
today’s personal wireless communication systems. Antenna
selection is a good approach to reduce a system’s cost while
maintain its performance. The bulk selection [16], [17] method
is adopted as a reference. Bulk selection method is a global
optimization method, which assumes there is a direct path
between any input port and output port.
The channel capacity with equal power emission strategy [18]
is adopted to evaluate performance of measured MIMO systems
covering the th band. Then channel capacity of wideband sys-
tems with bulk antenna selection is
(2)
where denotes the different combination of AP antenna el-
ements and is the set of the selectable antenna combina-
tions. Assuming there are 15 ports in the AP, each time the
total number of combinations is 455 and 3003 when 3 and 5
branches are selected respectively. On the UE side, all available
antennas are always used. is the channel capacity of
the MIMO systems with selected antennas combination over
the th band, and expressed as
(3)
450
Fig. 9. Configurations of the switching circuits: (a) 15-to-3 global selection
circuit; (b) pattern selection circuit; (c) 15-to-5 global selection circuit; (d) po-
larization selection circuit.
C. Simplified Pattern and Polarization Selection Methods
Although antenna selection is capable of reducing the cost
of the RF channels while maintaining the performance of the
MIMO systems, determinations of the forms of the antenna
array and implementations of the RF selection circuits are
not trivial [12], [19]–[21]. The existing research activities on
antenna selection little involve designs of antenna arrays and
selection circuits. The often used or assumed global selection
circuits require many RF switches and complicated RF circuits,
which are difficult to realize and may introduce inevitable high
insertion loss.
Two simplified selection circuits with low complexity and
cost are presented to reduce the complexity of global selec-
tion circuit and maintain a comparable performance, which are
pattern and polarization selection circuits. As shown in Fig. 9,
global selection methods select the best channels irrespective of
the polarizations and antenna types by a complex RF switch ma-
trix. While when pattern selection is applied, each port with the
same polarization in each tri-polarization antenna is connected
to a 5-to-1 RF switch. The pattern selection generates three
output ports. The pattern combinations approaching the most
channel capacities are then selected in the performance evalua-
tion. In polarization selection, the three ports in each tri-polar-
ization antenna are connected to a 3-to-1 RF switch. The polar-
ization selection generates five output ports.
The proposed antenna selection schemes consist of
-to-1 switches while the global selection cir-
cuits consist of an -to- switch matrix, here is the
number of the selected antennas and is the number of the
available receive antennas. The switch matrix is much more
complex than -to-1 switches.
IV. RESULT COMPARISONS
Because the spread of multi-path in elevation direction and
polarization rotation are significant in indoor scenarios, the per-
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
formance of the MIMO systems will depend not only on the
number of the antennas used in the base station but also on the
radiation pattern, polarization and array structure.
The whole spherical coverage characteristic and capability of
receiving any polarized impinging wave make the compact an-
tenna cube particularly suitable for indoor communications. We
examined the performance of the MIMO systems with the an-
tenna cube and various terminal antennas in different locales and
postures as followed. The performance of the selection MIMO
systems with antenna cube in AP is compared with that of the
referenced full MIMO systems with the often used uniformly
spaced vertical dipole arrays.
In the following studies, the measured data of all locations
illustrated in Fig. 6 is adopted. The number of locations is 10.
Four spots are measured at each location. Each measurement
includes 14 channels. The total size of the channel samples for
all following figures is and in each channel
sample, 23 frequency bins are measured with a frequency step
of 1 MHz to cover the 22 MHz channel bandwidth.
A. Average Normalized Receive Power
In wireless communication systems, the performance is af-
fected by the signal to noise ratio. Thus, the capability of col-
lecting more power is quite important to AP antennas. The av-
erage normalized receive power of the compact antenna cube
and the referenced dipole arrays with various antenna at UE are
listed in Table III, which is normalized according to the average
receive power on each port of referenced vertical 5-dipole array
at AP with vertical 3-dipole array at UE as
(4)
where , and are the numbers of the transmit and the
receive antennas of the normalized MIMO channels.
In Table III, normalized receive powers over different sets
of scenarios are listed. In the column of “All Scenarios”, the
locales UE1-10 are considered. In the columns of “LOS Sce-
narios” and “NLOS Scenarios”, the locales UE2, 5, 8, 10 and
UE1, 3, 4, 6, 7, 9 are taken into accounted, respectively. Fur-
ther, “Broad-sight” (locales UE5, 10) and “End-Fire” scenarios
(locales UE2, 8) are compared.
As shown, when vertical 3-dipole array is used at UE, the
average received power of the antenna cube, which is calculated
on each port over all locations, is similar to that of the referenced
dipole array.
However, in other situations, i.e., horizontal dipole array and
dual-polarization antenna are used at the UE, the received power
of the reference array at AP deteriorates. The average power
of the referenced dipole array over all scenarios is 6.14 and
6.01 dB with horizontal 3-dipole and is 3.14 and 3.00 dB
with dual-polarized antenna when and , re-
spectively. While the power of the antenna cube with selections
maintains in all circumstances.
When LOS and NLOS scenarios are considered separately,
the received power on ports of antenna cube with selection is
ZHENG et al.: COMPACT EIGHTEEN-PORT ANTENNA CUBE FOR MIMO SYSTEMS
TABLE III
AVERAGE NORMALIZED RECEIVE POWER OF AP ANTENNAS
Fig. 10. Capacity CCDFs of MIMO systems with cube using 15-to-3 global
and pattern selection circuits and vertical dipoles while and vertical
3-dipole array is applied in UE.
mostly larger than that of the referenced 5/3-dipole array. In few
situations, the received power of antenna cube is less than that
of the referenced dipole array, but the difference is quite small.
It also can be drawn that in almost all the situations, the re-
ceived power of antenna cube with global selection is slightly
stronger than that of antenna cube with simplified selection.
B. Channel Capacity and Eigenvalues
1) Vertical/Horizontal Three-Dipole Array is Employed as
UE Antennas: First, the scenarios that the vertical 3-dipole
array is applied as UE antennas are considered. The channel ca-
pacity complementary cumulative density functions (CCDFs)
of studied MIMO systems are shown in Fig. 10, when
is considered, the 50% outage capacity of full MIMO system
with antenna cube is 7.2 bits/s/Hz more than that of full MIMO
system with dipole array. The selection MIMO systems with
antenna cube also perform better than the full MIMO system
with dipole array in general, no matter whether global or
simplified pattern selection circuit is used. Considering the
451
Fig. 11. Capacity CCDFs of MIMO systems with cube using 15-to-3 global
and pattern selection circuits and vertical dipoles while and vertical
3-dipole array is applied in UE: (a) LOS and NLOS; (b) broad-sight and end-fire
scenarios.
50% outage capacity of the systems, the capacity of MIMO
systems using global selection is about 2.7 bits/s/Hz more than
the full MIMO systems with dipoles as AP antennas, while the
difference between the selection MIMO systems with global
and pattern selection circuits is less than 0.5 bits/s/Hz.
The comparisons between LOS and NLOS scenarios and be-
tween the broad-sight and end-fire scenarios (which all have
LOS path) are demonstrated in Fig. 11. For the average transmit
power normalization is performed with respect both LOS and
NLOS scenarios dependently, the MIMO systems in LOS sce-
narios performs better those in NLOS scenarios for the rela-
tive power gain advantage, which is different from the com-
monly used SNR normalization [15]. As shown in Fig. 11(a),
the channel capacity differences between MIMO systems with
cube and dipole array at AP are nearly same in LOS and NLOS
scenarios when the 50% outage capacity is considered, which
are 2.1 and 2.3 bits/s/Hz, respectively. But when the UE an-
tenna is in end-fire direction of the dipole array, as shown in
Fig. 11(b), the channel capacity deteriorates because of lower
received power which is shown in Table III.
When is considered, which is shown in Fig. 12, the
50% outage capacity of full MIMO system with antenna cube
is 4.5 bits/s/Hz more than that of full MIMO system with dipole
452 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 12. Capacity CCDFs of MIMO systems with cube using 15-to-5 global
and polarization selection circuits and vertical dipoles while and ver-
tical 3-dipole array is applied in UE.
Fig. 13. Capacity CCDFs of MIMO systems with cube using 15-to-5 global
and polarization selection circuits and vertical dipoles while and ver-
tical 3-dipole array is applied in UE: (a) LOS and NLOS; (b) UE in broadsight
and endfire of vertical dipoles at AP in LOS scenarios.
array. The capacity of MIMO systems using global selection
and polarization selection is 1.9 and 1 bits/s/Hz more than that
of the full MIMO systems with vertical 3-dipole array.
The comparison between LOS and NLOS scenarios and that
between the broad-sight and end-fire scenarios are shown in
Fig. 14. Eigenvalue PDFs of normalized MIMO channels with vertical
3-dipole array at UE and: (a) vertical 5-dipole array at AP; (b) antenna cube
using simplified selections at AP; (c) antenna cube using global selection at
AP; (d) full cube at AP, where .
Fig. 13. In Fig. 13(a), the 50% outage capacity differences be-
tween MIMO systems with cube and dipole array at AP are
1.1 and 0.7 bits/s/Hz for LOS and NLOS scenarios. As shown
in Fig. 13(b), it is obvious that the capacity of MIMO system
with referenced 5-dipole array is more susceptible to the envi-
ronment than that of the MIMO system with selected antenna
cube. The slops of the capacity curves of MIMO system with
dipole array in broad-sight and end-fire scenarios vary consid-
erably, but those of MIMO systems with cube antenna change
little. The capacity of MIMO channel with global selected an-
tenna cube is 0.1 and 2.3 bits/s/Hz more than that of MIMO
channel with vertical 5-dipole array.
The eigenvalue probability density functions (PDFs) of the
normalized MIMO channel are studied. The
normalization operation is according to as
(5)
The eigenvalue PDFs of MIMO systems with vertical
3-dipole at UE are illustrated in Fig. 14. In most situations,
the full MIMO system with antenna cube has three significant
eigenvalues with vertical 3-dipole array at UE, which is equal
to the number of transmit antennas.
When is considered, which is denoted with solid
symbols in Fig. 14. For the selection systems, about 50% of the
second maximum eigenvalues are larger than 0 dB, while for
ZHENG et al.: COMPACT EIGHTEEN-PORT ANTENNA CUBE FOR MIMO SYSTEMS
Fig. 15. Capacity CCDFs of MIMO systems with cube and dual polarization
antenna while the number of selected branch is: (a) (15-to-3 global and
pattern selection), (b) (15-to-5 global and polarization selection).
the MIMO system with dipole array, only about 20% of the 2nd
maximum eigenvalues are larger than 0 dB. Most the minimum
eigenvalues of MIMO system with dipole array are less than
10 dB, and are quite less than those of MIMO systems with
selections.
When is considered, which is denoted with hollow
symbols in Fig. 14. The eigenvalue PDFs of MIMO channels
with full cube are similar to those of the full MIMO system with
for full 15 RF branches are used. The full MIMO sys-
tems with antenna cube have three significant eigenvalues in
most situations. For the MIMO channels with global and po-
larization selected antenna cube and vertical 5-dipole array, the
differences between the distribution of the first maximum and
second maximum eigenvalues are small. The minimum eigen-
values of MIMO channels with antenna cube are mostly larger
than 10 dB, but a fairly large number of the minimum eigen-
values of MIMO channels with vertical 5-dipole array are less
than 10 dB. This shows the antenna cube is likely to provide
more effective subchannels.
Secondly, the MIMO systems with horizontal 3-dipole array
at UE are considered. In practical communication systems, per-
formance of the communication system often deteriorates be-
cause of undesirable factors, such as polarization mismatch and
453
Fig. 16. Eigenvalue PDFs of normalized MIMO channels with dual polariza-
tion antenna at UE and: (a) vertical 5-dipole array at AP; (b) antenna cube using
simplified selections at AP; (c) antenna cube using global selection at AP; (d)
full cube at AP, where .
so on, that is quite common in wireless communications for the
UE antennas are often rotated for the randomness of user behav-
iors. Considering the MIMO system with vertical dipoles at the
AP, the performance badly degrades with horizontal dipoles at
UE because the polarization mismatch seriously deteriorates the
received power as listed in Table III. However, for the MIMO
system with the cube, it works both well when either vertical
or horizontal dipoles at the UE. This means that the MIMO
systems with the antenna cube are more robust. When hori-
zontal 3-dipole is adopted at UE, the 50% outage capacity of
global selection MIMO systems with antenna cube is 4.6 and
3.9 bits/s/Hz better than those of the full MIMO systems with
vertical 3-dipole and 5-dipole arrays at AP, respectively.
2) Dual Polarization Antenna is Employed as UE Antenna:
In these scenarios, dual polarization antenna is employed in
UE. As shown in Fig. 15(a), the selection MIMO system with
antenna cube performs much better than the full MIMO system
with dipole array when , the 50% outage capacity
of global selection MIMO system with antenna cube is 2.9
bits/s/Hz more than that of the full MIMO system, and is nearly
the same with the capacity of MIMO systems with pattern se-
lection circuits, the 50% outage capacity of full MIMO system
with antenna cube is 5.4 bits/s/Hz more than the capacity of
the MIMO systems with dipole array, as shown in Fig. 15(a).
When , the MIMO systems with simplified polarization
selection circuit also perform better than the referenced full
MIMO systems. As shown in Fig. 15(b), the 50% outage
capacity of the MIMO systems with global and polarization
selection is 2.0 and 1.6 bits/s/Hz more than the capacity of
454
Fig. 17. Statistics of the selected antennas: (a) numbers of selected faces when
the referenced MIMO system when dual polarization antenna
is used in UE. The capacity of the full MIMO systems with
cube is 3.6 bits/s/Hz more than that of the MIMO system with
referenced vertical 5-dipole array.
The eigenvalue PDFs of MIMO systems with dual polariza-
tion antenna are illustrated in Fig. 16, where solid and hollow
symbols are used to denote situations of and re-
spectively. The eigenvalues of selection MIMO systems with
antenna cube at AP are located in higher range than that of
MIMO systems with vertical dipole. Furthermore, the second
maximum eigenvalues of MIMO systems with vertical 3-dipole
array are much less than that with vertical 5-dipole array while
the difference between eigenvalues of MIMO systems with an-
tenna cube when and are little. Almost all the
second maximum eigenvalues of MIMO systems with vertical
3-dipole array are less than 0 dB while 50% of the second max-
imum eigenvalues of selection MIMO systems with antenna
cube when are large than 0 dB even simplified selection
circuit is implied.
As above, the antenna cube performs well when linking with
different equipment antennas in various scenarios, and satis-
fies the requirements of MIMO systems especially with antenna
selection. It is noted that the capacity of the MIMO systems
with antenna cube maintains no matter what UE antennas are
adopted.
C. Statistics of the Selected Antennas
To further reveal the impact of antenna cube on channel
capacity of selection MIMO systems, the statistics of the se-
lected antenna in each MIMO channel realizations are shown in
Fig. 17. The number of the accounted MIMO channel realiza-
tions is 560. Particularly, when pattern or polarization selection
circuit is implied, the numbers of the selected specific ports or
the faces that the selected antennas lie on are equal because of
the constraints of the selection circuits.
When , the numbers of faces that the selected an-
tennas lie in are accounted and shown in Fig. 17(a). As shown
in Fig. 17(a), the tri-polarization antenna in the #1 (up) face is
more likely to be selected for a fairly large part of the trans-
mitted waves are reflected by the ceiling. The numbers of the
selected tri-polarization antennas vary little when different UE
antennas are used.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
; (b) numbers of selected specific antennas when .
The numbers of the specifically selected antenna in all the
faces are accounted and shown in Fig. 17(b) when .
The P2 ports have more probability to be selected when vertical
3-dipole array is used as UE antenna for the antennas have ver-
tical polarization radiation patterns and high gain.
V. CONCLUSION
This study proposes a planar tri-polarization antenna for
MIMO systems with antenna selection. The total volume of
the antenna cube is 0.76 0.76 0.76 . The designed pro-
totype was fabricated and tested, and measured data validating
simulation results were compared.
By comparing the channel capacity of MIMO systems be-
tween cube and several terminal antennas and the capacity of
MIMO systems between dipole array and the same terminal an-
tennas, the validation of the tri-polarization antenna cube for
MIMO system with antenna selection is approved. The mea-
surement reveals that the compact tri-polarization antenna cube
with the full coverage and polarization diversity is suitable for
MIMO systems with antenna selection indoors, and can provide
high quality communications in various scenarios with different
terminal antennas.
REFERENCES
[1] B. N. Getu and J. B. Andersen, “The MIMO cube—A compact MIMO
antenna,” IEEE Trans. Wirel. Commun., vol. 4, no. 5, pp. 1136–1141,
May 2005.
[2] B. N. Getu and R. Janaswamy, “The effect of mutual coupling on the
capacity of the MIMO cube,” IEEE Antennas Wirel. Propag. Lett., vol.
4, pp. 240–244, 2005.
[3] C. Y. Chiu and R. D. Murch, “Experimental results for a MIMO cube,”
in Proc. IEEE Antennas Propag. Soc. Int. Symp., 2006, pp. 2533–2536.
[4] C. Y. Chiu, J. B. Yan, and R. D. Murch, “24-port and 36-port antenna
cubes suitable for MIMO wireless communications,” IEEE Trans. An-
tennas Propag., vol. 56, no. 4, pp. 1170–1176, 2008.
[5] P. N. Fletcher, M. Dean, and A. R. Nix, “Mutual coupling in multi-
element array antennas and its influence on MIMO channel capacity,”
Electron. Lett., vol. 39, pp. 342–344, Feb. 2003.
[6] M. R. Andrews, P. P. Mitra, and R. deCarvalho, “Tripling the capacity
of wireless communications using electromagnetic polarization,” Na-
ture, vol. 409, no. 1, pp. 316–318, 2001.
[7] A. S. Konanur, K. Gosalia, S. H. Krishnamurthy, B. Hughes, and G.
Lazzi, “Increasing wireless channel capacity through MIMO systems
employing co-located antennas,” IEEE Trans. Microw. Theory Tech.,
vol. 53, no. 6, pp. 1837–1844, Jun. 2005.
[8] K. Itoh, R. Watanabe, and T. Matsumoto, “Slot-monopole antenna
system for energy-density reception at UHF,” IEEE Trans. Antennas
Propag., vol. 27, no. 8, pp. 485–489, Jul. 1979.
ZHENG et al.: COMPACT EIGHTEEN-PORT ANTENNA CUBE FOR MIMO SYSTEMS
[9] X. Gao, H. Zhong, Z. Zhang, Z. Feng, and M. F. Iskander, “Low-Pro-
file planar tri-polarization antenna for WLAN communications,” IEEE
Antennas Wirel. Propag. Lett., vol. 9, pp. 83–86, 2010.
[10] S. Sanayei and A. Nosratinia, “Antenna selection in MIMO systems,”
IEEE Commun. Mag., vol. 42, no. 10, pp. 68–73, Oct. 2004.
[11] A. Ghrayeb, “A survey on antenna selection for MIMO communication
systems,” in Proc. Int. Conf. Inform. Commun. Technol. (ICTTA), 2006,
pp. 2104–2109.
[12] A. F. Molisch, N. B. Mehta, H. Zhang, P. Almers, and J. Zhang, “Im-
plementation aspects of antenna selection for MIMO systems,” in Proc.
Int. Conf. Commun. Network. China (China Com), 2006, pp. 1–7.
[13] Y. Li, Z. Zhang, W. Chen, Z. Feng, and M. Iskander, “A dual-polar-
ization slot antenna using a compact CPW feeding structure,” IEEE
Antennas Wirel. Propag. Lett., vol. 9, pp. 191–194, 2010.
[14] M. Jensen and J. Wallace, “A review of antennas and propagation for
MIMO wireless communications,” IEEE Trans. Antennas Propag., vol.
52, no. 11, pp. 2810–2824, Nov. 2004.
[15] R. Tian, V. Plicanic, B. K. Lau, and Z. Ying, “A compact six-port
dielectric resonator antenna array: MIMO channel measurements and
performance analysis,” IEEE Trans. Antennas Propag., vol. 58, no. 4,
pp. 1369–1379, Apr. 2010.
[16] H. Zhang, A. F. Molisch, and J. Zhang, “Applying antenna selection in
WLANs for achieving broadband multimedia communications,” IEEE
Trans. Broadcast., vol. 52, no. 4, pp. 475–482, Dec. 2006.
[17] H. Zhang and R. U. Nabar, “Transmit antenna selection in MIMO-
OFDM systems: Bulk versus per-tone selection,” in Proc. IEEE Int.
Conf. Commun. (ICC), 2008, pp. 4371–4375.
[18] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wirel. Pers.
Commun., vol. 6, no. 3, pp. 311–335, Mar. 1998.
[19] J. Ahmadi-Shokouh, S. H. Jamali, S. Safavi-Naeini, and G. Z. Rafi,
“Switch loss and antenna directivity effects on MIMO antenna selec-
tion,” in Proc. Canadian Conf. Elect. Comput. Eng. (CCECE), 2008,
pp. 641–646.
[20] N. Honma, K. Nishimori, Y. Takatori, A. Ohta, and K. Tsunekawa,
“Antenna selection method employing orthogonal polarization and ra-
diation patterns for MIMO antenna,” in Proc. Eur. Conf. Antennas
Propag. (EuCAP), 2006, pp. 1–4.
[21] Z. Xu, S. Sfar, and R. S. Blum, “Receive antenna selection for closely-
spaced antennas with mutual coupling,” IEEE Trans. Wirel. Commun.,
vol. 9, no. 2, pp. 652–661, Feb. 2010.
Jianfeng Zheng received the B.S. and Ph.D. degrees
from Tsinghua University, Beijing, China, in 2002
and 2009, respectively.
He is currently an Assistant Researcher with the
State Key Laboratory on Microwave and Digital
Communications, Tsinghua University. His current
research interests include spatial temporal signal
processing, MIMO channel measurements and
antenna arrays for MIMO communications.
455
Xu Gao received the B.S. degree from Shandong
University, Jinan, China, in 2007, and the M.S.
degree from Tsinghua University, Beijing, China, in
2010. He is currently pursuing the Ph.D. degree from
the Missouri University of Science and Technology,
Rolla.
He is currently working with the EMC Lab,
Missouri University of Science and Technology.
His research interests include antenna design, wave
propagation, electromagnetic compatibility, RF
design, and computational electromagnetics.
Zhijun Zhang (M’00–SM’04) received the B.S.
and M.S. degrees from the University of Electronic
Science and Technology of China, Anhui, in 1992
and 1995, respectively, and the Ph.D. degree from
Tsinghua University, Beijing, China, in 1999.
In 1999, he was a Postdoctoral Fellow with the
Department of Electrical Engineering, University of
Utah, where he was appointed a Research Assistant
Professor in 2001. In May 2002, he was an Assistant
Researcher with the University of Hawaii at Manoa,
Honolulu. In November 2002, he joined Amphenol
T&M Antennas, Vernon Hills, IL, as a Senior Staff Antenna Development En-
gineer and was then promoted to the position of Antenna Engineer Manager.
In 2004, he joined Nokia Inc., San Diego, CA, as a Senior Antenna Design
Engineer. In 2006, he joined Apple Inc., Cupertino, CA, as a Senior Antenna
Design Engineer and was then promoted to the position of Principal Antenna
Engineer. Since August 2007, he has been with Tsinghua University, where he
is a Professor with the Department of Electronic Engineering. He is the author
of Antenna Design for Mobile Devices (Wiley, 2011).
Prof. Zhang is serving as an Associate Editor of the IEEE TRANSACTIONS
ON ANTENNAS AND PROPAGATION and the IEEE ANTENNAS AND WIRELESS
PROPAGATION LETTERS.
Zhenghe Feng (M’00–SM’08) received the B.S. de-
gree in radio and electronics from Tsinghua Univer-
sity, Beijing, China, in 1970.
Since 1970, he has been with Tsinghua University,
as an Assistant, Lecture, Associate Professor, and
Full Professor. His main research areas include
numerical techniques and computational electro-
magnetics, RF and microwave circuits and antenna,
wireless communications, smart antenna, and spatial
temporal signal processing.
456 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Printed MIMO-Antenna System Using
Neutralization-Line Technique for
Wireless USB-Dongle Applications
Saou-Wen Su, Member, IEEE, Cheng-Tse Lee, Member, IEEE, and Fa-Shian Chang
Abstract—A printed two-multiple-input multiple-output
(MIMO)-antenna system incorporating a neutralization line for
antenna port decoupling for wireless USB-dongle applications
is proposed. The two monopoles are located on the two opposite
corners of the system PCB and spaced apart by a small ground
portion, which serves as a layout area for antenna feeding network
and connectors for the use of standalone antennas as an optional
scheme. It was found that by removing only 1.5 mm long inwards
from the top edge in the small ground portion and connecting the
two antennas therein with a thin printed line, the antenna port
isolation can be effectively improved. The neutralization line in
this study occupies very little board space, and the design requires
no conventional modification to the ground plane for mitigating
mutual coupling. The behavior of the neutralization line was
rigorously analyzed, and the MIMO characteristics of the pro-
posed antennas was also studied and tested in the reverberation
chamber. Details of the constructed prototype are described and
discussed in this paper.
Index Terms—2.4 GHz WLAN antennas, multiple-input mul-
tiple-output (MIMO) antennas, neutralization line, printed
monopole antennas, wireless USB-dongle antennas.
I. INTRODUCTION
M ULTIPLE-INPUT multiple-output (MIMO) technology
using multiple transmit/receive antennas is considered
one of the most promising approaches to achieve higher data
rate with no additional spectrum required and at the same
time, to make use of the indoor multi-path propagation for
improving signal quality and reliability. Until quite recently,
the IEEE Standard Association has ratified the IEEE 802.11n
standard in September 2009 [1]. It can be expected that mul-
tiple antennas are demanded accordingly and may be closely
packed inside the devices due to limited space left for antennas.
Multi-antenna designs that require high decoupling between
the antenna ports are essential for the multi-radio, antenna
system development. Several methods for improving antenna
port isolation have been reported, including incorporating a
protruded ground plane between the antennas [2], inserting slits
Manuscript received December 23, 2010; revised April 22, 2011; accepted
August 16, 2011. Date of publication ; date of current version February 03, 2012.
S.-W. Su and C.-T. Lee are with the Network Access Strategic Business
Unit, Lite-On Technology Corp., Taipei County 23585, Taiwan (e-mail: stephen.
su@liteon.com).
F.-S. Chang is with the Department of Electronics, Cheng Shiu University,
Kaohsiung County 83347, Taiwan.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173450
0018-926X/$26.00 © 2011 IEEE
into the ground [3]–[5], arranging antenna shorting portions
facing each other [6], [7], manipulating radiation polarization
of the antennas [8]–[10] and so on. For the cost- effective
concern for USB-dongle applications that the printed antennas
are the most favorable designs, none of the above- mentioned
techniques for decoupling antenna ports are suited to the
requirements of printed antennas and little board space for
accommodating decoupling-structure layout. This motivates
us to implement another promising technique by incorporating
a neutralization line [11], [12] into the printed two-antenna
system for USB-dongle applications. A neutralization-line
technique allows the signals picked up from one antenna to the
other and produces an opposite coupling to the existing one
without the presence of neutralization line such that low mutual
coupling at certain frequencies is achieved [11].
The proposed design comprises two simple short-circuited
monopoles placed on the opposite corners of a FR4 substrate,
which can be considered the system PCB of a wireless USB
dongle. In between the monopoles is the small ground portion
reserved for the feeding network (for example matching cir-
cuits), I-PEX connectors for having standalone, coaxial-line-
feed antennas as an optional scheme, RF testing connectors for
manufacturing test and so on. Therefore, it is important to have
this portion as much untouched as possible and to avoid slit cuts
or a protruded ground therein. In this paper, the small ground
portion of the system ground plane was removed only 1.5 mm
long inwards from the top edge to accommodate a thin, printed
neutralization line that links to the monopoles relatively close to
the antenna feed ports. It was found that by adding the neutral-
ization line in the proposed design, the antenna port isolation can
be effectively enhanced, compared with the design with no neu-
tralization line. The antennas are not necessarily targeted on the
USB-dongle platform only but also on a wireless module-card
solution with the form factor of a USB dongle for possible wire-
less LCD TV applications. The experimental and simulation re-
sults of a constructed prototype are fully presented.
II. ANTENNA CONFIGURATION AND DESIGN CONSIDERATION
Fig. 1(a) shows the geometry of the proposed two-monopole-
antenna system formed on a single-layered FR4 substrate of
thickness 1 mm for MIMO applications. The size of the system
PCB selected in this study is 30 mm 65 mm, which can
be considered the form factor of the PCB for a wireless USB
dongle. The two monopoles are printed on the two opposite
corners of the PCB [shown on the top portion in Fig. 1(a)] and
spaced 14 mm apart with a small ground portion (13 mm 14
SU et al.: PRINTED MIMO-ANTENNA SYSTEM USING NEUTRALIZATION-LINE TECHNIQUE
Fig. 1. (a) Geometry of the two short-circuited monopoles linked by a neutral-
ization line printed on the top portion of a FR4 substrate. (b) Detailed dimen-
sions of the two- monopole-antenna system.
mm) therein between. Each monopole is designed in a clearance
area (no grounding layout and electric components 14 therein)
of size 8 mm 14.5 mm. The two monopoles are also identical
in size and symmetrically placed with respect to the PCB
center line [see symmetrical line in Fig. 1(a)]. Accordingly, it
is expected that the performance of each monopole should be
the same. Notice that the small ground portion is reserved for
the antenna feeding network and the I-PEX connectors for the
use of the standalone, coaxial-line-feed antennas as an optional
scheme. Therefore, it is important to keep this area as much
untouched as possible (no slit cut or protruded ground).
Detailed design dimensions are given in Fig. 1(b). The pre-
ferred dimensions for the prototype were attained by the rig-
orous parametric studies with the aid of the electromagnetic
simulator, Ansoft HFSS [13]. As seen in the figure, monopoles
1 and 2 are fed at ports 1 and 2 and short-circuited to the system
ground at points and with an L-shaped strip respectively.
Two short, 50- mini-coaxial lines with I-PEX connectors were
utilized for feeding the antennas in the experiments (see the
photo of a working sample demonstrated in Fig. 2). Notice that
the feed gap between the feed point (points and ) and the
system ground was fixed to be 1 mm in the study. Both the
457
Fig. 2. Photo of a constructed prototype fed by 50- mini-coaxial cables.
monopoles are of quart-wavelength resonant structures, and the
antenna operating frequencies can be determined by the reso-
nant- path length from the feed point to the open end of the
monopole. The input matching of the antenna is easily fine-
tuned by varying the length and the width of the short-circuiting
strip from point to for monopole 1 and point to for
monopole 2.
The two short-circuited monopoles were first designed indi-
vidually to obtain the optimal, achievable impedance bandwidth
for simplicity; the mutual coupling was not considered at this
stage. Then the two antennas were connected together by using
a neutralization line. This conducting line is of width 0.3 mm
and does not take up much available layout space of the system
PCB. In this design, the system ground was removed only 1.5
mm long inwards from the substrate’s edge. The antenna port
isolation was found to be effectively improved by linking the
two highly-coupled monopoles at proper locations near the feed
port at point for monopole 1 and point for monopole 2.
The locations chosen conform to the studies in [11], which re-
ports that a low impedance area (with minimum voltage but
maximum currents) is favorable. The occurrence of the isola-
tion properties is variable with respect to the location of the
connecting points and . The studies of this parameter will
be elaborated with the aid of Table I in Section III. Finally, both
the antenna frequencies are also affected with the incorporation
of the line, and each monopole needs fine-tuning to readjust its
required frequency band.
III. RESULTS AND DISCUSSION
A. Reflection Coefficients, Isolation, TARC, and Envelope
Correlation
A prototype of the proposed antenna as shown in Fig. 2
was first constructed and measured based upon on the design
and dimensions thereof described in Fig. 1. Fig. 3 shows the
measured reflection coefficients ( for monopole 1 and
for monopole 2) and the isolation between the
two monopoles, whose simulated counterparts are given in
Fig. 4(a). The isolation is only presented by the curves of
due to the symmetrical structure of the proposed design.
458 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

TABLE I
SIMULATED RESULTS OF THE MONOPOLES AS A FUNCTION OF STUDIED
IN FIG. 3. AND ARE, RESPECTIVELY, THE LOWER AND UPPER
EDGE FREQUENCIES OF THE 10 dB IMPEDANCE AND THE SAME
FOR BOTH MONOPOLES 1 AND 2; IS OF THE MAXIMUM
VALUE WITHIN THE BAND OF INTEREST

Fig. 3. Measured reflection coefficients ( for monopole 1 and for

monopole 2) and isolation between the two monopoles of the proposed
design; 8 mm.
Fig. 4. Simulated reflection coefficients ( for monopole 1 and for
monopole 2) and isolation between the two monopoles for (a) the
proposed design and (b) the reference case (with no neutralization line).
On average, the experimental data agree with the simulation
results, which were based on the finite element method (FEM).
The measured impedance matching of the two monopoles
resembles an inductor put in between and well matched to the
over the 2.4 GHz band is all below (about VSWR
two ports (that’s ports 1 and 2 here) and transfers the signals
of 2), which meets the demanded bandwidth specification for
from port 1 to port 2, resulting poor antenna port isolation. For
WLAN operation. The isolation between the antennas is all
the proposed design, it is interesting to observe that part of the
below about 19 dB. When there is no neutralization line (the
resistance values is negative [14], which indicates an opposed
reference design, see Fig. 8), the antenna port isolation rapidly
direction for the current flow in a virtual capacitor (because of
deteriorates by about 9 dB as seen in Fig. 4(b), compared with
negative imaginary part) placed between the ports to replace
the data in Fig. 4(a). This behavior suggests that the isolation
the two monopoles and the radiation air path. The increase in
can be effectively improved by incorporating the neutraliza-
the currents entering port 1 to the loading between the two ports
tion line into the design, although the two antenna ports face
also results in decreased voltage over the loading. In this case,
each other with the same radiation polarization (see Fig. 10).
the coupling field between the antenna ports is weak, leading
It is noticed that the antenna operating frequencies and the
to good antenna port isolation.
impedance bandwidth thereof are also affected by the use of the
To consider the constructive and destructive coupled signals
neutralization-line technique. Furthermore, various locations
on the combination of each antenna port’s reflected signals, the
of the connecting points and were also analyzed on the
total active reflection coefficient (TARC) [15]–[17] is also in-
monopole frequencies and the in-band isolation ;
vestigated. TARC is defined as the ratio of the square root of
the results are tabulated in Table I. The antenna frequencies are
total reflected power divided by the square root of total incident
seen to be less affected compared with the port isolation and
can still be within the 2.4 GHz frequency range. However, the
dip of the curve shifts from the lower to higher frequencies (1)
with an increase in the length . In this case, there exists an
optimal location for connecting both the monopoles. Fig. 5 power and can be considered the MIMO array radiation effi-
presents the impedance of the isolation studied in Fig. 4. For ciency for a multi-port antenna [16]. The calculations can be
the reference, the resistance impedance is stable and close to done by using (1) described in [17], where is the phase angle
50 with inductive reactance over the band. This behavior of port 2 excitation, from which it is straightforward to see that
SU et al.: PRINTED MIMO-ANTENNA SYSTEM USING NEUTRALIZATION-LINE TECHNIQUE
Fig. 5. Impedance of isolation between the two monopoles for the pro-
posed and the reference designs studied in Fig. 4.
Fig. 6. Calculated TARC for one monopole of the two-monopole antenna
system; each curve shows a reflection coefficient for an excitation with constant
amplitude but different phase angle of port 2 excitation in steps of 30 degrees
for 180 degrees.
TARC accounts for both coupling and random signals com-
bining. Fig. 6 presents the calculated TARC from the scattering-
matrix elements of , , , obtained in Fig. 4(a) for
monopole 1 with randomly phased excitation of monopole 2
with a set of seven excitation vectors. The curves retain the
original characteristics of the reflection coefficient of a single
antenna, but the impedance bandwidth is changed due to the ef-
fects of the mutual coupling and the incident waves with random
phases. Interesting to notice that TARC becomes worst when
the phase is equal to 180 (out- of-phase incident signals upon
monopole 2 with respect to monopole 1). The average TARC in
Fig. 6 shows that the in-band impedance is all below 10 dB
and the worst case calculated TARC is smaller than 7.3 dB
over the band.
Fig. 7 plots the calculated envelope correlation between
the two monopoles. The correlation was determined by the
parameters in (11) reported in [18] for sufficiently accurate re-
sults in many practical cases [19]. However, it should be noticed
that the equation is unsuitable for the case of zero mutual cou-
pling . The magnitude and phase of the parame-
ters were collected from the simulation data in Fig. 4(a). A brief
description of the calculation was discussed in [20]. From the
results, the values remain under 0.006 in the 2.4 GHz band and
are much smaller than 0.5 at the mobile station (the user end)
459
Fig. 7. Calculated envelope correlation for the two-monopole-antenna system.
[21]. In addition to the said method, the evaluation of the en-
velope correlation can be measured in a reverberation chamber
[22]. The correlation coefficient is measured, and the envelope
correlation is then obtained by (2)
(2)
B. Current Distribution, Near-Field, and Far-Field Radiation
Characteristics
The excited surface-current distributions of the design with
(the proposed) and without (the reference) the neutralization
line are studied in Fig. 8 to understand the behavior of the neu-
tralization line in relation to the two monopoles. The current-
magnitude scale is kept the same among all the cases. First,
the current distributions on the system ground plane and the
strength thereof are similar between the proposed and the refer-
ence. This behavior is different from the finding in [23], which
indicates the isolation between the antennas is related to the sur-
face currents on the ground. Instead, the mutual coupling in this
study is mitigated by introducing the counter-phased currents
against the excited antenna [11]. For example, when monopole
1 is excited (port 1 excitation) with the currents entering the an-
tenna, the current vector on the two-third of the neutralization
line closer to monopole 2 is in the direction toward monopole
1. In this case, the surface currents are out of phase with a null
located closer to monopole 1 on the line, such that the conduc-
tive currents are opposed to the excited antenna. In addition, it
can be seen that the currents enter the two monopoles in the
proposed (see currents from points to or to ) but in
opposite directions in the reference (currents entering port 2
from monopole 2). It means that there exists an close path, in-
cluding monopoles 1, 2, and the air between the open ends of
the two monopoles, from port 1 to port 2, and the signal cur-
rents entering port 2 are against those entering port 1 and out
of phase between the two ports for the proposed. Quite the con-
trary, for the reference, the signals currents enter port 1 but leave
port 2 from monopole 2, and the currents even enter the system
ground through the shorting strip. These can explain poor iso-
lation for the reference and better antenna port isolation for the
proposed. Second, for the reference, although the current-dis-
tributions compared with the proposed are not much different
on the system ground and identical on the monopole from point
460 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 8. Simulated surface-current distributions at 2442 MHz for the proposed
and the reference designs with port 1 and port 2 excitation, respectively.
and to the open end, the antenna isolation is poor because
without any means of coupling cancellation. Also, the currents
leave monopole 1 and enter monopole 2, respectively, at each
open end of the monopole. This suggests that the corresponding
antenna could receive the radiating signals of the excited an-
tenna through the near-field radiation in this design. Therefore,
to redirect the maximum near-field strength of the excited an-
tenna away from the port of the corresponding antenna should
facilitate port-to-port decoupling in a two-antenna system.
Fig. 9 presents the simulated reactive near-field radiation pat-
terns for monopole 1 excited at 2442 MHz for the proposed and
the reference designs. The radius with port 1 or the gap source
in the simulation as the center was set 14 and 22 mm, which
are the effective range that the near field of the excited antenna
(monopole 1) affects the corresponding one (monopole 2). The
radius, 14 mm, is the same length as that of the small ground
portion between the monopoles; in this case, the range is of the
port-to-port distance. Therefore, the radiation in Fig. 9(a) repre-
sents that when monopole 1 is operating, the coverage reaches
only to port 2 but not monopole 2 included. For the radius of 22
mm in Fig. 9(b), the range counts from port 1 to the right edge
of monopole 2. Compared with the reference, which shows the
large field strength aiming at port 2, the maximum strength of
the reactive near field in Fig. 9(a) is not in the direction of port
2 but pointing to the portion of the neutralization line with the
maximum currents seen in Fig. 8. This characteristic indicates
that the large currents on the neutralization line can draw the
near field in the case of the port-to-port radius and redirect the
field away from the receiving antenna port, which in turn results
in better port isolation. However, if the range of the reactive near
field exceeds the port-to-port distance, the near-field radiation
behaves similarly between the proposed and the reference as
can be seen in Fig. 9(b). This phenomenon is expected because
the current distributions on monopoles 1 and 2 from point and
to the open end are identical, the reactive near field excited
by monopole 1 can reach the open end of monopole 2 over the
air, such that the currents leave the open end of monopole 1 and
enter the open end of monopole 2. The reactive near fields with
other radius values larger than port-to-port range of 14 mm but
smaller than one wavelength (range criterion for reactive near
Fig. 9. Simulated reactive near-filed radiation patterns at 2442 MHz for the
proposed and the reference designs with the radius of (a) 14 mm and (b) 22 mm
with respect to port 1 (monopole 1) excitation.
field) at 2442 MHz were also examined. The results showed
similar patterns for both the proposed and the reference designs.
The over-the-air (OTA) performance of the antenna in free
space was studied. Fig. 10 shows the far-field, 2-D radiation pat-
terns at 2442 MHz, the center frequency of the 2.4 GHz band,
in and fields for monopoles 1 and 2. The patterns were
normalized with respect to the maximum field strength among
three major planes: the – , – , and – cuts. The omnidi-
rectional radiation patterns in this design lie in the – and –
planes, and the radiation for ports 1 and 2 tends to cover the com-
plementary space region (see – cuts). The polarization for the
two monopoles was observed the same. Fig. 11 presents the sim-
ulated, far-field 3-D radiation patterns of the antennas studied
in Fig. 10; the measured counterparts are given in Fig. 12. The
measurement was made by the ETS-Lindgren OTA test system
using the great-circle method in a CTIA authorized test labora-
tory [24]. Overall, the measured results are similar to the simu-
lation. It can be seen that the null radiation is located in the op-
posite half-spaces in the – planes. Other in-band frequencies
were also measured, and no inconsistency on the patterns was
noticed. Fig. 13 shows the peak antenna gain and the radiation
efficiency against frequency for the two monopoles. Again, be-
cause the two antennas are identical and symmetrically placed
with respect to the PCB center, the gain and the radiation effi-
ciency are about the same. The peak gain in the 2.4 GHz band
for the two antennas is at a constant level of about 2.1 dBi, and
the radiation efficiency is larger than about 70%. The gain mea-
surement here takes account of the antenna mismatching and is
the “realized gain” [25]. The radiation efficiency was obtained
SU et al.: PRINTED MIMO-ANTENNA SYSTEM USING NEUTRALIZATION-LINE TECHNIQUE
Fig. 10. Simulated 2-D radiation patterns at 2442 MHz for (a) monopole 1 and
(b) monopole 2 studied in Fig. 3.
Fig. 11. Simulated 3-D radiation patterns at 2442 MHz for monopole 1 (port 1
excitation) and monopole 2 (port 2 excitation) studied in Fig. 3.
by calculating the total radiated power of the antenna under test
(AUT) over the 3-D spherical radiation first and then dividing
461
Fig. 12. Measured 3-D radiation patterns (including the – and – cuts) at
2442 MHz for (a) monopole 1 and (b) monopole 2 studied in Fig. 3.
Fig. 13. Measured antenna gain and radiation efficiency for monopoles 1 and
2 studied in Fig. 12.
that total amount by the input power of 0 dBm (default value)
given to the AUT in the test laboratory.
Finally, to find out the antenna diversity gain, the proposed
design was tested in the Bluetest reverberation chamber [22],
462
Fig. 14. Cumulative probability distribution function of the two monopoles in
the reverberation chamber based on 3016 measured power samples for each
monopole (branch).
which emulates a rich scattering and fading environment fol-
lowing a Rayleigh distribution to represent a real MIMO envi-
ronment. The of the two monopoles were measured simul-
taneously by connecting each antenna (denoted as one branch)
to a four-port vector network analyzer. The subscript “1” and
“ ” of means port 1 connecting to the three transmitting
monopoles placed perpendicular to each other for three polar-
ization and port j connecting to each corresponding antenna in
the test (see Fig. 12, [26, Fig. 10]). The center frequency was
set at 2442 MHz with used, measured frequencies ranging from
2440 to 2444 MHz for sampling. Fig. 14 plots the cumulative
distribution function (CDF) of the measured power-transmis-
sion samples for the two branches (monopoles 1 and 2) against
the relative received power recorded. At a cumulative proba-
bility level of 1% (that’s, the sufficient quality 99% of the time),
the difference between the CDF of selection combining and the
best CDF between the two monopoles represents the apparent
diversity gain [26]. The apparent diversity gain was observed
about 9.8 dB in the test.
IV. CONCLUSION
A printed, two-monopole-antenna system decoupled by
using the neutralization-line technique has been demonstrated
to attain good antenna port isolation, and the constructed
prototype has been successfully constructed and tested. Each
antenna is of the same size and occupies a clearance layout area
of 8 mm 14.5 mm on the two opposite corners of the system
PCB with a small ground portion between the antennas. The
neutralization line in this design does not occupy much board
space of the system ground plane and only takes 1.5 mm long
inwards from the PCB edge in the small ground portion. In this
case, the antenna feeding network and the I-PEX connectors
can be all placed on that small ground portion for practical
applications. The results showed that the obtained antenna port
isolation is less than about 19 dB and is better than that of
the reference case with no neutralization line by about 9 dB.
The envelope correlation and the TARC were also studied
and derived from the parameters. The radiation patterns of
the two monopoles cover the complementary space regions
in general, and the antenna yields peak gain of about 2.1 dBi
with radiation efficiency exceeding about 70%. The impedance
of the isolation, the surface currents, and the near-fields were
analyzed in detail for the effects of the neutralization line used.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
The proposed design is also expected to be applied to wireless
module-card solution in the form factor of a USB dongle.
REFERENCES
[1] Wireless LAN specification to provide significantly improved data
throughput and range, The IEEE Standard Association [On-
line]. Available: http://standards.ieee.org/announcements/ieee802.
11n_2009amend-ment_ratified.html
[2] T. Y. Wu, S. T. Fang, and K. L. Wong, “Printed diversity monopole
antenna for WLAN operation,” Electron. Lett., vol. 38, pp. 1625–1626,
Dec. 2002.
[3] T. Ohishi, N. Oodachi, S. Sekine, and H. Shoki, “A method to im-
prove the correlation coefficient and the mutual coupling for diversity
antenna,” in IEEE Antennas Propag. Soc. Int. Symp. Dig., 2005, pp.
507–510.
[4] G. A. Mavridis, J. N. Sahalos, and M. T. Chryssomallis, “Spatial di-
versity two-branch antenna for wireless devices,” Electron. Lett., vol.
42, pp. 266–268, Mar. 2006.
[5] C. Y. Chiu, C. H. Cheng, R. D. Murch, and C. R. Rowell, “Reduction
of mutual coupling between closely-packed antenna elements,” IEEE
Trans. Antennas Propag., vol. 55, no. 6, pp. 1732–1738, Jun. 2007.
[6] K. L. Wong and J. H. Chou, “Integrated 2.4- and 5-GHz WLAN an-
tennas with two isolated feeds for dual-module applications,” Micro.
Opt. Technol. Lett., vol. 47, pp. 263–265, Nov. 2005.
[7] S. W. Su, J. H. Chou, and T. Y. Wu, “Internal broadband diversity
dipole antenna,” Microw. Opt. Technol. Lett., vol. 49, pp. 810–812,
Apr. 2007.
[8] Y. Ge, K. P. Esselle, and T. S. Bird, “Compact diversity antenna for
wireless devices,” Electron. Lett., vol. 41, pp. 52–53, Jan. 2005.
[9] S. W. Su, J. H. Chou, and Y. T. Liu, “Realization of dual-dipole-an-
tenna system for concurrent dual-radio operation using polarization di-
versity,” Micro. Opt. Technol. Lett., vol. 51, pp. 1725–1729, Jul. 2009.
[10] S. W. Su, “A three-in-one diversity antenna system for 5 GHz WLAN
applications,” Micro. Opt. Technol. Lett., vol. 51, pp. 2477–2481, Oct.
2009.
[11] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas, “Study
and reduction of the mutual coupling between two mobile phone PIFAs
operating in the DCS1800 and UMTS bands,” IEEE Trans. Antennas
Propag., vol. 54, no. 11, pp. 3063–3074, Nov. 2006.
[12] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas, “En-
hanced two-antenna structures for universal mobile telecommunica-
tions system diversity terminals,” IET Microw. Antennas Propag., vol.
2, pp. 93–101, 2008.
[13] Ansoft Corp., Pittsburgh, PA, “HFSS,” [Online]. Available:
http://www.ansoft.com/products/hf/hfss
[14] Wikipedia, the Free Encyclopedia, “Negative Resistance,” [Online].
Available: http://en.wikipedia.org/wiki/Negative_resistance
[15] M. Manteghi and Y. Rahmat-Samii, “Multiport characteristics of a
wide-band cavity backed annular patch antenna for multipolariza-
tion operations,” IEEE Trans. Antennas Propag., vol. 53, no. 1, pp.
466–474, Jan. 2005.
[16] D. W. Browne, M. Manteghi, M. P. Fitz, and Y. Rahmat-Samii, “Ex-
periments with compact antenna arrays for MIMO radio communica-
tions,” IEEE Trans. Antennas Propag., vol. 54, no. 11, pp. 3239–3250,
Nov. 2006.
[17] S. H. Chae, S. K. Oh, and S. O. Park, “Analysis of mutual coupling, cor-
relations, and TARC in WiBro MIMO array antenna,” IEEE Antennas
Wirel. Propag., vol. 6, pp. 122–125, 2007.
[18] S. Blanch, J. Romeu, and I. Corbella, “Exact representation of antenna
system diversity performance from input parameter description,” Elec-
tron. Lett., vol. 39, pp. 705–707, May 2003.
[19] V. Plicanic, Z. Ying, T. Bolin, G. Kristensson, and A. Derneryd, “An-
tenna diversity evaluation for mobile terminals,” in Proc. Euro. Conf.
Antennas Propag., 2006, pp. 1–3.
[20] S. W. Su, “High-gain dual-loop antennas for MIMO access points in
the the 2.4/5.2/5.8 GHz bands,” IEEE Trans. Antennas Propag., vol.
58, no. 7, pp. 2414–2419, Jul. 2010.
[21] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile com-
munications,” IEEE Trans. Veh. Technol., vol. 36, no. 11, pp. 149–172,
Nov. 1987.
[22] Bluetest, Gothenburg, Sweden, “Reverberation test systems,” [Online].
Available: http://www.bluetest.se/products/reverberation-test-systems
[23] K. L. Wong, J. H. Chou, S. W. Su, and C. M. Su, “Isolation between
GSM/DCS and WLAN antennas in a PDA phone,” Micro. Opt.
Technol. Lett., vol. 45, pp. 347–352, May 2005.
SU et al.: PRINTED MIMO-ANTENNA SYSTEM USING NEUTRALIZATION-LINE TECHNIQUE
[24] CTIA, the Wireless Association, Washington, DC, “CTIA Authorized
Test Laboratory,” [Online]. Available: http://www.ctia.org/busi-
ness_resources/certification/test_labs/
[25] J. L. Volakis, Antenna Engineering Handbook, 4th ed. New York:
McGraw-Hill, 2007, ch. 6, pp. 16–19.
[26] B. Furht and S. A. Ahson, Long Term Evolution: 3GPP LTE Radio and
Cellular Technology. Boca Raton, FL: CRC Press, 2009, ch. 12, pp.
441–443.
Saou-Wen Su (S’05–M’08) was born in Kaohsiung,
Taiwan, on November 11, 1977. He received the
B.S., M.S., and Ph.D. degrees in electrical engi-
neering from National Sun Yat-Sen University,
Kaohsiung, Taiwan, in 2001, 2003, and 2006,
respectively.
Since April 2006, he was with the Technology Re-
search and Development Center, Lite-On Technology
Corporation, Taipei, Taiwan, where he is currently
working with the Network Access Strategic Business
Unit. He built up the first RF Antenna Design Team,
Lite-On Technology Corporation and contributed numerous cutting-edge de-
signs to the company’s ODM projects, including enterprise/SMB access point,
router, Bluetooth headset/car kit, home entertainment device, media box, RF
module, etc. Many customized and standard antenna designs were successfully
mass produced. Currently, he has published over 80 refereed SCI journal pa-
pers and numerous international conference articles. He holds 25 U.S. and 26
Taiwan patents and many patents pending. His expertise is in the industrial RF
antenna designs for wireless AP/router, Bluetooth, WLAN, and MIMO appli-
cations, and previous researches prior to Lite-On Technology Corporation in-
cluded mobile-phone and wideband antenna designs.
Dr. Su was a recipient of a one-year full-time School Study Exchange Pro-
gram Scholarship to The University of Auckland, New Zealand from the Asian
2000 Foundation in 1998.
463
Cheng-Tse Lee (S’08–M’10) was born in Yilan,
Taiwan, in 1983. He received the B.S. degree in
electronic engineering from National Changhua
University of Education, Changhua, Taiwan, and
the M.S. and Ph.D. degrees in electrical engineering
from National Sun Yat-Sen University, Kaohsiung,
Taiwan, in 2005, 2007, and 2010, respectively.
He is currently with the Network Access Strategic
Business Unit, Lite-On Technology Corporation,
Taipei, Taiwan. His main research interests include
antenna designs for wireless communications,
especially for the planar antennas for mobile phone, laptop computer, and also
in microwave and RF circuit design. His expertise is in the industrial antenna
designs for WWAN, WLAN, Bluetooth and MIMO applications.
Fa-Shian Chang was born in Taoyuan, Taiwan,
on April 21, 1968. He received the B.S. degree in
physics from Chinese Military Academy, Kaoh-
siung, Taiwan, the M.S. degree in applied physics
from Chung Cheng Institute of Technology, National
Defense University, Tauyuan, Taiwan, and the Ph.D.
degree in electrical engineering from National Sun
Yat-Sen University, Kaohsiung, Taiwan, in 1990,
1995, and 2002, respectively.
Since November 1990, he has served in the mil-
itary at Chinese Military Academy, and became an
Assistant Professor with the Department of Electrical Engineering in 2005. He
built up the Advanced Technology Laboratory at Chinese Military Academy
and did several government-grant projects in the fields of base station antennas,
simulation system, and unman ground vehicle. Many of the designs were suc-
cessfully produced. After retiring from the colonel position, he taught at the
Department of Electronics, Cheng Shiu University, Kaohsiung, Taiwan, as an
Assistant Professor and was in charge of the Intelligent Vehicle Development
Laboratory. He has published over 25 refereed SCI journal papers and many
international conference articles. He holds 4 U.S. and 21 Taiwan patents and
many patents pending. His expertise is in the antenna designs for remote con-
trol robot, WLAN, and MIMO applications.
464 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Simple and Efficient Decoupling of Compact Arrays
With Parasitic Scatterers
Buon Kiong Lau, Senior Member, IEEE, and Jørgen Bach Andersen, Life Fellow, IEEE
Abstract—Compact arrays such as multiple antennas on a
mobile terminal suffer from low efficiency and high correlation
between antenna signals. In the present paper, a simple and rig-
orous procedure for decoupling two closely coupled antennas with
a parasitic scatterer is proposed. The parasitic scatterer, which
can be an additional antenna, acts as a shield between two active
antenna elements. In contrast to previous studies involving the use
of parasitic scatterer for decoupling antennas, we demonstrate
using antenna impedances the underlying decoupling mechanism
for two arbitrary antennas. By a proper choice of parameters, per-
fect matching and decoupling can be obtained for a given antenna
spacing without extending the overall area used, and without
introducing additional decoupling networks. The price to pay is a
reduction of bandwidth relative to that of widely spaced antennas,
which is the case for other decoupling methods as well. Simulation
and experimental results are used to substantiate the effectiveness
of the proposed design approach on a two-monopole array with
an antenna spacing of 0.1 wavelength. Finally, several practical
considerations of the proposal are also presented, including the
extension of the approach for more than two active antennas and
its implementation in mobile terminals.
Index Terms—Antenna array mutual coupling, impedance
matching, parasitic antennas.
I. INTRODUCTION
C ONVENTIONALLY, antenna arrays were used in radar
installations and satellite communications. In these appli-
cations, it is typical to separate adjacent antenna elements by
one half of a wavelength , in order to maximize array res-
olution without the problem of ambiguity [3]. The same conclu-
sions apply to the more recent application of antenna arrays at
base stations in wireless communications (see e.g., [4]).
However, the overall size of the array structure has become a
subject of current interest, following the widespread adoption
of multiple-input multiple-output (MIMO) technology in ex-
isting and future wireless communications standards [5]. One
reason for this is that the implementation of multiple antennas
in compact user terminals involves challenging design tradeoffs
[6]. For example, even though techniques exist to mitigate mu-
Manuscript received June 13, 2010; revised December 15, 2010; accepted
January 21, 2011. Date of publication October 25, 2011; date of current version
February 03, 2012. This work was supported by VINNOVA under grant 2008-
00970 and Vetenskapsrådet under grant 2006-3012. This paper was presented
in part at the International Workshop on Antenna Technology, Santa Monica,
CA, Mar 2–4, 2009 [1] and also in part as a patent application [2].
B. K. Lau is with the Department of Electrical and Information Technology,
Lund University, Sweden (e-mail: bklau@ieee.org).
J. B. Andersen is with the Antennas, Propagation and Radio Networking
(APNET) Section, Department of Electronic Systems, Faculty of Engineering
and Science, Aalborg University, Denmark (e-mail: jba@es.aau.dk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173440
0018-926X/$26.00 © 2011 IEEE
tual coupling and correlation among closely spaced antennas
[6], the achievable bandwidth is reduced when compared to
widely spaced antennas [7]. Nevertheless, antenna decoupling
techniques can be used to facilitate a smaller antenna separation
for a given set of performance requirements.
A. Existing Decoupling Techniques
One well-studied technique to decouple closely spaced an-
tennas is to apply the so-called multiport conjugate (MC) match
through introducing a separate impedance matching network
[6]–[15]. The MC match has been successfully demonstrated
for monopoles [8], [10], [11], [13], [14], dipoles [7], [9], patch
antennas [12] and planar inverted F antennas (PIFAs) [15].
Two drawbacks with implementing an additional network to
achieve decoupling are that ohmic losses are expected from the
decoupling network [14] and that the decoupling network can
increase the overall footprint of the multiple antenna system.
Other decoupling techniques, which are specific to antennas on
a common ground plane, include ground plane modifications
[16], [17] and use of neutralization line [18], [19].
More recently, the use of a parasitic element has been
proposed as an attractive alternative to decouple two closely
spaced antennas [20]–[23]. Akin to the MC match, it can
decouple different types of antennas, including dipoles [20],
[23], monopoles [21], [24], PIFAs [20], [24] and ultrawideband
(UWB) antennas [25]. In fact, the use of parasitic elements in an
antenna system is not new. Their previous applications, which
are unrelated to decoupling of multiple antennas, include:
• changing of antenna patterns [26]–[30];
• limiting current flow of antenna on a small ground plane
[31];
• enhancing bandwidth of the antenna structure [32]–[35];
• adding a resonant frequency band [36];
• increasing the reflection phase range of reflectarrays to be-
yond 360 [37].
One common feature in the existing literature on parasitic
decoupling is that the design procedure minimizes the coupling
coefficient in a best effort manner through sweeping the pa-
rameters of the parasitic element. As such, they are unlike the
MC match, which generates perfect decoupling at the desired
frequency for the given self and mutual impedances of the
closely coupled antennas. Another commonality of existing
parasitic decoupling literature, with the exception of [24], is
that the structure of the parasitic element does not resemble that
of the closely coupled antennas. For example, [20] proposes a
H-shape structure and a meander T-shape structure for decou-
pling dipoles and PIFAs, respectively, and a parameter sweep
is employed to design these structures.
LAU AND ANDERSEN: SIMPLE AND EFFICIENT DECOUPLING OF COMPACT ARRAYS WITH PARASITIC SCATTERERS 465

B. Proposed Parasitic Decoupling Technique

In this paper, we propose a simple and efficient parasitic
decoupling technique, which can perfectly decouple two ar-
bitrarily spaced antennas using a reactively loaded parasitic
antenna in between them. It will be shown that our approach
gives similar result as a MC matching network, but in a much
simpler realization while maintaining the overall size of the
antenna system. The proposed design procedure is simple and
rigorous, in that the objective is to tune the dimensions of both Fig. 1. A decoupling setup for a black box containing an arbitrary two-antenna
the active and parasitic antennas in order to satisfy a criterion structure (ports 1 and 3) and a parasitic scatterer (port 2). Port 2 is terminated
derived from antenna impedances. The criterion provides with an impedance load, whereas each of ports 1 and 3 is matched to a 50
feed cable.
perfect decoupling of the active antennas through the use of
a purely reactive load at the parasitic antenna. The reactive
load ensures lossless decoupling in the case of ideal elements.
II. THEORY OF DECOUPLING WITH A PARASITIC SCATTERER
Experimental results also show that the proposed technique
gives significantly better measured efficiency than the MC A. Derivation of Decoupling Procedure
match for two monopoles of 0.1 spacing [13].
Whereas [24] shows the possibility to decouple two active The theory of decoupling two arbitrary active antennas with
antennas by placing a reactively loaded parasitic antenna in a parasitic scatterer can be illustrated with the setup in Fig. 1.
between them, it relies on numerical optimization of only the The “black box” in the setup consists of two active antennas
(ports 1 and 3) and a parasitic scatterer (port 2) that acts as a
reactive load. No explicit information is provided on the un-
shield between the active antennas. The 3-port black box (or
derlying principle and mechanism, apart from the observation
network) of multiple antennas is also intended to highlight
that loading the parasitic antenna with different reactive loads
the fact that the closely coupled antenna and scatterer cannot
changes the gain patterns and coupling between the active an-
be considered as separate structures in general, e.g., they may
tennas. In this paper, we show that tuning the active and par-
share a common ground. The parasitic scatterer is terminated by
asitic antennas by changing their dimensions is necessary for the load impedance and the matching circuits (or matching
achieving perfect decoupling at the center frequency. network) connected to antennas 1 and 3 transform the antenna
The drawback of using any of the aforementioned techniques input impedance to the impedance of the feed cable (typically
for coupling compensation is the narrow bandwidth of the re- 50 ).
sulting antenna system, but this is unavoidable for antenna sys- The self and mutual impedances of the three-port array at the
tems with small antenna spacing [6]. Another consequence of center frequency ( , being the speed of light in
these approaches is a change of radiation pattern, but this should vacuum) are represented by and , respectively, where
only pose a minor problem in a rich scattering environment [38], , , 2, 3. We begin with the voltage and current
and in fact it is angle diversity which facilitates the decorrela- relationship of the setup
tion of the signals. A simpler solution with optimum uncoupled
port matching [39]–[41] is also a possibility, but the efficiency
is reduced compared with decoupling techniques. The use of (1)
parasitic scatterer or reflector to increase isolation of UWB an-
tennas (see [25] and references therein) has also been proposed. or in matrix notation , where and are the voltage
It is expected that a similar approach can be devised to enhance and current across the th antenna port. Moreover, due to reci-
the bandwidth of decoupled narrowband antennas by generating procity, , and .
multiple resonances in the parasitic element. The termination condition for the parasitic scatterer implies
For the purpose of demonstrating the effectiveness of that , which upon substitution into (1) and re-
our parasitic decoupling concept and giving insight into its arrangement gives the voltage and current relationships across
operation, we use electrical dipoles or monopoles as generic the ports of the active antennas
examples in this paper. However, the basic principle will
work for any antenna, since the method only relies on antenna (2)
impedances.
The paper is organized as follows: Section II introduces the
where
theoretical derivation of parasitic decoupling and the design
procedure, which is illustrated using the simple case of two
(3)
closely coupled dipoles. Section III shows the design approach
for monopole antennas in full wave simulations, and the results
(4)
are also verified in an experiment. Insights and practical issues
relating to the technique are discussed in Section IV. Section V
concludes the paper. (5)
466 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

To perfectly decouple the active antennas, we require that

, or equivalently

(6)

Treating the real and imaginary parts of (6) sep-

arately, and setting the load resistance to zero which will ideally
circumvent any ohmic loss in the loaded scatterer

(7)
Fig. 2. A decoupling setup with the dipole 2 acting as a parasitic scatterer
for the active dipoles 1 and 3. The parasitic scatterer is terminated with an
(8) impedance load.

where and .
Based on the above derivation, the design procedure for de-
coupling can be formulated into the following steps:
1) For a given closely coupled two-antenna array, insert a
third antenna between them as the parasitic scatterer.
2) Tune the three antennas so that criterion (7) is satisfied.
3) Calculate the reactance load for the parasitic scatterer
using (8).
4) Calculate the new input impedances of the active antennas
and using (3) and (5), respectively.
5) Calculate the required matching circuits to transform
and to 50 .

B. Illustrative Example of Design Procedure

Since the above derivation is purely based on antenna im-
pedances, the two antennas and the parasitic scatterer can be
arbitrary and need not be of the same type. However, the com- Fig. 3. Load (a) resistance and (b) reactance of the parasitic scatterer for perfect
decoupling versus the length of the dipole antennas.
monly used reference of dipole antennas are used to demonstrate
the decoupling procedure in the following numerical example.
The setup is given in Fig. 2, which is identical to Fig. 1, ex-
In general, the identical input impedance of dipoles 1 and 3
cept that the antennas are now explicitly shown. The center fre-
is not equal to the reference impedance of 50 when the load
quency is 900 MHz and the diameter of the dipoles is 2 mm.
reactance of antenna 2 is set to one of the two values
For simplicity, the dipole lengths are assumed to be identical
. It follows from (3) and (6) that
, such that (i.e., valid
for the thin dipoles used here). In general, allowing for different (9)
lengths will increase the flexibility of the design method. The
method-of-moments (MoM) Matlab scripts from [42] are used which in this case reduces to
to generate the antenna impedances. The spacing between the
two active dipoles is set at . (10)
For this example, the criterion (7) can be achieved by
adjusting the identical length of the dipole antennas . As due to the symmetry .
illustrated in Fig. 3(a), two solutions satisfy this criterion, i.e., The expression (10) implies that at the center frequency, the
and the corresponding load reactances in input impedance of the active antennas decreases correspond-
Fig. 3(b) are . Therefore, the proposed ingly when the spacing is reduced, due to the self and mutual
procedure can in theory achieve perfect and lossless decou- impedances approaching each other. Therefore, if becomes
pling, i.e., the scattering (or S) parameters , small, the required impedance transformation ratio to achieve
by ensuring that both conditions (7) and (8) are fulfilled. The 50 is high. For this example of , the input impedance
reactive load at the parasitic element may be realized by either of each of the two decoupled active port are as low as
lumped (e.g., inductor) or distributed (e.g., open-circuited and , respectively, for the two
transmission line) elements. In this example, lossless inductors solutions with . This complicates the
are used. matching and gives narrowband results. However, it is possible
LAU AND ANDERSEN: SIMPLE AND EFFICIENT DECOUPLING OF COMPACT ARRAYS WITH PARASITIC SCATTERERS
to use a more sophisticated matching network (such as a Cheby-
shev design) to enhance the bandwidth of by more
than a factor of two, if required, using a similar approach as for
matching single antennas [43].
The impedance matching circuit needed to transform the
impedance of each of the two decoupled active antennas (i.e.,
and ) to 50 is realized here with transmission lines
and a single open-circuited stub [44], although lumped ele-
ments [44] may be more attractive for circuit miniaturization,
especially at lower frequencies. Note that similar uncoupled
matching circuits are required for any realization of MC match,
except that in the present case the decoupling function of the
decoupler line [10] or the rat-race hybrid 180 coupler [11] in
the overall MC matching circuit is provided by the parasitic
scatterer.
The scattering parameters of the decoupled active antennas
using either of the two reactance load solutions are shown in
Fig. 4, where and are the scattering parameters of
the active antennas after the decoupling and 50 matching
steps. Lossless inductors are used in the MoM simulation to
provide the required reactance load at the parasitic scatterer.
As expected, perfect decoupling and matching is achieved at
the center frequency for either of the two solutions. However,
the solution with the shorter dipoles gives a more narrowband
behavior in , as can be expected from the higher reactance
load needed. As a reference case, the scattering parameters of
two half-wavelength dipoles that are individually conjugate
matched with their self-impedances (i.e., self impedance match)
are also shown. In this case, the antenna spacing of 0.1 is kept
and no parasitic scatterer is used. Comparing the decoupling
case with the reference case, it is clear that the decoupling
approach gives very good matching performance, albeit for a
relatively small bandwidth.
As another comparison, the scattering parameters for a real-
ization of MC match based on hybrid 180 coupler [11], [13] are
also provided in Fig. 4. Using this realization, the output ports
contain the odd and even modes. The isolation between the odd
467
Fig. 5. Monopole uniform linear array of three elements, with the coordinate
system used in the radiation pattern measurement.
and even modes has a large bandwidth and thus not shown here.
As in the reference case, the antenna spacing is 0.1 and no par-
asitic scatterer is used. It is observed that the odd mode of the
MC match, which has a smaller bandwidth than the even mode,
has been found to yield similar bandwidth performance to the
solution of the parasitic decoupling approach.
III. SIMULATION AND EXPERIMENTAL VERIFICATION
In this section, we present simulation and experimental re-
sults of the proposed decoupling approach at 900 MHz for the
monopole antenna setup shown in Fig. 5. The same monopole
array structure as in [40] is used, except that here we use three
monopoles, instead of only two. Monopole 2 is the parasitic
scatterer, whereas monopoles 1 and 3 are the active elements.
Monopole 3 is located at the center of the ground plane, whereas
monopoles 1 and 2 are separated by 0.1 and 0.05 from
monopole 3 along the negative -axis, respectively (i.e.,
). The ground plane of the monopole has the surface di-
mensions of 330 mm 250 mm, and it is made from FR4 ma-
terial of thickness 1.55 mm, with a thin copper coating on the
underside of thickness 35 . The dielectric constant and loss
tangent of the FR4 material at 900 MHz is 4.4 and 0.02, re-
spectively. The copper-coated FR4 ground plane is used due to
it being relatively lightweight and more rigid than pure copper
ground plane of comparable thickness. The monopole conduc-
tors are made from cylindrical copper wires of 2 mm in diam-
eter. Each of the two matching circuits, which is connected to
the feeding end of the copper conductor, is printed on a PTFE
printed circuit board (PCB) as transmission lines and open-cir-
cuited stubs. The PTFE PCB has a thickness of 0.8 mm and a
copper layer of 35 . At 900 MHz, The PTFE material has a
dielectric constant of 2.53 and a loss tangent of 0.0015.
The design procedure listed in Section II-A is applied to the
simulation model of the monopole array setup. The simulation
results of the monopole setup are obtained using the time-do-
main solver of CST Microwave Studio. For convenience of
tuning, we apply distributed elements for both the reactive load
at the parasitic element and the matching circuits at the active
antennas. In particular, an open-circuited transmission line on
a PCB is used as the reactive load and the matching circuits
consist of transmission lines and single open-circuited stubs.
These circuits are incorporated into the antenna simulation
468 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 6. Simulated and measured scattering parameters for active monopole an-
tennas 1 and 3.

through circuit co-simulation in CST Design Studio. For the

experimental verification, the scattering parameters of the
fabricated monopole array (with the corresponding distributed
decoupling and matching circuits attached) are measured with
a two-port vector network analyzer and the radiation patterns
are measured in a Satimo Stargate-64 measurement facility.
As in the case of dipoles, two reactance load solutions can be
found for perfect decoupling. However, we focus on the solu-
tion giving the larger bandwidth. The scattering parameters of
the decoupled (and matched) active monopoles from simulation
and measurement are given in Fig. 6. As can be seen, the simu-
lation and measurement results are in good agreement with each
other. Due to higher ohmic losses in practice than in simulation,
the bandwidths of the measured cases are slightly larger than the
simulated ones. Practical tuning likewise limits the exact repro-
duction of the isolation parameter.
The simulated and measured radiation patterns of the de-
coupled active elements are shown in Fig. 7. Again, the simu-
lated and measured results are in good agreement. It is noted simulated and measured cases are attributed to the presence of
that the simulated component of the plane in some losses and that it is difficult to practically obtain zero
Fig. 7(e) and (f) is not visible, since it is not within the given correlation.
range of pattern magnitudes. In addition, the simulated patterns The tolerance of efficiency measurement in the Satimo
of the two active elements exhibit non-exact mirror symmetry, facility at 900 MHz is specified to be 0.5 dB. The simulated
and this is because the center of the array is slightly displaced efficiencies of the parasitic decoupled monopoles are close
(i.e., by 0.05 ) from the center of the ground plane. As can be to 100%, whereas the measured efficiencies are about 70%
expected from the linearly polarized monopoles, the com- (-1.5 dB). The discrepancy is primarily the result of imperfect
ponent is dominant in the radiation patterns. Both fabrication of the antenna structure and the matching cir-
and planes reveal that the maximum gains of the two cuits, where the design of the experimental setup emphasizes
patterns point away from each other, towards the array endfires flexibility rather than precise construction (e.g., monopoles
( or 270 ), at a elevated angle of . The di- of different antenna spacing can be easily achieved on the
rectivity of the patterns is about 7.5 dBi, which is significantly same ground plane). As a reference, the measured efficiency
higher than that of a single monopole. This confirms that angle of a single monopole on the same ground plane is about 80%
diversity is strongly utilized in this setup. The simulated and (-1.0 dB), which is within the tolerance range of the decoupled
measured pattern correlation, assuming a 3D uniform angular monopoles’ efficiencies.
power spectrum (APS), is around 0.05 and 0.02, respectively. In contrast, for the same antenna spacing of for
In the ideal case of a lossless setup, perfect decoupling and the two-monopole setup which applies the MC match based
matching in the 3D uniform APS will lead to zero pattern on hybrid 180 coupler, the measured even and odd mode ef-
correlation [7], [45]. The slight discrepancies between the the- ficiencies at the center frequency are 75% and under 30%, re-
oretical zero correlation and the small correlation values in the spectively [13]. These efficiency values are for matching the
LAU AND ANDERSEN: SIMPLE AND EFFICIENT DECOUPLING OF COMPACT ARRAYS WITH PARASITIC SCATTERERS
Fig. 8. Simulated scattering parameters for the UTA, with and without the re-
actively loaded parasitic scatterer. Due to symmetry and
, due to reciprocity.
even and odd mode outputs of the hybrid coupler with transmis-
sion lines and single open-circuited stubs (i.e., the narrowband
matching solution). Recall that similar matching elements are
used to match the parasitic decoupled monopole ports. Com-
paring the achieved measured efficiency with parasitic decou-
pling and MC match, the parasitic decoupling approach is supe-
rior in terms of both average efficiency and balance of branch
power.
IV. FURTHER INSIGHTS AND PRACTICAL ASPECTS
A. Number of Antennas
For future systems, it is most realistic to first consider arrays
with only two elements, though an extension of our proposed
decoupling technique to the use of more elements is possible.
The technique will work for the case of three parallel dipoles
(dipoles 1 to 3) in a uniform triangular array (UTA) arrange-
ment, where one parasitic scatterer (dipole 4) in their centroid
is able to decouple the triangular array for any separation dis-
tance between the active antennas. Applying the same approach
from Section II-A in deriving (6) for decoupling two active an-
tennas, and assuming that the active dipoles are identical, the
corresponding expression for this three-dipole case is given by
. For dipoles with a diameter of 2 mm
and antenna spacing of among the active dipoles,
the scattering parameters as calculated using the MoM scripts
from [42] for two UTA cases (i.e., with and without a parasitic
dipole at the centroid) are illustrated in Fig. 8. As before, the
self-impedance match and half-wavelength dipoles are used for
the reference case without the parasitic dipole. As can be seen,
perfect decoupling is achieved at the center frequency when the
reactively loaded parasitic dipole is applied, whereas the no-par-
asitic case has a high coupling coefficient of between
a given antenna and each of its two adjacent antennas. Nonethe-
less, as in the case of two-element arrays, decreasing the sepa-
ration distance will result in a smaller bandwidth for the decou-
pled antennas. The reason that the decoupling technique applies
directly to the UTA case is that the symmetry of the array struc-
ture ensures that the coupling between any (active) antenna pair
is equal.
469
For three-element arrays of non-triangular arrangements, the
inherent asymmetry in the array structure introduces different
levels of coupling between different pairs of antennas, which
complicates the design of parasitic scatterer(s) for perfect de-
coupling. However, the performance of multiple antenna sys-
tems is usually limited by pair(s) of antennas with the smallest
antenna separation distance, such as for the case of uniform
linear arrays of three or more elements. Therefore, decoupling
the antenna pairs with the most severe coupling level can pro-
vide an approximate solution.
B. Application in Compact Terminals
One important application of decoupling techniques is
in achieving good performance for mobile terminals with
closely spaced antennas [6],[16]–[19]. Limited available space,
multiple-band operation and the need for co-existence with
other device components complicate the decoupling task
significantly. Preliminary simulation results confirm that the
proposed parasitic decoupling technique can perfectly decouple
dual-PIFA and dual-monopole antennas at 900 MHz for a
40 mm 100 mm ground plane, where the two active and one
parasitic antennas are placed at the two short edges and the
center of the ground plane, respectively. The parasitic antenna
is a PIFA in both cases.
C. MIMO Performance
It is known that a lossless decoupled and well matched
receive array is optimum not only from the viewpoint of
maximum power transfer from the antennas to the loads [9], it
also facilitates zero correlation in the uniform 3D APS [7], [45].
In fact, the decoupled array is likewise superior in received
power and correlation performance to coupled array in other
propagation environments [41]. Since MIMO performance
measured in terms of either capacity or diversity gain is a
function of correlation, branch power imbalance and available
power, arrays which are decoupled by any (lossless) method
will in general result in a better MIMO performance as well
(see e.g., [6], [9],[46]).
As an example, we consider the MIMO capacity for the loss-
less dipoles in Section II at the center frequency. For a
MIMO channel , the instantaneous channel capacity with
equal transmit power allocation can be expressed as [47]
(11)
where is the reference SNR and is the identity
matrix. Since the interest here is in antenna design, the refer-
ence propagation environment of independent and identically
distributed (IID) Rayleigh fading channel is assumed, i.e.,
the entries of are zero mean circularly symmetric complex
Gaussian random variables. Without loss of generality, the case
of receive antennas is examined. Then, the MIMO channel is
given by
(12)
where is the receive correlation matrix, which fully represents
the effects of the antennas on the channel, i.e., it characterizes
470
Fig. 9. Contour plot of the simulated total electric field (in dB) 5 mm above
the ground plane when active monopole antenna 1 is removed. The center of
the ground plane is at the origin and the coordinate system is shown in Fig. 5.
The field is shown for only the center region of the ground plane.
the efficiency, efficiency imbalance and correlation among the
receive antennas. In particular,
where is a normalized correlation matrix whose diagonal ele-
ments are 1 and the th element denotes the
complex correlation coefficient between the 3D radiation pat-
terns of the th and th antenna ports. denotes a diagonal ma-
trix given by
where is the total efficiency of the th antenna.
The correlation between the active antennas that has been de-
coupled with parasitic scatterer and matched to 50 is zero and
the total efficiency of each antenna is 100%. This means that
and the MIMO capacity is the same as that of the IID
Rayleigh channel. For and 10,000 Monte Carlo re-
alizations of the , the ergodic capacity is 11.3 bits per
second per Hertz (bits/s/Hz). In comparison, the correlation and
total efficiency of the reference case with self-impedance match
are 0.55 and 58% (-2.4 dB), respectively. This translates to an
ergodic capacity of 9.4 bits/s/Hz. Thus, the proposed decoupling
procedure gives an overall gain in capacity of 2 bits/s/Hz.
On the other hand, as also pointed out in Section II, the band-
width of the decoupled array can be significantly smaller than
that of a widely spaced array, depending on the antenna spacing
[7]. This implies that at a very small antenna separation, the ben-
efit of decoupling will be small, if the operating bandwidth sig-
nificantly exceeds the achieved antenna bandwidth.
D. Shielded Zone
As can be seen in the pattern plots in Fig. 7, the shielding
effect of the parasitic antenna ensures that the radiation of the
active antennas is directed away from each other. However, the
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
shielding effect is not only limited to far-field beamforming be-
tween the excited antenna and the parasitic element. Indeed,
since decoupling is a near-field phenomenon, one can expect
that there is a quiet zone within the shielded region, as can be
seen in the simulated total electric field distribution along the
center region of the large ground plane in Fig. 9. In Fig. 9, ac-
tive monopole antenna 1 is removed and the total electric field
is taken at a height of 5 mm above the ground plane. As can be
observed, the field at the former location of antenna 1 is over 10
dB lower than the value in the immediate vicinity of the excited
antenna 3.
Moreover, due to the decoupling phenomenon, removing
one of the two active antennas will only marginally affect the
impedance and radiation characteristics of the other active
antenna. This is confirmed in both simulations and measure-
ments, i.e., the remaining active antenna gives similar reflection
coefficient and radiation pattern as those shown in Figs. 6 and
7, respectively.
V. CONCLUSIONS
This paper takes up the task of decoupling closely coupled
antennas with parasitic scatterers. The main intention is to
provide the theoretical insights into the approach, which can
be applied to two arbitrary coupled antennas for an arbitrary
spacing. Example applications on reference antenna arrays of
(13) closely spaced dipoles or monopoles illustrate the procedure
and its effectiveness. Preliminary results confirm that the ap-
proach extends readily into more practical antenna elements,
such as those used in mobile terminals. However, the ability of
the parasitic scatterer approach to support multi-band operation
and its robustness to user effects are interesting subjects for
future studies.
(14)
ACKNOWLEDGMENT
The authors would like to thank L. Hedenstjärna of the De-
partment of Electrical and Information Technology, Lund Uni-
versity, for fabricating the antennas and matching circuits.
REFERENCES
[1] B. K. Lau and J. B. Andersen, “Unleashing multiple antenna systems
in compact terminal devices,” presented at the Int. Workshop Antenna
Technol. (IWAT2009), Santa Monica, CA, Mar. 2–4, 2009.
[2] B. K. Lau and J. B. Andersen, “Antenna System and Method of
Providing an Antenna System,” Swedish Patent Application (No.
0702307-0), Oct. 2007.
[3] H. Krim and M. Viberg, “Two decades of array signal processing re-
search: The parametric approach,” IEEE Signal Process. Mag., vol. 13,
no. 4, pp. 67–94, Jul. 1996.
[4] S. Andersson, B. Hagerman, H. Dam, U. Forssén, J. Karlsson, F. Kro-
nestedt, S. Mazur, and K. J. Molnar, “Adaptive antennas for GSM and
TDMA systems,” IEEE Personal Commun. Mag., vol. 6, no. 3, pp.
74–86, Jun. 1999.
[5] E. Dahlman, S. Parkvall, and J. Sköld, 3G Evolution: HSPA and LTE
for Mobile Broadband. London: Academic Press, 2008, pp. 267–298.
[6] B. K. Lau, “Multiple antenna terminals,” in MIMO: From Theory to Im-
plementation, C. Oestges, A. Sibille, and A. Zanella, Eds. San Diego:
Academic Press, 2011.
[7] B. K. Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch, “Impact
of matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag., vol. 54, no. 11, pp. 3225–3238, Nov. 2006.
[8] J. B. Andersen and H. H. Rasmussen, “Decoupling and descattering
networks for antennas,” IEEE Trans. Antennas Propag., vol. AP-24,
no. 6, pp. 841–846, Nov. 1976.
LAU AND ANDERSEN: SIMPLE AND EFFICIENT DECOUPLING OF COMPACT ARRAYS WITH PARASITIC SCATTERERS
[9] J. W. Wallace and M. A. Jensen, “Mutual coupling in MIMO wireless
systems: A rigorous network theory analysis,” IEEE Trans. Wireless
Commun., vol. 3, no. 4, pp. 1317–1325, Jul. 2004.
[10] S. Dossche, S. Blanch, and J. Romeu, “Optimum antenna matching to
minimise signal correlation on a two-port antenna diversity system,”
Elect. Lett., vol. 40, no. 19, pp. 1164–1165, Sep. 2004.
[11] S. Dossche, S. Blanch, and J. Romeu, “Decorrelation of a closely
spaced four element antenna array,” in Proc. IEEE Antenna Propag.
Soc. Int. Symp., Washington, DC, Jul. 3–8, 2005, vol. 1B, pp. 803–806.
[12] S. Dossche, J. Rodriguez, L. Jofre, S. Blanch, and J. Romeu, “Decou-
pling of a two-element switched dual band patch antenna for optimum
MIMO capacity,” in Proc. IEEE Antenna Propag. Soc. Int. Symp., Al-
buquerque, NM, Jul. 2006, pp. 325–328.
[13] C. Volmer, M. Sengul, J. Weber, R. Stephan, and M. A. Hein, “Broad-
band decoupling and matching of a superdirective two-port antenna
array,” IEEE Antennas Wireless Propag. Lett., vol. 7, pp. 613–616,
2008.
[14] C. Volmer, J. Weber, R. Stephan, K. Blau, and M. A. Hein, “An eigen-
analysis of compact antenna arrays and its application to port decou-
pling,” IEEE Trans. Antennas Propag., vol. 56, no. 2, pp. 360–370,
Feb. 2008.
[15] C. Volmer, J. Weber, R. Stephan, and M. A. Hein, “Mutual coupling
in multi-antenna systems: Figures-of-merit and practical verification,”
presented at the Eur. Conf. Antennas Propag. (EUCAP2006), Berlin,
Germany, Mar. 23–27, 2009.
[16] C. Y. Chiu, C. H. Cheng, R. D. Murch, and C. R. Rowell, “Reduction
of mutual coupling between closely-packed antenna elements,” IEEE
Trans. Antennas Propag., vol. 55, no. 6, pp. 1732–1738, Jun. 2007.
[17] Y. Gao, X. Chen, Z. Ying, and C. Parini, “Design and performance
investigation of a dual-element PIFA array at 2.5 GHz for MIMO ter-
minal,” IEEE Trans. Antennas Propag., vol. 55, no. 12, pp. 3433–3441,
Dec. 2007.
[18] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas,
“Enhanced two-antenna structures for universal mobile telecommu-
nications system diversity terminals,” IET Proc. Microw. Antennas
Propag., vol. 2, no. 1, pp. 93–101, Feb. 2008.
[19] C. Diallo, A. Luxey, R. Le Thuc, P. Staraj, and G. Kossiavas, “Diver-
sity performance of multiantenna systems for UMTS cellular phones in
different propagation environments,” Int. J. Antennas Propag., 2008.
[20] A. C. K. Mak, C. R. Rowell, and R. D. Murch, “Isolation enhancement
between two closely packed antennas,” IEEE Trans. Antennas Propag.,
vol. 56, no. 11, pp. 3411–3419, Nov. 2008.
[21] P. J. Ferrer, J. M. Gonzalez-Arbesu, and J. Romeu, “Decorrelation of
two closely spaced antennas with a metamaterial AMC surface,” Mi-
crow. Opt. Technol. Lett., vol. 50, no. 5, pp. 1414–1417, May 2008.
[22] S. Hong, K. Chung, J. Lee, S. Jung, S. Lee, and J. Choi, “Design of
a diversity antenna with stubs for UWB applications,” Microw. Opt.
Technol. Lett., vol. 50, no. 5, pp. 1352–1356, May 2008.
[23] A. Abe, N. Michishita, Y. Yamada, J. Muramatsu, T. Watanabe, and
K. Sato, “Mutual coupling reduction between two dipole antennas
with parasitic elements composed of composite right-/left-handed
transmission lines,” presented at the Int. Workshop Antenna Technol.
(IWAT2009), Santa Monica, CA, Mar. 2009.
[24] T. Michalski, V. Wienstroer, and R. Kronberger, “Beam forming ca-
pabilities of smart antennas on mobile terminal,” presented at the Eur.
Conf. Antennas Propag. (EUCAP2009), Berlin, Germany, Mar. 23–27,
2009.
[25] S. Zhang, Z. Ying, J. Xiong, and S. He, “Ultrawideband MIMO/di-
versity antennas with a tree-like structure to enhance wideband isola-
tion,” IEEE Antennas Wireless Propag. Lett., vol. 8, pp. 1279–1282,
2009.
[26] R. F. Harrington, “Reactively controlled directive arrays,” IEEE Trans.
Antennas Propag., vol. 26, no. 3, pp. 390–395, May 1978.
[27] R. M. T. Milne, “A small adaptive array antenna for mobile communi-
cations,” in Proc. IEEE Antennas Propag. Soc. Int. Symp., Vancouver,
Canada, Jun. 17–21, 1985, pp. 797–800.
[28] N. L. Scott, M. O. Leonard-Taylor, and R. G. Vaughan, “Diversity gain
from a single-port adaptive antenna using switched parasitic elements
illustrated with a wire and monopole prototype,” IEEE Trans. Antennas
Propag., vol. 47, no. 6, pp. 1066–1070, Jun. 1999.
[29] R. Schlub, D. V. Thiel, J. W. Lu, and S. G. O’Keefe, “Dual-band six-
element switched parasitic array for smart antenna cellular communi-
cations systems,” Elect. Lett., vol. 36, no. 16, pp. 1342–1343, Aug.
2000.
[30] P. Jarmuszewski, Y. Qi, and A. D. Stevenson, “Antenna With Near-
Field Radiation Control,” Canadian patent No. 2, 414, 124 A1, Sep.
12, 2004.
471
[31] K. Sato and T. Amano, “Improvements of impedance and radiation per-
formances with a parasitic element for mobile phone,” in Proc. IEEE
Antennas Propag. Soc. Int. Symp., San Diego, CA, Jul. 5–11, 2008.
[32] H. Nakano, R. Suzuki, and J. Yamauchi, “Low-profile inverted-f an-
tenna with parasitic elements on an infinite ground plane,” IEE Proc.
Microw. Antennas Propag., vol. 145, no. 4, pp. 321–325, Aug. 1998.
[33] T. H. Tsai, H. T. Peng, and K. Shih, “Built-in Multi-Band Mobile Phone
Antenna Assembly With Coplanar Patch Antenna and Loop Antenna,”
U.S. Patent Application Pub. No. 2004/0100410 A1, May 27, 2004.
[34] G. Johnson and B. Newman, “Single or Dual Band Parasitic Antenna
Assembly,” U.S. Patent No. 6, 456, 249 B1, Sep. 24, 2002.
[35] K. Q. da Costa, V. A. Dmitriev, and M. N. Kawakatsu, “Enlarging the
impedance matching bandwidth of wire and planar antennas using loop
parasitic elements,” presented at the Int. Workshop Antenna Technol.
(IWAT2009), Santa Monica, CA, Mar. 2009.
[36] J. C. Posluszny and R. C. Posluszny, “Parasitically Coupled
Folded Dipole Multi-Band Antenna,” U.S. Application Pub. No.
2006/0061515 A1, Mar. 23, 2006.
[37] L. Li, Q. Chen, Q. Yuan, K. Sawaya, T. Maruyama, T. Furuno, and
S. Uebayashi, “Novel broadband planar reflectarray with parasitic
dipoles for wireless communication applications,” IEEE Antennas
Wireless Propag. Lett., vol. 8, pp. 881–885, 2009.
[38] J. B. Andersen, J. O. Nielsen, G. F. Pedersen, G. Bauch, and M. Herdin,
“Room electromagnetics,” IEEE Antennas Propag. Mag., vol. 49, no.
2, pp. 27–33, Apr. 2007.
[39] J. B. Andersen and B. K. Lau, “On closely coupled dipoles in a random
field,” IEEE Antennas Wireless Propag. Lett., vol. 5, pp. 73–75, 2006.
[40] Y. Fei, Y. Fan, B. K. Lau, and J. S. Thompson, “Optimal single-port
matching impedance for capacity maximization in compact MIMO ar-
rays,” IEEE Trans. Antennas Propag., vol. 56, no. 11, pp. 3566–3575,
Nov. 2008.
[41] M. A. Jensen and B. K. Lau, “Uncoupled matching for active and pas-
sive impedances of coupled arrays in MIMO systems,” IEEE Trans.
Antennas Propag., vol. 58, no. 10, pp. 3336–3343, Oct. 2010.
[42] S. M. Makarov, Antenna and EM Modeling With MATLAB. New
York: John Wiley and Sons, 2002.
[43] R. Vaughan and J. A. Bach, Channels, Propagation And Antennas for
Mobile Communications. London: The IEE, 2003, pp. 511–511.
[44] D. M. Pozar, Microwave Engineering, 3rd ed. Hoboken, NJ: Wiley,
2005.
[45] S. Blanch, J. Romeu, and I. Corbella, “Exact representation of antenna
system diversity performance from input parameter description,” Elect.
Lett., vol. 39, no. 9, pp. 705–707, May 2003.
[46] J. W. Wallace and M. A. Jensen, “Termination-dependent diversity per-
formance of coupled antennas: Network theory analysis,” IEEE Trans.
Antennas Propag., vol. 52, no. 1, pp. 98–105, Jan. 2004.
[47] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
Communications. Cambridge: Cambridge University Press, 2003.
Buon Kiong Lau (S’00–M’03–SM’07) received the
B.E. degree (with honors) from the University of
Western Australia, Perth, Australia and the Ph.D.
degree from Curtin University of Technology, Perth,
in 1998 and 2003, respectively, both in electrical
engineering.
During 2000 to 2001, he worked as a Research En-
gineer with Ericsson Research, Kista, Sweden. From
2003 to 2004, he was a Guest Research Fellow at
the Department of Signal Processing, Blekinge Insti-
tute of Technology, Sweden. Since 2004, he has been
at the Department of Electrical and Information Technology, Lund University,
where he is now an Associate Professor. He has been a Visiting Researcher at
the Department of Applied Mathematics, Hong Kong Polytechnic University,
China, Laboratory for Information and Decision Systems, Massachusetts In-
stitute of Technology, and Takada Laboratory, Tokyo Institute of Technology,
Japan. His primary research interests are in various aspects of multiple antenna
systems, particularly the interplay between antennas, propagation channels and
signal processing.
Dr. Lau is an Associate Editor for the IEEE TRANSACTIONS ON ANTENNAS
AND PROPAGATION and a Guest Editor of the 2012 Special Issue on MIMO
Technology for the same journal. From 2007 to 2010, he was a Co-Chair of
Subworking Group 2.2 on “Compact Antenna Systems for Terminals” (CAST)
within EU COST Action 2100. Since 2011, he is a Swedish national delegate
and the Chair of Subworking Group 1.1 on “Antenna System Aspects” within
COST IC1004.
472 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Jørgen Bach Andersen (M’68–SM’78–F’92–
LF’02) received the M.Sc. and Dr.Techn. degrees
from the Technical University of Denmark (DTU),
Lyngby, Denmark, in 1961 and 1971, respectively.
In 2003 he was awarded an honorary degree from
Lund University, Sweden.
From 1961 to 1973, he was with the Electromag-
netics Institute, DTU and since 1973 he has been
with Aalborg University, Aalborg, Denmark, where
he is now a Professor Emeritus and Consultant. He
was head of a research center, Center for Personal
Communications, CPK, from 1993–2003. He has been a Visiting Professor
in Tucson, Arizona, Christchurch, New Zealand, Vienna, Austria, and Lund,
Sweden. He has published widely on antennas, radio wave propagation, and
communications, and has also worked on biological effects of electromagnetic
systems. He has coauthored a book, Channels, Propagation and Antennas
for Mobile Communications (IEE, 2003). He was on the management com-
mittee for COST 231 and 259, a collaborative European program on mobile
communications.
Professor Andersen is a former Vice President of the International Union
of Radio Science (URSI) from which he was awarded the John Howard
Dellinger Gold Medal in 2005. He is Associate Editor of Antennas and Wireless
Propagation Letters and Co-Editor the IEEE TRANSACTIONS ON ANTENNAS
AND PROPAGATION Special Issue on Multiple-Input Multiple-Output (MIMO)
Technology.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Reducing Mutual Coupling of MIMO Antennas With
Parasitic Elements for Mobile Terminals
Zhengyi Li, Zhengwei Du, Masaharu Takahashi, Senior Member, IEEE, Kazuyuki Saito, Member, IEEE, and
Koichi Ito, Fellow, IEEE
Abstract—Mutual coupling is a critical problem in the design of
MIMO antennas because it deteriorates the performance of MIMO
systems, which not only affects the antenna efficiency but also in-
fluences the correlation. Therefore, in this paper, using parasitic
elements to reduce mutual coupling is studied. By adding parasitic
elements a double-coupling path is introduced and it can create
a reverse coupling to reduce mutual coupling. As an example, a
dual-slot-element antenna with parasitic monopoles for mobile ter-
minals is described. The discussion on channel capacity shows that
the antenna can be considered as a good candidate for MIMO sys-
tems. Furthermore, based on the study of current distributions, it
is concluded that the technique is sensitive to relative positions be-
tween parasitic elements, and relative positions between active el-
ement and parasitic element. Finally, we also extend the technique
to a tri-element antenna.
Index Terms—Channel capacity, current distribution, MIMO
antennas, mutual coupling, parasitic elements.
I. INTRODUCTION
M ULTIPLE-INPUT MULTIPLE-OUTPUT (MIMO)
technology has been proposed for several years, which
significantly improves the performance of wireless commu-
nication systems [1]–[3]. In these systems, antennas play an
important role, since antenna’s features are inherently included
in the communication channel between the transmitter and the
receiver. Especially, mutual coupling between antenna ele-
ments not only affects the antenna efficiency but also influences
the correlation. At the base station, low mutual coupling is
easy to be realized where element separations are always many
wavelengths. However, for mobile terminals, acquiring low
mutual coupling will be difficult owing to limited volume.
At present, several antenna designs have been proposed to
reduce mutual coupling for mobile terminals. In [4]–[6], slot
technique was presented to achieve low mutual coupling. The
slot technique is explained as a slow wave structure, which
Manuscript received May 31, 2010; revised May 27, 2011, September 16,
2011; accepted September 23, 2011. Date of publication October 25, 2011; date
of current version February 03, 2012. This work was supported in part by the
National Basic Research Program of China under Grant 2009CB320205, in part
by the National Natural Science Foundation of China under Grant 60971005,
and in part by the Tsinghua-QUALCOMM Associated Research Plan.
Z. Li, M. Takahashi, and K. Saito are with the Research Center for Frontier
Medical Engineering, Chiba University, Chiba 263-8522, Japan (e-mail: zhy.li.
2008@gmail.com; omei@faculty.chiba-u.jp; kazuyuki_saito@faculty.chiba-u.
jp).
Z. Du is with the State Key Laboratory on Microwave and Digital Communi-
cations, Tsinghua National Laboratory for Information Science and Technology,
Department of Electronic Engineering, Tsinghua University, Beijing 100084,
China (e-mail: zwdu@mail.tsinghua.edu.cn).
K. Ito is with the Graduate School of Engineering, Chiba University, Chiba
263-8522, Japan (e-mail: ito.koichi@faculty.chiba-u.jp).
Digital Object Identifier 10.1109/TAP.2011.2173432
0018-926X/$26.00 © 2011 IEEE
473
can decrease the wavelength of the signal and thus increase the
separation between antenna elements [4]. Besides, a protruding
T-shaped stub in the ground plane was used to improve the
mutual coupling between antenna elements [7], [8]. Similarly,
a T-shaped and dual-inverted-L-shaped ground branch was
added to acquire low mutual coupling [9]–[11]. Then, in [12],
the ground branch technique was analyzed, and by adding a
ground branch, a dual-element Inverted-F antenna with low
mutual coupling was proposed. That study demonstrated that
the technique creates an additional coupling path to cancel out
the original coupling. In addition, neutralization technique was
proposed, which is adding a neutralization line between the
feeding strips or the shorting strips of the PIFAs [13]–[15]. As
the similar neutralizing principle, lumped circuits or neutraliza-
tion lines were applied between planar monopoles [16], [17].
Considering all these studies, most of them have a common
idea, which is adopting some structure to create reverse cou-
pling to reduce mutual coupling.
According to this common idea, we propose using parasitic
elements to reduce mutual coupling. Parasitic elements create
reverse coupling to reduce mutual coupling. The principle of
this technique is introduced in Section II. In Section III, as
an implementation, a dual-slot-element antenna with parasitic
monopoles is described. Compared to the dual-element In-
verted-F antenna in [12] operating at 2.4 GHz WLAN band
(2400–2480 MHz), the dual-slot-element antenna is designed
for the UMTS band (1920–2170 MHz), which will face the
problems of working at lower frequency and wider bandwidth.
By using a set of parasitic elements (two parasitic monopoles),
its bandwidth can cover the UMTS band with low mutual
coupling ( dB). Since its diversity performance has
been presented in our previous work [18], in this paper we
mainly focus on its MIMO performance by studying channel
capacity. Then, in Section IV, in order to deeply understand the
technique, the current distributions are validated and studied.
Furthermore, we extend the technique to a tri-element antenna
in Section V. Finally, in Section VI, some important conclu-
sions are drawn.
II. PRINCIPLE OF REDUCING MUTUAL COUPLING
A simplified model of reducing mutual coupling is illustrated
in Fig. 1. At first, there are two close antenna elements in a mo-
bile terminal as shown in Fig. 1(a). Element 1 is excited by a
current of . In the meanwhile, owing to small separation, a
current of is coupled on element 2, in which represents
coupling coefficient. Then, as given in Fig. 1(b), we add some
parasitic elements between the two antenna elements. As a re-
sult, there are two coupling paths: one is the original coupling
path ; the other is the double-coupling path, in which the
474 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 2. Geometry of the dual-slot-element antenna with dimensions in mm (top

Fig. 1. Simplified model of reducing mutual coupling: (a) two close antenna layer in black color and bottom layer in grey color).
elements; (b) two close antenna elements with parasitic elements.

current is firstly coupled from element 1 to parasitic elements

and secondly coupled from parasitic elements to element 2. It
should be noted that the number of parasitic elements is not
specified. Therefore, the coupling currents on parasitic elements
are given by

(1)

where N is the number of parasitic elements, and

are the corresponding coupling coefficients. Considering the
overall performance, an average coupling coefficient is used
to represent the coupling coefficients . Thus, the
average equivalent coupling current is expressed by . Be-
cause of the symmetric structure, a current of is coupled
on element 2 by the double-coupling path. Fig. 3. Simulated scattering parameters (“PM” means parasitic monopoles).
If we change the two coupling coefficients ( and ) by prop-
erly designing the antenna configuration, the mutual coupling
may be close to zero, which is expressed by line with one via-grounded end. On the bottom layer of the sub-
strate, the ground plane (75 mm 60 mm) is printed. In order
(2)
to reduce mutual coupling, two parasitic monopoles are added.
The metal strips of the parasitic monopoles were chosen to be
It means that parasitic elements can create reverse coupling to
of the same width of 1.5 mm as the feeding lines to simplify
reduce mutual coupling.
the design. For each parasitic monopole, one rectangular por-
tion is added at the corner to adjust matching. The length of
III. DUAL-SLOT-ELEMENT ANTENNA each slot element is 25.3 mm and the total length of each para-
sitic monopole from the point “O” to “A” is about 26 mm, both
Based on the principle of reducing mutual coupling in Sec-
implying a quarter-wavelength mode.
tion II, a dual-slot-element antenna with parasitic monopoles is
The simulated scattering parameters are plotted in Fig. 3, and
presented in this section.
as a comparison, the simulated performance without the two
parasitic monopoles is added in Fig. 3, both of which are sim-
A. Antenna Configuration and Scattering Parameters
ulated by High Frequency Structure Simulator (HFSS) [19]. It
The antenna is printed on a FR4 substrate board (95 mm 60 is found that by adding the two parasitic monopoles the mutual
mm), which has a thickness of 0.8 mm and a relative permittivity coupling is greatly improved, in which decreases from
of 4.4. As shown in Fig. 2, the antenna includes two symmetric dB to dB. Meanwhile, the operation band shifts a little to
slot elements, and each slot element is fed by a 50- microstrip lower frequency and the bandwidth increases somewhat.
LI et al.: REDUCING MUTUAL COUPLING OF MIMO ANTENNAS WITH PARASITIC ELEMENTS FOR MOBILE TERMINALS 475

Fig. 5. Measured radiation patterns in plane.

Fig. 4. Measured scattering parameters.

where and represent the field patterns of antenna

Based on the simulation, a prototype antenna is fabricated. element and , and represents incoming wave, de-
Its measured scattering parameters are shown in Fig. 4. Due to notes the conjugate operation, and represents expectation
the symmetric structure, the measured is almost the same as [21].
, and is thus not plotted in the figure for brevity. From Fig. 4, The mobile wireless environment defined in [22] has a series
the two slot elements are well matched in the whole UMTS band of reasonable assumptions: the fading envelope being Rayleigh
(1920–2170 MHz) where the dB criterion is distributed, the incoming wave arriving in horizontal plane only,
satisfied. In the meanwhile, across the band low mutual coupling the incoming wave’s orthogonal polarizations being uncorre-
( dB) is acquired. lated, the individual polarizations being spatially uncorrelated,
and finally the time-averaged power density per steradian being
B. Channel Capacity constant. Based on these approximations and the derivation in
In MIMO systems, usually, channel capacity is adopted to Appendix A, can be written as
evaluate system performance. When the transmitter does not
know the channel conditions, the power is equally divided to
each transmit antenna element and the channel capacity is given
by (6)

(3) where is the cross-polarization discrimination (XPD) of the

incoming wave. In this paper, is assumed to be 0 dB, which
where is the number of receive antenna elements, is is the average value in an urban fading environment [22], [23].
the number of transmit antenna elements, is the Therefore, we can calculate the channel capacity and its cumula-
identity matrix, is the average received signal-to-noise ratio tive distribution function (CDF) by using the measured radiation
(SNR), is the normalized channel ma- patterns in plane, which are shown in Fig. 5. These radia-
trix, and denotes the conjugate transpose [2]. tion patterns were measured at 2.1 GHz in an anechoic chamber
In fact, the entries of are correlated because of propaga- with one port excited and the other terminated to a 50- load.
tion environment and antenna elements. Therefore, one popular As in Table I, four kinds of system parameters are studied,
simplified model, the “Kronecker” model, is used to calculate “2 2 MIMO a”, “2 2 MIMO b”, “2 2 MIMO c”, and SISO
, which separates the correlation into two independent com- (single-input single-output). In “2 2 MIMO a”, is uncor-
ponents [20], [21] related, and thus, and are equal to identity matrix. In
“2 2 MIMO b”, the dual-slot-element antenna is equipped
(4) both in the transmitter and receiver so is equal to . In
“2 2 MIMO c”, the dual-slot-element antenna is also equipped
where is the receive correlation matrix, is the transmit
both in the transmitter and receiver but the correlated matrix
correlation matrix, and is a random matrix with independently
is calculated only with the magnitude of radiation patterns (as-
identical distributed complex Gaussian entries, and de-
suming a constant phase over all angles for both antenna ele-
notes the matrix square root. The th entry of is calcu-
ments).
lated by
Fig. 6 shows the CDFs of channel capacity in the four
conditions. Based on the results, firstly, it is found that the
MIMO performance with the dual-slot-element antenna is
promising since the CDF curve is very close to the uncor-
(5) related one. Secondly, relatively large distance between the
476
Fig. 6. CDFs of channel capacity in the four conditions (“2 2 MIMO a”, “2
TABLE I
FOUR KINDS OF SYSTEM PARAMETERS
CDF curves of “2 2 MIMO b” and “2 2 MIMO c” means
that the decorrelation of phases contributes more to system
performance compared to that of magnitude, which is similar
to the conclusion in [24]. Thirdly, even if we remove the effect
of phases and only adopt the magnitude of radiation patterns,
the system performance is still much better than that of SISO.
Besides, we also evaluated the antenna in Bluetest Reverbera-
tion chamber [25]. In the chamber, three perpendicular antenna
elements are equipped in the transmitter to generate isotropic
environment, and the dual-slot-element antenna is equipped in
the receiver. Therefore, a 3 2 MIMO system was setup. Fig. 7
shows the average channel capacities (marked as “3 2 MIMO
b”), in which 2100 channel realizations were measured (100
stirrer positions 21 frequency points centered at 2.1 GHz).
As a comparison, two other systems’ channel capacities are
also plotted: “3 2 MIMO a” ( is uncorrelated) and “3 1
MISO”. It is found that the channel capacity of “3 2 MIMO
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
2 MIMO b”, “2 2 MIMO c”, and SISO).
Fig. 7. Channel capacities in Bluetest Reverberation chamber (“3 2 MIMO
a”, “3 2 MIMO b”, and “3 1 MISO”).
b” is much higher than that of “3 1 MISO” though it is a little
lower than that of uncorrelated channel (“3 2 MIMO a”). In
addition, the average radiation efficiency was also measured in
Bluetest Reverberation chamber, which is 80% and 77% for el-
ement 1 and element 2, respectively.
According to the calculated channel capacity by using radi-
ation patterns and measured channel capacity in Bluetest Re-
verberation chamber, our antenna can be considered as a good
candidate for MIMO systems.
IV. CURRENT DISTRIBUTION VALIDATION AND STUDY
In Sections II and III, we adopted parasitic elements to reduce
mutual coupling and implemented the technique in a dual-slot-
element antenna. Therefore, in order to further understand the
technique, the current distributions are validated and studied.
A. Current Magnitude Distributions
The performance of reducing mutual coupling can also be
validated with surface current magnitude distributions. Fig. 8(a)
LI et al.: REDUCING MUTUAL COUPLING OF MIMO ANTENNAS WITH PARASITIC ELEMENTS FOR MOBILE TERMINALS
Fig. 8. The current magnitude distributions on the dual-slot-element an-
tenna (a) with the two parasitic monopoles and (b) without the two parasitic
monopoles (light color means large current density).
Fig. 9. The current vector distributions on the dual-slot-element antenna
without the two parasitic monopoles.
shows the current magnitude distributions at 2.1 GHz, in which
slot element 1 (the left slot element) is excited and slot element
2 (the right slot element) is terminated to a 50- load. As a
comparison, the current magnitude distributions without the two
parasitic monopoles are plotted in Fig. 8(b). From the figure, it
can be observed that the current distributions on the slot element
2 decrease dramatically by adding the two parasitic monopoles.
B. Current Vector Distributions
Firstly, the surface current vector distributions on the dual-
slot-element antenna without the two parasitic monopoles at 2.1
GHz are shown in Fig. 9, in which slot element 1 is excited and
slot element 2 is terminated to a 50- load. From the current
distributions, it is found that the coupling current on slot el-
ement 2 is out-of-phase with on slot element 1 (the current on
the right edge of slot element 1 is out-of-phase with the current
on the right edge of slot element 2). However, the currents on
two antenna elements will not always satisfy the out-of-phase
condition, when we change the distance between the two an-
tenna elements.
As a simplified example, the coupling coefficient of two half-
wavelength dipoles is examined. It is found that the currents on
the two dipoles are in-phase in some separations while out-of-
phase in other separations, which is illustrated in Fig. 10. In fact,
477
as in Fig. 11, based on the definition of scattering parameters,
the source current and coupling current can be expressed
by
(7)
(8)
where represents characteristic impedance and is equal to
50- in general. Thus, the coupling coefficient is given by
(9)
If we check the current distribution on dipole 1, it is the sum of
the source current and the reflected current, which is
(10)
Within antenna’s operation band, is relatively small, so
(7) is close to (10) and a approximate value of (the coupling
coefficient between current distributions on dipole 2 and dipole
1) is given by
(11)
The calculated magnitude and phase of coupling coefficient
by using (9) and (11) are plotted in Fig. 12. When the distance
between the two dipoles increases (within half-wavelength) the
magnitude of coupling coefficient would decrease sharply to
below 0.2. Beyond half-wavelength, the magnitude changes
very little. For the phase, the change is near linear, including
the in-phase (zero degree) and out-of-phase (180 degree and
degree). It should be noted that, the coupling coefficient
depends on antenna type, radiation patterns, and polariza-
tion, though the trend of it is similar. For example, for the
dual-slot-element antenna without parasitic monopoles in
Fig. 9, by using (11), the magnitude and phase of are 0.41
and 43 degree, respectively. They are so different from the
results of two half-wavelength dipoles in Fig. 12. With the
same magnitude the distance should be about 0.2 wavelength,
while it should be about 0.6 wavelength with the same phase.
Secondly, the surface current vector distributions on the
dual-slot-element antenna with the two parasitic monopoles at
2.1 GHz are illustrated in Fig. 13. By adding the two parasitic
monopoles, there are two coupling paths. One is the original
coupling path, by which is coupled on slot element 2
(black solid line with arrow). The other is double-coupling
path, in which the current is firstly coupled from slot element
1 to parasitic monopoles (white solid line with arrow), and
secondly coupled from parasitic monopoles to slot element
2, and as a result is generated on slot element 2 (black
dotted line with arrow). It is found that the current on the left
parasitic monopole is in-phase with the current on the right
edge of slot element 1, and the current on the right parasitic
monopole is out-of-phase with the current on the left parasitic
monopole. Therefore, on slot element 2, the double-coupling
current will be approximately 180 degree out-of-phase
with the original coupling current , leading to low mutual
coupling. It should be noted that if we change the distance
between the parasitic monopoles or the distance between one
478 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 10. Illustration of currents on two half-wavelength dipoles.

Fig. 12. Calculated (a) magnitude and (b) phase of coupling coefficient on
two half-wavelength dipoles.
Fig. 11. Definition of scattering parameters.

slot element and one parasitic monopole these phase relation-

ships will change, which is similar to the discussion about two
half-wavelength dipoles. Thus, the performance of reducing
mutual coupling will change. For example, we change the two
parasitic monopoles’ horizontal positions defined by parameter
with no other parameters changing, as in Fig. 14. The simu-
lated mutual couplings with different values of parameter are
given in Fig. 14. When the value of parameter is chosen to be
3 mm the performance of reducing mutual coupling degrades Fig. 13. The current vector distributions on the dual-slot-element antenna with
the two parasitic monopoles. (At first, the current vector distributions were gen-
by about 5–8 dB, compared to that of our antenna ( is equal to erated by the simulator. Then, we focused on the part of slot elements and para-
27 mm). Consequently, this technique, using parasitic elements sitic monopoles, and simplified the distributions to get a better illustration. How-
to reduce mutual coupling, is sensitive to relative positions ever, the current is very small on the right slot element since the mutual coupling
is greatly reduced. Therefore, two kinds of coupling current, the original cou-
between parasitic elements, and relative positions between
pling and the coupling due to parasitic monopoles, are used to replace the sum
active element and parasitic element. of them.).

V. TRI-ELEMENT ANTENNA
In this section, we extend the proposed technique to a tri-
element antenna operating at 1.9 GHz.
The antenna is printed on a FR4 substrate board (100
mm 112 mm), which has a thickness of 0.8 mm and a relative
permittivity of 4.4. As shown in Fig. 15, the antenna consists of
three slot elements, and each slot element is fed by a 50- mi-
crostrip line with one via-grounded end. On the bottom layer of
the substrate, the ground plane (75 mm 112 mm) is printed. In
LI et al.: REDUCING MUTUAL COUPLING OF MIMO ANTENNAS WITH PARASITIC ELEMENTS FOR MOBILE TERMINALS
Fig. 14. Simulated mutual couplings with different values of parameter .
Fig. 15. Geometry of the tri-element antenna with dimensions in mm (top layer
in black color and bottom layer in grey color).
order to reduce mutual coupling, four parasitic monopoles are
added. However, compared to the dual-slot-element antenna,
there are some differences in designing the tri-element antenna.
Firstly, the length of slot element 2 is different from that of
elements 1 and 3 because of different locations in the ground
479
Fig. 16. Simulated scattering parameters: (a) with PM; (b) without PM (“PM”
means parasitic monopoles).
plane. Secondly, the freedom of designing parasitic monopoles
2 and 3 is limited since they are too close to each other.
The simulated scattering parameters are plotted in Fig. 16,
and as a comparison, the simulated performance without the
four parasitic monopoles is also plotted in Fig. 16. Here,
is almost the same as , and is almost the same as , so
they are not plotted in the figure for brevity. From the figure, it
is found that by adding the four parasitic monopoles the mutual
couplings are greatly improved, in which and both de-
creases from dB to dB.
VI. CONCLUSION
Mutual coupling is a critical problem in the design of MIMO
antennas. At the base station, low mutual coupling is easy to
be realized where element separations are always many wave-
lengths. However, for mobile terminals, acquiring low mutual
coupling will be difficult owing to limited volume. Therefore, in
this paper, using parasitic elements to reduce mutual coupling is
480 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

studied. Its principle is that by a double-coupling path parasitic where is a constant related to the time-averaged power density.
elements create reverse coupling to reduce mutual coupling. It should be noted that normalization is operated in (5). Thus, we
As an example, a dual-slot-element antenna with parasitic can set to unity and acquire (6).
monopoles for mobile terminals is described. The discussion
on calculated and measured channel capacity shows that the an-
tenna can be considered as a good candidate for MIMO systems. ACKNOWLEDGMENT
In addition, we also extend the technique to a tri-element an-
tenna. More importantly, by studying the current distributions, The authors would like to thank S. Etoh, and N. Kitamoto of
it is found that the technique is sensitive to phase relationships, TOYO Corporation, Tokyo, Japan, for providing Bluetest Re-
which are related to relative positions between parasitic ele- verberation chamber and helping to measure channel capacity.
ments, and relative positions between active element and par- They would also like to thank Z. Feng of Tsinghua University,
asitic element. Beijing, China, for suggestions and valuable discussions and the
reviewers whose comments and suggestions helped to signifi-
cantly improve the paper.
APPENDIX
In this Appendix, we derive from (5) to (6) under the as- REFERENCES
sumption of mobile wireless environment defined in [22].
[1] G. J. Foschini, Jr., “Layered space–time architecture for wireless
communication in a fading environment when using multi-element
antennas,” Bell Labs Tech. J., pp. 41–59, Autumn 1996.
[2] G. J. Foschini, Jr. and M. J. Gans, “On limits of wireless communica-
tion in a fading environment when using multiple antennas,” Wireless
(5) Personal Commun., vol. 6, pp. 311–335, 1998.
[3] L. Zheng and D. N. C. Tse, “Diversity and multiplexing: A fundamental
tradeoff in multiple-antenna channels,” IEEE Trans. Inform. Theory,
Because of vol. 49, no. 5, pp. 1073–1096, May 2003.
[4] M. Karaboikis, C. Soras, G. Tsachtsiris, and V. Makios, “Compact
dual-printed inverted-F antenna diversity systems for portable wire-
(A1) less devices,” IEEE Antennas Wireless Propag. Lett., vol. 3, pp. 9–14,
2004.
(A2) [5] G. A. Mavridis, J. N. Sahalos, and M. T. Chryssomallis, “Spatial di-
versity two-branch antenna for wireless devices,” Electron. Lett., vol.
42, no. 5, pp. 266–268, Mar. 2006.
we can get [6] Y.-S. Shin and S.-O. Park, “Spatial diversity antenna for WLAN ap-
plication,” Microw. Opt. Technol. Lett., vol. 49, no. 6, pp. 1290–1294,
Jun. 2007.
[7] T.-Y. Wu, S.-T. Fang, and K.-L. Wong, “Printed diversity monopole
antenna for WLAN operation,” Electron. Lett., vol. 38, no. 25, pp.
1625–1626, Dec. 2002.
(A3) [8] G. Chi, L. Binhong, and D. Qi, “Dual-band printed diversity antenna
for 2.4/5.2-GHz WLAN application,” Microw. Opt. Technol. Lett., vol.
Since the incoming wave’s orthogonal polarizations are un- 45, no. 6, pp. 561–563, Jun. 2005.
[9] Y. Ding, Z. Du, K. Gong, and Z. Feng, “A novel dual-band printed di-
correlated, the expectation of orthogonal polarizations’ product versity antenna for mobile terminals,” IEEE Trans. Antennas Propag.,
is equal to zero and (A3) is simplified to vol. 55, no. 7, pp. 2088–2096, Jul. 2007.
[10] Z. Li, Z. Du, and K. Gong, “A novel wideband printed diversity antenna
for mobile phone,” in Proc. IEEE Antennas Propag. Soc. Int. Symp.
(AP-S 2008), San Diego, CA, 2008, pp. 1–4.
[11] X. Wang, Z. Du, and K. Gong, “A compact wideband planar diversity
antenna covering UMTS and 2.4 GHz WLAN bands,” IEEE Antennas
(A4) Wireless Propag. Lett., vol. 7, no. , pp. 588–591, 2008.
[12] A. C. K. Mak, C. R. Rowell, and R. D. Murch, “Isolation enhancement
between two closely packed antennas,” IEEE Trans. Antennas Propag.,
In addition, because the expectation is in terms of incoming vol. 56, no. 11, pp. 3411–3419, Nov. 2008.
wave, (A4) is derived to [13] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas, “Study
and reduction of the mutual coupling between two mobile phone PIFAs
operating in the DCS1800and UMTS bands,” IEEE Trans. Antennas
Propag., vol. 54, no. 11, pp. 3063–3073, Nov. 2006.
[14] A. Diallo, C. Luxey, P. Le Thuc, R. Staraj, and G. Kossiavas, “En-
hanced two-antenna structures for universal mobile telecommunica-
(A5) tions system diversity terminals,” IET Microw. Antennas Propag., vol.
2, pp. 93–101, Feb. 2008.
In fact, and is the time-averaged [15] A. Chebihi, A. Diallo, C. Luxey, P. Le Thuc, and R. Staraj, “User’s
head and hand influence on the diversity performance of neutralized
power density of incoming wave’s and components, and two-antenna systems for UMTS handsets,” in Proc. IEEE Antennas
XPD is defined by [26]. Furthermore, Propag. Soc. Int. Symp. (AP-S 2008), San Diego, CA, 2008, pp. 1–4.
[16] C.-Y. Lui, Y.-S. Wang, and S.-J. Chung, “Two nearby dual-band an-
the incoming wave arrives in horizontal plane only and the time- tennas with high port isolation,” in Proc. IEEE Antennas Propag. Soc.
averaged power density per steradian is constant so (A5) is sim- Int. Symp. (AP-S 2008), San Diego, CA, 2008, pp. 1–4.
plified to [17] Z. Li, K. Ito, Z. Du, and K. Gong, “Compact wideband printed diversity
antenna for mobile handsets,” in Proc. 2010 Asia-Pacific Radio Science
Conf. (AP-RASC’10), Toyama, Japan, 2010, pp. 1–4.
[18] Z. Li, Z. Du, and K. Gong, “A dual-slot diversity antenna with isolation
enhancement using parasitic elements for mobile handsets,” in Proc.
2009 Asia-Pacific Microwave Conf. (APMC 2009), Singapore, 2009,
pp. 1821–1824.
(A6) [19] HFSS [Online]. Available [Online]. Available: http://www.ansoft.com
LI et al.: REDUCING MUTUAL COUPLING OF MIMO ANTENNAS WITH PARASITIC ELEMENTS FOR MOBILE TERMINALS
[20] J. P. Kermoal, L. Schumacher, K. I. Pedersen, P. E. Mogensen, and
F. Frederiksen, “A stochastic MIMO radio channel model with exper-
imental validation,” IEEE J. Sel. Areas Commun., vol. 20, no. , pp.
1211–1226, Aug. 2002.
[21] L. Dong, H. Choo, R. Heath, Jr., and H. Ling, “Simulation of MIMO
channel capacity with antenna polarization diversity,” IEEE Trans.
Wireless Commun., vol. 4, no. 4, pp. 1869–1873, Jul. 2005.
[22] S. C. K. Ko and R. D. Murch, “Compact integrated diversity antenna
for wireless communications,” IEEE Trans. Antennas Propag., vol. 49,
no. 6, pp. 954–960, Jun. 2001.
[23] M. A. Jensen and Y. Rahmat-Samii, “Performance analysis of antennas
for hand-held transceivers using FDTD,” IEEE Trans. Antennas Prop-
agat., vol. 42, no. 8, pp. 1106–1113, Aug. 1994.
[24] V. Plicanic, B. K. Lau, A. Derneryd, and Z. Ying, “Actual diversity
performance of a multiband diversity antenna with hand and head ef-
fects,” IEEE Trans. Antennas Propag., vol. 57, no. 5, pp. 1547–1556,
May 2009.
[25] P.-S. Kildal and K. Rosengren, “Correlation and capacity of MIMO
systems and mutual coupling, radiation efficiency, and diversity gain
of their antennas: Simulations and measurements in a reverberation
chamber,” IEEE Commun. Mag., vol. 42, pp. 104–112, Dec. 2004.
[26] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile com-
munications,” IEEE Trans. Veh. Technol., vol. 36, no. 4, pp. 149–172,
Nov. 1987.
Zhengyi Li was born in Xi’an, China, in February
1982. He received the B.E. degree in information
engineering from Xi’an Jiaotong University, Xi’an,
China, in 2004, and the Ph.D. degree in electronic
engineering from Tsinghua University, Beijing,
China, in 2010.
Since 2010, he has been working as Postdoctoral
Researcher at Research Center for Frontier Medical
Engineering, Chiba University, Chiba, Japan. His re-
search interests include small antennas for body-cen-
tric wireless communications, MIMO antennas, and
reconfigurable antennas.
Dr. Li was the recipient of the 2010 Asia-Pacific Radio Science Conference
(AP-RASC’10) Young Scientist Award in 2010.
Zhengwei Du was born in Sichuan Province, China,
on August 21, 1971. He received the B.Sc., M.Sc.,
and Ph.D. degrees in engineering from the Univer-
sity of Electronic Science and Technology of China,
Chengdu, China, in 1992, 1995, and 1998, respec-
tively.
Since 1998, he has been with Tsinghua Uni-
versity, Beijing, China, as a Postdoctoral Fellow
(February 1998–October 1999), Research Assistant
(November 1999–July 2000), Associate Professor
(August 2000–December 2006), and currently, Full
Professor. His main research interests include ultra wideband/short pulse
electromagnetic, antenna, propagation, analysis of microwave/millimeter wave
planar structures, photonic bandgap (electromagnetic bandgap) circuits, and
electromagnetic compatibility/electromagnetic interference (EMC/EMI).
Prof. Du is Chairman of the Microwave Integrated Circuits and Mobile Com-
munications Professional Committee of the Microwave Society of China.
Masaharu Takahashi (M’95-SM’02) was born in
Chiba, Japan, in December 1965. He received the
B.E. degree in electrical engineering from Tohoku
University, Miyagi, Japan, in 1989, and the M.E.
and D.E. degrees in electrical engineering from the
Tokyo Institute of Technology, Tokyo, Japan, in
1991 and 1994, respectively.
From 1994 to 1996, he was a Research Associate,
and from 1996 to 2000, an Assistant Professor with
the Musashi Institute of Technology, Tokyo, Japan.
From 2000 to 2004, he was an Associate Professor
with the Tokyo University of Agriculture and Technology, Tokyo, Japan. He is
currently an Associate Professor with the Research Center for Frontier Medical
481
Engineering, Chiba University, Chiba, Japan. His main interests are electrically
small antennas, planar array antennas, and EM compatibility.
Dr. Takahashi is a Senior Member of the Institute of Electronics, Information
and Communication Engineers (IEICE), Japan. He was the recipient of the 1994
IEEE Antennas and Propagation Society (IEEE AP-S) Tokyo Chapter Young
Engineer Award.
Kazuyuki Saito (S’99-M’01) was born in Nagano,
Japan, in May 1973. He received the B.E., M.E., and
D.E. degrees in electronic engineering from Chiba
University, Chiba, Japan, in 1996, 1998 and 2001, re-
spectively.
He is currently an Associate Professor with the
Research Center for Frontier Medical Engineering,
Chiba University. His main interest is in the area
of medical applications of microwaves including
microwave hyperthermia.
Dr. Saito is a member of the Institute of Electrical,
Information and Communication Engineers (IEICE), Japan, the Institute of
Image Information and Television Engineers of Japan (ITE), and the Japanese
Society for Thermal Medicine. He was the recipient of the IEICE Antennas and
Propagation Society (AP-S) Freshman Award, the Award for Young Scientist
of the URSI General Assembly, the IEEE AP-S Japan Chapter Young Engineer
Award, the Young Researchers’ Award of the IEICE, and the International
Symposium on Antennas and Propagation (ISAP) Paper Award in 1997, 1999,
2000, 2004, and 2005 respectively.
Koichi Ito (M’81-SM’02-F’05) received the B.S.
and M.S. degrees from Chiba University, Chiba,
Japan, in 1974 and 1976, respectively, and the D.E.
degree from the Tokyo Institute of Technology,
Tokyo, Japan, in 1985, all in electrical engineering.
From 1976 to 1979, he was a Research Associate
at the Tokyo Institute of Technology. From 1979 to
1989, he was a Research Associate at Chiba Univer-
sity. From 1989 to 1997, he was an Associate Pro-
fessor at the Department of Electrical and Electronics
Engineering, Chiba University, and is currently a Pro-
fessor at the Department of Medical System Engineering, Chiba University.
From 2005 to 2009, he was Deputy Vice-President for Research, Chiba Uni-
versity. From 2008 to 2009, he was Vice-Dean of the Graduate School of En-
gineering, Chiba University. Since April 2009, he has been appointed as Di-
rector of Research Center for Frontier Medical Engineering, Chiba University.
In 1989, 1994, and 1998, he visited the University of Rennes I, France, as an In-
vited Professor. He has been appointed as Adjunct Professor to the University of
Indonesia since 2010. His main research interests include analysis and design
of printed antennas and small antennas for mobile communications, research
on evaluation of the interaction between electromagnetic fields and the human
body by use of numerical and experimental phantoms, microwave antennas for
medical applications such as cancer treatment, and antennas for body-centric
wireless communications.
Dr. Ito is a Fellow of the Institute of Electronics, Information and Commu-
nication Engineers (IEICE) of Japan, a member of the American Association
for the Advancement of Science, the Bioelectromagnetics Society (BEMS), the
Institute of Image Information and Television Engineers of Japan (ITE) and the
Japanese Society for Thermal Medicine. He served as Chair of the Technical
Group on Radio and Optical Transmissions, ITE from 1997 to 2001, Chair of the
Technical Committee on Human Phantoms for Electromagnetics, IEICE from
1998 to 2006, Chair of the Technical Committee on Antennas and Propagation,
IEICE from 2009 to 2011, Chair of the IEEE AP-S Japan Chapter from 2001
to 2002, TPC Co-Chair of the 2006 IEEE International Workshop on Antenna
Technology (iWAT2006), Vice-Chair of the 2007 International Symposium on
Antennas and Propagation (ISAP2007), General Chair of iWAT2008, Co-Chair
of ISAP2008, an AdCom member for the IEEE AP-S from 2007 to 2009, and an
Associate Editor for the IEEE Transactions on Antennas and Propagation from
2004 to 2010. He currently serves as a Distinguished Lecturer for the IEEE
AP-S, Chair of the IEEE AP-S Committee on Man and Radiation (COMAR), a
member of the Board of Directors, BEMS and a Councilor to the Asian Society
of Hyperthermic Oncology (ASHO). He has been appointed as General Chair
of ISAP2012 to be held in Nagoya, Japan, and as a member the IEEE Life Sci-
ences New Initiative (LSNI) Project Team.
482
A Compact Wideband MIMO Antenna With Two
Novel Bent Slits
Jian-Feng Li, Qing-Xin Chu, Senior Member, IEEE, and Tian-Gui Huang
Abstract—A compact wideband multiple-input- multiple-output
(MIMO) antenna is presented. The MIMO antenna consists of
two symmetric monopoles with edge-to-edge separation of nearly
at 2.5 GHz. Two novel bent slits are etched into the
ground plane. At the lower frequencies, the bent slits can reduce
the mutual coupling and have slight effect on the reflection coeffi-
cient. At the higher frequencies, the slits can be considered as slit
antennas to widen the impedance bandwidth because the two slits
are coupled fed by two 50- microstrip lines, respectively. Two
triangles are cut from the ground plane. In this way, the reflection
coefficient and isolation of the two slit antennas can be improved.
A bandwidth of 92.7% with 10 dB and 18
dB from 2.4 to 6.55 GHz is achieved. In order to provide quantifi-
cations for the performance of the MIMO antenna in real-world
usage conditions, the effects of human hand and head on the
performance of the MIMO antenna are investigated. The results
show that the MIMO antenna serving as a phone antenna can
provide spatial and pattern diversity to combat multipath fading.
Index Terms—Correlation coefficient, diversity antenna, diver-
sity gain, multiple-input multiple-output (MIMO) antenna, phone
antenna, total radiation power, wideband antenna.
I. INTRODUCTION
N OWADAYS, there is a demand to increase the data rate
of existing wireless communication systems. The appli-
cation of diversity techniques, most commonly assuming two
antennas in a mobile terminal, can enhance the data rate and re-
liability without sacrificing additional spectrum or transmitted
power in rich scattering environments [1], [2]. When a mul-
tiple-input multiple-output (MIMO) antenna is applied in a mul-
tifunctional portable device, wideband and high isolation are de-
manded. However, the design of a compact wideband MIMO
antenna with high isolation is an open issue.
Various techniques have been reported to enhance isolation
between the elements of a MIMO antenna. High isolation was
achieved by etching slits into the ground plane [3]. Ground
branches were applied in [4] to achieved low mutual coupling
within a narrow frequency band. Parasitic elements were added
to improve the port isolation of a MIMO antenna [5]. Since the
Manuscript received April 25, 2011; revised June 23, 2011; accepted July 16,
2011. Date of publication October 25, 2011; date of current version February 03,
2012. This work was supported by the Science Fund of China (60971052) and
“the Fundamental Research Funds for the Central Universities (2009IM0167)”.
The authors are with the School of Electronic and Information Engi-
neering, South China University of Technology, Guangzhou, Guangdong
510640, China (e-mail: li.jf01@mail.scut.edu.cn; qxchu@scut.edu.cn;
huang.tg@mail.scut.edu.cn).
Digital Object Identifier 10.1109/TAP.2011.2173452
0018-926X/$26.00 © 2011 IEEE
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
implementation of additional parasitic elements occupies a sig-
nificant space, this technique is not attractive for handset de-
vices. All of the aforementioned techniques deteriorate the re-
flection coefficient, and therefore a wide impedance bandwidth
with high isolation is difficult to obtain.
A compact wideband MIMO antenna with high isolation is
presented in this paper. The MIMO antenna consists of two
dual-branch monopoles with edge-to-edge separation of nearly
at 2.5 GHz, and the size of the monopole is 18 15
mm . Traditionally, slits were mainly used to reduce mutual
coupling of element antennas, but this approach usually dete-
riorate the reflection coefficient. In this design, two novel bent
slits are applied, which can avoid this problem. At the lower fre-
quencies, they can reduce the mutual coupling resulted from sur-
face currents and have slight effect on the reflection coefficient.
At the higher frequencies, the two bent slits can be considered
as two slit antennas because they are coupled fed by two 50-
microstrip lines, respectively, and they excide resonant modes
at 6.3 GHz to widen the impedance bandwidth. Two triangles
are cut from the ground plane to change the distribution of the
ground surface currents, and therefore the reflection coefficient
and isolation of the two slit antennas are improved. Moreover,
a metal strip located between the two monopoles is used to de-
crease the mutual coupling caused by near-field. Thus, the mu-
tual coupling between the two monopoles is further reduced. A
bandwidth of 92.7% with 10 dB and 18 dB
from 2.4 to 6.55 GHz is achieved, which covers 2.4/5.2/5.8-GHz
WLAN, 2.5/3.5/5.5-GHz WiMAX and the lower UWB band
(3.1–4.8 GHz) operation.
The radiation pattern, mean effective gain (MEG), radiation
efficiency, correlation coefficient and diversity gain (DG) are
calculated to evaluate the diversity performance of the proposed
MIMO antenna. To provide quantifications for the performance
of the MIMO antenna in real-world usage conditions, the effects
of human hand and head on the antenna performance including
radiation efficiency, mismatch efficiency, total radiation power
(TRP) and radiation pattern, are discussed. The geometry of the
proposed MIMO antenna is shown in Section II. In Section III,
the working mechanism of the MIMO antenna is investigated.
In Section IV, the MEG, radiation efficiency, correlation coef-
ficient and DG are calculated. The effects of human hand and
head on the performance of the MIMO antenna are discussed in
Section V. Finally, a conclusion is given in Section VI.
II. CONFIGURATION OF THE PROPOSED MIMO ANTENNA
The geometry of the proposed MIMO antenna is illustrated
in Fig. 1. The MIMO antenna consists of two symmetric dual-
branch (branch 1 with length of 30 mm and branch 2
LI et al.: COMPACT WIDEBAND MIMO ANTENNA WITH TWO NOVEL BENT SLITS
Fig. 1. Structure of the proposed MIMO antenna.
with length of 28 mm) monopoles. The MIMO antenna is
printed on the upper part of a partially grounded FR4 substrate
with dimensions 78 40 1.6 mm and relative permittivity
4.4. On the back surface of the substrate, the main rectangular
ground plane of 40 mm in width and 60 mm in length is printed.
In order to reduce the mutual coupling caused by the surface cur-
rents and improve the impedance matching, two bent slits with
a length of 22 mm (about at 6.3 GHz) are etched into the
ground plane. Moreover, each slit is coupled fed by a 50 mi-
crostrip line, and consequently the slits can be considered as slit
antennas. To reduce the effect of the slits on the lower part of the
impedance bandwidth, the slits are bent by 90 ; to suppress the
mutual coupling between the two slit antennas, they are set per-
pendicular to each other. Two triangles with height 3 mm
and width 4 mm are cut from the ground plane to change
the distribution of the ground surface currents. A metal strip of
size 18 1 mm , which is placed between the two monopoles,
is applied to decrease the mutual coupling caused by near-field.
III. DISCUSSION OF WORKING MECHANISM
The proposed MIMO antenna shown in Fig. 1 has been
fabricated and tested. In Fig. 2, it can be seen that the sim-
ulated and measured results are in a good agreement. The
results show a bandwidth of 92.7% with 10 dB and
18 dB from 2.4 to 6.55 GHz covering the following
bands: 2.4/5.2/5.8-GHz WLAN, 2.5/3.5/5.5-GHz WiMAX and
the lower UWB band (3.1–4.8 GHz) operation.
A. Dual-Branch Monopole Antenna
Investigation of the dual-branch monopole antenna shown in
Fig. 3 can help to understand the working mechanism of the
proposed MIMO antenna. The antenna presented in [6] had a
similar structure (see Fig. 3), but it is not suitable for wideband
application because the frequency interval between the two res-
onant modes is large. In Fig. 3, the length of branch 1 is
28 mm (shorter than at 2.6 GHz), the length of branch 2
is 26 mm (shorter than at 3.0 GHz), and the two
483
Fig. 2. Measured and simulated S-parameters of the proposed MIMO antenna.
Fig. 3. Configuration of the two-branch monopole antenna.
branches can produce a resonant mode at 2.6 GHz and a
resonant mode at 3.0 GHz, respectively. The inductive re-
actance of the dual-branch monopole antenna is compensated by
the coupling capacitance introduced by the small gap between
the monopole and the ground plane, and the electrical lengths
of the two branches are extended.
Fig. 4 shows the simulated reflection coefficients of the dual-
branch monopole antenna and two cases with branch 1 only or
branch 2 only. The case only with branch 1 generates a res-
onant mode at about 2.6 GHz, and a dual-resonance exci-
tation at about 5.5 GHz; the case only with branch 2 generates a
dual-resonance excitation at about 3.0 GHz and a res-
onant mode at about 5.8 GHz. When the two branches are inte-
grated together, a wide operating band of 89% with 10
dB extending from 2.4 to 6.25 GHz is produced. The dual-
branch monopole antenna with wideband and compact size is
suitable to be an element of a compact wideband MIMO an-
tenna.
B. Implementation of Novel Bent Slits
To analyze the working mechanism of the two novel bent slits
described in Section II, two more configurations will be inves-
tigated in this section. They are: 1) MIMO monopole antenna
with a conventional rectangular ground plane and 2) MIMO
monopole antenna with two bent slits etched into a conventional
rectangular ground plane. The S-parameters for the two config-
urations are given in Fig. 5.
It can be observed that the reflection coefficient of the
dual-branch monopole antenna (see Fig. 3) is better than that
484
Fig. 4. Simulated reflect coefficients of the two-branch monopole antenna and
the two cases with branch 1 only or branch 2 only.
Fig. 5. Simulated S-parameters of MIMO antennas with different configura-
tions of ground plane.
of the MIMO monopole antenna with a conventional rectan-
gular ground plane. Strong near-field coupling and surface
current coupling are induced when the two elements are placed
closely. Fig. 6 shows the current distribution at 4.5 GHz (the
central frequency of the operation bandwidth) of the MIMO
monopole antenna with a conventional rectangular ground
plane (monopole 1 is excited, while monopole 2 is terminated
with a 50 load). There are strong coupled currents observed
on the monopole 2 and the upper part of the ground plane. We
can believe that the strong coupled current is one reason of
the poor reflection coefficient and low isolation of the MIMO
monopole antenna with a conventional rectangular ground
plane. In order to improve the reflection coefficient and the
isolation, two bent slits are applied as shown in Fig. 7. As can
be observed in Fig. 5, the antenna with the two bent slits has
better reflection coefficient and higher isolation than the MIMO
antenna with a conventional ground plane. Thus, it can be
concluded that the bent slits improve the reflection coefficient
and also enhance the isolation effectively.
In [3], slits were already used to improve isolation, although
these slits had negative effect on the impedance matching, and a
wide impedance bandwidth was difficult to obtain. However, the
two novel bent slits applied in this paper can avoid this problem.
The slits are bent forming a 90 corner, and which can help
to reduce the side effect that the slits have on the current dis-
tribution at the lower part of the impedance bandwidth. Thus,
a small discrepancy of the lower part of the impedance band-
width can be seen between the two-branch monopole antenna
(see Fig. 4) and the MIMO antenna with the two bent slits (see
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 6. Average current distribution of the MIMO antenna with a conventional
rectangular ground plane at 4.5 GHz.
Fig. 7. Average current distributions of the MIMO antenna with two bent slits
etched into a conventional rectangular ground plane: (a) at 4.5 GHz and (b) 6.3
GHz.
Fig. 5). When comparing with Fig. 6, a large portion of the sur-
face currents presented in Fig. 7(a) is being trapped by the bent
slit located on the left-hand side of the ground plane. It demon-
strates that currents flowing from left-hand side of the ground
plane to right-hand side are substantially reduced. Therefore,
the mutual coupling cased by the ground surface currents is re-
duced. Moreover, the two slits can help to widen the impedance
bandwidth because they are coupled fed by the two 50- mi-
crostrip lines, respectively. Each bent slit is 22 mm in length
(about at 6.3 GHz), and resonant mode at about 6.3 GHz,
which does not appear in the cases without the two slits (see
Figs. 4 and 5), is excided by the two bent slits. The current dis-
tribution in Fig. 7 also provides evident that the resonant mode at
6.3 GHz (see Fig. 5) is excided by the bent slits. These currents
are mainly distributed along the slit located on the left-hand side
of the ground plane. Moreover, the two slits are set perpendic-
ular to each other to suppress the mutual coupling between them.
C. Discussion of Metal Strip and Cutting of Triangles
There is strong near-field coupling besides the ground surface
currents coupling. In order to reduce the near-field coupling,
the metal strip shown in Fig. 1 is used. The proposed MIMO
antenna depicted in Fig. 1 provides an 8 dB improvement of
isolation over the case with two bent slits etched into a con-
ventional rectangular ground plane (see Fig. 5). The metal strip
can be treated as a reflector of electromagnetic wave [7]. The
reflector separates the radiation patterns of the two monopoles
to decrease the unwanted mutual coupling resulted from the
near-field, thus the isolation is further enhanced, especially for
the frequencies near 6.3 GHz. From Figs. 2 and 5, it can also
LI et al.: COMPACT WIDEBAND MIMO ANTENNA WITH TWO NOVEL BENT SLITS 485

Fig. 9. Prototype of the proposed MIMO antenna: (a) front view and (b) back
view.

Fig. 8. Effects of and on the proposed MIMO antenna: (a) and (b)
.

be observed that the effect of the metal strip on the frequency

response of the reflection coefficient is small.
The cutting of two triangles of size from the ground
plane is to change the distribution of ground surface currents.
The effects of and on the reflection coefficient and the mu-
tual coupling are studied in Fig. 8(a) and (b), respectively. For
the frequencies near 6.3 GHz, the reflection coefficient and the
mutual coupling vary greatly with varying and . When
and are chosen as 3 and 4 mm, respectively, good impedance
matching and isolation for the proposed MIMO antenna can be
obtained as shown in Fig. 2.
Fig. 10. Simulated and simulated radiation patterns on (a) XY-plane and (b)
YZ-plane.
IV. DIVERSITY PERFORMANCE OF THE MIMO ANTENNA

The diversity performance of the proposed MIMO antenna is

evaluated by the MEG, radiation efficiency, correlation coeffi-
cient, and DG. DG is defined as the improvement in signal-to-
noise (SNR) ratio from the combined signal from all the an-
tennas of the diversity system relative to the SNR from a single
antenna [8].
The photos of the fabricated MIMO antenna are shown in
Fig. 9. Fig. 10 displays the simulated and measured 2-D radia-
tion patterns of the two monopoles (the tested monopole is ex-
cited, while the other one is terminated with a 50- load) at
4.5 GHz, and the gains have been normalized. The discrepancy
between the measured and simulated results comes principally
from our measurement setup, especially from the power loss
contribution of the feeding cables of the monopoles. In small
antenna measurements, it is usually difficult to efficiently choke
the feeding cables to avoid currents flowing on their outer part.
Moreover, these cables are very difficult to maintain perfectly
parallel to the length of the PCB due to the size and the prox-
imity of the ferrite chokes. Back-radiation patterns for the two
monopoles are achieved, and these special patterns can help to
reduce the mutual coupling caused by near-field. Thus, the pro-
posed MIMO monopole antenna shows good pattern diversity
characteristic.
To quantify the average received signal strength for each
monopole, MEGs are calculated based on a series of the as-
sumptions for mobile wireless environments defined in [4],
[9]. In this paper, a cross-polarization discrimination of 0
dB, which is the average in an indoor fading environment [9],
[10], is assumed. The MEGs and the ratios of to
at the center frequencies of 2.4/5.2/5.8-GHz
WLAN, 2.5/3.5/5.5-GHz WiMAX and the lower UWB band
(3.1–4.8 GHz) are calculated [4], [11], [12]. The MEGs, ratios
of to , radiation efficiencies , mismatch
486 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

TABLE I
PERFORMANCE OF THE PROTOTYPE OF THE PROPOSED MIMO ANTENNA

TABLE II
SIMULATED EFFECT OF THE HAND AND HEAD ON THE PROPOSED MIMO ANTENNA

diation patterns [4], [10], [16] and are denoted as and ,

respectively:

(2)

where

(3)
Fig. 11. Configuration of (a) MIMO antenna hand phantom, (b) MIMO an- in which and are the and components of the com-
tenna hand phantom SAM head. plex electric field radiation pattern, respectively, and the asterisk
presents the complex conjugate.
Based on the measured data, for the two kinds of propaga-
efficiencies and total efficiencies , which are obtained tion environments are calculated and listed in Table I. It can be
from the measured data, are listed in Table I. The maximum seen that is smaller than 0.01 for the both cases.
ratio of the to is smaller than 0.4 dB, and , In an ideal case (the efficiency of the two-element antenna
and are higher than 50% and 46%, respectively. is 100%, and the power levels of the signals received by the
For isotropic/uniform signal propagation environments, the two antennas are the same in the diversity system), DG using a
correlation coefficient and envelope correlation coefficient selection combiner is 10 dB, where the radio link reliability is
can be derived from the S-parameters [13], and they are denoted 99% [15]. In order to achieve the diversity gain as high as the
as and , respectively. Based on the conclusion in [14], ideal case, the received signals from the two elements should
and can be calculated by exhibit poor correlation and the discrepancy of the power levels
of the signals received by the two elements should be minimized
[15]. A good DG can be achieved with the proposed antenna,
because the signals received from the proposed MIMO antenna
satisfy the criteria [4], [10], and [16]
(1) and dB (4)
The expression contains the effect of on . The effects of and power unbalance of the two antenna
For the Gaussian signal propagation environments: and elements on DG should be taking into account. Findings in [15],
are the and components of the probability distribution [17] show that the degradation of DG due to is given by a
function of the incoming wave, respectively, and factor defined as
being Gaussian distributed, and and being uni-
form distributed. Thus, and can be evaluated from the ra- (5)
LI et al.: COMPACT WIDEBAND MIMO ANTENNA WITH TWO NOVEL BENT SLITS 487

Fig. 12. Simulated (SEMCAD) 3-D radiation patterns for: (a) MIMO antenna hand phantom with monopole 1 excided and monopole 2 terminated with a
standard 50- matching load; (b) MIMO antenna hand phantom with monopole 2 excided and monopole1 terminated with a standard 50- matching load; (c)
MIMO antenna hand phantom SAM head with monopole 1 excided and monopole 2 terminated with a standard 50- matching load; (d) MIMO antenna
hand phantom SAM head with monopole 2 excided and monopole 1 terminated with a standard 50- matching load.

The ratio of to of the antenna elements (as- radiation pattern of the MIMO antenna are studied. The input
suming only two elements) [15] is used to illustrate the side ef- power in this paper is normalized to 1 W or 30 dBm [19].
fect of power unbalance of the signals on the DG Fig. 11(a) shows the simulation model of the MIMO antenna
hand phantom; Fig. 11(b) shows the simulation model of the
(6) MIMO antenna hand phantom SAM (specific anthropo-
morphic mannequin [20]) head, representing the calling mode.
Based on the discussion above, DGs of the proposed MIMO In Fig. 11, the MIMO antenna and the system ground plane are
antenna are calculated by the formula covered by a plastic casing with 1 mm in thickness to avoid the
direct contact between the antenna and the hand/head model.
(7) From Tables I and II, it can be observed that and of
the MIMO antenna hand phantom are smaller than the MIMO
where is the diversity gain of the antenna in the ideal case. antenna in free space, and that is because some power is ab-
The calculated DGs, which are based on the measured data, are sorbed by the hand phantom. However, and of the
presented in Table I. MIMO antenna hand phantom are still larger than 27% and
70%, respectively. The SAM head also causes some power loss.
Compared with the MIMO antenna hand phantom, and
V. EFFECTS OF HAND AND HEAD
TRPs of the MIMO antenna hand phantom SAM head are
With the aid of the simulation software SEMCAD-X [18], smaller but still larger than 20% and 20 dBm, respectively. For
the effects of human hand and head on , TRP and the lower frequencies, the side effect of the hand phantom on
488 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
the impedance matching may be compensated by those of the
SAM head, Thus, of the proposed MIMO antenna hand
phantom SAM head is larger as compared to the proposed
MIMO antenna hand phantom. In Table II, discrepancies of
and TRPs can be seen between the two monopoles,
and that is because the relative positions of the hand and the
head for the two monopoles are different.
Fig. 12 plots the simulated 3-D radiation patterns at 2.5, 3.5,
and 5.5 GHz of the MIMO antenna hand phantom and the
MIMO antenna hand phantom SAM head. It can be seen
that distortions of the radiation patterns owing to the presence
of the hand phantom/SAM head are small, and back-radiation
patterns for the two monopoles are still achieved. All the results
show that the proposed antenna has good pattern diversity in
real-world usage conditions.
VI. CONCLUSION
A compact wideband MIMO monopole antenna has been in-
vestigated. The compact sizes of the monopole and the ground
plane are 18 15 mm and 60 40 mm , respectively. A band-
width of 92.7% with 10 dB and 18 dB
from 2.4 to 6.55 GHz is achieved. Two novel bent slits have
been used. At the lower frequencies, they can reduce the mu-
tual coupling resulted from the surface current, and their effect
on the reflection coefficient is slight. At the higher frequencies,
the bent slits are considered as slit antennas because they are
coupled fed by the two 50- microstrip lines, respectively. The
resonant modes at 6.3 GHz are excided by the two slit antennas
to widen the impedance bandwidth. A cutting of two triangles
from the ground plane has been introduced to change the dis-
tribution of ground surface currents. In this way, the isolation
of the two slit antennas is improved. A grounded metal strip lo-
cated between the two monopoles has been used to decrease the
mutual coupling caused by near-field, and the isolation of the
proposed antenna is further enhanced. The MEG, radiation effi-
ciency, correlation coefficient, and DG have been calculated to
evaluate the diversity performance of the proposed MIMO an-
tenna. The effects of human hand and head on the performance
of the proposed MIMO antenna have been investigated by sim-
ulation software SEMCAD-X. The results show that the pro-
posed antenna serving as a phone MIMO antenna can provide
spatial and pattern diversity to combat the multipath fading in
real-world usage conditions.
REFERENCES
[1] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wirel. Person.
Commun., vol. 6, no. 3, pp. 311–335, 1998.
[2] T. Bolin, A. Derneryd, G. Kristensson, V. Plicanic, and Z. Ying, “Two-
antenna receive diversity performance in indoor environment,” IEEE
Electron. Lett., vol. 41, no. 2, pp. 1205–1206, Oct. 2005.
[3] F. G. Zhu, J. D. Xu, and Q. Xu, “Reduction of mutual coupling be-
tween closely-packed antenna elements using defected ground struc-
ture,” Electron. Lett. 4, vol. 45, no. 12, pp. 601–602, Jun. 2009.
[4] Y. Ding, Z. W. Du, K. Gong, and Z. H. Feng, “A novel dual-band
printed diversity antenna for mobile terminals,” IEEE Trans. Ant.
Propag., vol. 55, no. 7, pp. 2088–2096, Jul. 2007.
[5] Z. Y. Li, Z. W. Du, and K. Gong, “A dual-slot diversity antenna with
isolation enhancement using parasitic elements for mobile handsets,”
in Proc. Asia Pacific Microw. Conf., 2009, pp. 1821–1824.
[6] J. R. Panda and R. S. Kshetrimayum, “A compact printed U-shaped
dual-band monopole antenna for wireless and RFID applications,” pre-
sented at the Appl. Electromagn. Conf., Kolkata, India, 2009.
[7] G. M. Chi, B. H. Li, and D. S. Qi, “Dual-band printed diversity antenna
for 2.4/5.2-GHz WLAN application,” Microw. Opt. Technol. Lett., vol.
45, no. 6, pp. 561–563, Jun. 2005.
[8] P. S. Kildal and K. Rosengran, “Correlation and capacity of MIMO
systems and mutual coupling, radiation efficiency, and diversity gain
of their aantennas: Simulations and measurements in a reverberation
chamber,” IEEE Commun. Mag., vol. 42, no. 12, pp. 102–112, Dec.
2004.
[9] S. C. K. Ko and R. D. Murch, “Compact integrated diversity antenna
for wireless communications,” IEEE Trans. Ant. Propag., vol. 49, no.
6, pp. 954–960, Jun. 2001.
[10] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile
communications,” IEEE Trans. Veh. Technol., vol. VT-36, no. 4, pp.
149–172, Nov. 1987.
[11] M. Karaboikis, C. Soras, G. Tsachtsiris, and V. Makios, “Compact
dual-printed inverted-F antenna diversity systems for portable wireless
devices,” IEEE Trans. Ant. Propag. Lett., vol. 3, no. 1, pp. 9–14, Dec.
2004.
[12] K. Rosengren and P. S. Kildal, “Radiation efficiency, correlation,
diversity gain and capacity of a six-monopole antenna array for a
MIMO system: Theory, simulation and measurement in reverberation
chamber,” IEE Proc. Microw. Ant. Propag., vol. 152, no. 1, pp. 7–16,
2005.
[13] S. Blanch, J. Romeu, and I. Corbella, “Exact representation of antenna
system diversity performance from input parameter description,” IEE
Electron. Lett., vol. 39, no. 9, pp. 705–707, May 1, 2003.
[14] H. Paul, “The significance of radiation efficiencies when using
S-parameterss to calculate the received signal correlation from two
antennas,” IEEE Ant. Wirel. Propag. Lett., vol. 4, no. 1, pp. 97–99,
Jun. 2005.
[15] Y. Gao, X. D. Chen, and Z. N. Ying, “Design and performance inves-
tigation of a dual-element PIFA array at 2.5 GHz for MIMO terminal,”
IEEE Trans. Ant. Propag., vol. 55, no. 12, pp. 3433–3441, Jun. 2007.
[16] M. P. Karaboikis, V. C. Papamichael, G. F. Tsachtsiris, and V. T.
Makios, “Integrating compact printed antennas onto small diver-
sity/MIMO Terminals,” IEEE Trans. Ant. Propag., vol. 56, no. 7, pp.
2067–2078, Jul. 2008.
[17] A. A. H. Azremi, J. Toivanen, and T. Laitinen, “On diversity perfor-
mance of two-element coupling element based antenna structure for
mobile terminal,” in Proc. Ant. Propag. Euro. Conf., 2010, pp. 1–5.
[18] SPEAG SEMCAD X 13.4, Schmid and Partner Engineering AGS [On-
line]. Available: http://www.semcad.com
[19] C. H. Li, E. Ofli, N. Chavannes, and N. Kuste, “Effects of hand
phantom on mobile phone antenna performance,” IEEE Trans. Ant.
Propag., vol. 57, no. 9, pp. 2763–2770, Sep. 2009.
[20] IEEE Recommended Practice for Determining the Peak Spatial-Av-
erage Specific Absorption Rate (SAR) in the Human Head From Wire-
less Communications Devices: Measurement Techniques, IEEE Stan-
dard 1528, Dec. 2003.
Jian-Feng Li was born in Maoming, Guangdong,
China. She is currently pursuing the Ph.D. degree
in electronic and information engineering from
South China University of Technology, Guangzhou,
Guangdong, China.
Her research interests include multi-band antenna
and MIMO phone antenna.
LI et al.: COMPACT WIDEBAND MIMO ANTENNA WITH TWO NOVEL BENT SLITS
Qing-Xin Chu (M’99–SM’11) received the B.S.,
M.E., and Ph.D. degrees in electronic engineering
from Xidian University, Xi’an, Shaanxi, China, in
1982, 1987, and 1994, respectively.
He is currently a full Professor with the School
of Electronic and Information Engineering, South
China University of Technology, Guangzhou,
Guangdong, China, where he is also head of Re-
search Institute of RF and Wireless Techniques. He
worked with the School of Electronic Engineering,
Xidian University, from 1982 to 2003, and was
the Vice-Dean and a full Professor of the school from 1997 to 2003. He
undertook his research with the Department of Electronic Engineering, Chinese
University of Hong Kong as a Research Associate from July 1995 to July 1997
and March to September 1998, and worked in the Department of Electronic
Engineering, City University of Hong Kong as a research fellow from February
to May 2001. He was Visiting Professors of the Department of Electronic
Engineering, Chinese University of Hong Kong from July to October 2002,
and the Department of Electronic Engineering, City University of Hong Kong
from December 2002 to March 2003. He visited the School of Electrical and
Electronic Engineering, Nanyang Technological University, Singapore from
July to October 2004, with the Tan Chin Tuan Exchange Fellowship Award,
and visited the Department of Electrical and Electronic Engineering, Okayama
University, Japan from January to March 2005, with the Fellowship awarded by
489
Japan Society for Promotion of Science (JSPS). His current research interests
include UWB antennas and RF components, active integrated antennas, spatial
power combining array, and computational electromagnetics.
Prof. Chu was a recipient of the first-class Educational Award of Shaanxi
Province in 2003, the top-class Science Award of Education Ministry of China
and second-class Science and Technology Advance Award of Shaanxi Province
in 2002, the top-class Educational Award of Shaanxi Province and the second-
class Award of Science and Technology Advance of Electronic Industry Min-
istry in 1995. He is a senior member of China Electronic Institute (CEI).
Tian-Gui Huang was born in Maoming, Guang-
dong, China, on May 5, 1986. He is currently
pursuing the M.E. degree in electronic and infor-
mation engineering from South China University of
Technology, Guangzhou, Guangdong, China.
His research interest includes UWB band-notch
antenna.
490 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Characteristic Mode Based Tradeoff Analysis of
Antenna-Chassis Interactions for Multiple Antenna
Terminals
Hui Li, Student Member, IEEE, Yi Tan, Buon Kiong Lau, Senior Member, IEEE,
Zhinong Ying, Senior Member, IEEE, and Sailing He, Senior Member, IEEE
Abstract—The design of multiple antennas in compact mobile
terminals is a significant challenge, due to both practical and fun-
damental design tradeoffs. In this paper, fundamental antenna de-
sign tradeoffs of multiple antenna terminals are presented in the
framework of characteristic mode analysis. In particular, interac-
tions between the antenna elements and the characteristic modes
and their impact on design tradeoffs are investigated in both theory
and simulations. The results reveal that the characteristic modes
play an important role in determining the optimal placement of an-
tennas for low mutual coupling. Moreover, the ability of antenna
elements to localize the excitation currents on the chassis can signif-
icantly influence the final performance. To demonstrate the effec-
tiveness of the proposed approach, a dual-band, dual-antenna ter-
minal is designed to provide an isolation of over 10 dB for the 900
MHz band without additional matching or decoupling structures.
A tradeoff analysis of bandwidth, efficiency, effective diversity gain
and capacity is performed over different antenna locations. Finally,
three fabricated prototypes verify the simulation results for repre-
sentative cases.
Index Terms—Antenna array mutual coupling, multiple-input
multiple-output (MIMO) systems, mobile communication.
I. INTRODUCTION
T HE phenomenal success of multiple-input mul-
tiple-output (MIMO) technology can be seen in its
critical role of enabling high data rates in Long Term Evolu-
tion (LTE), Worldwide Interoperability for Microwave Access
Manuscript received June 01, 2010; revised February 02, 2011; accepted June
20, 2011. Date of publication October 25, 2011; date of current version February
03, 2012. This work was supported in part by VINNOVA under Grants 2009-
04047 and 2008-00970 and also in part by a scholarship within EU Erasmus
Mundus External Cooperation Window Lot 14.
H. Li, is with the Department of Electrical and Information Technology,
Lund University, SE-221 00 Lund, Sweden and also with the School of Electro-
magnetic Engineering, Royal Institute of Technology, Sweden and the Center
for Optical and Electromagnetic Research, Zhejiang University, Hangzhou,
310058, China (e-mail: Hui.Li@eit.lth.se@eit.lth.se).
Y. Tan and B. K. Lau are with the Department of Electrical and Information
Technology, Lund University, SE-221 00 Lund, Sweden (e-mail: Yi.Tan@eit.
lth.se; Buon_Kiong.Lau@eit.lth.se).
Z. Ying is with Research and Technology, Corporate Technology Office, Sony
Ericsson Mobile Communications AB, SE-221 88 Lund, Sweden (e-mail: Ying.
Zhinong@sonyericsson.com).
S. He is with the Center for Optical and Electromagnetic Research, Zhe-
jiang University, Hangzhou, 310058, China and also with School of Electro-
magnetic Engineering, Royal Institute of Technology, SE-100 44 Stockholm,
Sweden (e-mail: sailing@kth.se).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173438
0018-926X/$26.00 © 2011 IEEE
(WiMAX) and IEEE802.11n. The key advantage of MIMO is its
potential to linearly increase channel capacity with the number
of antennas at both the transmitter and receiver, without sac-
rificing additional frequency spectrum and transmitted power
[1]. However, implementing multiple antennas in compact
terminal devices such as mobile phones is challenging, since it
involves both practical and fundamental design tradeoffs [2].
Practical tradeoff considerations include the allocation of more
antenna locations on the terminal and an increased likelihood
of one or more antennas being detuned by the hand or head of
the user. Nevertheless, most attention in the area has been on
the fundamental aspect of closely spaced antennas resulting in
an increase in spatial correlation and mutual coupling, which in
turn degrade the performance of MIMO systems as measured
by metrics such as efficiency, bandwidth, diversity gain and
capacity [2]–[5].
Recent results indicate that, in a rich scattering environment,
the MIMO performance of closely spaced antennas can be im-
proved by decoupling multiple antennas, with the tradeoff being
a smaller bandwidth [2], [5]. Decoupling techniques that are
suitable for multiple antennas on small printed circuit boards
(PCBs) are presented in [6]–[10]. Unfortunately, most of these
techniques focus on the relatively high frequency bands, in-
cluding the WLAN, DCS1800 and UMTS bands, and they have
not been demonstrated for the mobile bands below 1 GHz, such
as GSM900 and WCDMA850. To our understanding, there are
several reasons for this. First, since the wavelength at 900 MHz
is twice as long as that of 1800 MHz, for the same PCB, the
available electrical distance between the antennas is only half
of that at 1800 MHz. This complicates the isolation of the an-
tennas. Second, some decoupling techniques, such as neutral-
ization line [6] and quarter-wavelength slot filter [7], base their
mechanisms on wavelength. If the frequency decreases, the di-
mensions of these decoupling structures will increase corre-
spondingly, and they can take too much space on the PCBs.
Even more importantly, the mobile chassis, which only func-
tions as a ground plane for the antenna elements at high fre-
quency band, becomes the main radiator at the low frequency
bands [11]. Thus, different antenna elements share the same ra-
diator, making isolation worse (e.g., the prototype in [12]). This
aspect will be further explained in this paper.
The above discussion reveals that the influence of the mo-
bile chassis becomes the most critical factor for multiple an-
tenna terminals at the low frequency bands. However, nearly
all existing studies of the impact of chassis on antenna design
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS
focus their attention on single antenna design, e.g., [13]–[16].
In [13], a detailed study shows that self-resonant antenna el-
ements can be replaced by smaller non-resonant antenna ele-
ments (or “coupling elements”) that utilize the chassis as the
radiating structure, thereby reducing the volume of the mobile
terminal antenna. Mobile chassis can also be used to enhance
the bandwidth of terminal antennas [14], [15] and create mul-
tiple resonances [16]. A recent simulation study [17] concludes
that the impedance, bandwidth and radiation mode of an antenna
on a ground plane is often defined by the location of the antenna
and its feeding point, rather than the size of the ground plane.
While useful insights are provided by these studies, the re-
sults are based on the single antenna case, and hence may not
be directly applicable to multiple antennas implemented on the
same chassis. Some results for the multiple antenna case are pre-
sented in [18], where an oscillation of correlation coefficient is
observed when two antennas with a fixed separation distance
move along a large PCB with a length of . However, while
the phenomenon is attributed to characteristic mode [19] (also
called ‘chassis mode’ in [15], [16]), no further analysis is given.
To analyze the mobile chassis, different tools can be utilized,
such as resonator based equivalent circuit [20], flat dipole equiv-
alent circuit [21] and theory of characteristic mode [22]. Among
these tools, a characteristic mode based analysis is an efficient
way to gain physical insight into fundamental electromagnetic
properties of mobile chassis and yield valuable information on
antenna design. In particular, since characteristic modes are in-
dependent of excitation, and only depend on the shape of the
chassis [23], the characteristic radiation properties can be ob-
tained from mode analysis.
The purpose of this paper is to provide a framework in un-
derstanding fundamental design tradeoffs of multiple antennas
on mobile chassis at low frequency bands through characteristic
mode analysis and antenna simulations. The main contributions
are:
• Using the theory of characteristic mode and supporting
simulations to give insights into the mechanisms that
govern the location dependent performance of multiple
antennas on a mobile chassis.
• Relating the ability of a given type of antenna (i.e.,
monopoles and planar inverted F antennas (PIFAs)) to
localize the excitation current on the chassis to the perfor-
mance of multiple antennas.
• Achieving good MIMO performance, including isolation,
efficiency and diversity gain, by taking advantage of the
characteristics of antennas on the chassis, without the need
for additional matching or decoupling structures.
• Providing tradeoff analyses for the performance of mul-
tiple antenna terminals with respect to the locations of the
antenna elements, in terms of bandwidth, efficiency, corre-
lation, diversity gain and capacity.
It should be emphasized that even though this paper focuses
on the 900 MHz band, it is only intended to highlight the prac-
ticality of our proposal for typical mobile terminals sizes. The
underlying principles are based on electrical dimensions rather
than absolute dimensions, and hence more general. In other
words, chassis radiation of existing or future mobile devices
491
with different chassis sizes is significant in other frequency
bands.
The paper is organized as follows: Section II briefly reviews
the theory of characteristic mode and its application in providing
fundamental characteristics of chassis radiation. The resonant
characteristic electric fields, the eigenvectors and the modal sig-
nificance (MS) [23] of a chassis are used to give helpful infor-
mation on antenna design. A slot monopole on the chassis is also
investigated using the characteristic mode analysis. Antenna
simulations are carried out using an electromagnetic (EM) sim-
ulator in Section III. First, the properties of single monopoles
and single PIFAs with different locations on the chassis are ex-
amined. The results indicate that, for a given location on the
chassis, different types of antennas localize the chassis excita-
tion currents to different extents. This insight can be used to im-
prove the isolation between multiple antenna elements. Second,
multiple antenna cases are studied to investigate the effects of
chassis current localization on isolation for different antenna
combinations. In Section IV, performance tradeoff analyses of
different antenna locations on the mobile chassis is presented
with respect to bandwidth, efficiency, correlation, effective di-
versity gain and capacity. Based on the analysis, three dual-an-
tenna prototypes were fabricated, and their experimental results
are presented in Section V. Finally, some conclusions are given
in Section VI.
II. CHARACTERISTIC MODE ANALYSIS
The mode of an oscillating structure is a pattern of motion in
which the entire structure oscillates sinusoidally with the same
frequency. The frequencies of the modes are known as their
natural frequencies or resonant frequencies. A physical object
has a set of modes that depend on its structure, materials and
boundary conditions.
The characteristic mode analysis is a method used in electro-
magnetics, which gives insight into the potential resonant char-
acteristics of a structure by finding and examining the inherent
modes of the structure. The existence of the modes is indepen-
dent of the excitation. However, different kinds of excitations or
excitations at different locations/frequencies are expected to ex-
cite different modes, to satisfy different requirements. From this
perspective, the characteristic mode analysis can provide phys-
ical insight into the fundamental electromagnetic properties of
scattering objects and valuable information on antenna design.
The theory of characteristic mode was first introduced by
Robert Garbacz in [24] and later refined by Roger Harrington
in [19] and [25].
Considering a conducting body with surface , an external
electric field (or voltage) can induce a surface current on
it. This surface current will further generate a scattered field .
According to the boundary condition, the tangential component
on the surface of the conducting body satisfies
(1)
Equation (1) can be rewritten with an operator
(2)
492 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Following the approach in [19], characteristic mode is defined

by the eigenvalue equation expressed as

(5)

In our work, this linear equation is transformed to matrix equa-

tion using the method of moments (MOM) [26]. The MOM
relies on Rao-Wilton-Glisson (RWG) edge elements. The sur-
face of the metal antenna is divided into separate triangles (see
Fig. 1(d)). Each pair of triangles, having a common edge, consti-
tutes the corresponding RWG edge element. These RWG edge
elements correspond to the division of the antenna into small el-
ementary electric dipoles. In this sense, the impedance matrix
describes the interaction between different elementary dipoles.
If the edge elements m and are treated as small electric dipoles,
the element describes the contribution of dipole to the
electric current of dipole , and vice versa. This contribution is
calculated through electric field integral equation (EFIE). The
size of impedance matrix is equal to the number of the edge ele-
ments. With the matrix equation, the characteristic eigenvalues
, eigenvectors and characteristic electric fields of the
chassis are calculated with Matlab.
The chassis, with the dimensions of a typical candybar-type
mobile phone (100 mm 40 mm), is modeled by a perfectly
conducting board. Its eigenvalues over a frequency range from
0.5 GHz to 1.5 GHz are calculated in Matlab, and shown in
Fig. 1(a). A scattered plot is employed to generate the figure
and the chassis is meshed with 736 edge elements. As known
from [19], the smaller the magnitude of , the more important
the mode is for radiation and scattering problems, and
corresponds to a resonant mode. As observed from Fig. 1(a), the
lowest resonant frequency of our chassis is 1.35 GHz. The mode
in our work is numbered according to the order of occurrence
of its resonant frequency. The lower the resonant frequency, the
smaller is the mode number. The mode number of zero (or )
denotes a non-resonant mode.
To highlight the respective roles of characteristic modes and
external excitation in chassis radiation, the total current on the
surface of a conducting body can be expressed by the eigenvec-
tors as

(6)

Fig. 1. (a) The eigenvalues against frequency for the mobile chassis; (b) the
There are two factors determining the contribution of a cer-
modal significance (MS) against frequency for the mobile chassis; (c) the nor- tain (or th) mode to the total current distribution [23], i.e., the
malized magnitude of the total electric field of the first characteristic mode of modal-excitation coefficient
the chassis at 1.35 GHz; (d) the normalized eigenvector of the first characteristic
mode of the chassis at 1.35 GHz.
(7)

where the term can be considered as the electric field

and the modal significance (MS)
intensity on the surface due to the surface current . The op-
erator has the dimension of impedance, and the following
notations are hence introduced (8)

Whereas accounts for the external excitation, including the

(3) influence of its position, magnitude and phase, MS represents
(4) the inherent normalized amplitude of the characteristic modes.
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS 493

External excitation (e.g., port excitation) will not change the

characteristic modes of the conducting body; however, its lo-
cation on the structure is very important for the excitation of
certain characteristic mode(s).
The MS of each mode of the chassis is presented in Fig. 1(b).
It can be observed that the first mode is dominant at 1.35
GHz, while the other modes only contribute slightly to the cur-
rent distribution (with ). Thus, at the frequency band
around 1 GHz, we focus mainly on the first mode of the chassis
and its interaction with the antenna on the chassis.
The characteristic total electric field of the first mode on a
plane 5 mm above the chassis is evaluated at 1.35 GHz, and
shown in Fig. 1(c). The field is normalized to the maximum
value and presented in dB scale. The field corresponds to that
of a flat half-wavelength dipole [21]. It can be observed that
the electric field is stronger at the edges, especially at the cor-
ners, whereas it becomes very weak at the center of the chassis.
For the single antenna case, this insight reveals that the char-
acteristic mode can be most efficiently excited when a voltage
(a port excitation) is applied at the edge or at the corner. How-
ever, if multiple antennas at the same or closely similar frequen-
cies are integrated on the same chassis, the locations of the an-
tennas should be carefully considered. Concerning the mutual
coupling, if the antenna (e.g., a dipole) is responsive to electric
field, the place where the electric field is strong is not a good lo-
Fig. 2. (a) The eigenvalues against frequency for the mobile chassis with a slot
cation for more than one such antenna. On the other hand, if the monopole; (b) The normalized magnitude of the total electric field of the first
antenna (e.g., a small loop) responds mainly to magnetic field, characteristic mode of the slot monopole at 1.06 GHz.
the place with strong magnetic field is likewise not a good lo-
cation for more than one such antenna. Since most types of mo-
bile terminal antennas base their mechanisms on electric field, that of the chassis-only case in Fig. 1(c). Due to the resonance
we focus on electric field in this paper. of the slot monopole at 1.06 GHz, the positions of maximum
The normalized eigenvector (i.e., characteristic current) and minimum electric fields are slightly changed. The max-
of the first characteristic mode of the chassis is presented in imum electric field occurs at the edge with the slot monopole,
Fig. 1(d). A sinusoidal current distribution along the length and the minimum value shifts slightly from the center towards
of the chassis is observed, which shows that the length of the the monopole side. From the perspective of isolation, the best
chassis is approximately one half of a wavelength at the first location for another antenna should be in the region of the
resonant frequency. This current distribution is similar to that minimum electric field so that the characteristic mode will
calculated in the EM simulator for an excited monopole on the not be shared simultaneously by two antenna elements. This
chassis (see Figs. 7(a) and 8(c)). principle will be further analyzed and verified by the antenna
Once antennas are implemented on the chassis, the electrical simulations in the following section.
length of the chassis is correspondingly increased, and its res- The chassis with slot monopole is also simulated in the
onant frequency will be further reduced, making it closer to frequency domain solver of CST Microwave Studio, with the
900 MHz. To verify this hypothesis, the eigenvalues of the 100 monopole excited by a lumped port. The total electric field is
mm 40 mm chassis with a planar slot monopole [27] etched similar to the characteristic electric field in Fig. 2(b), and is
in it are calculated in Matlab and shown in Fig. 2(a). The dimen- thus not included here.
sions of the slot monopole are presented later in the paper (see
Fig. 8(b)). The lowest resonant frequency of 1.06 GHz is ob- III. ANTENNA SIMULATION
served in the figure. Moreover, when the antenna is excited with In this section, according to the results of the characteristic
a feed, its resonant frequency can be slightly changed. Another mode analysis, the excitation ports are applied to different loca-
observation from Figs. 1(a) and 2(a) is that the first resonance tions of the chassis. Full-wave antenna simulations are carried
of the chassis, with and without the slot monopole element, is out in the frequency domain using the CST Microwave Studio
relatively wideband, which appears to substantiate the findings software. Single monopole antenna and single PIFA cases are
of [17] that the chassis size is often less important than the loca- studied first, to identify the degree to which the radiation prop-
tions of the antenna element and its feed in determining single erties are influenced by different antenna types and locations.
antenna performance. Our choice is based on the fact that PIFAs and low profile vari-
The characteristic total electric field of the first mode of ants of monopole antennas are the most commonly used antenna
the slot monopole is evaluated at 1.06 GHz and presented in types in today’s mobile terminals. Based on these results, mul-
Fig. 2(b). As expected, the trend of the electric field is similar to tiple antenna cases are then analyzed. The size of the mobile
494 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 3. Geometry of quarter-wavelength top-loaded monopole and its locations

on the chassis in the single antenna case. The dimensions are: mm,
mm, mm. Diameter of the wire is 1 mm.

chassis is 100 mm 40 mm, which is identical to that in the

previous section. The chassis consists of a 0.1 mm top copper
layer and a 1.55 mm bottom FR4 layer with a permittivity of 4.7
and a loss tangent of 0.015. All antennas, except the top-loaded
monopoles, are designed for dual-band (900/1900 MHz) oper-
ation. However, the focus of the study in this and the following
sections are on the 900 MHz band. Note that even though the
dimensions of the monopoles and PIFAs used in the paper are
provided, they are specific to a given antenna location and are
slightly tuned to ensure good impedance matching (i.e., reflec-
tion coefficient dB) for other locations.

A. Single Antenna Cases

In this sub-section, four single antenna cases are studied:
quarter-wavelength top-loaded monopole at the center (MC)
or edge (ME) of the chassis, and PIFA at the center (PC) or Fig. 4. (a) Input impedances and (b) reflection coefficients of the simulated
edge (PE) of the chassis. The geometry of the monopole and top-loaded monopole at different locations with finite/infinite ground plane.
its location on the mobile chassis are shown in Fig. 3. Perfect
electric conductor (PEC) is assumed as the antenna material.
The top load (i.e., a circular plate with thickness of 0.1 mm) is
used to help match the monopole and slightly reduce its height,
without changing the radiation characteristics. The monopole
is first put at the edge of the chassis, and then moved to the
center with its dimension unchanged. The same monopole
antenna on an infinite ground plane (MIG) is also presented for
comparison. All the antennas are well matched at 0.92 GHz.
Real and imaginary parts of the simulated input impedance
and the magnitude of the reflection coefficients are shown in
Fig. 4(a) and (b), respectively. As can be seen in Fig. 4(a), the
input impedances of MC and MIG share a similar trend, whereas
that of ME is quite different. As explained in the previous sec-
tion, the characteristic mode for the eigenvalue is easily ex-
cited when the antenna is located at the edge of the chassis [13],
since the resonance of chassis is close to 0.92 GHz. The high Fig. 5. Geometries of the PIFA and its locations on the chassis in the single
antenna case. The dimensions are: mm, mm,
radiation resistance in the ME case is mainly due to the excita- mm, mm, mm, mm, mm,
tion of the characteristic mode, which increases the bandwidth mm, mm, mm.
dramatically (see Fig. 4(b)).
The PIFA cases are studied in a similar way. The geometry
and different locations of the single PIFA are shown in Fig. 5. from that of the characteristic mode, one more case is included,
The PIFA is integrated onto a hollow carrier (i.e., the shaded i.e., the PIFA in the center rotated by 90 and moved to the
regions), which is commonly used in mobile phones. The simu- longer edge of the chassis (RPC), as illustrated in Fig. 5(b). A
lated carrier has a thickness of 1 mm, a permittivity of 2.7 and a PIFA on an infinite ground plane (PIG) is used for comparison.
loss tangent of 0.007. To further separate the effect of the edge From Fig. 6(a), it can be observed that the input impedance of
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS
Fig. 6. (a) Input impedances and (b) reflection coefficients of the simulated
PIFA at different locations with finite/infinite ground plane.
PE is notably different from all other cases. With the help of the
chassis, both the input resistance and reactance become larger in
amplitude and flatter over frequency. As a result, the bandwidth
is significantly improved (see Fig. 6(b)).
To confirm that the larger bandwidth of PE is primarily the
result of characteristic mode excitation, rather than due to the
PIFA being at the chassis edge, the effect of the edge can be ex-
amined by comparing the bandwidths of RPC and PC. Since the
difference in bandwidth between the two cases is negligible, it
can be concluded that the edge effect is unlikely to have con-
tributed to the wideband behavior in the PE case.
In order to gain further insights into the influence of chassis
on the characteristics of monopole and PIFA at different loca-
tions, normalized current distributions are given for four distinct
cases in Fig. 7. The normalization is performed against the peak
current in each case. Two conclusions can be drawn from the
current distributions. First, when the antenna is at the short edge,
regardless of its type, the characteristic mode of the chassis is
excited (see Fig. 7(a) and (c)), and the current is distributed
over the whole chassis. This current distribution is similar as
the eigenvector of the first characteristic mode (see Fig. 1(d)),
especially for the ME case. This further verifies the strong exci-
tation of the first characteristic mode. When the antenna is at the
center, the current is more confined to the immediate vicinity of
the antenna. By comparing ME and MC in Fig. 4(b) and PE and
495
Fig. 7. Normalized magnitude of current distributions for: (a) monopole at the
edge, (b) monopole at the center, (c) PIFA at the edge, (d) PIFA at the center.
PC in Fig. 6(b), the characteristic mode excitation is observed to
offer significantly larger bandwidths for the ME and PE cases
than for the MC and PC cases, respectively. Second, the cur-
rent of PIFA is more localized than that of monopole at either
the center or the short edge. In other words, its radiation de-
pends less on the chassis than that of the monopole. Therefore,
in addition to the concept of characteristic mode, localization
of chassis current is also important in determining the isolation
between antenna elements: The more localized the induced cur-
rent on the chassis, the less is the current that couples from one
antenna port into other antenna port(s). Nonetheless, it should
be noted that an antenna with more localized current is an in-
dication of a smaller effective radiator, which inevitably leads
to a reduction of bandwidth. This aspect can be seen in the nar-
rower bandwidths of the PC and PE cases, relative to the MC
and ME cases, respectively. Further analysis of the localized
current phenomenon is provided in Section IV.
B. Dual-Antenna Cases
In this subsection, different combinations of antenna types,
including monopole-monopole (MM), PIFA-PIFA (PP) and
monopole-PIFA (MP), and antenna locations (at the edge(s)
and at the center) are studied to shed light on the effect of
characteristic mode and current localization on the isolation
level between two antenna elements on the chassis.
In all the simulations, two antennas (of monopole(s) and/or
PIFA(s)) are integrated onto the same chassis of dimensions
100 mm 40 mm. One antenna is fixed at one short edge, and
the other antenna is placed either at the opposite short edge
or at the center. A planar slot monopole [27] rather than the
top-loaded monopole is used as the fixed monopole at the edge,
considering antenna dimensions and matching problem. The
schematic drawing of the antenna setup and the geometry of
the slot monopole are shown in Fig. 8. The slot monopole is
etched into the ground plane on a substrate of FR4. It is fed
by a microstrip line (i.e., the dashed line in Fig. 8(b)) on the
other side of the substrate. The width of both of the slots is
mm. Good antenna matching is achieved by opti-
mizing the value of . The radiation pattern and polarization
of the slot monopole are only slightly different from those of the
top-loaded monopole when they are implemented on the mo-
bile chassis, since they both strongly excite the chassis, which
acts as a radiator. The normalized current distribution when the
slot monopole is excited is shown in Fig. 8(c). It is observed
496 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 8. (a) The schematic drawing of the locations of antennas in the dual-an-
tenna case. (b) The geometry of the slot monopole. The dimensions are:
mm, mm, mm, mm, mm,
mm, mm, mm, mm. (c) The normalized cur-
rent distribution of the slot monopole on the chassis.

that the degree of chassis excitation is similar as that when the

top-loaded monopole is used (Fig. 7(a)).
The simulation results are shown in Fig. 9. It is interesting
to note that, in all dual-antenna cases, the isolation improves
when one antenna is moved to the center (i.e., the two antennas
become closer). This phenomenon contradicts with intuition
and the common knowledge on the relationship between an-
tenna separation distance and port isolation. However, this
situation can be explained by the characteristic mode analysis
in Section II and the single antenna simulations above. When
the antenna elements are at the two edges, they excite the
chassis simultaneously. Since the chassis not only functions as
a ground plane, but also as the main radiator for both antenna
elements, the port isolation must be low. The mutual coupling,
in this case, not only comes from the field in free space and the
conventional ground plane current, but also from the radiation
of the shared chassis. Thus, it is difficult to achieve angle and
polarization diversities for the antenna elements in this setup.
Fig. 9. The simulated scattering parameters of dual-antenna terminals with dif-
ferent locations and combinations of antenna elements: (a) M-M combination,
(b) M-P combination, with PIFA at the edge or the center of the ground plane,
and (c) P-P combination.
When one antenna is moved to the center, the chassis is not
efficiently excited, and hence the current is more localized.
The chassis is only utilized as the main radiator by the antenna
(either the monopole or the PIFA) at the chassis edge. Conse-
quently, angle and polarization diversities can be more easily
achieved for the edge-and-center placement, which enhances
the isolation.
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS 497

Fig. 10. Normalized magnitude of current distributions for the M-P case: (a)
PIFA at the edge, monopole excited, (b) PIFA at the edge, PIFA excited, (c)
PIFA at the center, monopole excited, (d) PIFA at the center, PIFA excited.

It is also observed that different dual-antenna combinations

offer different degrees of improvement in isolation, when con-
sidering different placement options of a given combination.
The most dramatic improvement is achieved in the M-P combi-
nation, in which the monopole is the fixed antenna and the PIFA
is either at the edge or the center of the ground plane. This im-
provement is mainly due to the localized current achieved by
the PIFA when it is at the center. Another reason is that em-
ploying antennas of different types can reduce mutual coupling
to some extent by taking advantage of angle diversity in their
radiation patterns. The improvement in the isolation of the P-P
combination is better than that of the M-M combination, and
this is attributed to the localized current behavior of the PIFAs,
as shown in Fig. 7(c) and (d) for the single antenna cases.
By taking into consideration both characteristic mode and
current localization, isolation of over 10 dB is achieved in the
M-P combination, which can be considered low enough for ter-
minal applications involving frequencies lower than 1 GHz. In
addition, the antennas are easy to design and tune, since no ad-
ditional matching or decoupling structures are needed.
From the perspective of bandwidth, the monopole antenna
performs well, whereas the bandwidth of the PIFA is narrow, es-
pecially when it is placed at the center of the chassis. Thus, the
P-P combination is impractical, even though the isolation is im-
proved for the center-and-edge placement, in comparison to the
edge-and-edge placement. The M-M combination is likewise
impractical, due to the difficulties in implementing a low pro-
file monopole at the center: Its dimensions tend to be large and it
is difficult to achieve good matching. In addition, the isolation
between the monopoles is only 7 dB at the center frequency.
Consequently, the M-P combination is more attractive for mo-
bile terminal applications. As has been suggested in [12], the
monopole can be used as the main antenna to cover both down-
link and uplink frequencies, whereas the narrowband PIFA can
be used as a diversity antenna for only the downlink frequen-
cies.
Fig. 11. Performance tradeoffs of the dual-antenna terminal with respect to the
PIFA location in terms of the (a) relative bandwidth, (b) average efficiency, (c)
C. Discussions magnitude of complex correlation coefficient and average EDG, and (d) average
channel capacity.
For the M-P combination, normalized current distributions
are shown in Fig. 10. In Fig. 10(a) and (b), when the two an-
tennas are at the edges, the characteristic mode is easily excited, one antenna is excited, the other antenna is also strongly ex-
resulting in strong electric fields at the two edges. Therefore, if cited, which leads to poor isolation. In Fig. 10(c), the PIFA is
498 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 12. Prototypes with the slot antenna at one edge and (a) the PIFA at the
opposite edge, (b) the PIFA at the center, (c) the rotated PIFA at the center.

influenced to some extent, since the characteristic mode is ef-

ficiently excited by the monopole; whereas in Fig. 10(d), when
the PIFA is excited, due to its localized current, the chassis acts
primarily as a common ground rather than a radiator. Thus, good
isolation is achieved.
Rather than only the slot monopole, different monopole
types, including the most frequently used folded monopole
(such as the monopole in [12]), have also been simulated.
The trend of isolation enhancement is the same when the
PIFA moves away from the edge to the center location. The
slot monopole is used in our work due to the convenience of
fabrication and matching. Concerning the bandwidth of PIFA,
we note that the PIFA can be made tunable to cover different
bands according to different requirements.

IV. TRADEOFF ANALYSIS OF DUAL-ANTENNA TERMINALS

In this section, the performance tradeoffs of dual-antenna ter-
minals are investigated with respect to different locations of the
PIFA on the chassis for the M-P combination in Section III.B.
The different PIFA locations are meant to induce different levels
of characteristic mode excitation. The PIFA is moved gradually
from the edge to the center, in steps of 5 mm. is the distance
between the PIFA location and the edge (see Fig. 8(a)). When
the PIFA is at the edge, mm.
The relative bandwidth of the two antennas is shown in
Fig. 11(a). Here, the relative bandwidth is defined as the ratio
of the 6 dB impedance bandwidth to the center frequency. It Fig. 13. Measured scattering parameters for the slot antenna at one edge and
is obvious that the bandwidth of the monopole is much wider (a) the PIFA at the opposite edge (b) the PIFA at the center and (c) the rotated
than that of the PIFA, and it is almost constant regardless of the PIFA at the center.
PIFA’s location. That is one reason that the monopole is used as
the main radiator in the M-P combination. The relative band-
width of the PIFA falls quickly when it is moved away from the lyzed first. The highest radiation efficiency of the PIFA appears
edge, since the chassis no longer contributes significantly to the at the edge and decreases by 15% when it is moved to the center.
PIFA’s radiation. Indeed, it is observed that the bandwidth this This is because the chassis excitation helps to increase the radi-
almost unchanged when the PIFA is moved around the center ation resistance of the PIFA at the edge (see Fig. 6(a)).
location. According to expression
The efficiencies of both antennas, including the radiation ef-
ficiency at the center frequency and the average total efficiency (9)
over a given bandwidth, are presented in Fig. 11(b). Overall, the
efficiencies of both antennas increase when the PIFA is moved where is the radiation resistance and represents the
from the edge to the center. The radiation efficiency is ana- conduction-dielectric losses (which is almost constant over
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS
Fig. 14. Simulated and measured antenna patterns for the antenna system in
Fig. 12(a): (—) simulated measured , ( ) simulated
, ( ) measured . (a) Monopole, , (b) Monopole,
, (c) Monopole, , (d) PIFA, , (e) PIFA,
, (f) PIFA, .
the given range of frequency), efficiency can be high if the
radiation resistance is large. For the monopole, the radiation
efficiency does not change appreciably, since it always takes
advantage of the chassis to radiate efficiently.
The total efficiency is given by
(10)
When the PIFA is at the edge, the total efficiencies are rela-
tively low for both antennas, due to strong mutual coupling
(i.e., large ). As the PIFA moves away from the edge, the
total efficiency of the monopole increases greatly, regardless of
the bandwidth within which it is calculated. The total efficiency
of the PIFA increases as it is moved towards the center of the
chassis, if it is only measured at the center frequency. How-
ever, the trend changes when it is measured within a 30 MHz
bandwidth, because of its narrowing impedance bandwidth with
(see Fig. 11(a)). Therefore, the optimal position of the PIFA
can be different, depending on the efficiency bandwidth require-
ment.
499
Fig. 15. Simulated and measured antenna patterns for the antenna system in
Fig. 12(b): (—) simulated measured , ( ) simulated
, ( ) measured . (a) Monopole, , (b) Monopole,
, (c) Monopole, , (d) PIFA, , (e) PIFA,
, (f) PIFA, .
Correlation coefficient and diversity gain are important met-
rics for evaluating the performance of multiple antenna systems.
Fig. 11(c) presents the magnitude of complex correlation coef-
ficients at the center frequency, together with the average effec-
tive diversity gain (EDG) over the given bandwidths. EDG [12]
is defined by
(11)
where is the total efficiency of the antenna with the
highest efficiency and DG is the (apparent) diversity gain. In
this paper, DG is calculated with the maximum ratio combining
(MRC) method and taken at 1% probability. All the EDGs (over
different bandwidths) improve as the PIFA is moved towards the
center. Though the improvement becomes less obvious when
calculated over a larger bandwidth, a minimum enhancement of
2.3 dB can be observed within a bandwidth of 30 MHz. The
smallest correlation coefficient of 0.11 is achieved when
mm, which corresponds to the PIFA structure at nearly the
center location of the chassis.
The average channel capacity calculated under the equal
power (EP) and water-filling (WF) conditions for
500 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

dB is presented in Fig. 11(d). The WF procedure is performed

over the antenna elements. The Kronecker model and uni-
form 3D angular power spectrum (APS) is assumed. There is
no correlation between the (base station) transmit antennas,
whereas the (mobile terminal) receive antennas are correlated
according to their radiation patterns and the 3D APS. The
capacity is averaged over 10 000 i.i.d. Rayleigh realizations
at each frequency. The channels are normalized with respect
to the i.i.d. Rayleigh case, which means that the correlation,
total efficiency and power imbalance (efficiency imbalance)
are taken into account in the capacity evaluation. As refer-
ence cases, the average capacities for the 2 2 i.i.d. Rayleigh
channel with the EP and WF schemes are 11.29 bits/s/Hz and
11.32 bits/s/Hz, respectively.
Similar to the figure of EDG, the largest capacity is achieved
when mm, due to the low correlation and high efficiency
of the monopole when the PIFA is at the center. Because of
the PIFA’s narrow bandwidth, the power imbalance becomes se-
rious at frequencies away from the center frequency. This means
that one antenna (or spatial channel) is not efficiently used, and
thus the average channel capacity decreases. In general, the av-
erage capacity increases when the PIFA is moved to the center,
and the improvement is less obvious with an increase in band-
width. For the WF case, the power imbalance is accounted for in
the transmit power allocation, so the channel is more efficiently
used than in the EP case. Thus, the capacity improvement of WF
over EP increases with power imbalance, which increases with
bandwidth.

V. EXPERIMENTS AND DISCUSSIONS

The PIFAs and monopoles used by the dual-antenna pro-
totypes in Section IV are dual-band antennas and they cover
both 880 MHz–960 MHz and 1880 MHz–1990 MHz frequency
bands. Three prototypes are fabricated for the slot monopole
at one edge and (i) the PIFA at the opposite edge, (ii) the
PIFA at the center, and (iii) the rotated PIFA at the center (see
Fig. 12). The scattering (or S) parameters are measured with a
vector network analyzer and shown in Fig. 13. The isolation is
improved from 5 dB to 13 dB when the PIFA is moved to the
center. As a tradeoff, the bandwidth of the PIFA is reduced from
30 MHz to 12 MHz. In practice, PIFA can be made tunable to
cover different bands according to the given requirement. The
scattering parameters are almost unchanged when the PIFA is
rotated by 90 degrees.
The far field electric field patterns are measured in a Satimo
Stargate-64 antenna measurement facility. For cases (i) and (ii),
the patterns at the center frequency of the low band are shown
in Figs. 14 and 15, respectively. The patterns of the case (iii)
(as illustrated in Fig. 12(c)) are not included here, because the
pattern of the monopole is similar to that of the monopole in
Fig. 12(b), and the pattern of PIFA is similar to that of the PIFA
in Fig. 12(b) after a 90 rotation. The measured patterns agree
well with the simulated ones. The slight differences are caused
by influences of the feeding cables. The correlations calculated
from the measured patterns are 0.5, 0.18, and 0.19, respectively,
for the three prototypes in Fig. 12. Due to some cable influence
[12] and practical difficulties in measuring antennas with very
high correlation, the measured correlation for the case with the
Fig. 16. Measured efficiencies (a) low frequency band (b) high frequency band
for the slot monopole at one edge (ME) and (i) the PIFA at the opposite edge
(PE) (ii) the PIFA at the center (PC) and (iii) the rotated PIFA at the center
(RPC).
antennas at the edges (i.e., Fig. 12(a)) is slightly lower than the
simulated one.
The measured efficiencies over two operating bands are
shown in Fig. 16. In general, the measured efficiencies are
slightly lower than the simulated ones due to fabrication and
experimental tolerances. At the low frequency band, when
the PIFA is at the edge, the efficiencies of the monopole
are relatively low around the center frequency due to high
mutual coupling. When the PIFA is at the center, the effi-
ciencies of monopole approach that of a single monopole on
mobile chassis, whereas the PIFA efficiency becomes more
narrowband. At the high frequency band, the efficiencies of the
monopole are around 70% (-1.5 dB), since the coupling is not
significant. The highest efficiency of the PIFA is 75% (-1.2 dB),
and good efficiency is kept in downlink band. All these results
agree well with simulations.
VI. CONCLUSION
In this work, fundamental design tradeoffs of multiple an-
tennas on a mobile chassis are studied in the context of char-
acteristic mode excitation and the ability of antennas to localize
chassis currents. The goal is to provide useful information and a
LI et al.: CHARACTERISTIC MODE BASED TRADEOFF ANALYSIS OF ANTENNA-CHASSIS INTERACTIONS
design framework for optimal implementations of multiple an-
tennas on a mobile chassis according to different requirements.
The results for the 900 MHz band show that whereas the PIFA
has more localized currents than the monopole, especially when
it is at the center of the chassis, it is the monopole-PIFA combi-
nation that achieves the best isolation of over 10 dB (13 dB for
the measured case). Utilizing characteristic mode and chassis
current localization in the design of multiple antennas has the
advantage of not requiring any additional matching or decou-
pling structures. Three prototypes are fabricated and measured
to test three selected cases, and the results are found to be in
good agreement with those from simulations.
Since a mobile terminal user can significantly influence the
results obtained in this study, the effects of user on antenna-
chassis interaction is an interesting topic for future work.
ACKNOWLEDGMENT
The authors are thankful to Prof. J. B. Andersen of Aalborg
University and Prof. A. Karlsson of Lund University for helpful
discussions.
REFERENCES
[1] M. A. Jensen and J. W. Wallace, “A review of antennas and propagation
for MIMO wireless communications,” IEEE Trans. Antennas Propag.,
vol. 52, no. 11, pp. 2810–2824, Nov. 2004.
[2] B. K. Lau, , C. Oestges, A. Sibille, and A. Zanella, Eds., “Multiple
antenna terminals,” in MIMO: From Theory to Implementation. San
Diego, CA: Academic Press, 2011, pp. 267–298.
[3] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wireless Per-
sonal Commun., vol. 6, pp. 311–335, Mar. 1998.
[4] Z. Ying and D. Zhang, “Study of the mutual coupling, correlations and
efficiency of two PIFA antennas on a small ground plane,” in Proc.
IEEE Antennas Propag. Soc. Int. Symp., Washington, DC, July 2005,
pp. 305–308.
[5] B. K. Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch, “Impact
of matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag., vol. 54, no. 11, pp. 3225–3238, Nov. 2006.
[6] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas, “En-
hanced two-antenna structures for universal mobile telecommunica-
tions system diversity terminals,” IET Microw. Antennas Propag., vol.
2, no. 1, pp. 93–101, 2008.
[7] H. Li, J. Xiong, and S. He, “A compact planar MIMO antenna system of
four elements with similar radiation characteristics and isolation struc-
ture,” IEEE Antennas Wireless Propag. Lett., vol. 8, pp. 1107–1110,
2009.
[8] Y. Gao, X. Chen, Z. Ying, and C. Parini, “Design and performance
investigation of a dual-element PIFA array at 2.5 GHz for MIMO ter-
minal,” IEEE Trans. Antennas Propag., vol. 55, no. 12, pp. 3433–3441,
Dec. 2007.
[9] B. K. Lau and J. B. Andersen, “Simple and efficient decoupling of com-
pact arrays with parasitic scatterers,” IEEE Trans. Antennas Propag.,
submitted for publication.
[10] S. Dossche, S. Blanch, and J. Romeu, “Optimum antenna matching to
minimize signal correlation on a two-port antenna diversity system,”
Elect. Lett., vol. 40, no. 19, pp. 1164–1165, Sep. 2004.
[11] K. Solbach and C. T. Famdie, “Mutual coupling and chassis-mode cou-
pling small phases array on a small ground plane,” presented at the
Eur. Conf. Antennas Propag. (EuCAP), Edinburgh, U.K., Nov. 11–16,
2007.
[12] V. Plicanic, B. K. Lau, A. Derneryd, and Z. Ying, “Actual diversity
performance of a multiband diversity antenna with hand and head ef-
fects,” IEEE Trans. Antennas Propag., vol. 57, no. 5, pp. 1547–1556,
May 2009.
[13] J. Villanen, J. Ollikainen, O. Kivekas, and P. Vainikainen, “Coupling
element based mobile terminal antenna structure,” IEEE Trans. An-
tennas Propag., vol. 54, no. 7, pp. 2142–2153, July 2006.
501
[14] T. Taga and K. Tsunekawa, “Performance analysis of a built-in planar
inverted-F antenna for 800 MHz band portable radio units,” IEEE J.
Select. Areas Commun., vol. ACE-5, no. 5, pp. 921–929, 1987.
[15] W. L. Schroeder, A. A. Vila, and C. Thome, “Extremely small wide-
band mobile phone antenna by inductive chassis mode coupling,” in
Proc. 36th Eur. Microw. Conf., Manchester, U.K., Sep. 10–15, 2006,
pp. 1702–1705.
[16] W. L. Schroeder and C. T. Famdie, “Utilisation and tuning of the
chassis modes of a handheld terminal for the design of multiband
radiation characteristics,” in Proc. IEE Wideband Multiband Antennas
and Arrays, Sep. 7, 2005, pp. 117–122.
[17] S. R. Best, “The significance of ground-plane size and antenna location
in establishing the performance of ground-plane-dependent antennas,”
IEEE Antennas Propag. Mag., vol. 51, no. 6, pp. 29–42, Dec. 2009.
[18] C. Luxey and D. Manteuffel, “Highly-efficient multiple antenna
-system for small MIMO devices,” presented at the Int. Workshop
Antenna Technol. (IWAT), Lisbon, Portugal, Mar. 1–3, 2010.
[19] R. F. Harrington and J. R. Mautz, “Theory of characteristic modes for
conducting bodies,” IEEE Trans. Antennas Propag., vol. AP-19, no. 5,
pp. 622–628, Sep. 1971.
[20] P. Vainikainen, J. Ollikainen, O. Kivekas, and I. Kelander, “Res-
onator-based analysis of the combination of mobile handset antenna
and chassis,” IEEE Trans. Antennas Propag., vol. 50, no. 10, pp.
1433–1444, Oct. 2002.
[21] U. Bulus, C. T. Famdie, and K. Solbach, “Equivalent circuit mod-
eling of chassis radiator,” presented at the German Microw. Conf.
(GeMIC2009), Munich, Germany, Mar. 16–18, 2009.
[22] C. T. Famie, W. L. Schroeder, and K. Solbach, “Numerical analysis
characteristic modes on the chassis of mobile phones,” presented at the
Eur. Conf. Antennas Propag. (EuCAP), , France, Nov. 6–10, 2006.
[23] M. C. Fabres, E. A. Daviu, A. V. Nogueiram, and M. F. Bataller, “The
theory of characteristic modes revisited: A contribution to the design
of antennas for modern applications,” IEEE Trans. Antennas Propag.
Mag., vol. 49, no. 5, pp. 52–68, Oct. 2007.
[24] R. J. Garbacz, “A generalized expansion for radiated and scattered
field,” IEEE Trans. Antennas Propag., vol. AP-19, pp. 662–668, May
1971.
[25] R. F. Harrington and J. R. Mautz, “Computation of characteristic
modes for conducting bodies,” IEEE Trans. Antennas Propag., vol.
AP-19, no. 5, pp. 629–639, Sep. 1971.
[26] S. N. Makarov, Antenna and EM Modeling With MATLAB. New
York: Wiley-interscience, 2002.
[27] C. I. Lin and K. L. Wong, “Printed monopole slot antenna for internal
multiband mobile phone antenna,” IEEE Trans. Antennas Propag., vol.
55, no. 12, pp. 3690–3697, Dec. 2007.
Hui Li (S’08) received the Bachelor’s degree in
optical engineering from Tianjin University (TJU),
China, in 2007. She is currently working towards the
Ph.D. degree at the Royal Institute of Technology
(KTH), Sweden.
From 2007 to 2009, she pursued a Ph.D. degree
at the Joint Center (with KTH) for Optical and Elec-
tromagnetic Research, Zhejiang University (ZJU),
China. In 2010, she was awarded the EMECW
(Erasmus Mundus External Cooperation Window)
scholarship and visited the Department of Electrical
and Information Technology at Lund University for 13 months. In 2011, she
received the CSC (Chinese Scholarship Council) scholarship, and continued her
Ph.D. study in KTH. Her current research interests include compact antennas in
MIMO systems, antenna-user interactions, reconfigurable antennas, and RFID
antennas in wireless communications.
Yi Tan received the Bachelor’s degree in China, in
2003. He is currently working towards the Master’s
degree at Lund University, Sweden.
During 2004–2007, he was an RF Engineer in
Laird Technologies, Beijing, where he worked on
mobile antenna design and production. In 2008,
he started working towards a Master degree in
Lund University. In 2010, he took a year off from
his Master study to work as a Project Assistant
at the Department of Electrical and Information
Technology, Lund University.
502 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Buon Kiong Lau (S’00–M’03–SM’07) received the
B.E. degree (with honors) from the University of
Western Australia, Perth, Australia and the Ph.D.
degree from Curtin University of Technology, Perth,
in 1998 and 2003, respectively, both in electrical
engineering.
During 2000 to 2001, he worked as a Research En-
gineer with Ericsson Research, Kista, Sweden. From
2003 to 2004, he was a Guest Research Fellow at
the Department of Signal Processing, Blekinge Insti-
tute of Technology, Sweden. Since 2004, he has been
at the Department of Electrical and Information Technology, Lund University,
where he is now an Associate Professor. He has been a Visiting Researcher at
the Department of Applied Mathematics, Hong Kong Polytechnic University,
China, Laboratory for Information and Decision Systems, Massachusetts In-
stitute of Technology, and Takada Laboratory, Tokyo Institute of Technology,
Japan. His primary research interests are in various aspects of multiple antenna
systems, particularly the interplay between antennas, propagation channels and
signal processing
Dr. Lau is an Associate Editor for the IEEE TRANSACTIONS ON ANTENNAS
AND PROPAGATION and a Guest Editor of the 2012 Special Issue on MIMO
Technology for the same journal. From 2007 to 2010, he was a Co-Chair of
Subworking Group 2.2 on “Compact Antenna Systems for Terminals” (CAST)
within EU COST Action 2100. Since 2011, he is a Swedish national delegate
and the Chair of Subworking Group 1.1 on “Antenna System Aspects” within
COST IC1004.
Zhinong Ying (SM’05) is an expert of antenna tech-
nology in the Network Research Lab., Technology
office, Sony Ericsson Mobile Communication AB,
Lund, Sweden. He joined Ericsson AB in 1995. He
became Senior Specialist in 1997 and Expert in 2003
in his engineer career at Ericsson. His main research
interests are small antennas, broad and multi-band
antenna, multi-channel antenna (MIMO) system,
near-field and human body effects and measurement
techniques. He has been Guest Professor at Zhejiang
University, China since 2002. He has authored and
coauthored over 80 papers in various of journal, conference and industry
publications. He holds more than 70 patents and pending in the antenna and
mobile terminal areas. He contributed a book chapter to the well known Mobile
Antenna Handbook 3rd edition. He invented and designed various types of
multi-band antennas and compact MIMO antennas for the mobile industry.
One of his contributions in the 1990s is the development of non-uniform helical
antenna. The innovative designs are widely used in mobile terminal industry.
His patented designs have reached a commercial penetration of more than
several hundreds million products in worldwide.
Mr. Ying received the Best Invention Award at Ericsson Mobile in 1996
and Key Performer Award at Sony Ericsson in 2002. He was nominated for
President Award at Sony Ericsson in 2004 for his innovative contributions. He
served as TPC Co-Chairmen in International Symposium on Antenna Tech-
nology (iWAT), 2007, and served as session organizer of several international
conferences including IEEE APS, and a reviewer for several academic journals.
He was a member of scientific board of ACE program (Antenna Centre of Ex-
cellent in the European 6th Framework Programme) from 2004 to 2007.
Sailing He (M’92–SM’98) received the Licentiate
of Technology and the Ph.D. degree in electromag-
netic theory from the Royal Institute of Technology
(KTH), Stockholm, Sweden, in 1991 and 1992,
respectively.
Since then he has worked at KTH, Stockholm,
Sweden, as an Assistant Professor, an Associate
Professor, and a Full Professor. He is also with
Zhejiang University (ZJU, China) as a Distinguished
Professor of a special program organized by the
central government of China, as well as a joint
research center between KTH and ZJU. His current research interests include
electromagnetic metamaterials, optoelectronics, microwave photonics and
biomedical applications. He has first-authored one monograph (Oxford Univer-
sity Press) and authored/coauthored about 400 papers in refereed international
journals. He has given many invited/plenary talks in international conferences,
and has served in the leadership for many international conferences.
Prof. He is a Fellow of OSA (Optical Society of America) and SPIE (The
International Society for Optical Engineering).
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Multiple Antenna Systems With Inherently
Decoupled Radiators
Mauro Pelosi, Mikael Bergholz Knudsen, Member, IEEE, and Gert Frølund Pedersen
Abstract—In multiple antenna systems mutual coupling needs
to be minimized. We propose an alternative novel decoupling tech-
nique, investigating several multiple antenna configurations for
small handsets through measurements and numerical simulations.
The influence of different novel designs on performance metrics
such as total loss, antenna isolation and envelope correlation coef-
ficient are investigated. By varying antenna impedance bandwidth
and antenna location with respect to the handset, both planar
inverted F antenna (PIFA) and inverted F antennas (IFA) were
investigated in different UMTS frequency bands in proximity with
the user’s body. Results show that antennas may experience very
different losses and envelope correlation coefficients depending on
their relative position with respect to the handset, as the influence
of the user’s hand is not symmetrical in most cases. Narrow-band
antennas are inherently decoupled when integrated on the same
handset, while also other parameters such as frequency duplex
distance and interaction with the user’s body influence the mutual
coupling.
Index Terms—Hand phantom, isolation, multiple input multiple
output (MIMO) systems, small antennas.
I. INTRODUCTION
M OBILE terminals’ market is facing a growing demand for
a panoply of new services, as multiple communication
standards need to be supported. Small terminal antennas cannot
be designed as single-port components anymore, as they have to
coexist and interact with multiple radiators collocated on the same
printed circuit board (PCB). Having a good isolation and a low
envelope correlation coefficient is fundamental to achieve good
overall performance in multiple antenna systems for diversity and
MIMO purposes. However, improving isolation is one of the most
challenging tasks, as when conventional multiple antennas share
the same ground-plane at low frequencies, the PCB acts as the
main radiator; it is therefore difficult to achieve satisfactory an-
tenna isolation because of the high coupling.
The mutual coupling compromises the efficiency of a mul-
tiple antenna system, as a certain fraction of the available power
delivered to one antenna is not radiated but dissipated in the load
Manuscript received June 01, 2010; revised May 28, 2011; accepted July 25,
2011. Date of publication October 25, 2011; date of current version February 03,
2012. This work was supported by the Smart Antenna Front End (SAFE) Project
within the Danish National Advanced Technology Foundation framework.
M. Pelosi and G. F. Pedersen are with the Section of Antennas, Propagation
and Radio Networking (APNet), Department of Electronic Systems, Faculty
of Engineering and Science, Aalborg University, DK-9220 Aalborg, Denmark
(e-mail: mp@es.aau.dk; gfp@es.aau.dk).
M. B. Knudsen is with the Intel Mobile Communications Denmark Aps,
DK-9220 Aalborg Øst, Denmark (e-mail: mikael.knudsen@intel.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173437
0018-926X/$26.00 © 2011 IEEE
503
of the other radiator instead [1]. Moreover, antennas exhibiting
high mutual coupling are also expected to be highly correlated,
resulting in poor MIMO performance [2]. When antennas insist
on the same ground-plane structure they utilize it to enhance
their radiation [3].
However, the effects resulting from the combination of the
chassis and the antennas are very complicated, and sometimes
counterintuitive phenomena are observed. In fact reducing the
distance between antennas may have the unexpected effect of
lowering the mutual coupling, as the contribution of finite-sized
ground-planes needs to be taken into account [4]. Despite sev-
eral techniques have already been proposed in literature in order
to mitigate the effect of the mutual coupling in small handsets,
their practical implementation in a commercial product is not
mature yet. In fact the non-idealities and the additional com-
plexities in the proposed solutions often do not pay off enough
to justify their integration.
A. Decoupling Methods Review
Existing decoupling methods can be broadly grouped into
five categories:
1) Antenna Spacing and Angular Orientation Variation: The
earlier intuitive attempts to lower the mutual coupling between
antenna arrays have always been to keep a reasonable distance
between the radiating elements [1]. However a miniaturization
of the array size has become an indispensable feature in small
handsets, not allowing anymore a sufficient spacing between
array elements [5]. It has already been shown that mutual cou-
pling mainly depends on the distance between the open-ended
sides of PIFA antennas [6]. However, as the mutual coupling
also depends on the surface-current distribution on the PCB [7],
a larger spacing does not automatically results in better isola-
tion level [4]. As the relative angular orientation of the PIFAs is
also playing a role in the mutual coupling mechanism [6], it is
clear that a proper design process needs to include a methodical
investigation of the possible solutions, keeping always in mind
that sometimes the best design option in Free Space (FS) is not
the most appropriate when the effect of the human user is prop-
erly taken into account [8].
2) Decoupling Networks: The use of decoupling networks
for closely spaced antennas is proposed in several works
[10]–[14], and intends to introduce an additional impedance
matching network. Despite the proposed solutions provide
satisfactory results, often they fail to be to be feasible in
practical mobile devices [14]. In fact a decoupling network
exhibits ohmic losses due to the non-idealities in the compo-
nents [10] and can drastically reduce the usable space on the
handset. Moreover it seems that the provided decoupling is
504 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
very narrow-band in nature [11], following a trend similar to
Fano’s law for impedance matching [15], [16].
3) Parasitic and Coupling Elements: Another possibility
that has been well-studied is the use of distributed parasitic
elements in between the radiators that serve the same purpose
of a decoupling network realized with lumped components
[17]–[19]. By field cancellation, the element is able to artifi-
cially add a coupling path between the radiators [17]. Despite
this solution has the advantage of suffering less from ohmic
losses, its practical implementation is conditioned and limited
to particular configurations, as the decoupling can be very
sensitive to small changes in the close by environment of the
antennas [17]. The design procedure of these parasitic elements
differs from the analytical formulation of the aforementioned
decoupling networks. In fact the final structure is often a
sub-optimal result of an optimization process that tries to
minimize the footprint of the parasitic element and maximize
both decoupling bandwidth and isolation level. This approach
may lead to reduced physical insight, as the underlying cou-
pling mechanisms are often too complicated to be properly
investigated analytically.
4) Defected Ground-Plane Structures: The possibility of
using defected ground-plane structures has been explored in
several works [20]–[24]. This solution intends to embed filtering
capabilities in the PCB by modifying the ground-plane structure
itself. Despite the reached isolation levels might seem very
appealing, often the modifications in the ground-plane structure
are so severe and unrealistic [20] that no practical handset would
be able to bear it. In fact the industrial design of mobile phones
poses strict limitations on the allowed modifications, imposing
to the antenna designer several “forbidden areas” that cannot be
violated. Having slits in the middle of a typical bar-type mobile
phone with antennas at the top and bottom of the PCB is thereby
not feasible, while for clamshell models the slits might be put in
the hinge area with less problems.
5) Neutralization Lines: More recently the use of neutral-
ization lines between different antennas has been very popular
in literature [4], [25]–[28], providing many interesting designs.
The neutralization line proposes to create an additional cou-
pling path opposite to the original one [4]. This is achieved by
changing the shape, width and orientation of the neutralization
line with respect to the antennas, as again no general rule exists
due to the complicated coupling interaction of the ground-plane
with the antennas. This method can be identified as a general-
ization of the parasitic element decoupling technique, with the
possibility of positioning neutralization lines either between the
radiators [4] or between the ground-plane and the individual ra-
diators [27]. The solution offers a high degree of flexibility, al-
lowing also dual-band operation [28]. However, it is still un-
clear the real performance of this technique when the influence
of the user’s hand is properly taken into account [8], as the cor-
responding detuning might result in spoiling the very sensitive
decoupling mechanism. Moreover, at very low frequencies the
footprint would be too large [29].
B. Proposed Decoupling Method
More and more antennas have to be integrated in a limited
volume. The concurrent trend towards size reduction is counter-
acted by fundamental limitations for small antennas operating
in a given volume, so that a tradeoff between bandwidth, effi-
ciency and physical size has always to be established [30]. Con-
sequently, a small antenna size and high antenna efficiency can
coexist only sacrificing the impedance bandwidth. A bandwidth
reduction has always been seen as a drawback in conventional
antenna design, as this leads to the impossibility of covering all
the frequency range required by a given communication stan-
dard. However, if proper countermeasures are taken, this im-
pairment could become a benefit. In fact if the antennas are
narrow-band, their isolation is expected to be inherently better,
as they are electrically smaller.
In full-duplex radio communication systems the radio
transmitter is active at the same time as the radio receiver,
using only one radiator for both receiving and transmitting
frequency bands. If we imagine switching to a novel paradigm
where we separate receiving and transmitting antennas by
using two individual radiators exhibiting a higher isolation, the
requirements on the duplex filters could be potentially low-
ered. This approach is called hereinafter transceiver separation
mode (TSM). In this paper we will study the influence of the
impedance bandwidth on different performance metrics such
as total efficiency, antenna isolation and envelope correlation
coefficient through finite-difference time-domain (FDTD)
simulations and measurements. Numerical simulations offer
the advantage of performing huge parametric investigations,
which are very important in the preliminary design stage, as
they avoid the measurement biases and uncertainties [31]. By
using two antenna types, the PIFA and the IFA radiating in
different frequency bands, two distinct operating modes will
be investigated: the MIMO mode (MM) and the transceiver
separation mode respectively. In MIMO mode two antennas
resonating at the same frequency for a given UMTS band are
present at the same time on the handset. In transceiver sepa-
ration mode individual antennas only cover half of the duplex
for each corresponding UMTS band, the transmitting (TX) or
receiving (RX) half respectively. Recent studies have shown
that the antenna total efficiency strongly depends on the way
the mobile device is held by the human hand while the user’s
head has a minor impact [31], [32].
As handset antennas radiate more and more in close prox-
imity with the human body, its influence cannot be neglected
anymore. Moreover, there is a lack of knowledge concerning
the impact of the user’s body on antenna isolation and enve-
lope correlation coefficient, as hand phantoms representative of
average use have to be used. For this purpose the influence of
several hand phantoms will be studied, taking also into account
the user’s head.
The paper is organized as follows: Section I motivates the
proposed investigation providing also a review of existing de-
coupling methods and proposing a novel paradigm.
Section II describes the used antenna performance metrics
while Section III explains the parametric simulations setup.
Section IV shows some preliminary investigations which com-
pares simulated and measured handsets. Section V describes
the parametric simulations results, Section VI proposes design
guidelines and Section VII provides concluding remarks.
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS
II. DESCRIPTION OF THE ANTENNA PERFORMANCE METRICS
In our investigation the total efficiencies of the antennas have
been calculated in the following way:
(1)
(2)
where , and , are the radiation and
total efficiencies of the first and second antenna respectively,
while and are the total losses expressed in dB
(3)
(4)
The envelope correlation coefficient can be calculated in a
straightforward manner using solely the scattering parameters
according to the following equation [33]:
(5)
However this first solution is not reliable if the antenna
system exhibit losses [34], so that we adopted a more precise
formulation that computes the envelope correlation coefficient
through the complex radiation patterns. Under the assumption
of antennas operating in a uniform multi-path environment, we
can express the envelope correlation coefficient as [35] (see (6)
at the bottom of the page) where
(7)
(8)
, and , are the vertical and hori-
zontal polarized complex radiation patterns of antennas 1 and
2 respectively, while is the solid angle for a spherical coordi-
nate system. XPR stands for cross-polar discrimination and in
our case are equal to 1 because of the multi-path environment
hypothesis.
III. PARAMETRIC SIMULATIONS SETUP
The following paragraphs describe in detail the parametric
numerical investigations that were performed in this study.
A. FDTD Simulation Code
The numerical simulations were conducted using our
in-house FDTD code, choosing a space step size of 1 mm
and an energy based termination criterion. The FDTD code
505
is parallel and it uses the supercomputer facilities at Aalborg
University (Fyrkat) [36], allowing demanding simulations to
be run in a limited time.
B. Handset Characteristics
The handset ground-plane has been modeled as a Perfect
Electric Conductor (PEC) plate representing the typical dimen-
sions of a bar-type mobile phone (40 mm 100 mm). When
IFA antennas were used, the ground-plane size was reduced in
order to embed them in the same structure. Though we know
that the ground-plane contribution is significant especially at
low frequencies, this choice was made to keep the same overall
size of the handsets for all designs.
C. Antenna Pair Operating Mode
In our investigations each antenna pair on the handset was
used in two different operating modes:
1) MIMO Mode: In this case each radiator was resonating at
the same frequency for a given UMTS band.
2) Transceiver Separation Mode: In this case each individual
antenna only covered half of the duplex for the corresponding
UMTS band, the transmitting (TX) or receiving (RX) one re-
spectively.
D. Antenna Types
The PIFAs consisted of a main planar element parallel to the
ground-plane. In order to achieve a quarter-wavelength charac-
teristic, the planar element was connected to the ground-plane
by a shorting wire. The IFAs consisted of several metallic strips
coplanar to the ground-plane.
E. Antenna Impedance Bandwidth
1) Normal-band (NOB);
2) Medium-band (MEB);
3) Narrow-band (NAB).
We defined the antenna impedance bandwidth as the range of
frequencies having a reflection coefficient lower than ,
satisfying a criterion usually considered acceptable in most
handheld devices. Concerning the MIMO mode case, an-
tennas covering the full duplex in a given UMTS band were
named “normal-band.” If the antennas only covered half of
the frequency duplex they were named “medium-band,” while
“narrow-band” antennas represented the case in which only
slightly more than one UMTS channel was covered (8 MHz).
In the transceiver separation case, normal-band antennas cov-
ered only the TX or RX half duplex, while narrow-band ones
slightly more than one UMTS channel as in the aforementioned
MIMO mode case. Despite the previous definitions do not
reflect consolidated design approaches, they are needed to
(6)
506 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE I
FREQUENCY RANGE AND RESONANCE FREQUENCY FOR DIFFERENT UMTS
FREQUENCY BANDS
Fig. 1. Antenna pair configurations.
better comprehend the influence of the antenna impedance
bandwidth. It is then understood that antennas not covering the
full range of frequencies will need a frequency tuning stage in
a later practical implementation.
F. UMTS Frequency Band
In order to investigate the effect of different frequencies and
duplex distances, several UMTS bands were investigated:
1) Band I (BI);
2) Band II (BII);
3) Band V (BV).
The corresponding frequency ranges and resonance frequen-
cies can be found in Table I.
G. Antenna Location
The location of the antenna with respect to the handset is
a very important feature, as it influences many different per-
formance metrics. In our study four different locations were
chosen: Top Left (TL), Top Right (TR), Bottom Left (BL), and
Bottom Right (BR).
H. Antenna Pair Configurations
Each antenna pair configuration consists of a combination of
the antenna type and the location with respect to the handset
(Fig. 1). All antenna pairs are described in Table III.
I. Geometrical Features of Antennas
The geometry of the antennas was simple and without the use
of matching circuits. The impedance bandwidth of the PIFA an-
tennas was mainly determined by the height of the main plate
with respect to the ground-plane , while for the IFA antennas
by the distance between the top IFA strip and the ground-plane
. Table II shows the aforementioned features for different
UMTS bands and impedance bandwidths.
Fig. 2 shows the geometrical features of both PIFA and IFA
antennas for different UMTS frequency bands and impedance
bandwidths in MIMO mode. For brevity reasons only configu-
rations 3 and 5 are shown, while as in MIMO mode both an-
tennas are identical, only the antennas at the top of the handset
TABLE II
AND FOR DIFFERENT UMTS FREQUENCY BANDS AND IMPEDANCE
BANDWIDTHS
TABLE III
DESCRIPTION OF ALL ANTENNA PAIRS; ANTENNA PAIRS 1–39 ARE USED IN
MIMO MODE, WHILE 40–65 ONES IN TRANSCEIVER SEPARATION MODE
are displayed. The geometrical features of the other configura-
tions (C1, C2, C4) are very similar to the aforementioned an-
tennas, as only minor adjustments were needed to retune them
at the correct resonance frequency. Concerning the transceiver
separation mode, the geometry of the antennas was slightly ad-
justed to tune them at the corresponding TX or RX resonance
frequency, so that their layout is not shown. The shorting wire
for PIFA antennas was always positioned at the right top edge of
the main plate for bands I, II, while in band V it was located at its
top left edge; the feeding was very close to the shorting wire in
all cases (Fig. 2). In the IFA antenna case the distance between
the shorting and feeding strip was slightly modified in different
configurations to achieve a better impedance matching. In order
to resonate at band V PIFA antennas had a dielectric substrate of
relative permittivity , while IFA antennas a 1 mm thick
dielectric superstrate of .
J. Antenna Pair Proximity With Phantoms
The antenna pair on the handset was studied in FS and while
in proximity with the user’s hand. The influence of the user’s
head was investigated by using the Specific Anthropomorphic
Mannequin (SAM) phantom according to the standard right
cheek position [37]. Thanks to a recent grip study [32], it was
possible to generate detailed Computer Aided Design (CAD)
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS 507

Fig. 3. Hand phantoms (H1-H6).

TABLE IV
DESCRIPTION OF THE HAND PHANTOMS

Fig. 2. Antenna layout for several configurations; the ground-plane is shown

in grey, and only its top part is displayed. The layout is shown in scale; the small
grey round markers represent the feeding point, while the white ones represent
the shorting point.

models of hand phantoms consistent with the average use. The

dimensions of the hand phantoms were adjusted according
to a hand anthropometric study [38], while their dielectric
properties were chosen to comply with the homogeneous
material representing an average of different tissues [39]. Six
different grips were investigated (H1-H6), consisting of dif-
ferent palm-handset gaps (firm/soft grip styles) and index finger
locations as described in Fig. 3 and Table IV. As the position
of the index finger is one of the main factors in determining the
interaction of the antenna with the human hand [8], different
configurations were studied. As in the handset different heights Fig. 4. Simulated and manufactured handsets 1–4. Handset 1 had a white Sty-
rofoam brick only used as support .
were used for the PIFA antennas, the gap between the hand
phantoms and the PIFA main element was not constant. The
reason for this choice was to investigate the potential of a buffer
40 mm 100 mm. The geometrical properties of the handsets
air superstrate as in [40], which showed beneficial effects in
are illustrated in Fig. 4 and summarized in Table V that also
terms of total efficiency and proximity effects robustness.
shows results.
This means that there was always a 10 mm gap between the
ground-plane and the tip of the index finger.
B. Simulation and Measurement Results
IV. PRELIMINARY INVESTIGATIONS The following Figs. 5, 6 show a comparison between the sim-
ulated and measured scattering parameters , , and
A. Simulation and Measurement Setup for all 4 handsets. A fair agreement is observed despite a fre-
The purpose of this section is to validate the concept of quency shift is observed due to imperfections in the manufac-
narrow-band antenna decoupling by comparing a significant turing and the cable influence in the low band. Fig. 5 refers to
set of configurations in TSM mode, simulating, manufacturing the comparison between handsets 1 and 2 in the high band (BI),
and measuring four handsets in FS. Both high (Band I) and showing that when narrow-band antennas are used it is possible
low (Band V) UMTS frequency bands are studied, comparing to have a isolation. Fig. 6 compares handsets 3 and 4
normal-band and narrow-band antennas. The four handsets in the low band (BV), showing that of isolation are
consisted of a pair of PIFA antennas on a ground-plane of size still possible despite the high original coupling of normal-band
508
Fig. 5. Comparison of the simulated and measured scattering parameters for handsets 1, 2 (band I).
Fig. 6. Comparison of the simulated and measured scattering parameters for handsets 3, 4 (band V).
TABLE V
DESCRIPTION OF THE MANUFACTURED AND SIMULATED HANDSETS 1–4
WITH TOTAL EFFICIENCY. THE TOTAL EFFICIENCY IS SIMULATED IN FDTD
AND MEASURED IN ANECHOIC CHAMBER IN CORRESPONDENCE OF THE BEST
IMPEDANCE MATCH OF EACH RECEIVING ANTENNA (RX)
antennas. Table V shows the simulated and measured total effi-
ciencies for handsets 1–4.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Concerning band I, the total efficiencies are very good in both
normal-band and narrow-band cases. The high loss in the mea-
sured narrow-band antenna in band V can be explained by the
fact that ohmic losses tend naturally to increase as the band-
width shrinks, while other effects such as imperfect soldering,
contact losses and non-ideal dielectric substrate concur in low-
ering the total efficiency. This happens while keeping the mutual
coupling very low, which is instead the dominant loss factor in
the normal-band antennas.
V. PARAMETRIC SIMULATIONS RESULTS
A. MIMO Mode
1) Total Loss: In this paragraph the results concerning the
total loss are presented, combining in a single figure of merit
absorption, coupling, and mismatch loss according to (3). Fig. 7
shows the total loss in free space for different configurations and
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS
Fig. 7. Total loss in free space for MIMO mode; A1 and A2 refer to antenna 1
and 2 respectively, as shown in Fig. 1.
Fig. 8. Total loss in presence of hand phantom 1 for MIMO mode; A1 and A2
refer to antenna 1 and 2 respectively, as shown in Fig. 1.
UMTS bands, therefore only coupling and mismatch losses are
accounted for. Total loss is generally low in bands I, II, while
in band V it is higher due to the stronger coupling at lower fre-
quencies. As the impedance bandwidth of the antennas shrinks,
a reduction in the total loss is observed, as the coupling is lower
when antennas get narrow-band.
Fig. 8 shows the total loss when the handset is held by the
human hand. The hand phantom is H1, representing a typical
firm grip style with the position of the index finger in the prox-
imity of the top-right handset region. UMTS bands I, II exhibit
comparable loss as they are closer in frequency, while the loss
for band V is higher.
In configuration 1 the top-left antenna is less affected by the
proximity of the hand, as the index finger is not close. When
both antennas are at the bottom (C2), they suffer a similar loss
because of the effect of the palm. Similar consideration can
be done for C3, C4, and C5, where the antennas are more or
509
Fig. 9. Total loss in presence of phantom 1 for MIMO mode;
A1 and A2 refer to antenna 1 and 2 respectively, as shown in Fig. 1.
less affected depending on their relative position with respect
to the hand. By comparing bottom antennas in configurations
C4 and C5, it can be observed that IFA antennas experience the
lowest loss, as they exhibit a larger gap with respect to the palm.
Narrow-band antennas generally exhibit the lowest loss in bands
I, II, while the situation is different in band V. In fact the an-
tennas closer to the palm suffer also a larger frequency detuning,
which in the narrow-band case translates to a higher mismatch
loss. Despite narrow-band antennas exhibit the lowest coupling
loss, if they are strongly detuned their total loss may increase.
By comparing the total loss for both antennas (A1-A2), it can be
seen how the influence of the hand may significantly affect the
antennas in different ways. Similar tendencies were found for
the other hand phantoms’ configurations, where soft grip hands
experienced the smallest losses.
In Fig. 9 the SAM phantom is added to H1, showing a trend
similar to Fig. 8. The presence of the SAM does not further
degrade the antenna performance, showing a modest increase in
the total loss confirming that the influence of the hand is more
important than the head’s one.
2) Worst Case Isolation: We define the worst case isola-
tion as the maximum value of in dB over the frequency
range corresponding to a particular bandwidth (Fig. 10). This
peak value in MIMO mode is typically located in correspon-
dence of the antenna resonance frequency in free space, while it
can be shifted if the antennas are detuned because of the inter-
action with the human body. In Fig. 11 we compare the worst
case isolation for different normal-band antenna configurations.
The presence of the hand generally improves the isolation when
compared to the free space case. Soft grip hands have better iso-
lation than firm grip ones. This may be explained by the fact
that soft grip hands influence the detuning of the antennas in
asymmetric ways, thus leading to an increase in isolation. When
moving from band I to band II and V, the isolation gets worse;
this is due to the higher coupling at lower frequencies. Sim-
ilar conclusion can be derived for the medium-band case. Con-
cerning band V in Fig. 12 it is important to remark that the high
isolation is mainly determined by the detuning of the antennas.
510 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 10. Example of worst-case isolation.
Fig. 11. Worst-case isolation in free space and in presence of hand phantoms
1–6 with normal-band antennas in MIMO mode.
In configurations 4 and 5 soft grip hands have worse isolation
than firm grip ones. This may be explained by noting that soft
grip hands exhibit a larger gap between the antennas and the
palm of the hand, leading therefore to less detuning and worse
isolation.
In Fig. 13 it can be seen how narrow-band antennas generally
lead to a better isolation. There is little difference between the
case with only hand and hand with SAM, as most of interaction
is driven by the hand alone.
3) Envelope Correlation Coefficient: In Fig. 14 it can be seen
the envelope correlation coefficient in free space for dif-
ferent configurations depending on the impedance bandwidth
of the antennas. is typically lower in bands I, II, and when
narrow-band antennas are used. In band V the envelope correla-
tion coefficient is higher, as in this frequency range the coupling
is stronger. When narrow-band antennas are used they exhibit
higher than their normal-band counterparts. This may be due
to the fact that besides mutual coupling, also the level of mis-
match influences the envelope correlation coefficient, so that a
bad mismatch may further correlate the antennas.
Fig. 12. Worst-case isolation in free space and in presence of hand phantoms
1–6 with narrow-band antennas in MIMO mode.
Fig. 13. Worst-case isolation for different configurations in MIMO mode.
In Fig. 15 it is shown the influence of the hand phantom 1
on the envelope correlation coefficient. A higher is gener-
ally observed, especially in the configurations with larger total
loss. In configuration 2 narrow-band antennas exhibit a smaller
loss, so that the envelope correlation coefficient gets smaller. In
Fig. 16 normal-band antennas are compared, studying the in-
fluence of both hand and SAM phantoms. Concerning bands I,
II, there is small difference in the envelope correlation coeffi-
cient between different phantoms’ configurations, while in band
V the SAM has a larger impact. Similar considerations can be
done for Fig. 17, where the envelope correlation coefficient is
still low for bands I, II but higher for band V.
B. Transceiver Separation Mode
1) Total Loss: In Fig. 18 the total loss is displayed in the free
space case for different configurations. The total loss is always
lower for narrow-band antennas, as they exhibit better isolation
and therefore less coupling loss.
The benefit of narrow-band antennas is more evident at lower
frequencies, where the coupling loss is higher. In Fig. 19 the in-
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS
Fig. 14. Envelope correlation coefficient in free space in MIMO mode.
Fig. 15. Envelope correlation coefficient in presence of hand phantom 1 in
MIMO mode.
fluence of the user’s hand is shown using hand phantom 1 (H1).
It can be seen that narrow-band antennas have always a total
loss lower than their normal-band counterparts. This beneficial
effect is more evident in band V, as both coupling and absorption
loss are smaller because of a reduced interaction with the hand
phantom. There is a similar trend for the remaining hand phan-
toms, with firm grip styles (H1-H3) having larger total losses
than soft grip ones (H4-H6). Depending on the position of the
index finger and the proximity of the palm of the hand the dif-
ferent configurations are diversely affected.
2) Worst Case Isolation: Fig. 20 displays the worst-case
isolation for normal-band antennas for different configurations.
Considerations similar to the MIMO mode can be made; the
coupling increases from band I to band V, as we are moving
towards lower frequencies. The presence of the user’s hand is
generally beneficial to the isolation, while depending on the grip
style the coupling may change. Comparing Figs. 17 and 18 we
511
Fig. 16. Envelope correlation coefficient with normal-band antennas in MIMO
mode.
Fig. 17. Envelope correlation coefficient with narrow-band antennas in MIMO
mode.
can see how narrow-band antennas always have the best isola-
tion when compared to normal-band ones.
VI. FURTHER DESIGN INSIGHTS
Optimizing an antenna design to meet today’s requirements
is becoming more and more challenging. Beside single antenna
specifications such as total radiated power (TRP) or total
isotropic sensitivity (TIS), no concise figure of merit (FOM)
exists so far to assess the performance of multiple antenna
systems [41]. The standardization efforts of the just concluded
COST Action 2100 have suggested throughput as one of the
FOMs to focus on [42]. However, considering a given rich
scattering environment a high throughput may either depend on
smart signal processing or on a well designed antenna system.
Despite the need of cross-layer optimization, it is still important
to properly understand which parameters influence MIMO
performance. A good performance for a MIMO array requires
that all the antennas should exhibit both similar and high
512 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 18. Total loss in free space in transceiver separation mode; TX and RX
represent the transmitting and receiving antennas respectively, and their location
with respect to the handset is described in Table III.
Fig. 19. Total loss in presence of hand phantom 1 in transceiver separation
mode; TX and RX represent the transmitting and receiving antennas respec-
tively, and their location with respect to the handset is described in Table III.
mean effective gain (MEG). Furthermore the cross correlation
between the array branches should be low. If the MEGs are
either low and dissimilar, the cross correlation plays a minor
role as even being low would not help in recovering the MIMO
performance because of the fact that the other performance
metrics are already compromised. On the other hand, if both
branches exhibit good and similar MEGs, the cross-correlation
role is more determinant. However correlation is very seldom
an issue in small handsets, as the presence of the human user
always has decorrelating effects [43]. As cellular systems are
inherently interference limited [44], the antenna array can still
provide some benefits by offering beamforming and interfer-
ence cancellation capabilities in case the two branches do not
have similar MEGs. All the aforementioned considerations
imply that the most important things in a MIMO array for
small handsets is the MIMO-favorable channel, the received
Fig. 20. Worst-case isolation in free space and in presence of hand phantoms
1–6 with normal-band antennas in transceiver separation mode.
Fig. 21. Worst-case isolation in free space and in presence of hand phantoms
1–6 with narrow-band antennas in transceiver separation mode.
signal level and the similarities of the total efficiencies in both
branches. If for example one antenna suffers severely from
absorption loss because of the influence of the hand while the
other one is less affected, this might result in very poor MIMO
performance independently of correlation level [45], reducing
MIMO to a bare antenna selection scheme. Having a good
signal level requires to have high total antenna efficiencies,
meaning that various parameters such as mismatch, coupling
and absorption loss need to be jointly minimized.
Mismatch loss depends on the antenna impedance being dif-
ferent from the characteristic reference impedance, and typi-
cally a well-designed antenna does not need additional matching
networks. In literature we find several examples of adaptive
impedance matching techniques that claim to provide signifi-
cant benefits [46]–[48], but often they focus on worst case sce-
narios that have a low probability to happen in realistic mobile
phone usage. As mismatch loss is most of times in the order
of 0.5–1.5 dB for typical configurations including the influence
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS
of the user’s body [8], an additional device to tune back the
impedance match perfectly to 50 would not be justified be-
cause of the additional complexities and losses. As shown in
several contributions absorption loss is more significant than
mismatch loss [8], [32], however little work has been done so far
in order to mitigate this problem. In fact absorption loss is often
seen as inevitable and common to any mobile device, while the
recent events of the iPhone 4 “antenna-gate” have made aware
even the general public of the fact that any antenna design must
be done considering the effect of the human user and especially
the hand’s influence. A good antenna design results in an im-
proved link budget that has positive effects on both battery du-
ration and network operators. Unfortunately there is no general
rule in designing an antenna system for a small handset, as this
is often a strict tradeoff between the conflicting expectations of
antenna engineers and industrial designers. It is then very im-
portant to conduct proper usability studies before fixing the lo-
cation of the antennas, as particular “death-grips” may result in
a complete spoiling of the antenna performance. It has already
been shown that keeping the fingers or the palm of the hand as
far as possible from the antenna region is always beneficial [8],
but often is very hard to optimize for talk and data modes at the
same time. It is good practice to embed a certain degree of ro-
bustness in the antenna radiation mechanism [48], so that the
disturbance caused by the close by environment can be mini-
mized. This is achievable by confining the near fields of the an-
tennas, by choosing a proper location and finally by respecting
small clearance areas around the antenna structure [40]. It is also
important to minimize coupling loss as much as possible as its
contribution to the total efficiency is significant especially at the
lower frequencies. However the choice of any decoupling tech-
nique needs to be considered carefully, as the additional com-
plexities might results in a poor net benefit.
VII. CONCLUSION
In this paper we investigated through measurements and
FDTD simulations two different cases, the MIMO mode and
the transceiver separation mode. By using different antenna
types for UMTS bands I, II, and V, it was possible to study the
influence of the impedance bandwidth on several performance
metrics such as total loss, isolation and envelope correlation
coefficient in MIMO mode. The influence of the user’s body
has been investigated using different hand grips and the SAM
phantom. By changing the location of the antennas with respect
to the handset several configurations have been investigated.
Concerning the transceiver separation mode, a novel paradigm
has been introduced, using individual antennas to cover the
transmitting and receiving bands of the corresponding fre-
quency duplex with the aim of lowering the requirements of the
duplex filters by providing a better isolation. Results confirm
that absorption loss is the most relevant factor in determining
the total loss. The user’s hand is the main responsible for ab-
sorption loss, while the head does not contribute significantly.
Firm grip hand phantoms have higher losses than soft grip
ones, as the most important features are the position of the
index finger and the palm-handset gap; a buffer air superstrate
is confirmed to be always beneficial in reducing the interaction
of the antennas with the user’s hand. While mismatch loss
513
is not an issue in bands I, II, it is more important in band V
when narrow-band antennas are detuned. The location of the
antennas on the handset is very important. In fact depending
on the location with respect to the handset each antenna may
experience significantly different loss because of the proximity
of the human hand. In fact the gap between the radiating
element and the palm or the index finger can drastically change
the total loss of the antennas. The envelope correlation coeffi-
cient has been derived using the far field pattern approach for
different configurations, showing in free space low values for
bands I, II and higher values at lower frequencies where the
coupling was stronger. The gets generally smaller when the
impedance bandwidth shrinks, but if a high level of mismatch
is present the envelope correlation coefficient may rise. When
the total loss increases because of the user’s body interaction
an increase in the is observed, especially in bands I, II.
In band V the situation is different, as the smaller coupling is
beneficial to the envelope correlation coefficient. Concerning
the isolation, many different factors were found to concur in the
isolation potential of one antenna pair. The coupling is stronger
at lower frequencies as the ground-plane becomes the main
radiator, while a smaller frequency duplex distance correspond
to a lower isolation. When the antenna impedance bandwidth
shrinks, the isolation tends to improve, and this effect is more
evident in the transceiver separation mode. If the antennas
exhibit a bad impedance matching at a given frequency, they
will have a better isolation. The location and orientation of the
antennas concur in determining the antenna isolation; however
because of the radiation on a finite size ground-plane a smaller
antenna spacing does not always translate in a smaller mutual
coupling. Using two different antenna types may help in getting
a better isolation, while the close proximity of the human
hand with the antennas has generally a beneficial effect, as a
better isolation can be achieved. However this improvement
has to be traded-off with other negative phenomena such
as frequency detuning and absorption loss increase. In brief
narrow-band antennas exhibit attractive features that may lead
to novel design paradigms that may lower the requirements
of the duplex filters. Though high losses have been observed
in the manufactured and measured handsets in the low band,
more research is needed to understand the loss mechanism,
especially when narrow-band antennas are used, as the high
current flow may increase the ohmic losses. The combination of
MIMO mode with the transceiver separation mode is currently
under investigation. As a general conclusion it is important to
remark that the antenna design cannot ignore the influence of
the human body, as the success of multiple antenna systems
strongly depend on having high efficiency, small coupling and
low envelope correlation coefficient.
ACKNOWLEDGMENT
The authors wish to express their gratitude to the Danish
Center for Scientific Computing (DCSC).
REFERENCES
[1] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile com-
munications,” IEEE Trans. Veh. Technol., vol. 36, no. 4, pp. 149–172,
Nov. 1987.
514 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[2] A. Derneryd and G. Kristensson, “Signal correlation including antenna
coupling,” Electron. Lett., vol. 40, no. 3, pp. 157–159, Feb. 2004.
[3] O. Kivekäs, J. Ollikainen, T. Lehtiniemi, and P. Vainikainen, “Effect
of the chassis length on the bandwidth, SAR, and efficiency of in-
ternal mobile phone antennas,” Microw. Opt. Technol. Lett., vol. 36,
pp. 457–462, Mar. 2003.
[4] A. Diallo, C. Luxey, P. L. Thuc, R. Staraj, and G. Kossiavas, “En-
hanced two-antenna structures for universal mobile telecommunica-
tions system diversity terminals,” IET Microw. Antennas Propag., vol.
2, pp. 93–93, 2008.
[5] P. Vainikainen, J. Holopainen, C. Icheln, O. Kivekäs, M. Kyrö, M. Mu-
stonen, S. Ranvier, R. Valkonen, and J. Villanen, “More than 20 antenna
elements in future mobile phones, threat or opportunity,” in Proc. 3rd
Eur. Conf. on Antennas Propag. (EuCAP’2009), Mar. 23–27, 2009, pp.
2940–2943.
[6] H. Carrasco, H. D. Hristov, R. Feick, and D. Cofré, “Mutual coupling
between planar inverted-F antennas,” Microw. Opt. Technol. Lett., vol.
42, no. 3, pp. 224–227, Aug. 2004.
[7] K.-L. Wong, J.-H. Chou, S.-W. Su, and C.-M. Su, “Isolation between
GSM/DCS and WLAN antennas in a PDA phone,” Microw. Opt.
Technol. Lett., vol. 45, pp. 347–347, 2005.
[8] M. Pelosi, O. Franek, M. B. Knudsen, G. F. Pedersen, and J. B. An-
dersen, “Antenna proximity effects for talk and data modes in mobile
phones,” IEEE Antennas Propag. Mag., vol. 52, no. 3, pp. 15–27, Jun.
2010.
[9] B. K. Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch, “Impact
of matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag., vol. 54, no. 11, pp. 3225–3238, Nov. 2006.
[10] C. Volmer, J. Weber, R. Stephan, K. Blau, and M. A. Hein, “An eigen-
analysis of compact antenna arrays and its application to port decou-
pling,” IEEE Trans. Antennas Propag., vol. 56, no. 2, pp. 360–370,
Feb. 2008.
[11] B. K. Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch, “Impact
of matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag., vol. 54, no. 11, pp. 3225–3238, Nov. 2006.
[12] J. B. Andersen and H. Rasmussen, “Decoupling and descattering net-
works for antennas,” IEEE Trans. Antennas Propag., vol. 24, no. 6, pp.
841–846, Nov. 1976.
[13] S.-C. Chen, Y.-S. Wang, and S.-J. Chung, “A decoupling technique for
increasing the port isolation between two strongly coupled antennas,”
IEEE Trans. Antennas Propag., vol. 56, no. 12, pp. 3650–3658, Dec.
2008.
[14] R. A. Bhatti, S. Yi, and S. Park, “Compact antenna array with port de-
coupling for LTE-standardized mobile phones,” IEEE Antennas Wire-
less Propag. Lett., vol. 8, pp. 1430–1433, 2009.
[15] R. Fano, “Theoretical limitations on the broadband matching of arbi-
trary impedances,” J. Franklin Inst., vol. 249, no. 1, 2, pp. 57–3, 1950.
[16] G. Shaker, S. Safavi-Naeini, and N. Sangary, “Q-Bandwidth relations
for the design of coupled multi-element antennas,” in Proc. IEEE An-
tennas Propag. Society Int. Symp. (APSURSI’2009), Jun. 1–5, 2009,
pp. 1–4.
[17] A. C. K. Mak, C. R. Rowell, and R. D. Murch, “Isolation enhancement
between two closely packed antennas,” IEEE Trans. Antennas Propag.,
vol. 56, no. 11, pp. 3411–3419, Nov. 2008.
[18] P. Ferrer, J. Arbesu, and J. Romeu, “Decorrelation of two closely
spaced antennas with a metamaterial AMC surface,” Microw. Opt.
Technol. Lett., vol. 50, no. 5, May 2008.
[19] A. Abe, N. Michishita, Y. Yamada, J. Muramatsu, T. Watanabe, and
K. Sato, “Mutual coupling reduction between two dipole antennas
with parasitic elements composed of composite right-/left-handed
transmission lines,” in Proc. IEEE Int. Workshop on Antenna Technol.
(iWAT’2009), Mar. 2–4, 2009, pp. 1–4.
[20] C. Chi-Yuk, C. Chi-Ho, R. D. Murch, and C. R. Rowell, “Reduction
of mutual coupling between closely-packed antenna elements,” IEEE
Trans. Antennas Propag., vol. 55, no. 6, pp. 1732–1738, Jun. 2007.
[21] G. Yue, C. Xiaodong, Y. Zhinong, and C. Parini, “Design and per-
formance investigation of a dual-element PIFA array at 2.5 GHz for
MIMO terminal,” IEEE Trans. Antennas Propag., vol. 55, no. 12, pp.
3433–3441, Dec. 2007.
[22] C. Younkyu, J. Seong-Sik, D. Ahn, C. Jae-Ick, and T. Itoh, “High iso-
lation dual-polarized patch antenna using integrated defected ground
structure,” IEEE Microw. Wireless Compon. Lett., vol. 14, no. 1, pp.
4–6, Jan. 2004.
[23] H. Li, J. Xiong, and S. He, “A compact planar MIMO antenna system of
four elements with similar radiation characteristics and isolation struc-
ture,” IEEE Antennas Wireless Propag. Lett., vol. 8, pp. 1107–1110,
2009.
[24] F.-G. Zhu, J.-D. Xu, and Q. Xu, “Reduction of mutual coupling be-
tween closely-packed antenna elements using defected ground struc-
ture,” Electron. Lett., vol. 45, no. 12, pp. 601–602, Jun. 2009.
[25] A. Diallo, C. Luxey, P. Le Thuc, R. Staraj, and G. Kossiavas, “En-
hanced two-antenna structures for universal mobile telecommunica-
tions system diversity terminals,” IET Microw. Antennas Propag., vol.
2, no. 1, pp. 93–101, Feb. 2008.
[26] A. Chebihi, C. Luxey, A. Diallo, P. Le Thuc, and R. Staraj, “A
novel isolation technique for closely spaced PIFAs for UMTS mobile
phones,” IEEE Antennas Wireless Propag. Lett., vol. 7, pp. 665–668,
2008.
[27] H. Minseok and C. Jaehoon, “Small-size printed strip MIMO antenna
for next generation mobile handset application,” Microw. Opt. Technol.
Lett., vol. 53, no. 2, pp. 48–352, 2011.
[28] I. Dioum, A. Diallo, C. Luxey, and S. M. Farsi, “Compact dual-band
monopole antenna for LTE mobile phones,” in Proc. Loughborough
Antennas Propag. Conf. (LAPC), Nov. 8–9, 2010, pp. 593–596.
[29] J. Holopainen, O. Kivekäs, C. Icheln, and P. Vainikainen, “Internal
broadband antennas for digital television receiver in mobile terminals,”
IEEE Trans. Antennas Propag., vol. 58, no. 10, pp. 3363–3374, Oct.
2010.
[30] D. M. Pozar, “New results for minimum Q, maximum gain, and polar-
ization properties of electrically small arbitrary antennas,” in Proc. 3rd
Eur. Conf. on Antennas Propag. (EuCAP 2009), Mar. 23–27, 2009, pp.
1993–1996.
[31] L. Chung-Huan, E. Ofli, N. Chavannes, and N. Kuster, “Effects of
hand phantom on mobile phone antenna performance,” IEEE Trans.
Antennas Propag., vol. 57, no. 9, pp. 2763–2770, Sep. 2009.
[32] M. Pelosi, O. Franek, M. B. Knudsen, M. Christensen, and G. F. Ped-
ersen, “A grip study for talk and data modes in mobile phones,” IEEE
Trans. Antennas Propag., vol. 57, no. 4, pp. 856–865, Apr. 2009.
[33] S. Blanch, J. Romeu, and I. Corbella, “Exact representation of antenna
system diversity performance from input parameter description,” Elec-
tron. Lett., vol. 39, no. 9, pp. 705–707, May 2003.
[34] P. Hallbjorner, “The significance of radiation efficiencies when using
S-parameters to calculate the received signal correlation from two an-
tennas,” IEEE Antennas Wireless Propag. Lett., vol. 4, pp. 97–99, 2005.
[35] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile com-
munications,” IEEE Trans. Veh. Technol., vol. 36, no. 4, pp. 149–172,
Nov. 1987.
[36] [Online]. Available: http://www.dcsc.aau.dk/index.php?id=26
[37] Recommended practice for determining the peak spatial-average spe-
cific absorption rate (SAR) in the human body due to wireless com-
munications devices: experimental techniques, IEEE Standard 1528,
2003.
[38] T. M. Greiner, “Hand anthropometry of US army personnel,”
Natick/TR-92/011, 1991.
[39] C. Gabriel, “Tissue equivalent material for hand phantoms,” Phys.
Med. Biol. PMB-52, pp. 4205–4210, 2007.
[40] M. Pelosi, O. Franek, M. B. Knudsen, and G. F. Pedersen, “Influence of
dielectric loading on PIFA antennas in close proximity to user’s body,”
Electron. Lett., vol. 45, no. 5, pp. 246–247, Feb. 2009.
[41] P. Kyosti, J.-P. Nuutinen, and T. Jamsa, “MIMO OTA test concept with
experimental and simulated verification,” in Proc. 4th Eur. Conf. on
Antennas Propag. (EuCAP’2010), Apr. 12–16, 2010, pp. 1–5.
[42] R. Verdone, “Pervasive Mobile & Ambient Wireless Communica-
tions,” in COST 2100 Final Report. Berlin: Springer, 2011.
[43] B. Yanakiev, J. Ø. Nielsen, M. Christensen, and G. F. Pedersen, “An-
tennas in real environments,” in Proc. 5th Eur. Conf. on Antennas
Propag. (EuCAP’2011), Apr. 11–15, 2011, pp. 1–5.
[44] J. G. Andrews, C. Wan, and R. W. Heath, “Overcoming interference
in spatial multiplexing MIMO cellular networks,” IEEE Wireless
Commun., vol. 14, no. 6, pp. 95–104, Dec. 2007.
[45] A. A. H. Azremi, J. Ilvonen, R. Valkonen, J. Holopainen, O. Kivekas,
C. Icheln, and P. Vainikainen, “Performance analysis of broadband
coupling-Element-Based multiantenna structure for mobile terminal
with hand effects,” in Proc. IEEE 21st Int. Symp. on Personal Indoor
and Mobile Radio Communications (PIMRC’2010), Sep. 26–30, 2010,
pp. 1111–1116.
[46] P. Ramachandran, Z. D. Milosavljevic, and C. Beckman, “Adaptive
matching circuitry for compensation of finger effect on handset an-
tennas,” in Proc. 3rd Eur. Conf. on Antennas Propag. (EuCAP’2009),
Mar. 23–27, 2009, pp. 801–804.
[47] E. L. Firrao, A.-J. Annema, and B. Nauta, “An automatic antenna
tuning system using only RF signal amplitudes,” IEEE Trans. Circuits
Syst. II: Expr. Briefs, vol. 55, no. 9, pp. 833–837, Sep. 2008.
PELOSI et al.: MULTIPLE ANTENNA SYSTEMS WITH INHERENTLY DECOUPLED RADIATORS
[48] J. Anguera, A. Andujar, A. Camps, C. Puente, and C. Picher, “Mitiga-
tion of the finger loading effect in handset antennas,” in Proc. 3th Eur.
Conf. on Antennas Propag. (EuCAP’2010), Apr. 12–16, 2010, pp. 1–4.
Mauro Pelosi was born in 1982 and is from
Picinisco, Italy. He received the B.Sc. and M.Sc.
degrees (summa cum laude) in telecommunications
engineering from University of Cassino, Cassino,
Italy, in 2004 and 2007, respectively, and the M.Sc.
degree in electrical engineering and the Ph.D.
degree in wireless communications from Aalborg
University, Aalborg, Denmark, in 2006 and 2009,
respectively.
Currently, he is a Postdoctoral Fellow at the De-
partment of Electronic Systems, Aalborg University.
He is also Deputy Project Manager for the Smart Antenna Front End (SAFE)
Project sponsored by the Danish National Advanced Technology Foundation.
His research interests include computational electromagnetics, innovative mul-
tiple antenna systems and antenna proximity effects with focus on the influence
of the user’s body. He is also involved in the COST Action 2100 on “Perva-
sive Mobile & Ambient Wireless Communications” and the ICT COST Ac-
tion IC1004 on “Cooperative Radio Communications for Green Smart Envi-
ronments.”
Mikael Bergholz Knudsen (S’99–M’01) was born
in 1964. He received the B.S. degree in electrical en-
gineering from Aarhus Teknikum, Denmark, in 1989,
and the M.S. and Ph.D. degrees from Aalborg Uni-
versity, Denmark, in 1992 and 2001, respectively.
In 1993, he joined Maxon Telecom A/S, Aalborg,
Denmark, where he designed RF circuitry for both
analog and digital mobile phones. From 1998 to
2001, he worked as an industrial Ph.D. student for
Siemens Mobile Phones A/S, Denmark, while he at
the same time studied at Aalborg University, CPK.
515
He is now with Intel Mobile Communications Denmark, where he is engaged
in the system design and development of RF transceiver chips for mobile
phones. His areas of interest include RF system design and handset antenna
performance including more than one antenna.
Gert Frølund Pedersen received the B.Sc. E.E. de-
gree (with honors) in electrical engineering from Col-
lege of Technology, Dublin, Ireland, in 1991, and the
M.Sc. E.E. and Ph.D. degrees from Aalborg Univer-
sity, Aalborg, Denmark, in 1993 and 2003, respec-
tively.
At present, he is a Full Professor heading the An-
tennas, Propagation and Radio Networking (APNet)
Group at Aalborg University. His research has fo-
cused on radio communication for mobile terminals
including small antennas, antenna-systems, propaga-
tion and biological effects and has more than 100 publications including more
than 15 patents. He has also worked as consultant within small antennas and
developed more than 50 dedicated designs for small mobile terminals starting
with the first internal antenna for mobile phones in 1993 with very low SAR,
First internal triple-band antenna in 1998 with low SAR and high efficiency and
various antenna diversity systems rated as the most efficient on the market. Re-
cently he has been involved in establishing the method to measure over the air
(OTA) communication performance for mobile terminals adopted by 3 GPP for
measurements also including the antenna. Further he is involved in small ter-
minals for 4G including several antennas (MIMO systems) and ultrawideband
antennas to enhance the data communication.
Prof. Pedersen is the Chairman of COST 2100 SWG 2.2 which is working
on the coming OTA Standard for multi-antenna terminal testing for 3 GPP and
CTIA. He is also the Project Manager for the Smart Antenna Front End (SAFE)
Project sponsored by the Danish National Advanced Technology Foundation.
516 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
A Pattern Reconfigurable U-Slot Antenna and Its
Applications in MIMO Systems
Pei-Yuan Qin, Y. Jay Guo, Senior Member, IEEE, Andrew R. Weily, Member, IEEE, and
Chang-Hong Liang, Senior Member, IEEE
Abstract—A new compact pattern reconfigurable U-slot an-
tenna is presented. The antenna consists of a U-slot patch and
eight shorting posts. Each edge of the square patch is connected
to two shorting posts via PIN diodes. By switching between the
different states of the PIN diodes, the proposed antenna can
operate in either monopolar patch or normal patch mode in
similar frequency ranges. Therefore, its radiation pattern can be
switched between conical and boresight patterns electrically. In
addition, the plane with the maximum power level of the conical
pattern can be changed between two orthogonal planes. Owing to
a novel design of the switch geometry, the antenna does not need
dc bias lines. The measured overlapping impedance bandwidth
of the two modes is 6.6% with a center
frequency of 5.32 GHz. The measured radiation patterns agree
well with simulated results. The antennas are incorporated in a
2 2 multiple-input-multiple-output (MIMO) orthogonal fre-
quency division multiplexing (OFDM) system to demonstrate the
improvement in system capacity. In the real-time MIMO-OFDM
channel measurement, it is shown that compared to omnidirec-
tional antennas, the pattern reconfigurable antennas can enhance
the system capacity, with 17% improvement in a line-of-sight
(LOS) scenario and 12% in a non-LOS (NLOS) scenario at a
signal-to-noise ratio (SNR) of 10 dB.
Index Terms—Microstrip antennas, multiple-input-multiple-
output (MIMO), reconfigurable antennas, slot antennas.
I. INTRODUCTION
I N the past few years, reconfigurable antennas have re-
ceived significant attention due to their ability to improve
the performance of wireless communication systems [1]–[10].
Typical parameters of an antenna that can be reconfigured are
frequency, radiation pattern, polarization or combinations of
the above. Pattern reconfigurable antennas have the potential to
Manuscript received June 10, 2010; revised May 10, 2011; accepted June 06,
2011. Date of publication October 25, 2011; date of current version February
03, 2012. This work was supported by the DIISR Australia-China special fund
CH080270.
P.-Y. Qin is with the Science and Technology on Antenna and Microwave
Laboratory, Xidian University, Xi’an, Shaanxi 710071, China and also with
the Department of Electronic Engineering, Macquarie University, North Ryde,
NSW 2109, Australia and CSIRO ICT Centre, Epping, NSW 1710, Australia
(e-mail: Peiyuan.qin@csiro.au).
Y. J. Guo and A. R. Weily are with the CSIRO ICT Centre, Epping, NSW
1710, Australia.
C.-H. Liang is with the Science and Technology on Antenna and Microwave
Laboratory, Xidian University, Xi’an, Shaanxi, 710071, China.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173439
0018-926X/$26.00 © 2011 IEEE
avoid noise sources by changing the null position, to save en-
ergy by better directing the signal toward intended users and to
provide larger coverage by redirecting the main beam [7]–[10].
Among the reported pattern reconfigurable antenna designs,
several [8]–[10] feature the capability to switch between
boresight and conical patterns. A conical radiation pattern is
generally one for which the maximum directivity is off bore-
sight (where boresight corresponds to the direction normal to
the plane containing the antenna) and the pattern shape resem-
bles a cone. In [8], a wide-band L-probe circular patch antenna
with dual feeds was presented. Unfortunately, an integrated
matching network consisting of switches needs to be designed
in order to reconfigure the radiation pattern electrically. In [9]
and [10], single feed pattern reconfigurable square-ring patch
antennas were designed with air gaps to increase the impedance
bandwidth, and dc bias networks were used to drive the PIN
diodes. In most conventional reconfigurable antenna structures,
dc bias lines are required to control the switching elements,
which can make the whole antenna structure more complicated
or even degrade the antenna performance. This hinders the
wide applications of reconfigurable antennas. Therefore, a
reconfigurable antenna structure without bias lines is desired.
Very recently, reconfigurable antennas have found new ap-
plications in adaptive multiple-input-multiple-output (MIMO)
systems [11]–[20], enabling the dynamic change of radiating
characteristics of each antenna element according to the usually
fast changing channel conditions. Generally, there are two
methods to increase the MIMO system capacity by employing
reconfigurable antennas. The first is to reduce the correlation
of sub-channels by using polarization or pattern reconfigurable
antennas. Specifically, this is implemented by switching be-
tween different configurations of reconfigurable antenna arrays
according to the varying channel conditions. The polarization
or pattern diversity in some of the configurations can be used
to realize low correlation of the sub-channels. The second is
to increase the signal power received by switching the antenna
radiation patterns according to the channel information. In
[12]–[15], the capacity of a MIMO system was improved
by using polarization reconfigurable antennas to reduce the
sub-channel correlation. In [16]–[20], pattern reconfigurable
antennas were incorporated in MIMO systems to enhance the
system capacity. In [16], a MIMO system with pattern recon-
figurable antennas was tested in an anechoic chamber with
artificial objects acting as the scatterers that made up a multi-
path environment. The system capacity increase was mostly
attributed to the increase of average receiver signal-to-noise
ratio (SNR) by changing the main beam direction. In [17]–[20],
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS 517

the capacity of a MIMO system was increased by exploiting

antenna pattern diversity to introduce sub-channel decorre-
lation to the MIMO system. For the experiments in [17], the
reference antenna was set to be one configuration of the recon-
figurable antenna. It is known that an omnidirectional antenna
can receive rich multipath, which can lead to low sub-channel
correlation [20]. Therefore, a capacity comparison between
systems with reconfigurable and omnidirectional antennas is
necessary for highlighting the effect of pattern diversity on the
sub-channel correlation.
In this paper, a new pattern reconfigurable microstrip U-slot
patch antenna using eight PIN diodes is proposed. Eight shorting
posts are implemented around the patch to change the operating
mode of the antenna from monopolar patch mode to normal
patch mode. In addition, the two modes are designed to res-
onate in similar frequency ranges. Therefore, the proposed an-
tenna can electrically reconfigure the radiation pattern between
conical and boresight patterns with an overlapping impedance
bandwidth. Furthermore, the plane with the maximum power
level of the conical pattern can be varied between two orthog-
onal planes when the antenna operates in the monopolar patch
mode.
To demonstrate the benefit of using the proposed reconfig-
urable antennas to increase the system capacity, four antennas
have been employed in a 2 2 MIMO-OFDM demonstrator.
Omnidirectional antennas are used as a reference for capacity
comparison. Channel measurements conducted in both line-of-
sight (LOS) and non-LOS (NLOS) indoor environments show
significant capacity enhancement.
Part of this work has been described in [21]. The present paper
extends the work in [21] significantly by providing the antenna
design principles and showing the equivalent circuits of the an-
tenna with PIN diodes. We also present a parametric study on
Fig. 1. Configurations of the patch antenna with PIN diodes (a) antenna in Ref.
the antenna resonant frequency and describe the measured re- [23] with two shorting posts; (b) antenna with four shorting posts; (c) antenna
flection coefficient, radiation pattern, gain and efficiency of the with eight shorting posts.
antenna. In addition, a discussion on the measured realized gain
of the antenna is given. Furthermore, the pattern reconfigurable
antenna is applied to a MIMO- orthogonal frequency division The effect of the loss of PIN diodes on the antenna gain is
multiplexing (OFDM) system and the capacity comparison be- discussed at the end of Section IV.
tween systems with pattern reconfigurable antennas and omni- This paper is organized as follows. In Section II, the operating
directional antennas is shown. principle and structure of the proposed antenna are described.
Compared with antennas in [8]–[10], the proposed antenna Section III presents parametric studies of the antenna. Simu-
has three main advantages. Firstly, only a single bias-tee, which lated and measured performances of the antenna are provided
superimposes the bias voltage on the RF signal, is needed to in Section IV. In Section V, the effects of the pattern reconfig-
control the PIN diodes in the proposed antenna. Consequently, urable antenna on the capacity of a 2 2 MIMO-OFDM system
the complex bias network for PIN diodes or the matching net- in indoor environments are analyzed. The paper concludes in
work for dual feeds is not required as part of the printed antenna Section VI with a summary and suggestions for further work.
structure, which greatly simplifies the device. Secondly, the pro-
posed antenna is compact and of low profile since it is designed II. ANTENNA DESIGN
on a single layer microwave substrate.
Thirdly, compared to antennas in [8]–[10] which can switch A. Design Guidelines
between two radiation patterns, the proposed antenna has Microstrip patch antennas excited in the normal patch mode
three different patterns in a similar frequency band. The larger for boresight radiation and monopolar mode for conical radia-
number of patterns gives the proposed antenna more flexibility tion have been reported in [22] and [23], respectively. In [23],
to improve the system capacity of a wireless link. However, a the monopolar mode is excited by two shorting posts located
disadvantage of the proposed antenna is the use of eight PIN to the left and right of the feeding point, which is shown in
diodes, the loss of which will reduce the antenna realized gain. Fig. 1(a). If PIN diodes are used to connect the shorting posts
518
Fig. 2. Simulated normalized radiation pattern (a) antenna in Ref. [23]; (b)
antenna in Fig. 1(b).
and the microstrip patch, it is possible for this single antenna
to operate in monopolar patch mode or normal patch mode by
switching between the different states of the PIN diodes. How-
ever, according to [23], the resonant frequency of the monopolar
mode is roughly a factor of 2.5 below the fundamental normal
patch mode. In order to design an antenna with two modes res-
onating at similar frequency ranges and having a reasonably
wide frequency bandwidth, we have taken three measures in the
design process.
Firstly, compared with the antenna in [23], we use two
shorting posts at either side of the feed point. In order to
simplify the bias network, the shorting posts are implemented
around the edge of the patch and connected to the patch via
PIN diodes, which is shown in Fig. 1(b). It is well known that
the resonant frequency of a patch antenna loaded with reactive
components can be varied depending on the type of reactance
used [24], [25]. When the shorting posts are connected to the
microstrip patch, the antenna operates in the monopolar patch
mode. In this case, the increase in the number of shorting posts
will increase the resonant frequency. On the other hand, when
all the shorting posts are disconnected from the patch, the
antenna operates in the normal patch mode. In this case, the
increase in the number of shorting posts will reduce the reso-
nant frequency. Therefore, the frequency difference between
the two modes becomes smaller as the number of the shorting
posts increases. The effect of the number of shorting posts on
the antenna resonant frequency will be detailed in Section III.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 3. Schematics of the pattern reconfigurable U-slot antenna.
Secondly, we examine the normalized far-field radiation pat-
terns for the antennas operating in monopolar patch mode in
Fig. 1(a) and (b), which is given in Fig. 2(a) and (b), respec-
tively. The antennas are analyzed using the time domain solver
of CST Microwave Studio [26]. For the antenna in Fig. 1(a), as
is described in [23], two identical conical patterns are located in
two orthogonal planes . For the antenna in Fig. 1(b),
the maximum power level of the conical pattern in the plane
are 6 dB greater than that in the plane . The
qualitative explanation of this behavior is that the two shorting
posts connected to each edge of the patch can be treated as a
shorting wall that suppresses the E-field at the centre of that
edge. Therefore, the maximum power level in the plane
is lower than that in the plane . In order to have another
similar conical pattern with the maximum power level located
at the plane for the antenna in Fig. 1(b), four shorting
posts are inserted into the substrate around the other two edges
of the microstrip patch, which can be seen in Fig. 1(c).
Finally, as the probe-fed microstrip patch antenna has a
narrow impedance bandwidth that precludes its use in typical
communication systems, a U-slot is etched on the patch to
increase its impedance bandwidth [27].
B. Antenna Structure
The layout of the proposed pattern reconfigurable U-slot an-
tenna is shown in Fig. 3. A U-slot is inserted into a square patch
of dimensions . Each side of the patch is connected with
two shorting posts via PIN diodes. The radius of the shorting
posts is 0.7 mm. The feeding probe connected to the U-slot
patch through the ground plane and substrate is offset from the
top edge of the patch by . Since the length of the PIN diode
is less than the width of the gap , conducting ring pads are
placed around the shorting posts to enable attachment of the PIN
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS
TABLE I
DIMENSIONS OF THE PATTERN RECONFIGURABLE U-SLOT ANTENNA
Fig. 4. Equivalent circuit for PIN diode (a) forward bias (b) reverse bias.
diodes. The parameters and dimensions of the antenna are given
in Table I.
Beam lead PIN diodes (MA4AGBLP912) are used as the
switching elements. The equivalent circuit used in the simula-
tion software is presented in Fig. 4. According to the PIN diode
datasheet [28], the resistor is 4 in the forward bias state
and the capacitor in the parallel circuit is 0.025 pF in the
reverse bias state. The resistor is 10 representing the
net dissipative resistance of the diode in the reverse bias state.
For the zero bias state, the value of is almost infinity which
is equal to an open circuit and the loss caused by the resistor
is negligible.
The orientation of the diodes is also shown in Fig. 3. As all
PIN diodes are mounted across the ground and the center patch,
only a bias tee attached to the SMA connector is needed to con-
trol the PIN diodes. When the bias voltage is supplied from the
coaxial probe, opposite bias conditions are applied to diodes in
group A and B due to their reversed orientation. When the dc
voltage is zero, all diodes are turned off. In this case, the an-
tenna operates in the normal patch mode and radiates a bore-
sight pattern (State I). When the dc voltage is negative, diodes
in group B are on, and the other diodes are off. In this case,
the antenna has four shorting posts connected and can be re-
garded as a monopolar patch antenna, which radiates a conical
pattern with the maximum power level in the z-y plane (State II).
Changing the polarity of the dc voltage from negative to posi-
tive, diodes in group A are on, and all the other diodes are off.
In this case, a similar conical pattern can be observed with the
maximum power level in the z-x plane (State III). The possible
radiation patterns of the reconfigurable U-slot antenna and the
corresponding diode states are summarized in Table II.
III. PARAMETRIC STUDY
Three important parameters which affect the input reflection
coefficients of the two modes of the proposed reconfigurable an-
tenna are the radius of the shorting posts, the distance between
519
TABLE II
THREE STATES OF THE PATTERN RECONFIGURABLE U-SLOT ANTENNA
shorting posts and the patch , and the number of the shorting
posts. Also it is well known that the U-slot plays an important
role on the performance of the patch antenna. Since a parametric
study of the U-slot has already been reported [29], we only ex-
amine the effects of the former three parameters in this paper.
The other parameters remain constant and their values are given
in Table I. As the shorting posts have the same effect on the re-
flection coefficients of States II and III for the monopolar patch
mode, only the results for State II are presented for the para-
metric analysis.
In order to better analyze the effect of the parameters, the
equivalent circuits of the antenna with PIN diodes for States I
and II are given in Fig. 5(a) and (b), respectively. In the equiv-
alent circuits, a parallel resonant circuit is used to model
the patch antenna with a U-slot. The purpose of the electrical
models in Fig. 5 is to give a physical insight into the behavior of
the antenna for the parametric analysis, but not to exactly pre-
dict the antenna input impedance. A similar method has been
used in [30] for the parametric analysis of a microstrip patch
antenna.
Fig. 5(a) shows the equivalent circuit of the antenna operating
in the normal patch mode. The microstrip patch with a U-slot is
represented by a circuit . For the normal patch
mode, only the PIN diodes and the shorting posts that are at-
tached to the radiating edges are considered in the equivalent
circuit, and the effects of the two PIN diodes and shorting posts
at each edge are combined together to simplify the equivalent
circuit. As the PIN diodes are all zero bias, according to the PIN
diode datasheet, is almost infinity. Therefore, the equivalent
circuit of the PIN diode in this state is an inductor in se-
ries with a capacitor . The parasitic capacitance between the
shorting posts and the patch is modeled by a capacitor . The
shorting post is represented by a shunt inductor . The imag-
inary part of the input admittance from the reference plane
on the right hand side of Fig. 5(a) is investigated by using CST
Microwave Studio. Simulation results show that it is capacitive
with a capacitance within the antenna operating frequency
range.
Fig. 5(b) shows the equivalent circuit of the antenna oper-
ating in the monopolar patch mode. As with the circuit of the
normal patch mode, the microstrip patch with a U-slot is also
represented by a circuit . In the monopolar patch
mode (State II), four PIN diodes (diodes A) are reverse bias.
According to the analysis in the last paragraph, the total effect
520 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 5. The equivalent circuits of the proposed antenna for (a) State I (normal patch mode) and (b) State II (monopolar patch mode).

of the shorting posts with PIN diodes in zero bias state is ca-
pacitive. The only difference between the zero and reverse bias
states is the value of the net dissipative resistor of the PIN
diodes. Therefore, a capacitor is used to model the total ca-
pacitive effect of the shorting posts connected with diodes A. A
resistor is used to represent the losses in the four PIN diodes.
On the other hand, four shorting posts are connected to the patch
by PIN diodes B that are in the forward bias state. Those PIN
diodes are modeled by an inductor in series with a resistor
. In addition, the effects of the shorting posts and the para-
sitic capacitance between the patch and the shorting posts are
represented by an inductor and a capacitor , respectively.
The imaginary part of the input admittance from the refer-
ence plane on the right hand side of Fig. 5(b) is investigated.
Simulation results show that it is inductive with an inductance
within the antenna operating frequency range.

A. The Radius of Shorting Posts

Fig. 6(a) and (b) show the effects of the radius of shorting
posts on the resonant frequencies of the monopolar patch and
normal patch modes, respectively. It is observed that the res-
onant frequency of the monopolar patch mode increases with
the radius of the shorting posts. This is due to the fact that
when the radius increases the inductance from the shorting posts
Fig. 6. Simulated performance of the proposed antenna as a function of (a)
reduces, which makes decrease; hence, the resonant fre- input reflection coefficient of monopolar patch mode; (b) input reflection coef-
quency of the parallel circuit in Fig. 5(b) increases. For the ficient of normal patch mode.
normal patch mode, the resonant frequency remains almost un-
affected by the radius change of shorting posts. This is because
that in Fig. 5(a) is almost stable within the changing range B. The Distance Between Shorting Posts and the Patch
of the shorting post radius, which is evidenced by the simulation Fig. 7(a) and (b) show the effects of on the resonant fre-
results. Therefore, the radius of the shorting posts can change quencies of the monopolar patch and normal patch modes, re-
the resonant frequency of the monopolar patch mode but has spectively. As seen from Fig. 7, the resonant frequency of the
little effect on that of the normal patch mode. monopolar patch mode increases when is reduced. However,
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS
Fig. 7. Simulated performance of the proposed antenna as a function of
(a) input reflection coefficient of monopolar patch mode; (b) input reflection
coefficient of normal patch mode.
the resonant frequency of the normal patch mode decreases with
. This can be attributed to the fact that when decreases, the
parasitic capacitance between the shorting posts and the patch
increases. The increased capacitance reduces the induc-
tive effect from the reference plane in Fig. 5(b), but in-
creases the capacitive effect from the reference plane in
Fig. 5(a) and the in Fig. 5(b). The decreased and the
increased have reverse effects on the resonant frequency of
the monopolar patch mode (the resonant frequency of a parallel
circuit). However, the simulation results show that the ef-
fect of outweighs that of , which leads to an increase
of the resonant frequency of the monopolar patch mode. On the
other hand, the larger capacitance will decrease the resonant
frequency of the parallel circuit in Fig. 5(a), which will shift the
resonant frequency of the antenna in normal patch mode to a
lower value.
C. The Number of the Shorting Posts
In the design guidelines (Part A of Section II), it is stated that
an increase in the number of shorting posts will increase the
resonant frequency of the monopolar patch mode, but reduce
that of the normal patch mode. The reason is as follows. For the
monopolar patch mode, the increased number of shorting posts
means that for the diodes in forward bias state, more inductors
521
Fig. 8. Photograph of the pattern reconfigurable U-slot antenna.
are paralleled, which makes the effective inductance de-
crease. And for the diodes in reverse bias state, more capaci-
tors are paralleled, which makes the effective capacitance
increase. Simulation results show that the decrease of the total
inductance outweighs the increase of the total capacitance on
the resonant frequency of the circuit in Fig. 5(b). There-
fore, the resonant frequency of the monopolar patch mode is
increased when the number of the shorting posts increases. On
the other hand, for the normal patch mode, since the PIN diodes
connected to the radiating edges are all zero bias, the increased
number of shorting posts means more capacitors are paralleled,
which makes increase; hence, the resonant frequency of the
normal patch mode (the resonant frequency of a parallel
circuit in Fig. 5(a)) will be decreased.
IV. SIMULATED AND MEASURED RESULTS OF THE ANTENNA
Based on the above analysis, a pattern reconfigurable U-slot
antenna was designed and fabricated. The proposed antenna was
etched on a 3.175-mm-thick RT/Duroid 5880 substrate (dielec-
tric constant , ). A photograph of the
fabricated prototype is shown in Fig. 8. Figs. 9 and 10 show
the simulated and measured reflection coefficients versus fre-
quency for three different states of the antenna, respectively.
Compared with the simulated results the measured resonant fre-
quencies for State II and III are slightly higher. This discrepancy
can be mostly attributed to the inaccuracies in the fabrication of
the shorting posts. As is shown in Section III, the resonant fre-
quency of the monopolar mode is quite sensitive to the radius
and position of the shorting posts. However, the simulated over-
lapping impedance bandwidth of the three
states is 6.5% with a center frequency of 5.24 GHz. The cor-
responding measured bandwidth is 6.6% centered at 5.32 GHz,
which agrees reasonably well with the simulated results.
Radiation patterns were measured for the three states of the
proposed antenna using a spherical near-field (SNF) antenna
measurement system. Simulated and measured normalized
522
Fig. 9. Simulated input reflection coefficient for the proposed antenna.
Fig. 10. Measured input reflection coefficient for the proposed antenna.
radiation patterns are compared for both co-polarization and
cross-polarization. Figs. 11 and 12 display the z-x and z-y
plane radiation patterns for the three states of the antenna at
5.3 GHz, respectively. For State I, boresight radiation patterns
with a maximum cross-polarization level of 20 dB are shown
in Figs. 11(a) and 12(a). For State II, a symmetrical conical
pattern with the maximum power level in the z-y plane directed
at (elevation angle) 44 is plotted in Figs. 11(b) and 12(b). For
State III, an asymmetrical conical pattern with the maximum
power level in the z-x plane directed at (elevation angle) 45 is
drawn in Figs. 11(c) and 12(c). It can be seen from Fig. 11(c)
that the pattern is asymmetrical and there is 1 dB difference
between the left and right maximum power level of the conical
pattern. This is due to the position of the probe feed. Simulation
results show that if we put the probe feed at the center of
the patch, the difference between the left and right maximum
power level in Fig. 11(c) will become smaller. However, in that
case the overlapping impedance bandwidth of the two modes
will be reduced. This can be viewed as a compromise for the
antenna to provide good overlapping impedance bandwidth and
radiation patterns. In Fig. 11, the simulated cross-polarization
patterns for the three states are not given since they are very
small compared to the measured ones.
Additionally, the realized gain was measured using the gain
comparison technique [31]. The losses of the cable and bias tee
have been calibrated out of the gain measurement. The mea-
sured gains for the three states are plotted in Fig. 13. The mea-
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
sured efficiencies of State I, State II and State III at 5.3 GHz are
86.6%, 45.1% and 45.4%, respectively, where the measured ef-
ficiency is obtained from the difference between the measured
gain and directivity. From Fig. 13, it can be seen that the mea-
sured gain of State III is 0.5 dB greater than that of State II,
which can be mostly attributed to the asymmetry of the conical
pattern of State III and slight variations in the resistance of in-
dividual diodes. Furthermore, at 5.3 GHz the gain of State I is
approximately 3 dB greater than those of State II and III, which
is mainly due to the loss of the PIN diodes. On the one hand,
for State II and III, four PIN diodes, each acting as a 4 re-
sistor , are attached to the antenna, but for State I all PIN
diodes are turned off. To examine the effect of , we have
simulated several different values of at 5.3 GHz. Simula-
tion results show that when decreases to zero the realized
gain increases by 2 dB and 1.75 dB for State II and III, respec-
tively. On the other hand, for State II and III, diodes with finite
value also have more losses than the diodes in zero bias state
which is approximately lossless. Simulation results show that at
5.3 GHz when changes from 10 to infinity the realized
gain increases by 0.32 dB and 0.27 dB for State II and III, re-
spectively. Therefore, the loss of State II and III is much greater
than that of State I and the corresponding gain is much lower.
In order to increase the gains of State II and III, low loss ele-
ments such as radio frequency microelectromechanical system
(RF MEMS) switches could be used. However, the disadvan-
tages of using currently available RF MEMS are the higher cost
and lower reliability than PIN diodes.
V. APPLICATION TO MIMO-OFDM SYSTEMS
Since the proposed pattern reconfigurable antenna is compact
and can vary its radiation pattern without dc bias network, it
can be easily applied to a MIMO system. The performance of
a MIMO system is affected by the spatial correlation which is
a function of the channel characteristics and the antenna array
properties. Therefore, the antenna elements can be treated as
additional parameters of the MIMO system. In this section, a
typical example for the application of pattern reconfigurable an-
tennas in a MIMO-OFDM system is described.
A. Antenna Array in MIMO-OFDM System
In previous sections, the design of a single pattern reconfig-
urable antenna is presented. In this sub-section, we incorporate
the pattern reconfigurable antennas into a 2 2 MIMO-OFDM
system.
At each end of the MIMO-OFDM system, there are two re-
configurable antennas working as a two-element array. How-
ever, this does not imply that the array operates as a traditional
beamforming array. Rather, each antenna has its own radiation
characteristics. As each reconfigurable antenna has three states,
there are nine configurations of the antennas at each end. For
each measurement, the configuration of the receive antennas is
the same as the transmit antennas. Table III gives nine configu-
rations of the reconfigurable antennas.
When two antennas are located close to each other, the mu-
tual coupling can produce pattern distortion and should be elim-
inated. Therefore, the spacing of the two antennas at both ends
is set to be one wavelength in order to keep the mutual coupling
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS 523

Fig. 11. Measured and simulated z-x plane normalized radiation patterns of the proposed antenna at 5.3 GHz (a) State I, (b) State II, (c) State III.

Fig. 12. Measured and simulated z-y plane normalized radiation patterns of the proposed antenna at 5.3 GHz (a) State I, (b) State II, (c) State III.

TABLE III
NINE CONFIGURATIONS OF THE RECONFIGURABLE ANTENNAS AT TRANSMIT
END

Fig. 13. Measured gain of the proposed antenna.

it provides a measure of antenna diversity performance. The

to an acceptable level. The measured results in Fig. 14 show that antenna diversity will be better if the correlation coefficient is
the mutual coupling is lower than 20 dB for all configurations lower. The envelope correlation coefficient can be calculated
at this spacing. The mutual coupling results for configurations using the farfield radiation patterns of the antennas [32]
4, 7 and 8 are not shown since they have the same results as the
configurations 2, 3 and 6, respectively.
For the reference antenna, in our study, commercially avail-
able omnidirectional antennas (Sky-Cross SMA-5250-UA) are
used that have omnidirectional patterns with peak gain at 2.2 (1)
dBi. The spacing of the two antennas is also set to be one wave-
length to keep the mutual coupling below 20 dB.
When evaluating the MIMO antenna array performance, the where is a complex vector indicating the electric field
envelope correlation coefficient is another critical parameter as radiated from the element. The envelop correlation coeffi-
524
Fig. 14. Measured mutual coupling coefficient for different antenna configu-
rations.
TABLE IV
ENVELOPE CORRELATION COEFFICIENT OF NINE CONFIGURATIONS OF THE
ARRAY
TABLE V
ENVELOPE CORRELATION COEFFICIENT OF THREE STATES OF THE ANTENNA
cient between the patterns generated at the two ports of the an-
tenna array for different configurations (Table III), as well as be-
tween the patterns excited by a single antenna for different states
(Table II), is estimated by using (1) according to the simulated
radiation patterns. Table IV gives the envelope correlation co-
efficient for different antenna configurations at 5.25 GHz. It is
evident that the envelope correlation coefficients of the config-
urations (2, 3, 4, 6, 7, 8) having pattern diversity are lower than
those (1, 5, 9) with the same antenna patterns, and they are com-
parable to the results of the antennas in [33] that employ both the
pattern and polarization diversity. Table V shows the envelope
correlation coefficient for different antenna states at 5.25 GHz.
It is observed that all the coefficients between different antenna
states are below 0.5, which satisfy the criterion for enabling the
antenna to provide a good level of diversity [32].
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 15. (a) Layout of the indoor MIMO-OFDM testing environments; (b) ori-
entation of the reconfigurable antennas.
B. MIMO-OFDM Demonstrator
Actual MIMO-OFDM channel coefficients were measured
by the MIMO-OFDM hardware demonstrator developed by
CSIRO ICT Centre (Sydney, Australia) [34]. It operates at 5.25
GHz and supports an operational bandwidth of 40 MHz. The
receive antennas are connected to an antenna array positioner
controlled by a computer. The channel training sequence is
designed to estimate the frequency response over 117 OFDM
subcarriers in a 40 MHz bandwidth with a subcarrier spacing
of 312.5 kHz.
C. Measurement Location and Process
The 2 2 MIMO-OFDM channel measurement was con-
ducted in the CSIRO ICT Centre indoor environment, which
consists of both concrete and gypsum-board walls, glass win-
dows and wooden doors. The channel was measured in LOS
over a 5 m distance and NLOS over an 8 m distance. The layout
of the LOS and NLOS testing scenarios is shown in Fig. 15(a).
For the LOS scenario, both the transmitter and receiver are lo-
cated in the same laboratory (Lab 1) equipped with some metal
bookshelves and cabinets. For the NLOS scenario, the trans-
mitter and receiver are placed in two adjacent laboratories (Lab
1 & 2).
Each reconfigurable antenna is placed vertically as shown in
Fig. 15(b). In the LOS scenario, the reconfigurable antennas
at transmit (Tx1) and receive ends are configured to face each
other. In the NLOS scenario, the reconfigurable antennas at
transmit (Tx2) and receive ends are configured to face the neg-
ative and negative directions, respectively. The reference
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS
Fig. 16. MIMO-OFDM demonstrator with reconfigurable antennas.
antennas are located with the omnidirectional pattern aligning
with the horizontal plane (x-y plane).
For each measurement, the antenna positioner moves the re-
ceive antenna array to 10 location samples with 0.05 wavelength
increments. The position of the transmit antennas is fixed.
In each scenario only the antenna configuration is changed.
Any other factors that can possibly vary the channel character-
istics, such as the number of the scatterers and their positions,
remain the same. The measurement is performed in the night
to avoid the human activities, so the testing environment is en-
tirely static. A photograph of the MIMO-OFDM demonstrator
with reconfigurable antennas at the receive end and the dc power
supply is shown in Fig. 16.
D. Channel Measurement and Capacity Estimation
The MIMO-OFDM channel is characterized by its coefficient
matrix , the element of which is the complex ratio
of the signal output from the th receive antenna over the signal
input to the th transmit antenna at the th OFDM subcarrier
and the th location sample. The Shannon capacity of a MIMO-
OFDM channel is given by [34]
(2)
where is the system capacity in bits/second/Hz, is the
number of sub-carriers, n is the number of location samples,
is the number of receive antennas, is the number of transmit
antennas, is the identity matrix, is the
normalized channel matrix, the superscript denotes the con-
jugate transpose, and SNR is the average signal-to-noise-ratio
over all receive array elements. The capacity of a wideband
channel is the average value of the capacities over all subcar-
riers of the MIMO-OFDM system and the 10 location samples.
It is convenient to use the normalized channel matrix so that
the capacity of the channel can be derived as a function of the
average SNR per receive antenna over all OFDM subcarriers,
525
receive antenna location samples and MIMO sub-channels. The
normalization is performed as follows:
(3)
where denotes the Frobenius norm of the channel matrix.
In the MIMO-OFDM channel measurement, the peak gain of
the omnidirectional antenna is 2.2 dBi. But the gain of different
states of the reconfigurable antennas varies from 2.8–6.4 dBi at
5.25 GHz. The antenna gain of each configuration is preserved
in . The lower gain of the omnidirectional antenna can lead
to a lower value of . Therefore, compared to the proposed
antenna, if all channel matrices are normalized to a common
factor, the reference antenna will suffer a capacity loss due to
the smaller .
In this section, we aim to show that the proposed antenna has
the ability to improve the system capacity by using pattern di-
versity to reduce the sub-channel correlation. If the gain of the
antenna is taken into account, the extent of the enhancement
of the system capacity derived only from the pattern diversity
will not be explicitly shown. Therefore, the gain effect on the
capacity should be eliminated in order to realize a fair compar-
ison. In order to separate the effect of pattern diversity on the
system capacity from the effect of antenna gain, two different
channel matrix normalization methods are employed.
The first is that the channel matrix of each antenna configu-
ration is normalized independently (Method I). In this case, the
antenna gain is not included in the capacity calculation. There-
fore, only the effect of the sub-channel correlation is kept in the
capacity evaluation [35]. The second is that the channel matrix
of each antenna configuration is normalized with respect to that
of the reference antenna (Method II). In this way, not only the
sub-channel correlation but also the relative received power dif-
ference, is preserved in the capacity calculation.
Figs. 17 and 18 show the system capacity percentage
improvement for different antenna configurations based on
channel normalization Method I and Method II, respectively.
For each scenario, the capacity improvement is the difference
in capacity between the system with one configuration of
reconfigurable antennas which gives the largest capacity and
the system with reference antennas. Then the improvement is
normalized with respect to the capacity of reference antennas.
For channel matrix normalization Method I, it can be ob-
served from Fig. 17(a) that compared with omnidirectional an-
tennas, the proposed pattern reconfigurable antennas improve
the system capacity by 17% and 12% for the LOS and NLOS
cases, respectively, at an SNR of 10 dB. This is because that
with antenna pattern diversity the correlation of sub-channels is
reduced. Therefore, the system capacity is improved, especially
for the environments with insufficient scatterers. In our experi-
ments the best configurations, which lead to the largest system
capacity improvement, are the configuration six and eight for
the LOS and NLOS scenarios, respectively. An explanation of
this is as follows. Although the antennas with omnidirectional
patterns in the horizontal plane are considered to be a good solu-
tion for MIMO systems as they can receive rich multipath in that
526
Fig. 17. Percentage improvement of the reconfigurable antenna MIMO system
capacity versus SNR for Method I. The benchmark is (a) omnidirectional an-
tenna and (b) antenna configuration 1.
Fig. 18. Percentage improvement of the reconfigurable antenna MIMO system
capacity versus SNR for Method II. The corresponding omnidirectional antenna
MIMO system capacity is used as the benchmark.
plane, it turns out that there are still many multipath components
outside the azimuth plane which can be received by other states
of the reconfigurable antennas. Therefore, system capacity im-
provement is possible by using pattern reconfigurable antennas.
In addition, since antenna configuration 1 (Table III) acts as
a two-element rectangular patch antenna array with boresight
radiation of each element, a comparison of the capacity of the
system with configuration 1 and the best configurations is con-
ducted, which is given in Fig. 17(b). It can be seen that at an
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
SNR of 10 dB, the system capacity is improved by 18% and
13% for the LOS and NLOS scenarios, respectively.
Furthermore, as is seen from Fig. 17, the percentage improve-
ment of the system capacity for the LOS scenario is greater than
that for the NLOS scenario. This is due to the fact that generally
the sub-channel correlation in an indoor LOS environment is
larger than that in an NLOS environment. Therefore the extent
of improvement of the system capacity by reducing the correla-
tion of MIMO sub-channels is larger for the LOS scenario.
For channel matrix normalization Method II, the best config-
uration is antenna configuration one (Table III) in terms of the
system capacity for both scenarios. It can be seen from Fig. 18
that at an SNR of 10 dB, the peak improvement is 285% and
264% for the LOS and NLOS scenarios, respectively, which are
much higher than those in Fig. 17. The reason that antenna con-
figuration one outperforms other configurations as well as om-
nidirectional antennas to such a great extent lies in the higher
gain of the antennas in configuration one. Specifically, the
of the system with configuration one is much greater than that
with other configurations. By Method II, this effect is taken into
account because the channel matrix of each antenna configura-
tion is normalized to a common value. In this way, the normal-
ized channel matrix of the antenna configuration one is scaled
up by a large factor due to the normalization; hence, the im-
provement of the capacity is significantly larger.
From channel matrix normalization Method I, it is known that
the pattern diversity of the proposed antenna can be exploited to
improve the system capacity by reducing the sub-channel cor-
relation, which is evidenced by the fact that antenna configura-
tions six and eight outperform the other configurations and the
omnidirectional one in the LOS and NLOS scenarios, respec-
tively. When the gain of different configurations is considered,
however, it is found that the effect of antenna gain outweighs
the effect of the channel decorrelation on the system capacity.
Therefore, the antenna configuration one which has the largest
antenna gain becomes the best in both scenarios. In this situa-
tion, the pattern diversity loses its effect on system capacity due
to the large gain difference of different antenna states.
In order to employ the pattern diversity of the proposed an-
tenna in a practical way that the gain difference is included in
the capacity calculation, we need to improve the realized gain
of antenna States II and III. As discussed at the end of the
Section IV, compared to State I, the lower gain of State II and III
is mostly attributed to the losses of PIN diodes. Therefore, low
loss switches, such as RF MEMS, should be used to make the
gains of States II and III close to that of State I. An updated an-
tenna prototype with RF MEMS could be designed and adopted
in our future work.
Reference [17] also demonstrates the capability of a pattern
reconfigurable antenna to improve the MIMO system capacity.
Compared to [17], our work has three main differences. First,
the mechanism of the pattern reconfigurability of the antenna in
[17] uses mutual coupling of two closely spaced antennas. The
pattern diversity will be reduced if the antenna separation is in-
creased. Therefore, the spatial diversity, which is another effec-
tive scheme [36] to improve the system capacity, can not be effi-
ciently employed together with the pattern diversity in a MIMO
system. However, as the way to reconfigure the pattern of our
QIN et al.: A PATTERN RECONFIGURABLE U-SLOT ANTENNA AND ITS APPLICATIONS IN MIMO SYSTEMS
antenna is changing the antenna operating modes, there is no
space limitation on our antenna. Second, in our experiments, at
an SNR of 10 dB, the capacity percentage improvement is 17%
and 12% for the LOS and NLOS scenarios, respectively. These
results are larger compared to the average 10% improvement
reported in [17]. Admittedly, since the capacity measurement is
heavily dependent on the testing environment, the improvement
could be significantly different for other test scenarios. Third,
as has already been mentioned in the introduction, the reference
antenna in [17] is chose to be one configuration of the antenna.
While in our work, omnidirectional antennas are used for refer-
ence. Since omnidirectional antennas can receive rich multipath
in the horizontal plane, this comparison can highlight the effect
of pattern diversity on MIMO system capacity.
VI. CONCLUSION
A new design of a pattern reconfigurable patch antenna is
presented. Shorting posts around the patch are used to change
the antenna operating modes. The antenna can switch between
three different radiation patterns by employing PIN diodes. It is
compact but can realize an overlapping frequency bandwidth of
6.6% with a center frequency of 5.32 GHz for the three states by
etching a U-slot into the patch. Compared to most conventional
pattern reconfigurable antennas, the proposed antenna does not
need additional dc bias lines to control the PIN diodes, which
greatly simplifies the antenna structure. Due to the simple struc-
ture and pattern reconfigurability, the antenna has the ability to
improve the performance of a wireless communication system
considerably.
To demonstrate the antenna’s capability of increasing
the system capacity, four antennas are applied to a 2 2 [13] P. Y. Qin, Y. J. Guo, and C. H. Liang, “Effect of antenna polarization
MIMO-OFDM system and the indoor channel measurement is
conducted. Capacities of the system with pattern reconfigurable
antennas and reference omnidirectional antennas are compared
based on two channel matrix normalization methods. The
comparison indicates that, for normalization Method I, the
system capacity is improved by 17% and 12% for the LOS and
NLOS scenarios, respectively, at an SNR of 10 dB by using
the pattern reconfigurable antennas. When the gain effect is
considered, the configuration with the greatest antenna gain
produces the largest system capacity. Therefore, in order to use
the pattern diversity to enhance the system capacity effectively
in a practical way, the gain of antenna States II and III needs to
be improved to approach that of State I.
Future research includes integrating RF MEMS switches
with the proposed antenna to increase the gain of antenna States
II and III. Furthermore, work will focus on designing combined
pattern and polarization reconfigurable antennas to achieve
higher system capacity improvement.
ACKNOWLEDGMENT
The authors would like to thank M. Shen for assisting with the
attachment of PIN diodes and Dr. H. Suzuki, Dr. Z. Chen and
Dr. W. Ni for the discussion on the subject of MIMO commu-
nication. In addition, we appreciate the constructive comments
from the anonymous reviewers.
527
REFERENCES
[1] G. H. Huff, J. Feng, S. Zhang, and J. T. Bernhard, “A novel radia-
tion pattern and frequency reconfigurable single turn square spiral mi-
crostrip antenna,” IEEE Microw. Wireless Compon. Lett., vol. 13, pp.
57–59, Feb. 2003.
[2] A. R. Weily, T. S. Bird, and Y. J. Guo, “A reconfigurable high-gain
partially reflecting surface antenna,” IEEE Trans. Antennas Propag.,
vol. 56, no. 11, pp. 3382–3390, Nov. 2008.
[3] P. Y. Qin, A. R. Weily, Y. J. Guo, T. S. Bird, and C. H. Liang, “Fre-
quency reconfigurable quasi-Yagi folded dipole antenna,” IEEE Trans.
Antennas Propag., vol. 58, no. 8, pp. 2742–2747, Aug. 2010.
[4] S. V. Hum and H. Y. Xiong, “Analysis and design of a differen-
tially-fed frequency agile microstrip patch antenna,” IEEE Trans.
Antennas Propag., vol. 58, no. 10, pp. 3122–3130, Oct. 2010.
[5] P. Y. Qin, A. R. Weily, Y. J. Guo, and C. H. Liang, “Polarization re-
configurable u-slot patch antenna,” IEEE Trans. Antennas Propag., vol.
58, no. 10, pp. 3383–3388, Oct. 2010.
[6] F. Yang and Y. Rahmat-Samii, “A reconfigurable patch antenna using
switchable slots for circular polarization diversity,” IEEE Microw.
Wireless Compon. Lett., vol. 12, no. 3, pp. 96–98, Mar. 2002.
[7] S. Zhang, G. H. Huff, J. Feng, and J. T. Bernhard, “A pattern reconfig-
urable microstrip parasitic array,” IEEE Trans. Antennas Propag., vol.
52, no. 10, pp. 2773–2776, Oct. 2009.
[8] S. L. S. Yang and K. M. Luk, “Design a wide-band L-probe patch an-
tenna for pattern reconfigurable or diversity applications,” IEEE Trans.
Antennas Propag., vol. 54, no. 2, pp. 433–438, Feb. 2006.
[9] S. H. Chen, J. S. Row, and K. L. Wong, “Reconfigurable square-ring
patch antenna with pattern diversity,” IEEE Trans. Antennas Propag.,
vol. 55, no. 2, pp. 472–475, Feb. 2007.
[10] W. L. Liu, T. R. Chen, S. H. Chen, and J. S. Row, “Reconfigurable
microstrip antenna with pattern and polarization diversities,” Electron.
Lett., vol. 43, no. 2, pp. 77–78, Jan. 2007.
[11] M. A. Jensen and J. W. Wallace, “MIMO wireless channel modeling
and experimental characterization,” in Space-Time Processing for
MIMO Communications, A. B. Gershman and N. D. Sidiropoulos,
Eds. West Sussex, U.K.: Wiley, 2005.
[12] P. Kyritsi, D. C. Cox, R. A. Valenzuela, and P. W. Wolniansky, “Effect
of antenna polarization on the capacity of a multiple element system in
an indoor environment,” IEEE J. Sel. Areas Commun., vol. 20, no. 6,
pp. 1227–1239, Aug. 2002.
diversity on MIMO system capacity,” IEEE Antennas Wireless Propag.
Lett., vol. 9, pp. 1092–1095, Dec. 2010.
[14] B. A. Cetiner, E. Akay, E. Sengul, and E. Ayanoglu, “A MIMO system
with multifunctional reconfigurable antennas,” IEEE Antennas Wire-
less Propag. Lett., vol. 5, pp. 463–466, Dec. 2006.
[15] A. Grau, J. Romeu, M. J. Lee, S. Blanch, L. Jofre, and F. D. Flaviis, “A
dual-linearly-polarized MEMS-reconfigurable antenna for narrowband
MIMO communication systems,” IEEE Trans. Antennas Propag., vol.
58, no. 1, pp. 4–17, Jan. 2010.
[16] J. D. Boerman and J. T. Bernhard, “Performance study of pattern recon-
figurable antennas in MIMO communication systems,” IEEE Trans.
Antennas Propag., vol. 56, no. 1, pp. 231–236, Jan. 2007.
[17] D. Piazza, N. J. Kirsch, A. Forenza, R. W. Heath, Jr., and K. R. Dan-
dekar, “Design and evaluation of a reconfigurable antenna array for
MIMO systems,” IEEE Trans. Antennas Propag., vol. 56, no. 3, pp.
869–881, Mar. 2008.
[18] P. Mookiah, D. Piazza, and K. R. Dandekar, “Reconfigurable spiral
antenna array for pattern diversity in wideband MIMO communica-
tion systems,” presented at the IEEE Antennas and Propag. Society
Int. Symp., Jul. 2008.
[19] H. K. Pan et al., “Increasing channel capacity on MIMO system
employing adaptive pattern/polarization reconfigurable antenna,”
presented at the IEEE Antennas and Propag. Society Int. Symp., Jun.
2007.
[20] D. Piazza, M. D’Amico, and K. R. Dandekar, “Performance improve-
ment of a wideband MIMO system by using two-port RLWA,” IEEE
Antennas Wireless Propag. Lett., vol. 8, pp. 830–834, 2009.
[21] P. Y. Qin, A. R. Weily, Y. J. Guo, C. H. Liang, and Y. Cai, “A pattern
reconfigurable U-slot patch antenna,” presented at the IEEE Antennas
and Propag. Society Int. Symp., Jul. 2010.
[22] I. J. Bahl and P. Bhartia, Microstrip Antennas. New York: Artech
House, 1980.
[23] C. Delaveaud, P. Leveque, and B. Jecko, “New kind of microstrip an-
tenna: The monopolar wire-patch antenna,” Electron. Lett., vol. 30, no.
1, pp. 1–2, Jan. 1994.
528 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[24] P. Bhartia and I. J. Bahl, “Frequency agile microstrip antennas,” Mi-
crow. J., pp. 67–70, Oct. 1982.
[25] D. L. Sengupta, “Resonant frequency of a tunable rectangular patch
antenna,” Electron. Lett., vol. 20, no. 15, pp. 614–615, Jul. 1984.
[26] CST Studio SuiteTM 2009 Computer Simulation Technology, Ger-
many.
[27] T. Huynh and K. F. Lee, “Single-layer single-patch wideband mi-
crostrip antenna,” Electron. Lett., vol. 31, no. 16, pp. 1310–1312, Aug.
1995.
[28] M/A-COM Data Sheet for MA4AGBLP912 Beam Lead PIN Diode.
[29] S. Weigand, G. H. Huff, K. H. Pan, and J. T. Bernhard, “Analysis and
design of broadband single-layer rectangular U-slot microstrip patch
antennas,” IEEE Trans. Antennas Propag., vol. 51, no. 3, pp. 457–468,
Mar. 2003.
[30] J. Anguera, E. Martinez, C. Puente, C. Borja, and J. Soler, “Broad-
band dual-frequency microstrip patch antenna with modified Sierpinski
fractal geometry,” IEEE Trans. Antennas Propag., vol. 52, no. 1, pp.
66–72, Jan. 2004.
[31] C. A. Balanis, Antenna Theory: Analysis and Design, 3rd ed. New
York: Wiley, 2005.
[32] R. Vaughan and J. Andersen, “Antenna diversity in Mobile Communi-
cations,” IEEE Trans. Veh. Technol., vol. VT-36, no. 4, pp. 149–172,
Nov. 1987.
[33] Z. N. Chen and T. S. P. See, “Diversity and its applications in ultra-
wideband antennas,” presented at the IEEE Int. Workshop on Antenna
Tech., Santa Monica, CA, Mar. 2009.
[34] H. Suzuki, T. V. Tran, and I. B. Collings, “Characteristics of MIMO-
OFDM Channels in indoor environments,” J. Wireless Commun. Netw.,
vol. 2007, no. 19728, Jan. 2007.
[35] M. A. Jensen and J. W. Wallace, “A review of antennas and propagation
for MIMO wireless communications,” IEEE Trans. Antennas Propag.,
vol. 52, no. 11, pp. 2810–2824, Nov. 2004.
[36] C. B. Dietrich, Jr., K. Dietze, J. R. Nealy, and W. L. Stutzman, “Spatial,
polarization, and pattern diversity for wireless handheld terminals,”
IEEE Trans. Antennas Propag., vol. 49, no. 9, pp. 1271–1281, Sep.
2001.
Pei-Yuan Qin was born in Liaoning, China, in 1983.
He received the Bachelor of Engineering degree
in electronic engineering from Xidian University,
Xi’an, China, in 2006. He is currently a joint Ph.D.
student of Xidian University, China and Macquarie
University, Australia.
From 2008 to 2010, he was a Ph.D. student vis-
itor to the CSIRO ICT Centre, Australia funded by
the China Scholarship Council (CSC). His research
interests are in the areas of reconfigurable antennas
and filters, reconfigurable reflectarrays, and MIMO
communications.
Y. Jay Guo (SM’96) received the Bachelor Degree
and Master Degree from Xidian University, Xi’an,
China, in 1982 and 1984, respectively, and the Ph.D.
degree from Xian Jiaotong University, China, in
1987. He was awarded an honorary Ph.D. degree in
1997 by the University of Bradford, U.K., for his
world leading research in Fresnel antennas.
Currently, he is the Research Director of Advanced
Broadband Networks and Services Theme, CSIRO
ICT Centre, Australia, and the Director of the Aus-
tralia China Research Centre for Wireless Communi-
cations. From August 2005 to January 2010, he served as the Research Director
of the Wireless Technologies Laboratory, CSIRO ICT Centre. Prior to joining
CSIRO, he held various senior positions in a number of international companies
including Fujitsu, Siemens and NEC, managing the development of advanced
technologies for the third generation (3G) mobile communications systems. He
is an Adjunct Professor at Macquarie University, Australia, and a Guest Pro-
fessor at the Chinese Academy of Science (CAS) and Shanghai Jiaotong Univer-
sity. He has published three technical books Fresnel Zone Antennas, Advances
in Mobile Radio Access Networks, and Ground-Based Wireless Positioning, 59
journal papers and over 80 refereed international conference papers. He holds
18 patents in wireless technologies.
Dr. Guo is a Fellow of IET. He was the recipient of the Australian Engi-
neering Excellence Award and the CSIRO Chairman’s Medal. He has served
on the organizing and technical committees of numerous national and interna-
tional conferences. He was Chair of the Technical Program Committee (TPC)
of 2010 IEEE Wireless Communications and Networking Conference (WCNC)
and 2007 IEEE International Symposium on Communications and Information
Technologies (ISCIT), and is the TPC Chair of IEEE ISCIT2012. He has been
the Executive Chair of Australia China ICT Summit since 2009. He was a Guest
Editor of the IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, Special
Issue on Antennas and Propagation Aspects of 60–90 GHz Wireless Communi-
cations. Currently, he is serving as a Senior Guest Editor for the IEEE JOURNAL
ON SELECTED AREAS IN COMMUNICATIONS, Special Issue on Challenges and
Dynamics for Unmanned Autonomous Vehicles.
Andrew R. Weily (S’96–M’01) received the B.E. de-
gree in electrical engineering from the University of
New South Wales, Australia, in 1995 and the Ph.D.
degree in electrical engineering from the University
of Technology Sydney (UTS), Australia, in 2001.
From 2000 to 2001, he was a Research Assistant
at UTS. He was a Macquarie University Research
Fellow then an ARC Linkage Postdoctoral Research
Fellow from 2001 to 2006 with the Department of
Electronics, Macquarie University, Sydney, NSW,
Australia. In October 2006, he joined the Wireless
Technology Laboratory at CSIRO ICT Centre, Sydney. His research interests
are in the areas of reconfigurable antennas, EBG antennas and waveguide
components, leaky wave antennas, frequency selective surfaces, dielectric
resonator filters, and numerical methods in electromagnetics.
Chang-Hong Liang (M’80–SM’83) was born in
Shanghai, China, in December 1943. He gradu-
ated from Xidian University (Formerly Northwest
Telecommunications Institute), Xi’an, China, in
1965 and continued his graduate studies until 1967.
From 1980 to 1982, he worked at Syracuse Univer-
sity, New York, as a Visiting Scholar. Since 1986, he
has been a Professor and Ph.D. student advisor in the
School of Electronic Engineering, Xidian University.
He has published numerous papers and five books.
He has wide research interests, which include com-
putational microwave and computational electromagnetics, microwave network
theory, microwave measurement method and data processing, lossy variational
electromagnetics, electromagnetic inverse scattering, and electromagnetic com-
patibility.
Prof. Liang is a fellow of Chinese Institute of Electronics (CIE) and a Director
of the Academic Committee of National Key Lab of Antenna and Microwave
Technology. He has received the titles of “National Distinguished Contribution”
and “National Excellent Teacher.” He is the Editor-in-Chief of the Journal of
Xidian University.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Multiple Element Antenna Efficiency and its Impact
on Diversity and Capacity
Jane X. Yun, Student Member, IEEE, and Rodney G. Vaughan, Fellow, IEEE
Abstract—A desirable characteristic of a multiple element
antenna (MEA) is to be compact, but a smaller size tends to lead
to higher ohmic and mutual coupling losses. A metric for the
efficiency of the MEA would help clarify the tradeoffs between
compactness and performance. In a MIMO/diversity antenna, the
total efficiency seen at each port directly affects the signal-to-noise
ratio (SNR) in the diversity branch. The SNR after diversity
combining governs the performance of the diversity antenna
system. In this paper, MEA efficiencies is therefore discussed
and formulated in the context of mutual coupling and diversity
combining. The impact of MEA efficiency on the diversity gain
and the information theoretical capacity is also formulated and
demonstrated using measurements of example MEAs. With these
formulations, an equivalent number of idealized (lossless, uncor-
related, uncoupled, equal power) branches can be found for an
MEA, and this defines the diversity order and the capacity order
of the MEA. With this metric, the performance of different MEAs
can be compared.
Index Terms—Antenna correlation, capacity, diversity gain,
MEA total efficiency, multiple element antenna (MEA), mutual
coupling.
I. INTRODUCTION
E FFICIENCIES of single element antennas are well under-
stood and defined in IEEE Standards [1], [2]. But for an
MEA used for MIMO communications, definitions of efficien-
cies are not yet clarified. Since power is transferred between
elements by mutual coupling, the antenna efficiencies of MEA
elements are not only decided by the element loss, but also by
the mutual coupling and the diversity combining method which
includes the terminations. A formulation of the MEA efficiency
needs to include these effects, and it should apply to both the
transmit case and the receive case.
The efficiency of each element directly impacts the MEA
transmitting and receiving capability. In an MIMO system, the
signal-to-noise ratio (SNR) of a receive branch is proportional
to the efficiency of the element. The combination of the branch
SNRs governs the antenna diversity gain and its capacity perfor-
mance. How the MEA efficiency impacts the MEA performance
in a Rayleigh fading scenario is discussed in [3], but an impor-
tant part of the MEA efficiency for compact designs,—mutual
Manuscript received June 24, 2010; revised January 26, 2011; accepted
September 23, 2011. Date of publication October 25, 2011; date of current
version February 03, 2012.
The authors are with the School of Engineering Science, Simon Fraser
University, Burnaby, BC V5A 1S6, Canada (e-mail: xyun@sfudotca;
rvaughan@sfudotca).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173444
0018-926X/$26.00 © 2011 IEEE
529
coupling loss,—is omitted in order to study the impact of ef-
ficiency caused by the element ohmic loss and impedance mis-
match loss only. Also, averaged efficiency (cf.[3]) or the highest
efficiency among all the elements (cf.[4]) is used, instead of con-
sidering the impact of each element when the total efficiency
of each element is different. Often, MIMO antennas must be
mounted on complex platforms. The available space can be crit-
ical and the designer may be forced to use different types of ele-
ments. In fact, different types of elements are often deployed in
order to reduce correlation and mutual coupling. Different el-
ement types normally have different radiation efficiencies. In
some compact implementations, this difference can be up to
about 6 dB.
This paper develops efficiency formulas for MEAs. The for-
mulation reveals the duality of the embedded element trans-
-mit efficiency and the embedded element receive efficiency. The
multiple conditions on the circuit model impedances can be
identified in order to make these efficiencies the same. Also, we
use the terms total transmit/receive efficiency of the embedded
element for the total efficiency of the embedded element. The
loss due to mutual coupling, which can be a dominant part of
the antenna loss, changes with the type of diversity combining
and the terminations of the elements. Therefore, the formula-
tion of the efficiencies considers the embedded element losses
with different diversity schemes. The analysis of the losses is
demonstrated with measurements of an example MEA.
An MEA total efficiency, in the form of a matrix containing
the total efficiency of each embedded element, accounts for the
impact of individual element efficiencies on the diversity gain
and capacity. By comparing these performance metrics before
and after the MEA total efficiency is included, the impact of
antenna efficiency is separated from the impact of antenna cor-
relation. The formulation is not limited to Rayleigh channels but
Rayleigh channels are used in the examples.
The rest of paper is laid out as follows. Section II develops
the efficiency formulas for an MEA. Section III takes this ef-
ficiency and incorporates it into the MEA diversity gain, and
Section IV does the same for the information theoretical ca-
pacity. Section V concludes the paper.
II. EMBEDDED ELEMENT EFFICIENCIES OF MEA
For single element antennas, the radiation efficiency is
expressed in terms of the transmit case. It is well established
and featured in Standards [1], [2]. From a simplified cir-
cuit model, the radiation efficiency is normally written as
, where is the radiation
resistance and is the ohmic resistance of the antenna. To
include the impedance mismatch, the antenna total efficiency
can be expressed in the form where
530 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 1. Transmission line circuit model of the th embedded element of an
-element MEA, when the th element is transmitting and the other elements
are present and terminated. The transmit source resistance is , and the other
elements are loaded with their transmitting source resistances, denoted
(a) is the th transmitting element and (b) is one of the other terminated elements
(taking the th element as an example here). For simultaneous combining, there
are similar circuits to (b). is the power accepted by the network
to the right of the dashed line, from the th transmitting element. The source
reactance and the antenna self reactance are assumed to be tuned out for sim-
plicity.
and is the reflection coefficient at the
antenna port.
However, for MEAs, efficiencies are impacted by mutual
coupling, the diversity combining, and the terminations of
the elements. Mutual coupling causes power transfer from
one element to other elements. This transferred power is then
dissipated in different ways: in the terminating resistances; as
ohmic losses in the elements and the transmission lines; and as
radiation from the elements. In practice, there are also ohmic
loss and radiation from the MEA support structure, and this is
part of the “embedded element”. In a simplified circuit model
these powers can be included into the element’s radiation and
ohmic losses, respectively.
In terms of the termination of the elements, diversity schemes
can be categorized as simultaneous (including Optimal Com-
bining (OC), Maximum Ratio Combining (MRC), Equal Gain
Combining (EGC), and Selection Combining (SC)) and non-
simultaneous combining (Switch Combining (SwC)), as dis-
cussed in Appendix A. The MEA efficiency is formulated based
on the analysis of the losses for both combining schemes in
the transmit and receive cases, respectively. The analysis of the
losses is confirmed with the measurement of an example of a
transmitting MEA.
A. Transmitting MEAs
Fig. 1 gives the transmission line circuit model of the th em-
bedded element of an -element MEA, when only the th ele-
ment is transmit excited and all the other elements are present,
not transmit excited, and terminated with their transmit source
resistances. The dashed line represents the transmit port of the
th embedded element. To the left hand side of the dashed line
is the th transmit source, and the right hand side is the em-
bedded element, which includes the th transmitting element in
Fig. 1(a) and the other elements terminated and excited by
mutual coupling. One of these terminated elements is depicted
in Fig. 1(b). The reactance of the power source and the antenna
self reactance are, in principle, able to be tuned out, so reac-
tances are not included for brevity, although their inclusion is
straightforward.
In Fig. 1(a), the th element is excited with the transmit source
, which is the only external power source of the th em-
bedded MEA element. The source resistance is , and the
resistance representing the ohmic loss of the transmission line
connecting the power source to the th element is . The ra-
diation resistance, , is associated with the radiation of the
th element in the presence of other elements. is the asso-
ciated ohmic loss of this element. The current causes power
loss in the other elements if there is non-zero mutual coupling.
The total power lost from the th element to all the other ele-
ments through mutual coupling is represented via the sum of
the voltage sources in the circuit model in
Fig. 1(a).
. In Fig. 1(b), each of the non-transmit-excited elements is
excited by mutual coupling. Taking the th element as an ex-
ample, the total mutual coupling excitation on this element is
. Some of the power transferred by mutual
coupling is then radiated by this element via , and the
rest dissipates as ohmic loss in and (the transmit
source resistance of the th element) and .
It is emphasized that the currents and in Fig. 1 are the
currents caused by the transmit excitation of the th element
only. For simultaneous combining, all the elements are excited
by their own transmitters, so there are circuits similar to that
in Fig. 1 for the MEA, and the total current in each element is
the sum of these circuit currents, by superposition. For non-si-
multaneous combining, the unselected elements are open cir-
cuited, i.e., , from the circuit model view-
point. In a physical situation with real-world devices this may
not mean a zero current, but the circuit model still includes any
mutual coupling. This is because the selected element is treated
as embedded, so that any mutual coupling with the unselected
elements is included through the embedded element resistances
and .
1) For Simultaneous Combining (All Elements Connected
to Fixed Transmit Loads): For the th embedded element, al-
though only the th element is transmit excited, the radiation
is not only from this element but also from the other elements
through mutual coupling.
The th embedded element transmit efficiency for si-
multaneous combining is the ratio of the total power radiated
from all the elements (this is denoted below as )
to the power accepted by the network (to the right of the dashed
line in Fig. 1) from the transmit source (to the left of the dashed
line), denoted . So the embedded element transmit effi-
ciency of the th element is written as
(1)
where
(2)
YUN AND VAUGHAN: MULTIPLE ELEMENT ANTENNA EFFICIENCY AND ITS IMPACT ON DIVERSITY AND CAPACITY 531

and is the total power

radiated by all the elements from the excitation of the th
element; is the ohmic loss in all the elements;
is the loss in all
the loads of the non-transmit-excited elements.
is the ohmic loss in all the transmission
lines of the non-transmit-excited elements.
When the situation reduces to a single element, the above
transmit efficiency is the same as the classical single antenna
radiation efficiency .
The total transmit efficiency of the embedded element at the
th transmit port includes the matching efficiency at the transmit
port, , as well as its embedded transmit efficiency, as
Fig. 2. Transmission line circuit model of th embedded element of the -ele-
(3) ment MEA, when the th element is receiving and the other elements are present
and terminated. (a) is one of the terminated elements (taking the th element as
where an example here). For simultaneous combining, there are similar circuits
as (a). (b) is the th receiving element. is the power absorbed by the
network (to the left of the dashed line) from the receiving port (the dashed line).
(4)

and is the reflection coefficient at the port of the th element.

1) For Simultaneous Combining (All Elements Connected to
2) For Non-Simultaneous Combining With all Unselected El-
Fixed Receiving Loads): The th embedded element receive ef-
ements Open Circuited: In this case, the currents of the unse-
ficiency is defined here as the ratio of the power deliv-
lected elements in the circuit model are zero, so the radiated
ered to the loads of all the elements (the wanted power) to the
power and losses induced on these elements are zero. The th
absorbed power of the MEA owing to the reception of the th
embedded element transmit efficiency is then in the same form
element, as
as the classical single antenna radiation efficiency:

(5)
(7)
where
where
(6)

B. Receiving MEAs
In Fig. 2, the dashed line represents the receive port of the (8)
th embedded element. The left hand side of the dashed line
is the th receiving embedded element, including the receiving
and is the absorbed power at the embedded element;
element and the other elements. Again, there are similar
circuits to Fig. 2(a) for the non-receiving elements, but only the and is the radiated power from all the non-
th element is shown here. receiving elements and is the ohmic loss on
To the right hand side of the dashed line in Fig. 2(b), the these elements, both excited by mutual coupling. As indicated
voltage source, , provides the received power on the in Fig. 2(b), The reflected power is
receive port, and it is the only external source for this em- the total received power minus the absorbed power , so
bedded element circuit model. is the power absorbed it is not part of . It changes the total receive efficiency
(discussed in [5] and [6]) by the th embedded element (to the of the embedded element.
left of the dashed line). The absorbed power includes the power The total receive efficiency of the embedded element at the th
dissipated in all the receiving loads (which is considered as receive port , analogously to (3) and (4), is
the wanted power); the power dissipated as the ohmic losses
of all the elements and transmission lines (which is “wasted” (9)
power); and the power radiated by all the non-receiving ele-
ments (which is excited by mutual coupling and considered as where, as above
“wasted” power).
If there is an impedance mismatch at the receive port (the (10)
dashed line in Fig. 2), some of the received power is reflected in
a transmission line style, back to the th antenna element, where For a single element receive antenna, (7) reduces to
it is re-radiated by and dissipated . . This receive efficiency is not the same
532 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

as the radiation efficiency. In fact it is not an antenna param-

eter, because in the receive case the metric is to measure how
much power is delivered to the load (which is not part of the
antenna) rather than to radiation. However, (7) and (8) are sim-
ilar to (1) and (2), except that and , and
and , are interchanged respectively. The efficiency for
the receive case and the transmit case, although having a dif-
ferent physical basis, can be related in the sense of dual equa-
tions. The receive and transmit efficiencies are often assumed to
be the same in MIMO communications system studies, but for
this assumption, specific conditions are required related to the
impedance matching, as follows.
Fig. 3. (a) Photo and (b) schematic of the hollow cube MEA example.
Consider an MEA which is impedance matched across the
dashed line in Figs. 1 and 2 for both the transmit and receive
case, so
for all , and (2) and (8) are equivalent. It follows
that in (1) equals to in (7) only when
and the same combining
detail is deployed. Under the above conditions, an embedded
element receive efficiency can be found from the embedded
element transmit efficiency, for the same MEA with the same
combining detail.
2) For Non-Simultaneous Combining With all Unselected
Elements Open Circuited: Similar to the non-simultaneous
transmit case, since the unselected elements are open circuited,
the power lost on these elements is zero in the circuit model.
The embedded element receive efficiency is, therefore, in the
same form as that for a single element antenna: Fig. 4. (a) Matching efficiency, total efficiency and (b) transmit efficiency of
each embedded element in the slot cube MEA example for simultaneous (simu)
and non-simultaneous (non-simu) diversity schemes.
(11)

where
(12)
In the above formulas, and measure how much
wanted power is radiated/received per unit accepted/absorbed
power by the embedded element. On the other hand, these also
indicate how much power is wasted through mutual coupling in
the forms of ohmic loss and re-radiation. It is difficult to phys-
ically measure and (although they could be simu-
lated) with their definitions in (1), (2) and (7), (8). Instead, with
(3), (4), (9) and (10), they can be found from the embedded el-
ement total efficiency and matching efficiency, both of which
can be measured. In Section III, an example MEA is measured
to demonstrate this process. The results demonstrate the impact
of the diversity scheme on the power radiated and lost in the
embedded elements.
C. Demonstration of MEA Efficiencies With an Example MEA
Fig. 3(a) is a 12-element hollow slot cube MEA [7] without
the feed lines, and Fig. 3(b) demonstrates the schematic coaxial
feed for the 10th element. The MEA is measured for simulta-
neous combining (one port is excited and the remaining ports are
terminated with 50 ) and non-simultaneous combining (one
port is excited and the rest are open circuited). The reflection
coefficient of each port is measured with a multiport Network
Analyzer for convenience (Agilent E5071B and E5091A [8])
and these results are given in [7]. The derived from (4)
is given in Fig. 4(a). The total efficiency of each embedded el-
ement is measured with a 5 meter spherical near field chamber
(SG64 [9]) and the results are included in Fig. 4(a). The de-
rived from (3) for each diversity scheme are compared in
Fig. 4(b).
In Fig. 4, for simultaneous combining, each embedded ele-
ment of the MEA has wideband match to 50 [7], so is
close to unity. The mutual coupling between any two elements
is low but not negligible (transmission coefficients
are mostly lower than , as given in [7], and the reflec-
tion coefficients and losses are low). With a large number of the
MEA elements , the total power lost from the excited
element to the loads of the non-excited elements (by mutual cou-
pling) is significant. Therefore, as can be seen from (1) and (2),
the resulting is low, which causes a low . This
is demonstrated with the measured of each port in
Fig. 4(a).
For non-simultaneous combining, the non-excited elements
are open circuited, so, based on (5) and (6), there is no power lost
to the loads of these elements through mutual coupling (in the
circuit model), and is much higher, as shown in Fig. 4(b).
is thereby mainly decided by . In this partic-
ular example, is lower than in the simultaneous com-
bining case, owing to a worse impedance match that the em-
bedded element has for the non-simultaneous combining case.
The measured and result in a high
which confirms the low mutual coupling loss suggested in (5)
and (6).
YUN AND VAUGHAN: MULTIPLE ELEMENT ANTENNA EFFICIENCY AND ITS IMPACT ON DIVERSITY AND CAPACITY 533

D. MEA Total Efficiency Matrix
As noted above, the total efficiency for each embedded ele-
ment can be different. A diagonal matrix is used to include all
the total efficiencies of embedded element of an MEA, viz.,
effective branches, and this is not necessarily the same as the
number of elements in the MEA.
For the example of Rayleigh channels, the probability den-
sity function of the SNR of the MRC signal from the equivalent
diversity branches can be expressed as [14]

.. .. .. .. (13)
. . . .

and this is here referred to as the MEA total efficiency.

In this section, the MEA efficiencies were formulated and ap- (15)
plied to the analysis of how the radiated power, received power
and losses are impacted by mismatch, mutual coupling, diver- where is the number of poles in the Laplace space, and the
sity combining, and the associated terminations of the MEA. In residues are
Section III, the impact of the MEA total efficiency on the di-
versity and capacity performances is presented. The receive ef-
ficiency, as presented above, is difficult to measure. It is often
assumed, as below, that the MEA receive efficiency is the same (16)
as its transmit efficiency for simplicity, although they are dif-
ferent quantities.
and is the order of the pole, meaning that there are uncorre-
III. IMPACT OF MEA TOTAL EFFICIENCY ON DIVERSITY GAIN lated branches with the same mean SNR, . The CDF is
The diversity performance of an MEA is measured by its
(17)
diversity gain [10], , which has several definitions as
introduced in Appendix B. In this Section, a signal processing
and this simplifies when the orders of all the poles are ones, as
approach [10]–[12] is used to calculate the diversity gain for
different combining methods. For a lossless MEA with Max-
imum Ratio Combining (MRC), the calculation is summarized
in [10]–[12] as follows. First, the correlation coefficient matrix
of the MEA, , is estimated. The correlation is elaborated in
Appendix C. The estimate can be from: time series analysis
of physical measurements; embedded element patterns and
a modeled propagation channel; or, in certain conditions, (18)
from network impedances or S-parameters [13]. The modeled
channel includes probability density functions for the polarized
A. Formulation of the Impact of on the Diversity Gain
incoming wave angular spectra. Then can be orthogonalized
using singular value decomposition (SVD), here interpreted as The approach of (14)–(17) includes the impact of the antenna
returning the non-zero singular values, denoted correlation coefficient, but it corresponds to a lossless MEA
with MRC. When the antenna is lossy, this approach needs to
(14) be modified. Taking receive diversity as an example, loss in the
embedded MEA elements reduces the receive SNR of each re-
so is an column of the non-zero singular values, . ceive branch. The impact of this reduced SNR on can be
Since the antenna correlation coefficients are essentially unaf- included by scaling the correlation coefficients with the receive
fected by the antenna losses (see Appendix C), the singular total efficiency, see (19) shown at the bottom of the page, and
values do not contain any information regarding the loss. There- now
fore, the singular values, which correspond to the mean SNRs
of the signals that would be received by the equivalent lossless,
uncorrelated antenna branches. is the number of non-zero (20)

.. .. .. .. (19)
. . . .
534
Fig. 5. (a) Computed and (b) measured ([3]@ 2008 IEEE) CDF vs. SNR of a 2-element orthogonal half wavelength lossy dipole MEA with Selection Combining.
Equation (19) defines a scaled correlation coefficient matrix,
, (in either the transmit or receive mode) which is
weighted by the positive square root of of the MEA
from both sides. It is important to note that represents
the reduced SNR on each receive branch, rather than a reduced
correlation. Note also that the diagonal of is no longer
unity because of the scaling. Likewise, the scaled singular value
column matrix, , contains the impact of the embedded
element total efficiencies on the equivalent lossless and uncor-
related branches of the MEA. With (20), (15), (16), and (17),
the CDF of the MRC combined SNR with the impact of MEA
total efficiency can be found. Other combination techniques
can follow the same method, should this be required. For
example, the CDF of Selection Combining (SC) in a Rayleigh
environment is
B. Reduction of a Dipole MEA With SC
To demonstrate and verify the above formulas including the
impact of MEA efficiency on , the CDF of an MEA with
SC, was computed with (19)–(21), and compared with a pub-
lished measurement result in [3]. The MEA comprised two or-
thogonal half wavelength lossy dipoles. The total efficiencies of
the embedded elements are 0.51 and 0.59 [3], respectively. The
elements are assumed uncorrelated, as suggested in [3], so the
correlation coefficient matrix is a 2 2 identity matrix.
In Fig. 5(a) the solid curves are the SC CDFs of ideal-
ized MEAs (uncorrelated and lossless) having 1 to 2 antenna
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
branches from the left to the right. The dashed curve is the com-
puted SC CDF of the 2-element orthogonal half wavelength
lossy dipole MEA, which agrees very well with the measured
result in Fig. 5(b) from [3]. The possible sources of the differ-
ence are the measurement uncertainties of the efficiency and
the finite correlation of an implemented antenna.
C. of Element-Symmetric and -Asymmetric MEA
There is a special class of MEA structures which are sym-
metric in the sense that each element sees an identical struc-
ture of elements around it. The analysis of these structures be-
comes simplified because if all the terminating impedances are
the same, then under the same conditions, all the elements will
see the same impedance on transmit, and will invoke the same
total radiation and losses, and the same efficiencies. So an av-
eraged (across the branches) efficiency can be used, and
(21) and can be simplified as
(22)
Conversely, for a general MEA, there is asymmetry in the
sense that the elements will have different losses and efficien-
cies. It is of interest to compare the difference in diversity per-
formance of the element-asymmetric MEA, with the behavior
of the element-symmetric one.
Fig. 6 gives the computed SC CDF curves of three ele-
ment-symmetric MEAs in Fig. 6(a) and the curves of the same
MEAs but with asymmetric element efficiencies in Fig. 6(b).
Each MEA comprises three uncorrelated half wavelength
YUN AND VAUGHAN: MULTIPLE ELEMENT ANTENNA EFFICIENCY AND ITS IMPACT ON DIVERSITY AND CAPACITY
Fig. 6. Selection combining CDFs of (a) uncorrelated, symmetric (b) uncorrelated, asymmetric (c) correlated, symmetric and (d) correlated, asymmetric 3-element
half wavelength dipole MEAs.
dipole elements with the element efficiencies given in the
figures. To verify the impact of correlation on element-asym-
metric MEAs, the CDFs of correlated MEAs are also computed
in Fig. 6(c) and (d).
Fig. 6 gives the computed SC CDF curves of three ele-
ment-symmetric MEAs in Fig. 6(a) and the curves of the same
MEAs but with asymmetric element efficiencies in Fig. 6(b).
Each MEA comprises three uncorrelated half wavelength dipole
elements with the element efficiencies given in the figures. To
verify the impact of correlation on element-asymmetric MEAs,
the CDFs of correlated MEAs are also computed and presented
in Fig. 6(c) and (d).
In Fig. 6(a), with the decrease of the element efficiency, the
CDF curve translates to the left of the figure towards the lower
SNR direction, so decreases. The spacing between any two
adjacent dashed curves is 1 dB, which corresponds to the differ-
ence of the element efficiencies. The computed CDF curve for
the MEA with element efficiency of agrees well with
the published measured result of the same MEA in [3], and the
agreement helps to verify the formulation from measurements.
In Fig. 6(b), the averaged (in decibel) efficiency of each
asymmetric MEA is , , and , respectively,
and equals to the element efficiency of the symmetric MEAs in
Fig. 6(a). It is obvious that the CDF curves of the corresponding
symmetric and asymmetric MEAs are the same. This observa-
tion in turn indicates that it is reasonable to use averaged [3]
535
(rather than the highest [4]) element efficiency for the MEA
in a simple way.
Fig. 6(c) and (d) are for the same symmetric and same
asymmetric MEAs as in Fig. 6(a) and (b), respectively, but
the elements are correlated. A correlation coefficient matrix
is used as an example.
The computed results show that, although the correlation
changes the slope of the CDF curves, the spacing between the
adjacent curves is again 1 dB, and the averaged (in decibel) ef-
ficiency can still be used to find the asymmetric MEA diversity
performance in a simple way. The last observation suggests
that the requirement of having uncorrelated elements for using
(dB-)averaged element efficiency in [3] does not seem to be
necessary.
D. Diversity Gain Reduction of a Slot Cube MEA Due to the
Impact of the MEA Total Efficiency
The above discussions of the impact of MEA efficiency on
the diversity performance are based on idealized dipole models,
although some results are verified with published measurement.
Here, the MRC diversity performance of the measured hollow
slot cube MEA [7] in a Rayleigh channel with the uniform, un-
correlated incoming wave distribution is presented.
This MEA with simultaneous combining (such as MRC) has
low mismatch loss, and low correlation coefficients for the uni-
form Rayleigh scenario [7], but the high mutual coupling loss
536 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 7. Scaled singular values of the 12 element hollow slot cube MEA.
indexes the equivalent uncorrelated antenna branches.
Fig. 8. MRC CDFs for SNR less than the abscissa for ideally uncorrelated and
lossless branches ( , solid curves), and for the hollow slot cube MEA
when the impact of antenna correlation coefficient is included (circle-dashed
curve) and when efficiency is also included (star-dashed curve).
causes a low MEA total efficiency as shown in Fig. 4. The corre-
lation coefficient matrix of the MEA is obtained with the method
of embedded element radiation patterns [15] as mentioned in
Appendix B. Based on the MEA efficiency and correlation co-
efficient matrix, the singular values are attained with (20) and
depicted in Fig. 7, which represent the different mean power in
each equivalent lossless and uncorrelated branch.
In Fig. 8, the solid curves are the MRC CDFs of idealized
MEAs (uncorrelated and lossless) having 1 to 12 antenna
branches, from the left to the right. The circle-dashed curve
is for the 12-element slot cube at its resonance with MRC,
obtained with (14) and (18) including the impact of antenna
correlation coefficient but not including the impact of the MEA
total efficiency. It overlaps with the curve for the 11-branch
ideal MEA, so the diversity order of this slot cube MEA is
slightly reduced (by about one ideal branch), owing to the finite
correlation among slot elements. Taking the first left curve for
1 branch antenna as the reference antenna, and a probability of
0.5%, the diversity gain is , which is 1 dB less
than the ideal case.
With (18)–(20), the impact of the MEA total efficiency is in-
cluded in the CDF of the MEA (star-dashed curve). It is parallel
to the circle-dashed curve, and is near the curve for the 6-branch
ideal MEA. Now the diversity gain of the MEA is
at the probability of 0.5%. Comparing the star-dashed
and circle-dashed curves, it can be seen that the antenna corre-
lation coefficient reduces the MEA diversity order by one ideal
branch ( is reduced by 1 dB), then the MEA total efficiency
reduces the diversity order by five more ideal branches ( is
further reduced by 4 dB).
This example demonstrates how the impact of correlation and
MEA efficiency on the MEA diversity performance can be sep-
arated. The impact of the correlation is to change the slope of the
CDF curve, and the impact of MEA efficiency is to translate the
CDF curve. For this example of the hollow slot cube MEA, the
mutual coupling loss dominates the reduction in diversity gain
and the correlation is low so its impact is small. Other antenna
examples may be differently impacted by these factors.
This 12 element MEA with MRC has a diversity order of
only 6, so it is equivalent to a 6-element idealized (uncorrelated
and lossless) diversity antenna. With this, the transmission reli-
ability at 0.5% probability is still improved by a diversity gain
of 25 dB.
IV. IMPACT OF MEA TOTAL EFFICIENCY ON CAPACITY
The MIMO channel capacity [16], [17] (the term capacity
is favoured by information theorists, and capacity efficiency
by communications engineers; here we also use capacity for
brevity) for a static unknown channel can be expressed as
(23)
where is the identity matrix; is the number of
the transmit antenna; and is a normalized (instantaneous)
channel transfer matrix. The mean capacity is averaged over the
channel variations, written . Note that this
is different to the capacity of the mean channel. The mean ca-
pacity from this formula is known to be close to the upper limit
given by the parallel (eigen) channel formulation for known
channels [10]. This capacity is an information theoretic form
derived from mutual information, and is in bits per channel use.
Imperfect antennas, filters, etc., and communications system as-
pects reduce this drastically to an achievable communications
capacity efficiency (in bits/sec/Hz). But it is convenient here to
isolate the impact of antenna effects on the information theo-
retic capacity for an insightful evaluation of MIMO antennas.
The Kronecker model [18], [19] is a simplistic method to
combine the transmit and receive antenna correlation coeffi-
cients into the channel transfer matrix. The accuracy of this ap-
proach has been questioned [20], and [21] reports that it is not
applicable to the polarimetric case. But in the same spirit as the
capacity, the Kronecker model is nevertheless a convenient sim-
plification for comparatively evaluating MEAs.
With this model, the channel can be taken as zero mean, unit
variance, complex Gaussian i.i.d., and the instantaneous white
channel matrix, denoted , is modified with the correla-
tion coefficient matrices of the transmit and receive antennas,
as
(24)
where and are the correlation coefficient matrices of
the transmit and receive antenna, respectively, in a given propa-
gation channel model. The transmit correlation coefficient ma-
trix is taken to be the same as that from its well defined receive
operation.
To include the impact of the receive and transmit MEA total
efficiencies on the MIMO capacity, the scaled correlation coef-
ficient matrices defined in (19) are used in the same way,
YUN AND VAUGHAN: MULTIPLE ELEMENT ANTENNA EFFICIENCY AND ITS IMPACT ON DIVERSITY AND CAPACITY
Fig. 9. MIMO Capacity of the hollow slot cube MEA.
(25)
The capacity is now
(26)
If an ideal transmit antenna (unity total efficiency for each el-
ement and zero correlation) is used, i.e., , the
impact of the imperfections of the receive antenna can be sin-
gled out. Here, the receive MEA efficiency scales the transmis-
sion matrices to , as
(27)
or alternatively stated, the SNR is scaled, and the resulting ca-
pacity, , is
(28)
Fig. 9 illustrates the impact of MEA total efficiency on the
averaged capacity of the slot cube MEA in a Rayleigh channel.
Statistical channel information is contained in the correlation
coefficient matrix (Appendix C) so the capacity for other prop-
agation scenario models can also be found.
The solid curves are the averaged capacity for the unknown,
ideal, Rayleigh channels of order of 1 to 12. The dashed curve
is for the case when an ideal 12-element transmit antenna is
used, and the slot cube MEA example with unity efficiency (in
(27) and (28)) is receiving. This curve is between the MIMO
capacity order of 11 and 12. In other words, the correlation co-
efficient between the slot cube elements decreases its MIMO
capacity slightly, so this lossless form of the slot cube MEA
is equivalent to an ideal MIMO antenna with 11.2 branches at
. comparison of different MEA designs. An element-asymmetric
The circle-dashed curve includes the MEA total efficiency in
(27) and (28) for using the slot cube at the receiver and an ideal
537
12-element MEA at the transmitter. The curve has a MIMO ca-
pacity order of about 9 when the SNR is higher than 15 dB, and
a MIMO capacity order from 7 to 8 for lower SNR. A more sig-
nificant drop in MIMO capacity is caused by the MEA antenna
efficiency. For an SNR of 20 dB, this MEA example is equiva-
lent to an 8.5-branch ideal MEA in terms of MIMO capacity.
The cross-dashed curve is for the slot cube MEA example
at both transmitter and receiver. It is emphasized that here the
transmit and receive total efficiencies are assumed to be the
same for simplicity, but this is not a limitation of the formu-
lation. Here, the capacity is computed with (25) and (26), and it
is obviously decreased further. From the figure, the pair of slot
cube MEAs has the capacity of a pair of 6-branch ideal MEAs
for an SNR of 20 dB.
Note that diversity gain and MIMO capacity vary differently
at given probabilities for , and at given SNR for capacity.
Overall, this MEA is essentially equivalent to a 6 branch ideal-
ized MEA in terms of both and capacity.
V. CONCLUSION
In this paper, formulas of MEA efficiencies with fixed diver-
sity combining are developed in the context of MIMO commu-
nications performance. The MEA embedded element efficiency
is derived for both the transmit and the receive cases. This al-
lows identification of the conditions for the transmit and receive
efficiencies to be the same. The total efficiency of the embedded
element is also presented. The MEA total efficiency is expressed
as a diagonal matrix of the total efficiencies of the embedded ele-
ments. This matrix is for calculating the impact of the MEA total
efficiency on the communications performance in terms of the
diversity gain and capacity in modeled propagation scenarios.
The reductions in the diversity gain and capacity, resulting from
the finite correlation coefficient and the MEA total loss (ohmic
and through mutual coupling) are separated in order to analyze
the different performance degradations. The diversity order and
MIMO capacity order of an MEA are expressed as an equiv-
alent number of idealized branches, i.e., lossless, equal gain,
uncorrelated and uncoupled. This metric allows a performance
MEA example and two element-symmetric MEAs are used as
examples to get a feel for the efficiency metrics.
538 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
APPENDIX A
DIVERSITY COMBINING
Combining can be implemented at either the transmitter or re-
ceiver, and it may act to change the terminations of the elements.
One class of combining implementation has all the elements
connected to fixed loads (viz., the transmitters or receivers), and
it is here called simultaneous combining because all the branch
signals are used simultaneously. It includes Optimal Combining
(OC), Maximum Ratio Combining (MRC), Equal Gain Com-
bining (EGC), and Selection Combining (SC) implementations
where all the elements have fixed terminations.
With simultaneous combining, power coupled from one
MEA element to other elements can be “collected” at the
receive loads in the receive case, or radiated by other ele-
ments in the transmit case. Such power must be considered
“useful”—except in the receive case of SC with fixed termi-
nations because we do not have access to the received power
in the unselected terminations. For the case of the re-radiation,
the power radiated via mutual coupling is part of the embedded
element radiation pattern, and is therefore “useful”. Meanwhile,
there is “wasted” power in the form of ohmic loss in all the
elements.
Another class of combining implementation can be called
non-simultaneous combining, in which the signals are not com-
bined simultaneously. It includes Switched Combining (SwC),
and the implementation is to have only one element selected
(i.e., terminated) at a time and the unused elements open cir-
cuited. With this non-simultaneous combining, we can simplis-
tically model the situation as having no mutual coupling since
there is no current flowing through the open circuits. In some im-
plementations, there may be physical non-zero currents because
the physical “open circuit” termination may in fact be reactive
and/or have finite resistances, but the circuit model (with ideal
open circuits) is still valid because the elements are considered
as embedded.
Therefore for MEAs, the efficiency of one element needs to
be considered as being embedded in the MEA, i.e., the effi-
ciency must include the presence of the other elements and the
supporting structure, as well as the diversity combining scheme
in the sense that the embedded efficiency depends on the termi-
nations of the other elements. In this way, the MEA efficiencies
are defined for different diversity combining schemes.
For simultaneous combining (OC, MRC, EGC, SC etc.), an
embedded transmit element is taken to mean that this element is
connected to its transmit source while all the other elements are
terminated with their transmit source impedances. Likewise, an
embedded receive element means that this element is receiving
and all the other elements are terminated with their receive load
impedances.
For non-simultaneous combining (i.e., SwC), an embedded
transmit element means that this element, in the circuit model,
is connected to its transmit source while all the other unselected
elements are open circuited. So an embedded receive element
circuit model includes this receiving element, and all the other
unselected elements are open circuited.
APPENDIX B
DIVERSITY GAIN
The diversity performance of an MEA is measured by
its diversity gain [10], . There are several definitions
for the diversity gain. For example, in digital communi-
cations, an error rate can be modeled along the lines of
, or plotted on the usual log-log
scale, log , for
high SNR. The slope of the BER function gives the diversity
gain (and is a coding gain). But such a slope metric
alone, while simple, is insufficient to capture the complete gain
garnered from diversity action. A more complete definition
is conveniently read from cumulative density function (CDF)
plots of the SNR, but clarification is required for the definition
of the CDF quantities. Here, the diversity gain is defined as the
improvement in the averaged normalized SNR, here expressed
in dB
(B-1)
where is the instantaneous SNR of the combined received
signal. is the mean SNR on one MEA receiving branch. Simi-
larly, and are respectively the instantaneous and mean
SNR received by a single element reference antenna. This shows
that the diversity gain is the improvement in receive SNR of the
diversity MEA over the SNR of a reference single branch.
Note that this definition of diversity gain depends on the
choice of the probability and the choice of reference antenna
element. There are several options for the reference element.
For example, it can be one of the MEA’s elements which is al-
ready mounted on the supporting structure or terminal, but with
the other elements removed [22]. In this case, the diversity gain
gives the SNR improvement from a single element system to a
multiple element diversity systems in the same device. Another
option is to take the element with the highest average SNR as
the reference antenna, while the other elements remain present.
Or the reference element can be a separate antenna, such as
a dipole, or for analysis or simulation, an idealized element
such as one with a lossless, isotropic pattern. The choices of
probability and reference antenna need to be specified in a study
of diversity performance in order to allow direct comparison of
results.
Factors affecting and include the propagation channel
characteristics, diversity combining method, the MEA total ef-
ficiency, and the correlation coefficient between each element.
The total efficiency of each element impacts the SNR value of
the branch which includes that antenna element. Changing prop-
agation channel characteristics may change the antenna cor-
relations, but do not change the MEA efficiencies. Using the
embedded element patterns and modeled propagation scenario
statistics, the antenna correlations can be found from the ap-
proach of [15] for diversity performance analysis; and likewise
for the capacity analysis in different propagation models in the
Section IV.
In a measurement, can be estimated from the time series
samples from real world propagation environments, e.g., urban,
indoor, etc. This is the best estimation technique in the sense that
the real world behavior is being sampled. But it is a complex
and expensive measurement, and care must be taken with the
averaging, i.e., the experiment must cover all the propagation
environments, with appropriate weightings, for an accurate en-
semble estimate. Moreover, the mobile antenna must be moved,
e.g., translated, rotated, etc., in a realistic way, and be deployed
YUN AND VAUGHAN: MULTIPLE ELEMENT ANTENNA EFFICIENCY AND ITS IMPACT ON DIVERSITY AND CAPACITY
appropriately—a major exercise for handheld terminals which
are held to the head and also operated in the hands, in both cases
with many different poses [23]. In such measurements, an obser-
vation from one environment may not match well with another
observation from a different environment, or with an ensemble
average.
Another physical measurement method is to create an artifi-
cial incoming wave distribution, typically striving for uniform
and uncorrelated scattering [24]. This method needs no antenna
patterns as required by the approach of [15]. It is useful for mea-
suring the mean effective gain of an antenna when it is not pos-
sible to measure the patterns and when the uniform propagation
scenario can be assumed present all the time.
APPENDIX C
ANTENNA CORRELATION
Antenna correlation usually refers to the time series correla-
tion of the received signals (or rather the modulation imposed
by the multipath channel on a transmitted carrier signal). This is
hard to measure accurately, requiring extensive data collection
and time series analysis, and the results from such a measure-
ment can only be repeated in a statistical sense. The accuracy
is further complicated since such a measurement features signal
plus noise. If the signal cannot be separated from the noise (usu-
ally impossible) then the time series measurement is affected by
ohmic loss in the antenna, because the SNR is governed by the
loss. This is clear from the limiting case of zero efficiency where
a measured signal correlation will reflect only the (un)correla-
tion of the noise at the receiver. If the signal can be separated
from the noise, then the correlation coefficient of the signals can
be calculated and will be independent of the losses, or antenna
efficiency.
Under certain conditions, this correlation coefficient can be
estimated by the normalized inner product of the embedded
antenna element patterns, using probability density functions
(pdfs) to model the polarized incident waves. This measure is
deterministic, and flexibility is possible using the distributions
of the incoming waves. The measured pattern correlation coef-
ficient is not strongly affected by the ohmic loss in the antenna
(because the pattern measurement requires good SNR, the an-
tenna efficiency cannot be zero), but may well be affected by
the changes to the environment (i.e., the pdfs), including the
presence of a user. In this paper, we refer to the correlation as
the signal correlation, and consider that the pdf-weighted pat-
tern-calculated correlation is a good approximation to this.
REFERENCES
[1] “IEEE Standard Test Procedures for Antennas,” in IEEE, Standards.
Piscataway, NJ: IEEE, pp. 149–1979.
[2] “IEEE standard definitions of terms for antennas,” in IEEE, Stan-
dards. Piscataway, NJ: IEEE, pp. 145–1993.
539
[3] J. F. Valenzuela-Valdés, M. A. García-Fernández, A. M. Martínez-
González, and D. A. Sánchez-Hernández, “The influence of efficiency
on receive diversity and MIMO capacity for Rayleigh-Fading chan-
nels,” IEEE Trans. Antennas Propag., vol. 56, no. 5, pp. 1444–1450,
May 2008.
[4] K. Karlsson, J. Carlsson, I. Belov, G. Nilsson, and P.-S. Kildal, “Opti-
mization of antenna diversity gain by combining full-wave and circuit
simulations,” in Proc. EuCAP 2007, Nov. 2007, pp. 1–5.
[5] J. B. Andersen and A. Frandsen, “Absorption efficiency of re-
ceiving antennas,” IEEE Trans. Antennas Propag., vol. 53, no. 9, pp.
2843–2849, Sep. 2005.
[6] J. B. Andersen and R. G. Vaughan, “Transmitting, receiving and scat-
tering properties of antennas,” IEEE Antennas Propag. Mag., vol. 45,
no. 4, pp. 93–98, Aug. 2003.
[7] J. X. Yun and R. G. Vaughan, “Slot MIMO cube,” presented at the IEEE
Antennas Propag. Soc. Int. Symp., Toronto, ON, Canada, 2010.
[8] [Online]. Available: http://www.agilent.com/
[9] [Online]. Available: http://www.satimo.com/
[10] R. G. Vaughan and J. B. Andersen, “Channels, propagation and an-
tennas for mobile communications,” in Inst. Elect. Engr. Electromag-
netics Waves Ser.. London, U.K.: , 2003, vol. 50.
[11] W. C. Jakes, Microwave Mobile Communications. Piscataway, NJ:
IEEE Press, 1993.
[12] R. G. Vaughan, “Requirements for dversity/MIMO with some eval-
uation techniques,” in Antennas for Wireless Systems. New York:
Wiley, 2007, ch. 15, pp. 405–445.
[13] J. X. Yun, “Design and evaluation of slot multiple element antennas,”
Ph.D., Simon Fraser Univ., Burnaby, BC, Canada, 2011.
[14] M. Schwartz, W. R. Bennett, and S. Stein, Communication Systems and
Techniques. New York: McGraw-Hill, 1966.
[15] R. G. Vaughan and J. B. Andersen, “Antenna diversity in mobile com-
munications,” IEEE Trans. Veh. Technol., vol. 36, no. 4, pp. 149–172,
Nov. 1987.
[16] J. H. Winters, “On the capacity of radio communications systems
with diversity in a Rayleigh fading environment,” IEEE J. Sel. Areas
Commu., vol. SAC05, no. 5, pp. 871–878, Jun. 1987.
[17] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wireless Pers.
Commun., vol. 6, pp. 311–335, 1998.
[18] A. L. Moustakas, H. U. Baranger, L. Balents, A. M. Sengupta, and S.
H. Simon1, “Communication through a diffusive medium: Coherence
and capacity,” Science, vol. 287, Jan. 2000.
[19] D. Chizhik, F. Rashid-Farrokhi, J. Ling, and A. Lozano, “Effect of an-
tenna separation on the capacity of BLAST in correlated channels,”
IEEE Commun. Lett., vol. 4, no. 11, pp. 337–339, Nov. 2000.
[20] H. Özcelik, M. Herdin, W. Weichselberger, J. Wallace, and E. Bonek,
“Deficiencies of the Kronecker MIMO radio channel model,” Electron.
Lett., vol. 39, pp. 1209–1210, Aug. 2003.
[21] J. P. Kermoal, L. Schumacher, F. Frederiksen, and P. E. Mogensen,
“Polarization diversity in MIMO radio channels: Experimental valida-
tion of a stochastic model and performance assessment,” in Proc. 54th
IEEE VTS Fall VTC 2001, 2001, vol. 1, pp. 22–26.
[22] V. Plicanic, B. K. Lau, A. Derneryd, and Z. Ying, “Actual diversity
performance of a multiband diversity antenna with hand and head ef-
fects,” IEEE Trans. Antennas Propag, vol. 57, no. 5, pp. 1547–1556,
May 2009.
[23] M. Pelosi, O. Franek, M. B. Knudsen, M. Christensen, and G. F. Ped-
ersen, “A grip study for talk and data modes in mobile phones,” IEEE
Trans. Antennas Propag., vol. 57, no. 4, pp. 856–865, Apr. 2009.
[24] P. S. Kildal, K. Rosengren, J. Byun, and J. Lee, “Definition of effec-
tive diversity gain and how to measure it in a reverberation chamber,”
Microw. Opt. Technol. Lett., vol. 34, no. 1, pp. 56–59, Jul. 2002.
J. X. Yun, photograph and biography not available at the time of publication.
R. G. Vaughan, photograph and biography not available at the time of publica-
tion.
540 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
On the Accuracy of Equivalent Circuit Models for
Multi-Antenna Systems
Jon W. Wallace, Member, IEEE, and Rashid Mehmood, Student Member, IEEE
Abstract—The equivalent circuit model of a general PEC an-
tenna array is derived, based on a rigorous method of moments
(MOM) formulation, indicating that network analysis is exact from
the standpoint of electromagnetic wave theory. It is found that
the network parameters (Z, Y, or S-parameters) for the transmit
mode can be used for exact prediction of the receive-mode array
response. Numerical and experimental examples illustrate the va-
lidity of the analytical results.
Index Terms—Antenna arrays, equivalent circuits, modeling,
moment methods, multiple-input multiple-output (MIMO) sys-
tems, mutual coupling.
I. INTRODUCTION
E QUIVALENT circuit models, also referred to as network
models, have gained attention for modeling antenna ar-
rays and multiple-input multiple-output (MIMO) systems, al-
lowing circuit effects such as amplifier noise, matching and re-
configurability to be studied [1]–[5]. Such models are compu-
tationally attractive, since the transmit and receive arrays can
be represented as equivalent circuits, requiring only a modest
number of full-wave simulations or measurements, after which
circuits of varying complexity are analyzed with efficient cir-
cuit-level simulation. In [6] it was formally proven that antenna
arrays in the transmit and receive mode can be modeled with
network analysis and that the same impedance matrix (with the
exception of a transpose) can be used for both modes. In [7], the
effect of mutual coupling on adaptive arrays is studied by con-
sidering an equivalent network model, yielding a simple linear
relationship between the loaded and open-circuit voltages on a
receive array and a beamformer for optimal signal to interfer-
ence and noise ratio (SINR) is derived. More recently, [8] gives
the equivalent circuit of a single receive antenna and [9] pro-
vides an equivalent circuit for a receiving array.
In contrast to work endorsing simple equivalent circuit
models, there is also work that questions the use of such
models. For example, [10] suggests that the compensation
method in [7] is suboptimal since the network model only
assumes a single basis function per antenna and an improved
compensation method based on a full moment method model
of the array is proposed. In [11] the use of Norton or Thévenin
Manuscript received June 15, 2010; revised November 03, 2010; accepted
November 15, 2010. Date of publication May 10, 2011; date of current version
February 03, 2012.
The authors are with the School of Engineering and Science, Jacobs Uni-
versity Bremen, Bremen, Germany (e-mail: wall@ieee.org; r.mehmood@ieee.
org).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2152339
0018-926X/$26.00 © 2011 IEEE
equivalent circuits for receiving antennas is challenged, since
power absorbed by the internal impedance of the network
does not have a strict physical meaning and a different model
is proposed based on a constant power source. A physically
appealing model is also proposed in [12] that includes the
transmit antenna in the equivalent receive model. Recent work
[13], [14] also questions the use of array transmit-mode mutual
impedances for the receive mode, which is troubling from the
standpoint of reciprocity.
The purpose of this paper is to provide a straightforward but
rigorous analysis of antenna arrays based on method of mo-
ments (MOM), illustrating that equivalent circuit models are
exact and that receive-mode behavior of an array can be exactly
predicted by usual transmit-mode quantities. Although these ob-
servations are basically equivalent to those in [6], the MOM
analysis here has a number of advantages: the development is
simpler and intuitive, the transmit and receive modes do not
need to be considered separately and the MOM discretization
provides an exact definition for the ports. The analysis provides
valuable insight on the operation of equivalent circuit models,
such as the connection of transmit and receive mode, the number
of basis functions (or degrees of freedom) required to represent
currents accurately on an antenna array and potential sources of
inaccuracy in network models. Several examples are provided
based on full-wave simulations and direct measurement that val-
idate the analytical results.
The paper is organized as follows: Section II derives Z, Y and
S-parameter equivalent circuit models of antenna arrays in the
transmit/receive mode from MOM. Section III provides numer-
ical examples that demonstrate the observations of the MOM
analysis, followed by an experimental example in Section IV.
Section V provides concluding remarks.
II. METHOD OF MOMENTS ANALYSIS OF MULTI-ELEMENT
ARRAYS
In this section, we rigorously analyze general antenna arrays
accessible at a finite set of ports by applying the method of
moments. Although we restrict our attention to antennas com-
posed of perfect electric conductor (PEC) surfaces, we do so
only for the sake of simplicity and the method can be naturally
extended to dielectric and magnetic materials, finite conduc-
tivity, etc. This exercise provides valuable intuition on the con-
nection between network analysis and full-wave analysis and
the requirements for good agreement. Furthermore, we prove
that mutual interactions of the transmit mode, receive mode and
combined transmit/receive mode can be captured with a single
equivalent model that involves the usual (transmit mode) mu-
tual impedance matrix.
WALLACE AND MEHMOOD: ON THE ACCURACY OF EQUIVALENT CIRCUIT MODELS FOR MULTI-ANTENNA SYSTEMS 541

Fig. 2. Equivalent antenna array circuit model in transmit/receive mode.

Appropriately partitioning (4) into elements at the ports (P)

and on the PEC antenna array structure (A),
Fig. 1. Geometry of (a) a general antenna array and (b) a wire dipole array and
example method of moments segmentation.
(5)
(6)
A. Governing Equations
Fig. 1 depicts a general antenna array in a free-space medium where due to PEC surfaces.
where signals can be driven and/or measured at ports. An
externally applied incident field may also be present, denoted B. General Transmit/Receive Mode
by , which is general (plane wave, spherical wave, etc.). The general mode (transmit and receive) of the antenna is
Total field at observation point is given in terms of current considered. From (6) the currents on the antenna array are
density on the antenna according to [15]
(7)
(1)
and plugging into (5) gives the voltages on the ports in terms of
where is the dyadic Green’s function for free space. the port currents only, or
MOM discretization is performed using basis expansion
, or

(2) (8)
This resulting system is equivalent to the network model de-
picted in Fig. 2. Thus, the operation of the array can be com-
puted exactly if we know the open circuit voltages for all
followed by the projection onto the weighting functions incident fields of interest as well as the impedance matrix .
Radiated fields from the antenna can be computed using (2),
where the partitioning yields

(3)

(9)
which gives the usual linear relationship

(4)
where is field scattered by the antenna for (open
We choose and to have identical local support circuit) and is interpreted as the radiation pattern of port
, such that and represent the current and voltage on the for polarization for unit input current when the other ports
element, respectively. are open-circuited. Typically, we are most interested in far-fields
The basis functions are ordered so that the first corre- of the array and the expression can be simplified to
spond to port terminals, where and for
are the port currents and voltages. Many choices of the and (10)
are possible, but a simple choice is a filament properly
connecting the two port terminals, where is the current on In the transmit mode, , and
the filament and is the contour integral for voltage between so that the radiated fields are just a superposition of the open-
the terminals. Note that basis and weighting functions for ele- circuit embedded patterns weighted by the port currents. For
ments on the PEC structure are chosen to be tangential to these the receive mode (and combined mode) the scattered field term
surfaces. must also be included.
542 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

current . If we use a simple Hertzian dipole for the reference

antenna, the field present at the origin is

(11)

where is the orientation and the effective length of the ref-

Fig. 3. Computation of for far-field sources (a) direct problem and (b) re- erence antenna and and are the wavenumber and intrinsic
ciprocal equivalent.
impedance of the background medium.
Next, consider the reciprocal system in Fig. 3(b), which will
give the same value of . The field incident on the reference
C. Comparison of Transmit and Receive Mode antenna is
Our analysis shows that the only difference in using the
antenna in transmit versus receive mode is that for the pure (12)
transmit case, so that a linear system is obtained,
whereas for the receive case. Also, note that is the The open-circuit voltage on the Hertzian dipole is
usual transmit mutual impedance and there is no approximation
when using this for the receive case. This is in contrast to the
development in [13] that questions the use of transmit mutual
impedances for the receive mode.
We also compare with the result in [6], where it was shown (13)
that the receive mode mutual impedance matrix is the transpose
of that for the transmit mode. Although a reciprocal antenna
and medium are considered here, the same equivalent circuit
as in Fig. 2 would be obtained for non-reciprocal (general bian- Thus, the receive-mode open-circuit voltages for plane-wave in-
isotropic) materials. The impedance matrix does not transpose cidence can be computed using the transmit-mode open-circuit
due to or what is connected to the ports, indicating that embedded radiation patterns.
must be the same for both the transmit and receive mode. For receive array calibration, the direct arrangement in
To remedy the apparent dilemma, it must be noted that [6] Fig. 3(a) may be preferred. Terminating the array with load
invokes the Lorentz reciprocity theorem, where for non-recip- impedance matrix and defining to be the vector of currents
rocal materials the system must be changed to a new system with flowing into the antenna ports,
complementary materials [15]. Given that the original transmit
impedance matrix is , we denote the transmit impedance ma- (14)
trix of the complementary system as . In [6] it was actu- (15)
ally shown that whereas is used for the transmit case,
should be used for the receive case. However, since the orig- Proper combination gives
inal and complementary systems are related by reciprocity,
and alone (no transpose) can be used for the receive (16)
system. Also, [6] shows that the load matrix is op-
timal for receive power transfer, but this is the same as a Hermi- indicating that the open-circuit voltage can be computed using
tian match to the original physical transmit impedance, or a measurable voltage across known loads .
. For S-parameters with a real normalizing impedance, it can
be shown that this condition is the same as , which was E. Degrees of Freedom of an Port Array
proven to provide maximum power transfer in [2]. As the number of MOM basis functions used to discretize
the antennas grows large, one would expect that an equally large
D. Computation of number of parameters would be needed to represent the current
distribution on the antennas. Here we illustrate that often a
The exact network model requires knowledge of for more concise representation is possible.
all incident fields of interest. In many cases, external Consider the array in transmit mode where
sources are far away from the array and only needs to . According to (7), is a weighted sum of
be found for plane-wave incidence. In this case, reciprocity at most linearly independent vectors spanned by the
arguments can be used to obtain in terms of the radiated columns of , regardless of what is attached to
far-fields for the transmit-only mode. the ports (loads or sources). Therefore, only independent
Consider the configuration depicted in Fig. 3(a) for finding basis functions are needed to completely represent currents on
, which is the open-circuit voltage induced on the array the array. Next consider the receive mode with a fixed .
due to a plane wave originating from direction . Here, a The incident field simply adds one additional basis vector in
transmit reference antenna is placed at coordinate , which (7), meaning that can be decomposed into a weighted sum
is in the far-field of the array near the origin and driven with of fixed basis vectors, regardless of port termination.
WALLACE AND MEHMOOD: ON THE ACCURACY OF EQUIVALENT CIRCUIT MODELS FOR MULTI-ANTENNA SYSTEMS
For the receive mode when is not fixed, the number
of required basis vectors is at most , where is
the number of linearly independent that can exist, which
may be large in practice. However, in some special cases, the
are linearly dependent. For example, if the array consists
of minimally scattering thin-wire dipoles [16] and is a
plane wave coming from direction and oriented parallel to the
dipoles, for each antenna is a scan angle dependent scalar
times a constant vector. According to (7), the current on each
antenna requires at most one basis vector in addition to the
transmit mode vectors, meaning up to vectors are needed
to represent all current distributions on the array. In [10] it is
correctly observed that one basis function per antenna is insuf-
ficient to represent antenna currents on a dipole array with par-
allel plane-wave excitation. However, our analysis shows that
only two basis functions are needed per antenna, as long as they
span the correct subspace.
F. Alternative Model Parameterizations
Although equivalent, there are times when other network pa-
rameterizations are desirable. Multiplying both sides of (8) by
results in the admittance formulation
(17)
where is the admittance matrix and is the current flowing
into the ports due to the incident field when all ports are short-
circuited. Substituting from (17) into expression (9) for radi-
ated fields,
(18)
where are the transmit-mode short-circuit embedded pat-
terns where the pattern (column) is obtained by placing a
unit voltage source across the port and zero volts (short cir-
cuit) across the other ports.
The scattering parameter (S-parameter) formulation relates
the ingoing waves and outgoing waves on the network ports,
which are related to the port voltages and currents according to
(19)
where is an arbitrary normalizing impedance. Substituting
and from (19) into (8) and rearranging yields
(20)
where is the outward traveling wave due to the incident
field for matched circuits (loads with impedance ) connected
to all ports and is the transmit S-parameter matrix of array.
Substituting (20) into (19),
(21)
543
which substituted into (9) yields
(22)
where is the scattered field due to incident field with
matched loads on all ports and are the matched circuit
patterns.
G. Network Analysis of Reconfigurable Antennas
Network analysis is an attractive solution for analyzing re-
configurable antennas with a large number of possible states,
since the number of required full-wave simulations can be kept
to a minimum. In this section, we illustrate that if out of
ports are terminated with loads (such as reconfig-
urable elements), this simply creates a new effective array with
ports having modified network parameters and new radi-
ation/reception patterns. Also, we use this framework to show
that the impedance matrix of a non-reciprocal system is the same
for transmit and receive mode and only the transmit and recep-
tion patterns are different.
Consider a reconfigurable antenna array with total
ports, where ports are terminated with loads (such as recon-
figurable elements) and are left accessible to be connected
to transmit or receive RF chains. Letting vector ports 1 and 2
correspond to the accessible and reconfigurable terminations,
respectively, we have
(23)
Terminating port 2 with a network having the impedance matrix
, we have and solving for
(24)
which means that the loaded antenna system forms a new circuit
with impedance matrix and open-circuit voltage . The
far-field of the array can be computed using (10) where
(25)
and and are radiation pattern matrices for the
accessible and loaded ports, respectively. Eliminating
(26)
indicating that the loaded array can be treated as a usual array
as before except with modified open-circuit radiation patterns.
544 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

It is instructive to consider the case where the unloaded array

and channel are reciprocal , but the loading is not
. The unloaded open-circuit voltage can still be
computed from (13), or

(27)

Substituting into (24),

(28)
(29)
Fig. 4. Current distribution on a single dipole antenna for different loading,
where is the predicted value using the transmit-mode current distribution and
is from direct receive-mode simulation.
Thus, although the radiation pattern in (26) is
different from the reception pattern , the same
impedance matrix in (24) is used for both transmit and III. NUMERICAL EXAMPLES
receive modes. This section provides numerical examples, not only to val-
idate the previous analytical results, but also to help provide
H. Potential Sources of Inaccuracy intuition on the behavior of the multi-antenna systems with
Although (8) and other network representations are exact, it coupling.
is instructive to consider how improper application may lead to
potential inaccuracy. A. Single Antenna: Transmission vs. Reception
1) Isolated vs. Embedded Element Patterns: A common sim- First, we study a simple single-antenna MOM simulation. Al-
plifying assumption in equivalent circuit models of antenna ar- though somewhat trivial, this case demonstrates the important
rays is the use of isolated transmit and receive patterns, rather principle that the difference in transmit and receive current dis-
than precise embedded patterns. For an array composed of iden- tributions on an antenna is exactly predicted by the open-circuit
tical elements, this approach only requires a single pattern to be current distribution.
found followed by pattern multiplication with the array factor. The Numerical Electromagnetics Code Version 2 (NEC2)
The operation of isolated elements is defined from (5) and (6) [17] was used for simulations of a single -directed dipole with
by replacing with the block matrix , where length and radius . The antenna was
first simulated in the transmit mode with a 1 V source placed
(30) across the terminals and the resulting current on the antenna
otherwise
and -directed far-zone E fields versus azimuth angle
and the function maps the basis function index to were stored. Second, the antenna was simulated in receive
its associated antenna port . Assuming isolated antennas in the mode with a plane wave coming from azimuth angle with
transmit mode, implies that the current for loads
on an entire open-circuit antenna is zero. In the receive mode, giving antenna current .
a similar substitution of is made for in (8), which Fig. 4 plots the current distribution on the antenna for
leads to the observation that when is computed, only the different loads obtained in two ways. First, the load cur-
currents induced by for are used and currents rent in the receive mode combined with the open-cir-
on other antennas are treated as zero. cuit current is used to compute the
Antennas that exhibit negligible current when open-cir- current everywhere on the antenna, or
cuited, or , are referred to as minimally scattering , where
antennas [16], where the principle example is thin wire dipoles. is the load position. In this first case, the current can be com-
Clearly for general antennas, will not be close to , puted for any receive load using just two basis functions.
meaning that significant currents can flow on the surface of an Second, the current is taken directly from receive mode simu-
open-circuited antenna. Since these currents will affect and lations .
, assuming isolated patterns can lead to significant error in The result indicates nearly perfect agreement in the current
network computations. distribution for the receive mode simulation for different loads
2) Current Distribution on Loads and Sources: The network and the value predicted from transmit-mode quantities. This also
model in Fig. 2 allows arbitrary loads and sources to be con- confirms that for the chosen incident field, the current distribu-
nected to the ports. However, note that the model is only exact tion on the antenna is the sum of only two independent basis
when the current densities on the ports have the same distri- functions.
bution as the basis functions that were chosen for computing
. When the current distributions on the ports are significantly B. Two Dipole Simulation
altered, the matrices , and are also different, Next, we consider the case of two dipoles and show that
leading to error. load voltages in the receive mode are exactly predicted by
WALLACE AND MEHMOOD: ON THE ACCURACY OF EQUIVALENT CIRCUIT MODELS FOR MULTI-ANTENNA SYSTEMS 545

Fig. 6. Directional gain of a 7-element parasitic dipole array (thick line) and
fractional deviation of gain when network analysis is used (thin lines), where
Fig. 5. Voltage on antenna 1 for simulations of coupled 2-dipole simulations, is the number of fixed-length segments used for the load.
computed using direct receive (RX) mode simulations, using transmit (TX) sim-
ulations with embedded patterns (EP) and TX simulations using isolated pat-
terns (IP).

transmit-mode only quantities, regardless of dipole separation.

NEC2 simulations were performed for two -directed dipoles
separated by distance with and .
First, the array is analyzed in the transmit mode, where the
quantities and are found by performing sim-
ulations, where for the simulation, port is driven with
a 1 V source across the terminals and short circuits (PEC)
are placed across the terminals of the other elements. The
resulting vector of terminal currents is denoted , giving Fig. 7. Load voltages for the terminated 7-element array for plane wave excita-
tion arriving from azimuth angle . Solid lines show voltages computed using
the column of the admittance matrix . Far-fields are transmit-mode quantities with network analysis and points show values from
and the far-field pattern direct receive-mode simulations.
is stored for the azimuthal plane ,
yielding the column of . After performing a sim-
ulation for each port, the impedance matrix is computed with C. Simulation of a Parasitic Array
and is found from (18) as .
In this section we demonstrate the behavior of a larger array
Second, the array is analyzed directly in the receive mode by with parasitic loading. We also study the effect of a varying port
terminating each antenna with and running a sepa- current distribution that can affect the accuracy of the network
rate simulation for a plane wave arriving in the plane from analysis computations.
azimuthal angle with . Moment method simulations of a 7-element uniform linear
The load voltage on one of the antennas in the receive mode array of dipoles were performed using NEC2 identical to the
versus the arrival angle is obtained three different ways. In the antennas in Sections III-A and III-B, except a fixed inter-ele-
first case, the open-circuit voltage of the array for arrival angle ment spacing of was used. We also consider the
is computed using the embedded transmit patterns ac- case of a distributed load occupying of the segments at
cording to (13), after which the loaded voltage for arbitrary load the middle of the antenna, allowing the port current distribution
impedance can be computed. In the second case, we apply to be changed.
the same procedure, except that the open-circuit voltage is com- First, we demonstrate how the size of the load can affect the
puted using the transmit patterns of isolated elements (i.e., one accuracy of network computations. Fig. 6 depicts the directive
antenna is driven with 1 V and the other is removed). In the gain of the array for (thick black line) obtained
third case, the voltages from direct receive-mode simulations from a single transmit-mode NEC2 simulation when the center
are used. element (antenna 4) is driven with an active source, antennas
Fig. 5(a) and (b) show the amplitude and phase, respectively, 1–3 are terminated with and antennas 5-7 are termi-
of the voltage on antenna 1 for and . The result nated with . Next, directional gain was computed
shows exact agreement between the transmit and receive mode with network analysis using and and the results of
cases when the embedded patterns are used to obtain . Also, Section II-G. The fractional error of the network analysis so-
since the results are not normalized, the agreement validates the lution compared to the direct solution is shown as thin lines in
scaling constant in (13) relating to the embedded open-cir- the plot. For small , network analysis and the direct solu-
cuit transmit patterns. When isolated patterns are used, however, tion give nearly identical results, whereas for increasing
a small amount of error (mainly in the amplitude) is created for moderate error is obtained.
dipoles separated by less than , which may or may not be Finally, we illustrate again that transmit-mode quantities
tolerable depending on the application. can precisely predict the receive-mode response. Fig. 7 plots
546 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 8. Photos of the parasitic reconfigurable antenna measured in a compact

anechoic chamber: (a) parasitic array where ports 1 and 2 are connected to the
receiver and a varactor diode load (b) reference patch antenna used to illuminate
the array (c) reconfigurable varactor diode circuit.

the terminal voltage on three of the antennas for a single Fig. 9. Measurement cases taken in an anechoic chamber to illustrate antenna
computations.
plane wave, where antennas 1–4 and 5–7 are terminated with
and , respectively and . The
receive load voltage was computed with network analysis
the diode and was measured, where
using transmit mode quantities (solid lines) and with direct
is the complex amplitude of the constant incident wave fed to
receive-mode NEC2 simulations for plane waves at specific
the patch and the superscript “(a)” denotes configuration (a) in
angles (points). Nearly exact agreement is obtained, indicating
Fig. 9.
that the receive-mode behavior can be computed from the usual
Next, we show that the antenna system can be character-
transmit-mode quantities.
ized using pure transmit-mode quantities. Consider the recon-
IV. EXPERIMENTAL EXAMPLE figurable antenna to be a two-port element, where Port 2 is ter-
minated with the varactor diode load having the bias-depen-
This section provides a simple experiment that confirms that dent reflection coefficient . Using (20) and noting that
transmit-mode quantities can be used to predict receive mode , the outgoing wave from Port 1 of the an-
behavior in real antenna systems. The chosen antenna is a re- tenna is
configurable parasitic antenna that has application in adaptive
matching and pattern synthesis. (31)
Fig. 8(a) depicts the antenna system to be analyzed, con-
sisting of one monopole antenna connected to a receiver (Port Reflection of the varactor circuit with respect to is per-
1) and a coupled parasitic monopole terminated with a varactor formed with a 1-port VNA measurement to obtain .
diode having variable 0–5 V reverse bias (Port 2), where the var- S-parameters of the 2-port monopole array are found using
actor diode circuit is depicted in Fig. 8(c). The monopoles are the usual transmit-mode configuration in Fig. 9(b), giving the
spaced by 1 cm which is approximately at the operating values , and
frequency of 2.2 GHz. Antenna measurements were performed as required in (31).
at 2.2 GHz in the anechoic chamber depicted Next, the source waves and must be determined in the
where a low transmit power of 0 dBm was used to ensure lin- presence of the transmitting patch antenna for each illumination
earity of the varactor diode. The reference antenna for the exper- angle . The quantity is found using the reciprocal arrange-
iment was a patch antenna mounted on a chamber wall, depicted ment in Fig. 9(c), where for each angle Monopole 1 is excited
in Fig. 8(b). with an incident wave having complex amplitude , Monopole
Measurements were performed with a two-port 2 is terminated with and the wave received by the patch is
Rohde&Schwarz VNB20 vector network analyzer outside measured. This measurement gives
of the chamber, connected to the antennas with phase-stable
cables. To avoid having to reroute cables between the various (32)
measurements, three cables were run from the VNA location
to the antenna ports inside the chamber, where two 3 m cables
were run to the monopole antenna ports and one 6 m cable to indicating that can be obtained by just scaling the transmit-
the patch antenna. Note that the VNA was calibrated for each mode quantity . Similarly, the arrangement in Fig. 9(d) is
configuration to obtain S-parameters with respect to the ends used to obtain . Substituting into (31),
of the cables.
First, the receive antenna was characterized directly in the (33)
pure receive mode using the arrangement in Fig. 9(a), where
for each rotation angle , biases Fig. 10 compares direct receive-mode measurement of
of were successively placed on and the value computed with network analysis
WALLACE AND MEHMOOD: ON THE ACCURACY OF EQUIVALENT CIRCUIT MODELS FOR MULTI-ANTENNA SYSTEMS
Fig. 10. Received signal on a parasitically controlled reconfigurable antenna
obtained by direct measurement in the receive mode (RX) and predicted using
network analysis and measured transmit-mode quantities (TX), where is the
rotation angle of the antenna and is the reverse bias placed on the varactor
diode element.
in (31) using transmit-mode quantities, where cases of both
the bias and sweep angle being held constant are shown. Good
agreement is obtained and the small discrepancies are to be ex-
pected due to separate measurements being performed, where
disconnection and reconnection of a single cable caused as
much as and variation in measured S-parameters.
V. CONCLUSION
This paper has analyzed antenna arrays consisting of gen-
eral PEC surfaces based on a rigorous method of moments
(MOM) formulation, showing that equivalent circuit models
are exact from the standpoint of electromagnetic wave theory.
The results also indicate that receive-mode operation of arrays
is exactly predicted by employing the usual transmit-mode
network parameters (Z, Y, or S-parameters) and an excita-
tion-dependent source term. For the case of reciprocal antennas
and far-field sources, this source term is completely determined
by the transmit-mode embedded radiation patterns of the array.
Two sources of possible error in equivalent network models
were identified, namely using isolated instead of embedded
element patterns and modeling the incorrect current mode at
the ports. Numerical and experimental examples demonstrated
the validity of the analytical results and the effect of the sources
of error.
REFERENCES
[1] C. Waldschmidt, S. Schulteis, and W. Wiesbeck, “Complete RF
system model for analysis of compact MIMO arrays,” IEEE Trans.
Veh. Technol, vol. 53, pp. 579–586, May 2004.
[2] J. W. Wallace and M. A. Jensen, “Mutual coupling in MIMO wireless
systems: A rigorous network theory analysis,” IEEE Trans. Wireless
Commun, vol. 3, pp. 1317–1325, Jul. 2004.
[3] M. L. Morris and M. A. Jensen, “Network model for MIMO systems
with coupled antennas and noisy amplifiers,” IEEE Trans. Antennas
Propag, vol. 53, pp. 545–552, Feb. 2005.
547
[4] B. K. Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch, “Impact
of matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag, vol. 54, pp. 3225–3238, Nov. 2006.
[5] V. Papamichael and C. Soras, “MIMO antenna modelling using the
effective length matrices,” Progr. Electromagn. Res. C, vol. 10, pp.
111–127, Oct. 2009.
[6] A. T. D. Hoop, “The -port receiving antenna and its equivalent elec-
trical network,” Philips Res. Rep., vol. 30, pp. 302–315, 1975.
[7] I. Gupta and A. Ksienski, “Effect of mutual coupling on the perfor-
mance of adaptive arrays,” IEEE Trans. Antennas Propag, vol. 31, pp.
785–791, Sep. 1983.
[8] J. Van Bladel, “On the equivalent circuit of a receiving antenna,” IEEE
Antennas Propag. Mag., vol. 44, pp. 164–165, Feb. 2002.
[9] P.-S. Kildal, “Equivalent circuits of receive antennas in signal pro-
cessing arrays,” Microwave Opt. Technol. Lett., vol. 21, pp. 244–246,
May 1999.
[10] R. S. Adve and T. K. Sarkar, “Compensation for the effects of mutual
coupling on direct data domain adaptive algorithms,” IEEE Trans. An-
tennas Propag, vol. 48, pp. 86–94, Jan. 2000.
[11] A. W. Love, “Comment on the equivalent circuit of a receiving an-
tenna,” IEEE Trans. Antennas Propag, vol. 44, pp. 124–125, Oct. 2002.
[12] W. Geyi, “Derivation of equivalent circuits for receiving antenna,”
IEEE Trans. Antennas Propag, vol. 52, pp. 1620–1623, Jun. 2004.
[13] H. T. Hui, “A new definition of mutual impedance for application in
dipole receiving antenna arrays,” IEEE Antennas Wireless Propag.
Lett, vol. 3, pp. 364–367, Mar. 2004.
[14] H.-S. Lui, H. T. Hui, and M. S. Leong, “A note on the mutual-coupling
problems in transmitting and receiving antenna arrays,” IEEE Antennas
Propag. Mag., vol. 51, pp. 171–176, Oct. 2009.
[15] J. A. Kong, Electromagnetic Wave Theory, 2nd ed. New York: Wiley,
1990.
[16] W. Wasylkiwskyj and W. Kahn, “Theory of mutual coupling among
minimum-scattering antennas,” IEEE Trans. Antennas Propag, vol. 18,
pp. 204–216, Mar. 1970.
[17] [Online]. Available: http://www.nec2.org
Jon W. Wallace (S’99–M’03) received the B.S.
(summa cum laude) and Ph.D. degrees in electrical
engineering from Brigham Young University (BYU),
Provo, UT, in 1997 and 2002, respectively.
He received the National Science Foundation
Graduate Fellowship in 1998 and worked as a
Graduate Research Assistant at BYU until 2002.
From 2002 to 2003, he was with the Mobile
Communications Group, Vienna University of
Technology, Vienna, Austria. From 2003 to 2006,
he was a Research Associate with the BYU Wireless
Communications Laboratory. Since 2006, he has been an Assistant Professor
of electrical engineering at Jacobs University, Bremen, Germany. His current
research interests include MIMO wireless systems, physical-layer security,
cognitive radio and UWB systems.
Dr. Wallace is serving as an Associate Editor of the IEEE TRANSACTIONS ON
ANTENNAS AND PROPAGATION and is a Co-Guest Editor of the Special Issue on
Multiple-Input Multiple-Output (MIMO) Technology.
Rashid Mehmood (S’05) received the B.Sc. degree
(cum laude) in communication systems engineering
from the Institute of Space Technology (IST),
Pakistan, in 2007 and the M.Sc. degree in electrical
engineering from Jacobs University Bremen (JUB),
Bremen, Germany, in 2010.
From 2007 to 2008, he worked as a Research As-
sociate at IST and supervised various undergraduate
laboratories. From 2008 to 2010, he worked as a Re-
search Assistant in several laboratories at JUB and
external companies. His current research interests in-
clude reconfigurable aperture antennas, antenna optimization and wireless and
optical communications.
Mr. Mehmood was a recipient of the 2009 IEEE AP-S Undergraduate Re-
search Award.
548 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

A Low-Cost MIMO Channel Sounder Architecture

Without Phase Synchronization
Daniele Pinchera, Member, IEEE, and Marco Donald Migliore, Member, IEEE

Abstract—A MIMO channel sounder that does not require phase

measurements is proposed. The system is based on parasitic an-
tennas, with a single active element connected to a low expensive
amplitude only receiver and a proper phase retrieval algorithm
whose computation burden is compatible with today personal com-
puters. Numerical simulations confirm the effectiveness of the ap-
proach.
Index Terms—Channel estimation, microwave propagation,
MIMO, mutual coupling, phase estimation.

I. INTRODUCTION Fig. 1. Scheme of a standard MIMO sounder.

T HE effectiveness of a MIMO system is strictly related to

the number of orthogonal communication sub-channels
that can reliably transmit information. Consequently, the char-
acterization of the MIMO channel is of paramount importance
for the evaluation of the performance of the communication
system. In order to measure the channel matrix, high-accuracy
channel sounders have been developed, for example [1]–[4].
The method usually adopted for the estimation of the -th
element of the channel matrix is to measure the ratio between
the phasor associated to the harmonic signal received by the
-th receiving element and the phasor associated to the har-
monic signal transmitted by the -th transmitting element of the
MIMO system. A critical point of this strategy is the exact mea-
surement of the difference between the phase of the transmitted
and the received signal. The phase noise in a MIMO sounder is
indeed an important issue [5], [6]. Low phase noise can be ob-
tained using a very stable time-reference in the transmitter and
receiver system, like the use of atomic clocks [1] or GSM [4].
In a different approach the reference is obtained directly by the
received signal after a proper demodulation [2]. However, even
this solution is complex, since it requires, in practice, the devel-
opment of a complete MIMO system.
In this paper a radically different approach, presented for the
first time in [7], is proposed to overcome the problem of phase
synchronization using amplitude only measurements.
The key point of our solution is the use of parasitic loads in-
stead of RF switches (see Figs. 1 and 2), with three main ad-
vantages: first, the strong reduction of the sounder cost, since
parasitic loads are cheaper than RF switches; second, since the
possible number of combinations of the impedance states can be
Manuscript received May 31, 2010; revised March 16, 2011; accepted April
09, 2011. Date of publication October 25, 2011; date of current version February
03, 2012.
The authors are with the DAEIMI, University of Cassino, Cassino 03043,
Italy (e-mail: pinchera@unicas.it; mdmiglio@unicas.it).
Digital Object Identifier 10.1109/TAP.2011.2173433
Fig. 2. Scheme of the Parasitic MIMO Channel Sounder (PMCS).
very large we could use these redundant data to improve the re-
construction of the channel; third, we could use a simple scalar
detector as a receiver, since it is possible to retrieve the phase
information of the signals by means of a proper elaboration of
the measurements. Broadly speaking, the idea at the basis of the
proposed method is to use the simplest (and cheapest) possible
hardware (e.g., parasitic loads and a simple scalar detector), at
a cost of a greater computation burden compared to other ap-
proaches. For example, a possible hardware consists of a gener-
ator, connected to the transmitting parasitic antenna array, and
a (scalar) spectrum analyzer connected to the receiving para-
sitic antenna array. The electronic switches in transmission and
reception are controlled by a microcontroller, while a low-cost
wireless connection (for example ZigBee or Wi-Fi) is used to
control the state of the switches at the transmitter side.
It has to be noticed that the cost and complexity of RF
switches increase very rapidly with the number of terminations
needed; the use of single controllable impedances allows an
almost linear increase of the cost and complexity of the mea-
suring system with the number of employed parasitic loads.
As it will be shown in the following, the performances of
the scheme that we propose are slightly inferior than the per-
formances of existing schemes, but the strong reduction of the

0018-926X/$26.00 © 2011 IEEE

PINCHERA AND MIGLIORE: A LOW-COST MIMO CHANNEL SOUNDER ARCHITECTURE WITHOUT PHASE SYNCHRONIZATION
hardware cost could make our solution attractive in many appli-
cations.
It is interesting to note that this idea resembles, at least in the
basic philosophy, a recent trend in data acquisition and elabo-
ration, characterized by the simplification of the detector at the
cost of a more complex data elaboration [8].
The paper is organized in the following Sections. The
problem of the measurement of the MIMO channel matrix
using vector and amplitude only strategies is discussed in
Section II. Section III is devoted to the description of the
amplitude only MIMO channel sounder architecture, while the
phase-retrieval algorithm is described in Section IV. Numerical
simulations in case of measurement of a large MIMO channel
matrix (25 25 elements) and small channel matrix (4 4
elements) are reported in Section V. Finally, in Section VI
conclusions and observations are reported.
II. THE MEASUREMENT OF THE MIMO CHANNEL MATRIX
Let us consider a MIMO system constituted by transmit-
ting and receiving elements. In the following harmonic sig-
nals at frequency will be considered, and the time-dependence
, wherein , will be understood and dropped.
The channel matrix of such a system relating the vector
of the receiver currents and the vector of the transmitter
voltages, can be written as [9]:
(1)
where and are the impedance matrices of the receiving
and transmitting array, is the matrix relating the currents on
the transmitting antennas to the currents on the receiving an-
tennas (MIMO propagation matrix in the following), and
and are the diagonal matrices of the internal impedances of
the generators and the receiver equivalent impedances, respec-
tively. The matrices and are known quantities,
that do not depend on the particular channel realization. Our
goal is the estimation of the propagation matrix .
In standard channel sounders the elements of are measured
using a vector (amplitude and phase) receiver. The estimation of
requires a simple inversion procedure. Under suitable condi-
tioning of the and matrices, the propa-
gation matrix can be stably recovered. This method requires
a number of transmitters and receivers equal to the length of the
vectors and respectively, or a proper switching system at
both communication sides. Furthermore, it requires phase syn-
chronization, which is a cumbersome problem. In case of short
distance the phase reference can be transmitted by cables or op-
tical fibers, but this affects the flexibility of the system. An ef-
fective solution is the use of highly stable time reference at both
the transmitter and receiver side.
A further approach to solve the problem of phase estimation
is to completely avoid the measurement of the phase, and to es-
timate it from some functional relationships in the amplitude
only received signals. This approach is used, for instance, in
antenna near-field measurement systems, in which the phase is
estimated from amplitude only measured data. The basic idea
of the phase-retrieval methods in near-field measurements is to
549
Fig. 3. Graphical representation of the linear mapping .
Fig. 4. Scheme of the alternate projection algorithm; the solution is the inter-
section point (if exists) between the two sets and ; in absence of intersection
point, the solution is the point which minimizes the distance between the two
sets; the non-convexity of the sets causes the presence of local minima in which
the algorithm can be trapped.
collect a set of amplitude only data, which is (hopefully) com-
patible with only the true phase distribution, avoiding false so-
lutions [11].
Before introducing the details of the proposed technique, it
is useful to clarify the general strategy of the method using a
simple geometrical approach.
A matrix maps a -dimensional hypersphere
having unit radius, into a -dimensional hyperellipsoid. The
length of the semi-axes of the hyperellipsoid are equal to the
singular values of . Consequently, from a geometrical point
of view, the problem is the estimation of its principal axes. This
requires the identification of the hyperellipsoid apart from an
arbitrary rotation around the origin. For the sake of simplicity
and clarity, in the following the matrix channel will be sup-
posed a 2 2 real symmetric matrix having full rank. Extension
to complex non-hermitian matrices is straightforward, but less
geometrically intuitive.
With reference to Fig. 3, the ellipsoid can be identified by
varying on the unit-circle of the plane, and observing
the output . In particular, standard MIMO sounders
choose along the coordinate axes of the system , i.e.,
and . The values of
allow to identify the ellipsoid, and hence the singular
values.
In order to give an intuitive idea of the problems related to
amplitude-only measurements, let us suppose that we measure
only the length of the vectors and . The amount of in-
formation obtained from the length of the 2 output vectors is
only one-half compared to the amount of information associated
to . Consequently, only two measurements are not capable to
550 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

uniquely identify the ellipses. In order to increase the number the transmitting and one at the receiving side, and, respectively,
of data, we need to observe the output of under other input and parasitic elements on the transmitting and re-
vectors . This would require a beamforming at the transmitter, ceiving side. The parasitic elements are loaded with controllable
that would increase the complexity and the cost of the sounder. impedances that can assume, for instance, two different values
A possible approach to obtain a larger set of vectors is to ( and ). Note that the synchronization channel shown
start with a single vector, let be, and to apply an orthogonal in figure is used only to synchronize the switches and, conse-
transformation at the transmitter side that rotates the vector quently, the value of the loads used at the transmitter and re-
. By using different orthogonal transformations ceiver antenna, and not for a phase-synchronization of the trans-
, we can obtain different outputs . mitter and receiver oscillators. For these reasons, transmitter-re-
Let us consider the received vector . In order to simplify the ceiver synchronization requires a relatively simple hardware.
receiver we suppose that we are able to measure only one entry The relationship between the current on the active antenna at
of , for example only the first entry. In order to measure all the receiver and the voltage on the active antenna at the trans-
the components the vector, we can apply a further orthogonal mitter is [9]:
transformation at the receiver side that rotates the vector
. By using different orthogonal transformations
, we obtain different outputs . (2)
Using the above described approach, the system becomes
, wherein is an operator that se- where is the diagonal
lects only the first entry of . matrix containing the internal impedance of the generator
Note that the receiver must know which is used by the of the transmitting active antenna and the load impedances
transmitter. This requires a form of synchronization between of the transmitting parasitic antennas
transmitter and receiver. This synchronization is fundamentally in the state defined by the binary vector .
different from the phase-synchronization required in other This means that the -th parasitic element will be connected
MIMO sounders, and can be implemented in a quite simple to an impedance if and to an
way using low-rate wireless systems. impedance if . Similarly we have
The number of measured data depends on the number of that is the diagonal
and that can be synthesized. Consequently, a key point in matrix containing the internal impedance of the receiver
the application of the method in amplitude-only measurements attached to the active antenna and the parasitic load impedances
is related to the possibility of obtaining a large number of dif- for the receiving parasitic antennas in
ferent and in a cheap and fast way. Regarding this the state defined by the binary vector , and
point, it is worth stressing that the method works also in case and are two vectors
of not-orthogonal and matrices, provided that they are taking into account the fact that we cannot access to the signals
sufficiently well conditioned, of course at the cost of a higher on the parasitic elements.
computational complexity. In practice, we have access to a single element of the matrix
This work proposes a low-cost easily-implementable way to . However, we have the possibility to modify
obtain such matrices, and an effective algorithm to estimate the the matrices and and, consequently, we can ob-
singular values from amplitude-only data. serve different outputs associated to different matrices.
In particular, the identification of an effective strategy to ob- Let us suppose that the number of parasitic load com-
tain pieces of information large enough to estimate the phase binations employed is at the receiver and at the
in a stable way will be discussed in the next Section, while the transmitter; in such a way we can define the matrices
problem of identifying the functional relationships between the and
measured data, and an effective algorithm to retrieve the phase wherein and ,
from the set of measured data is the object of Section IV. and and are respectively the -th and -th binary
vector defining the combination of the parasitic load imped-
III. PARASITIC MIMO CHANNEL SOUNDER (PMCS) ances for the receiving and the transmitting antenna, obtaining
ARCHITECTURE
(3)
Let us consider the data acquisition problem. As discussed
in Section II, standard sounder configurations allow to mea- wherein the elements of are the ratio between and
sure independent quantities. This set of data is in- for the considered combinations of transmitting and receiving
sufficient for amplitude only measurements. Consequently, we loads. Roughly speaking, we make the control system on the
must change the sounder hardware to perform a larger number transmitting antenna cycle on the set of parasitic load com-
of measurements. binations while we maintain the parasitic load combination at
The solution that we have investigated to increase the number the receiver fixed, then we change the parasitic load combina-
of acquired data consists in the use of parasitic antennas [12], tion at the receiver and we cycle through all the parasitic load
[13]; these antennas comprise an active element, connected to combination at the transmitter, and so on: in this way the ampli-
the transmitter or the receiver, and a number of parasitic ele- tude of the signal received will be proportional to the entries of
ments placed close to the active element and terminated on con- the matrix read by rows. It is necessary to underline that the
trollable loads (Fig. 2). The system uses only one transceiver at acquisition of the matrix has to be done within a coherence
PINCHERA AND MIGLIORE: A LOW-COST MIMO CHANNEL SOUNDER ARCHITECTURE WITHOUT PHASE SYNCHRONIZATION 551

time of the channel, otherwise the reconstruction of the propa- to estimate the propagation matrix and to do so we need to
gation matrix would be mistaken. reconstruct the phase of the elements of the matrix .
Under the hypothesis that the impedance combinations are Let us consider the quantity
chosen in order to obtain well-conditioned and ma-
trices, the above equation allows to estimate . (8)
Let be the set of complex valued matrices having
dimension, the set of complex valued wherein if the Frobenius matrix norm and is the pro-
matrices having dimension, the set of jector operator on the set of matrices compatible with the
matrices compatible with the employed parasitic system (in choice of the parasitic loads, namely:
the following PSM—parasitic system matrices) and
the linear operator
(9)
(4)
and the matrix is obtained as
wherein and are fixed matrices that depend on the
choice of the parasitic combinations. is a one-to-one, and (10)
hence invertible, mapping under suitable hypotheses on the
matrices and . In particular, if and are full with a phase matrix, i.e., a matrix in which each element is
rank squared (e.g., ) matrices, the eval- in the form and “ ” is the Hadamard element by element
uation of is straightforward, being . matrix product.
If we can evaluate the Moore-Penrose One null of is surely achieved when . The
pseudo-inverse of and , let and be [10], uniqueness of the solution, i.e., that no other phase matrix gives
a matrix lying in the set of matrices compatible with the para-
(5) sitic loads, is still an open problem. Nevertheless, we have un-
(6) dertaken a long numerical investigation, during which no other
nulls were found when the following conditions
and consider the linear mapping
defined in the following way (11)
(12)
(7)
were satisfied. It is not easy to find a proof of (11) and (12),
Note that since and , the null space since non linear equation are involved in the amplitude-only
of will be of dimension . Consequently, the problem. In order to better understand this behavior we could
increasing of the number of parasitic configurations increases observe that the set of equations we are dealing with consists of
the number of dimensions of the null space and only a subspace non linear real equations in real variables;
of the complex matrices of elements corresponds to if the equations were linear we would need at least
matrices of a parasitic system defined by and . and in order to achieve an unique solution, but
The possibility of redundant measurements by means of a since the equations are non linear we need a slightly larger set
large number of parasitic load combinations opens a further in- of equations, thus giving a rough explanation of the obtained
teresting perspective. In fact, use of “redundant” information al- relationships. Numerical evidence of (11) and (12) will be given
lows to simplify the hardware required for the sounder avoiding in Section V.C.
the transmission of the phase information by means of a proper Identification of the null of the objective function (8) can be
phase retrieval algorithm able to reconstruct the lost informa- solved in terms of intersections of two sets. The first set is the
tion. set of all the matrices compatible with the parasitic loads.
The algorithm adopted to solve the problem is a mixed alter- The second set is the set of all the matrices compatible with the
nate projections-random search procedure. The use of such an measured data (e.g., whose elements have amplitude equal to
algorithm, as described in the following, allows to obtain a rea- the elements of ), let be this set. The solution
sonable convergence speed, and a good robustness with respect is the point .
to the local minima problem. In presence of noise-affected data the intersection point gen-
erally does not exist. In this case the condition is
IV. AMPLITUDE-ONLY MIMO CHANNEL SOUNDER PHASE substituted by the condition where is a proper
RETRIEVAL ALGORITHM threshold that is proportional to the additive Gaussian noise at
In the following we will assume, for the sake of simplicity, the receiver (see Appendix).
that we have a perfect knowledge of the system, i.e., there is The minimization problem can be solved by means of the al-
no uncertainty on the definitions of the matrices and ternate-projection algorithm [14], [15]. The alternate-projection
(this assumption will be dropped in the sensitivity analysis of algorithm allows to find the intersection point among sets using
Section V). Furthermore, we suppose that only the amplitude successive orthogonal projections on the sets or, if the intersec-
of the elements of the matrix is known; the amplitude only tion point does not exist, the solution which minimizes the dis-
measured data will be collected in the matrix ; our aim is tance between the two sets.
552 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
The projection operator on the space of the set has already
been defined in (9); the projection operator on the set is:
(13)
where operates on the single entries of the matrix M,
so that . The op-
erator simply substitutes the amplitude of the elements
of the matrix with the measured data, without affecting the
phase of the element.
The alternate projection algorithm implemented can thus be
described by the following steps:
1) Generate a random phase matrix ;
2) Reset iteration-counter;
3) Evaluate ;
4) Increase iteration-counter by 1;
5) Evaluate ; scheme the estimated channel matrix is then compared to the
6) Evaluate ; “real” channel matrix by means of the following metric:
7) If go to step 4;
8) If go to step 1 otherwise stop.
A brief discussion of the algorithm is now in order. When
the two sets are not convex, as in our case, local minima arise,
and the alternate-projection algorithm can be trapped in false
solutions. Once the counter has reached the maximum number
of allowed iterations we can check if the algorithm achieved
, otherwise we can start again, generating another
random phase matrix. The choice of the threshold will be dis-
cussed in the Appendix.
It could be possible that the alternate projection algorithm re-
quires hundreds of trials for but, as will be shown in the next
section, the phase retrieval algorithm remains perfectly compat-
ible with the computing power available with modern PCs. It
has also to be underlined that if we increase the number of par-
asitic load combinations at the transmitter and receiver so that
and we achieve the convergence of the
algorithm in typically less than ten trials.
Once a correct reconstruction of the phase of is performed,
it is trivial to recover the propagation matrix by means of (7).
V. NUMERICAL EXAMPLES
In the following a number of numerical examples will be dis-
cussed, in order to show the flexibility of the proposed method.
In a first set of simulations a 25 25 MIMO system will be used;
in a second set of simulations a 4 4 MIMO system will be con-
sidered.
The numerical test procedure has been the following. A
random channel propagation matrix , simulating a rich scat-
tering environment has been simulated; then the measurement
matrix has been computed, adding to it a white Gaussian
noise matrix ; then the phase of the matrix has been
reconstructed by means of the algorithm presented in the pre-
vious paragraph, obtaining . Once obtained , the estimated
channel matrix is given by:
(14)
Fig. 5. Scheme of the 25 element antenna array.
To evaluate the reconstruction capability of our measurement
(15)
Values of lower than can be considered good for
most purposes.
A. 25 25 MIMO System
The 25 elements MIMO antenna is constituted by 25 wire
elements displaced on a plane in a regular grid of 5 rows by 5
columns (Fig. 5). The central element (marked 0) is the active
one, the remaining 24 elements are parasitic. The inter element
distance on the grid is , so that the overall area of the array
is .
The impedance of the parasitic element has been chosen to be
and . This two values have been
chosen in order to guarantee a good conditioning of the matrices
to be pseudo-inverted (7); it has to be noticed that these values
are close to the values obtainable by pin-diode switches we de-
veloped in a previous work [16]. It could have been possible to
consider impedances with a higher number of states, in order to
increase the possible number of parasitic load combinations, but
we chose to investigate the performances achievable by means
of the simplest parasitic load.
Since the MIMO system is perfectly symmetric we choose
the same parasitic load combinations at both the transmitter and
the receiver. The number of parasitic load combinations for this
antenna, even if the load can assume only two values, is very big,
more than 16 millions; this means that if we want to use a set
of parasitic load combinations at the receiver, the number of
possible sets would be really huge , so a brute
force search of the best possible set of load combinations is
practically impossible.
In order to choose the best combinations a proper genetic
algorithm (GA) [17] has been implemented. The genetic algo-
rithm tries to find the best set of load combinations in order
to reject as much as possible the measurement noise; this is ob-
tained minimizing the condition number of the matrix . The
description of the GA employed is in the Appendix.
PINCHERA AND MIGLIORE: A LOW-COST MIMO CHANNEL SOUNDER ARCHITECTURE WITHOUT PHASE SYNCHRONIZATION 553

Fig. 6. Performance of the parasitic sounder with amplitude only measure-

ments. Fig. 8. Sensitivity analysis for variable and in the case
.

B. Sensitivity Analysis

The results shown up to this point were obtained supposing a

perfect knowledge of the matrices and ; it is thus impor-
tant to understand the stability of the reconstruction procedure
we are proposing with respect to the knowledge of the system.
In order to answer this question we have considered a sensi-
bility analysis, where we supposed an imperfect knowledge of
the impedance matrices and , needed for the reconstruc-
tion of the channel matrix. In particular in the reconstruction of
the channel matrix we used parasitic matrices and cal-
Fig. 7. Performance of the parasitic sounder with vector measurements. culated using the following impedance matrices:

(17)
In Fig. 6 the average of on 100 channel snapshots, for a
(18)
different number of parasitic load combinations, is depicted as
a function of the average SNR at the receiver, defined as:
where and are complex Gaussian variable matrices.
The imperfection on the knowledge of and will be
(16) modeled by the quantities:

where is the matrix of the AWGN noise on the measure of (19)

at the receiver. It has to be underlined that each combination of
the parasitic loads at the transmitter and at the receiver would
result in a different SNR, but for the reconstruction purposes
we look at the whole measurement matrix , so combinations
related to particularly low or high SNRs do not influence the
reconstruction.
It is interesting to observe that there is a great improvement
when passing from 50 parasitic load combinations to 75, while
the improvement is much smaller when passing from 75 to 100
combinations. The performances of the parasitic sounder are
worse than that of the standard sounder, i.e., we require a higher
SNR in order to achieve the same performances, but we would
like to stress the point that the standard sounder requires more
expensive hardware.
For the sake of comparison, in Fig. 7 we have depicted the
performances of the parasitic sounder when vector measure-
ments are employed. It is possible to see that the gap with respect
to the standard sounder is reduced, since we do not need any-
more to estimate the phase from measures, but this solution is
less attractive, since a phase synchronization between the trans-
mitter and the receiver is required.
(20)
For the sake the of simplicity we will suppose
.
The results of the sensitivity analysis are reported in Fig. 8;
according to the reported results, a relative error of
is required to achieve a satisfactory reconstruction. This error
level is comparable with the error made by VNAs commonly
used to measure the impedance matrix of an array; moreover,
since this error is deterministic, it could be possible to set up a
specific calibration procedure in order to further reduce the error
on the knowledge of the system.
A similar sensitivity analysis has been done also on the con-
trollable impedances, obtaining similar results; it has not been
reported here since the uncertainty on the measure of single im-
pedances is usually smaller than the uncertainty on the measure-
ment of an impedance matrix, so we focused on the impedance
matrix to understand the stability of the reconstruction method
proposed.
554
Fig. 9. Number of correct reconstructions for a variable .
Fig. 10. Average number of starting points needed for a variable
C. Convergence and Computational Considerations
In Section VI we have given two “rule of thumb”, (11) and
(12), that should guarantee the convergence of the algorithm.
In Fig. 9 we provide a numerical evidence of such equations:
we have considered the reconstruction of 100 channel snapshots
with a variable , and we have depicted the number
of reconstructed channel matrices that converged to the exact
channel matrices. The graph has a step behavior, with a turning
point around , where we achieve the correct
result in 57% of the considered cases; for the
convergence is guaranteed, and for we have
never converged to the exact solution. This result was obtained
with an of 50 and perfect knowledge of the parasitic
matrices, but similar results, except for the slope of the curve
around the turning point, have been obtained in other cases.
Another aspect that has to be taken into account is the com-
putational burden of the algorithm. First of all, we need to un-
derline that the reconstruction of the channel matrix can be done
off-line once the measured data have been collected, so the com-
putation of the channel matrix can be done on a PC without par-
ticular computation concerns.
Beside this consideration, we have to observe that one of the
key points of the procedure we are proposing is the generation of
a random phase matrix in order to provide a starting point for the
alternate projection algorithm. In principle nobody can assure us
on the number of starting points we need to consider in order to
achieve a solution; for this reason in Fig. 10 we have reported
the average number of starting points that the algorithm needed
to generate in order to reconstruct 100 channel snapshots.
It has to be noticed that the number of starting points is pretty
high when , but it rapidly decreases, so we need
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 11. Performance of the PMCS for a variable at the receiver.
an average number of 8.3 starting points when ,
and an average number of 1.5 starting points when
. For the sake of completeness, we have to say that the av-
erage number of iterations of the alternate projection algorithm,
before it is able to detect if it has converged to a good solution,
or if it needs to generate another starting point, is approximately
300.
D. 4 4 MIMO System
This section has been introduced in order to show that a
PMCS could also be implemented in MIMO systems with a
. relatively low number of antennas. In this case, our transmitting
and receiving antenna is constituted by a linear array of 4 wire
elements, with spacing between the elements. The active
element is one of the inner ones, while the other three are para-
sitic. The parasitic impedances are equal to those considered in
the previous array, thus resulting in a maximum of 8 parasitic
load combinations.
In Fig. 11 it is possible to see the performances of the PMCS
with 8 parasitic load combinations both at the transmitter and at
the receiver and non-LOS rich scattering environment. The per-
formances are lower than the ones obtained in the case of the
25 25 MIMO channel sounder, since the number of parasitic
load combinations that can be used is smaller, but for a high
SNR at the receiver the channel matrix reconstruction is satis-
factory. It is important to stress that the increase of the SNR at
the receiver requires an increase of the length of the measure-
ments, so it is not very difficult to achieve such a SNR, provided
that the entire set of 8 8 channel measurement is performed
within the coherence time of the channel.
VI. CONCLUSION
A novel MIMO sounder, based on the use of parasitic ele-
ments, has been discussed. The sounder is low cost and, ac-
cording to our simulations, allows to obtain a good estimate of
the channel using phaseless measurements.
The basic idea is to use parasitic MIMO antennas with a
single active element connected to an amplitude only receiver
and a proper phase retrieval algorithm. The system requires a
very simple hardware (sinusoidal tone generator, controllable
impedances, and a scalar detector), and a computational burden
that is compatible with current personal computers.
The building and testing of a prototype PMCS is in progress.
The actual implementation of the sounder will be presented in
a future paper, as a part of a much larger investigation on the
PINCHERA AND MIGLIORE: A LOW-COST MIMO CHANNEL SOUNDER ARCHITECTURE WITHOUT PHASE SYNCHRONIZATION
MIMO channel sounding topic; we plan to study also the use of
more complex loads, in order to improve the performances in
terms of noise rejection and stability.
editor for their helpful corrections and suggestions, that have
As last observation, the decreasing of the cost of elabora-
allowed a significant improvement of the quality of the paper.
tion makes the use of strongly different measurement strategies
possible, accepting the acquisition of data indirectly related to
the quantity of interest, provided that the acquired data con-
tain enough information on the quantity of interest. Along this
track, the connections between information theory and electro-
magnetics could play an important role for the next generation
electromagnetic measurement systems.
APPENDIX
A. On the Noise Threshold
In order to derive a simple expression for the threshold as
a function of the receiver noise, let us suppose that we have
obtained the optimal phase matrix , such that . 2009.
If we consider now the measurement matrix reconstructed
with such a phase, we would have:
(21)
where N is the matrix of the AWGN at the PMCS receiver; so
when we compute (8) we would obtain
(22)
where is the variance of the noise at the receiver and we
have used the property , valid for any pro-
jection. The threshold could thus be chosen to be equal to
. vol. 1, pp. 159–163.
B. The Genetic Algorithm Employed
The genetic algorithm used for the selection of the set of par-
asitic load combinations considers “individuals” described a the
vector of discrete values, in the range . 3, pp. 358–371, Nov. 1994.
The population is made of 150 individuals and the selection is
by tournament; overlapping of generations is also implemented.
The crossover operator works by joining the vector of two se-
lected individuals into a single vector of discrete values;
a new individual is then created randomly extracting a set of
discrete values from joined vectors (“Join and Extraction”
scheme).
A mutation can occur on an element of the new vector, that is
changed to a random one in the range , with a prob-
ability . In order to prevent stagnation of the algorithm,
some new random individuals are also added at each iteration
(“Migration scheme”). The algorithm is then terminated after a
fixed number of iterations, usually in the range 1000–10000.
555
ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers and
REFERENCES
[1] J. W. Wallace, M. A. Jensen, A. L. Swindlehurst, and B. D. Jeffs, “Ex-
perimental characterization of the MIMO wireless channel: Data ac-
quisition and analysis,” IEEE Trans. Wireless Commun., vol. 2, no. 2,
pp. 335–343, Mar. 2003.
[2] V. M. Kolmonen, J. Kivinen, L. Vuokko, and P. Vainikainen, “5.3-GHz
MIMO radio channel sounder,” IEEE Trans. Instrum. Meas., vol. 55,
no. 8, pp. 1263–1269, Aug. 2006.
[3] V. M. Kolmonen, P. Almers, J. Salmi, J. Koivunen, K. Haneda, A.
Richter, F. Tufvesson, A. F. Molisch, and P. Vainikainen, “A dynamic
dual-link wideband MIMO channel sounder for 5.3 GHz,” IEEE Trans.
Instrum. Meas., vol. 59, no. 4, pp. 873–883, 2010.
[4] J. M. Molina-Garcia-Pardo, J. V. Rodriguez, and L. Juan-Llacer,
“MIMO channel sounder based on two network analyzers,” IEEE
Trans. Instrum. Meas., vol. 57, no. 9, pp. 2052–2058, 2008.
[5] A. Taparugssanagorn and J. Ylitalo, “Characteristics of short-term
phase noise of MIMO channel sounding and its effect on capacity
estimation,” IEEE Trans. Instrum. Meas., vol. 58, no. 1, pp. 196–201,
[6] A. A. Aboudaand, H. M. El-Sallabi, and S. G. Haggman, “Reducing
impact of phase noise on accuracy of measured MIMO channel ca-
pacity,” IEEE Antennas Wireless Propag. Lett., vol. 6, pp. 419–422,
2007.
[7] D. Pinchera and M. D. Migliore, “A phaseless parasitic MIMO-channel
sounder,” in Proc. IEEE Antennas and Propagation Society Int. Symp.,
2008, pp. 1–4.
[8] D. Takhar, V. Bansal, M. Wakin, M. Duarte, D. Baron, J. Laska, K. F.
Kelly, and R. G. Baraniuk, “A compressed sensing camera: New theory
and an implementation using digital micromirrows,” presented at the
Comp. Imaging, IV SPIE Electronic Imaging, San Jose, Jan. 2006.
[9] M. D. Migliore, D. Pinchera, and F. Schettino, “Improving channel
capacity using adaptive MIMO antennas,” IEEE Trans. Antennas
Propag., vol. 54, no. 11, pt. 2, pp. 3481–3489, Nov. 2006.
[10] G. H. Golub and C. F. Van Loan, Matrix Computations. Baltimore,
MD: The Johns Hopkins Univ. Publishers, 1983.
[11] M. D. Migliore, F. Soldovieri, and R. Pierri, “Far-field antenna pattern
estimation from near-field data using a low-cost amplitude-only mea-
surement setup,” IEEE Trans. Instrum. Meas., vol. 49, no. 1, pp. 71–76,
2000.
[12] M. Wennstrom and T. Svantesson, “An antenna solution for MIMO
channels: The switched parasitic antenna,” in Proc. 12th IEEE Int.
Symp. on Personal, Indoor and Mobile Radio Commun., Sept. 2001,
[13] M. D. Migliore, D. Pinchera, and F. Schettino, “A simple and robust
adaptive parasitic antenna,” IEEE Trans. Antennas Propag., vol. 53,
no. 10, pp. 3262–3272, Oct. 2005.
[14] A. Levi and H. Stark, “Image restoration by the method of generalised
projections with application to restoration from magnitude,” J. Opt.
Soc. Am. A, vol. 1, no. 9, pp. 932–943, 1984.
[15] O. M. Bucci, G. D’Elia, G. Mazzarella, and G. Panariello, “Antenna
pattern synthesis: A new general approach,” Proc. IEEE, vol. 82, no.
[16] D. Pinchera, J. W. Wallace, M. D. Migliore, and M. A. Jensen, “Exper-
imental analysis of a wideband adaptive-MIMO antenna,” IEEE Trans.
Antennas Propag., vol. 56, no. 3, pp. 908–913, Mar. 2008.
[17] A. Fraser and D. Burnel, Computer Models in Genetics. New York:
McGraw-Hill, 1970.
Daniele Pinchera (S’05–M’08) received the Dr.
Eng. degree (summa cum laude) in telecommunica-
tion engineering and the Ph.D. degree in information
and electronic engineering from the University of
Cassino, Italy, in 2004 and 2008, respectively.
He is currently working as a Postdoctoral Re-
searcher in the Faculty of Engineering, University of
Cassino. His current research is in the fields of smart
antennas and MIMO systems, large array synthesis,
compressed sensing, sensor networks and industrial
and medical applications of microwaves.
556 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Marco Donald Migliore (M’04) received the Laurea
degree (honours) in electronic engineering and the
Ph.D. in electronic engineering from the University
of Naples, Naples, Italy.
He is currently an Associate Professor at Univer-
sity of Cassino, where he teaches electromagnetic
fields, propagation in urban area and microwave
measurements. His research interests are adap-
tive and MIMO antennas, antenna measurement
techniques, studies of connection between electro-
magnetic theory and information theory, industrial
and medical applications of microwaves. He is author or coauthor of more than
100 papers in books, journals or international conferences. He was a Visiting
Professor at the University of San Diego in California in 2007 and 2008, and a
speaker at the summer research lecture series of the UCSD CALIT2 Advanced
Network Science (ANS) in 2008.
Prof. Migliore is a member of SiEm (Societ Italiana di Elettromagnetismo)
and the Electromagnetics Academy.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Impact of Incomplete and Inaccurate Data Models
on High Resolution Parameter Estimation in
Multidimensional Channel Sounding
Markus Landmann, Martin Käske, and Reiner S. Thomä, Fellow, IEEE
Abstract—Multidimensional channel sounding aims to estimate
the geometrical structure of multi-path wave propagation in terms
of directions of arrival/departure, Doppler shift, time delay, and
complex polarimetric path weights. Maximum likelihood param-
eter estimation based upon an underlying data model is used to
achieve high-resolution of the path parameters and, thus, renders
possible an antenna independent channel characterization. How-
ever, any mismatch of the underlying data model to physical re-
ality imposes limits to accuracy and reliability of the estimation. To
cope with the limited resolution capability of the setup we are using
a propagation data model that does not only contain discrete de-
terministic components but also a non-resolvable stochastic part.
Joint estimation of both components considerably enhances the
estimation quality and finally allows the interpretation as spec-
ular and diffuse contribution of multi-path propagation respec-
tively. However, besides of noise influence, the achievable resolu-
tion is further limited by the accuracy of the data model that de-
scribes the measurement setup. Since the antenna characteristics
are very susceptible to calibration and modeling errors, the di-
rectional estimates are most error-prone. We refer to the antenna
array calibration procedure and discuss common pitfalls in high-
resolution multi-path direction estimation that are related to inac-
curate and/or incomplete device data model. Depending on the type
of the antenna array (linear, circular) this will inherently produce
biased and artificially spread angular estimates. Only with precise
knowledge of the model errors the stochastic part can be identified
as diffuse propagation component vs. modeling error.
Index Terms—Antenna array calibration, direction of ar-
rival (DoA) estimation, high resolution parameter estimation,
multi-path channel characterization, multi-path cluster, mul-
tiple-input multiple-output (MIMO) channel sounding, po-
larimetry.
I. INTRODUCTION
M ULTIDIMENSIONAL channel sounding is the key
technology for experimental analysis of electromag-
netic wave propagation in mobile radio [1], [2]. Measurement
campaigns are conducted in representative deployment sce-
narios [3]. The results help to understand the physical propa-
Manuscript received August 05, 2010; revised March 28, 2011; accepted
September 14, 2011. Date of publication October 25, 2011; date of cur-
rent version February 03, 2012. This work was supported in part by the
Deutsche Forschungsgemeinschaft (DFG), Bundesministerium für Bildung
und Forschung (BMBF) and in part by the European Union (EU).
M. Landmann is with the Fraunhofer Institute for Integrated Circuits IIS
Helmholtzplatz 2, 98693 Ilmenau, Germany.
M. Käske and R. Thomä and are with Ilmenau University of Technology,
98693 Ilmenau, Germany (e-mail: martin.kaeske@tu-ilmenau.de).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173446
0018-926X/$26.00 © 2011 IEEE
557
gation mechanisms and to recognize fundamental information
theoretic limitations of multiple-input multiple-output (MIMO)
channels in real propagation environments. The data is further
used to deduce and parameterize channel models, and to eval-
uate transmit and receiver signal processing schemes. Once the
channel response has been recorded in its multiple dimensions
(spatial, temporal, and frequency) by the measurement system,
the data undergoes extensive analysis. Gaining explicit knowl-
edge of the geometric structure of wave propagation in terms
of direction of arrival (DoA), direction of departure (DoD),
time of arrival (ToA), Doppler shift, and complex polarimetric
path weights is a prerequisite for most of the propagation
study and data analysis procedures. A wide variety of high
resolution parameter estimation (HRPE) is available [4] to
estimate these structural parameters by fitting an appropriate
data model to the recorded data. Only if we can assume the
underlying data model to be physically correct, complete, and
precise, the achievable estimation accuracy and resolution is
given by the Cramér-Rao lower bound (CRLB), which defines
the fundamental limitations on the parameter variance in case
of noise influence. In practice however, the model will never
be perfect. There are various reasons for this deficiency.
Although the data model has to be accurate enough to repre-
sent the essential effects of electromagnetic wave propagation
and the influence of the measurement system it must be simpli-
fied in order to make it tractable by the HRPE. Regarding the
propagation data model, in most cases a finite number of dis-
crete narrow-band plane waves is assumed describing the spec-
ular components of propagation or the “propagation paths.” For
example, the same data model is applied by the subspace based
estimation of signal parameters via rotational invariance tech-
niques (ESPRIT) method which becomes even more stringent
in this case since ESPRIT additionally presumes identical an-
tenna responses and shift invariant array structures. Very often
omnidirectional antennas are presumed and the influence of po-
larization is not considered [5]. Also whole dimensions such
as elevation or DoD are frequently neglected. In general, max-
imum likelihood (ML) HRPE estimation based upon iterative
search procedures such as SAGE are more flexible which al-
lows defining more sophisticated data models. Nevertheless, the
same simplifications of the propagation data model are often ap-
plied also with ML HRPE estimation (see, e.g., [6]–[8]).
Since the real propagation environment is not finite discrete
but rather a continuum of structures and materials, a purely dis-
crete data model will not suffice. Moreover, polarization cannot
be neglected since it is strongly related to the antenna char-
558 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
acteristics and influenced by scattering and reflection. There-
fore, in this paper we are using a hybrid polarimetric propa-
gation data model which is composed of a superposition of a
finite number of discrete deterministic contributions that are
attributed as specular components (SC) and statistically dis-
tributed dense multipath components (DMC) mainly resulting
from non-resolved diffuse scattering. It was shown that this hy-
brid propagation model matches the received data much better
than the SC-only model [9]. Depending on the propagation en-
vironment the diffuse propagation may carry a considerable part
of the received power which can be in the same order of mag-
nitude as the specular components. Both the SC and the DMC
parts are jointly estimated by the iterative ML HRPE estimation
framework RIMAX which uses a combination of space alter-
nating and gradient search for parameter optimization [10].
In addition to the propagation phenomena the data model
has to include precise knowledge of the measurement device
in all the relevant dimensions to separate its influence from
the measured data. We call this part the device data model.
Since device calibration data are determined by measurements,
there are many practical issues that limit its accuracy. These
are mainly related to the complexity and repeatability of the
measurements. Accuracy may be limited by the available time
and effort, by a lack of precise references, and by noise and
phase drift during the calibration measurement procedure. De-
termination of the antenna characteristics (the array manifold
respectively) is probably the weakest point in this respect which
makes directional estimates most error prone. One reason for
that is the lack of a precise echo-free calibration environment.
Any “anechoic chamber” will give rise to parasitic reflections
which will especially be generated by the installed calibration
equipment itself, namely the antenna positioner. Even if the
equipment is carefully covered by absorbing material, the accu-
racy of the measured radiation patterns is clearly limited which
also precludes the antenna frequency response to be included
into the total frequency response calibration. In this paper
we will describe the effective aperture distribution function
(EADF) method which is not only a memory-efficient descrip-
tion of the radiation patterns but easily allows interpolation and
gradient calculation for data model parameter optimization as
well. Moreover, due to its compactness in the spatial domain
it additionally allows detection and reduction of calibration
errors from parasitic reflections by spatial gating (which is
comparable to time domain gating).
However, the practical situation as reflected by the available
literature seems to be even more severe. It is often observed that
inadequate calibration is performed which will seriously affect
the performance. The common pitfall is to use incomplete an-
tenna responses which implicitly result from ignoring orthog-
onal polarization and/or full solid angle characterization of the
antenna elements. For example, single linear polarimetric and
azimuth-only responses are often considered despite of the true
spread of impinging waves in polarization and elevation and ir-
respective of the real antenna characteristic. Elevation charac-
teristics are often neglected since “only azimuth is of interest”
or only azimuth and elevation cuts of the full solid angle an-
tenna radiation characteristic are used for simplicity. Polariza-
tion is often reduced to single linear if only single port antennas
(such as dipoles) are available. This, however, completely ig-
nores that any real antenna is sensitive to both orthogonal po-
larizations which severely limits the polarization discrimination
ratio especially for off-main radiation direction. Only a few pub-
lications are found trying to avoid these insufficient assumptions
when estimating the radio channel parameters [11]–[13]. In this
paper we will show that in cases of such a serious data model
mismatch the parameter estimator will yield completely wrong
results since it tries to match the wrong (reduced) data model
to the received data. The estimated results will approximate the
avoidable model error by spurious components, thus pretending
multi-path components that are non existing in reality [14]. Ob-
viously, this is primarily a problem of the data model and widely
independent of the actual estimation procedure applied. The
practical consequences are far reaching. If the estimated propa-
gation parameters are used for further analysis such as clustering
algorithms, channel capacity calculations, and parameterization
of geometry-based stochastic channel models (GBSCM) espe-
cially in the angular domain (e.g., in [15]–[17]) we have to make
sure that a full solid angle polarimetric data model is used since
otherwise the results may be rendered useless. Moreover, the
limited accuracy of the antenna array (AA) calibration measure-
ments has to be considered to define the available dynamic range
with respect to the reliability level of the estimated paths. This is
of crucial importance also for the estimation and interpretation
of the hybrid data model described. Only the precise knowledge
and control of the modeling error level allows for identifying the
estimated DMC as either a result of diffuse propagation or just
as a mere approximation of the modeling error.
In the following we will at first describe the hybrid propaga-
tion data model and give an overview of the RIMAX param-
eter estimation framework. Then we will introduce the antenna
data model based on the EADF and describe how it is deter-
mined by a calibration measurement in a well defined anechoic
environment. We will discuss the influence of unavoidable cal-
ibration errors resulting from parasitic reflections and give an
idea about the achievable model accuracy in a practical situa-
tion. Moreover, we will discuss the decisive factors to evaluate
the resulting DoA estimation performance. Then we will dis-
cuss the common pitfalls resulting from using only subsets of
the full solid angle polarimetric antenna response, show the se-
vere estimation errors that will arise in this case, and explain
the reasons of their occurrence. Since practical antennas are in-
volved, we follow an experimental approach based on measure-
ments and simulations. All the parasitic antenna effects such
as mutual coupling, antenna element position error, gain and
phase errors are considered as systematic effects and do not
need a separate discussion since they are already included in the
measured antenna calibration vector. Hence, a further statistical
analysis of these effects, e.g., as given in [18] is not adequate
and also not tractable since all those effects cannot be handled
by a closed form analytical model. On the other hand, the results
given here are strictly valid only for the specific system used.
The aim of the paper is toward emphasizing the physical under-
standing of propagation and device model mismatch, pointing
out the resulting estimation error mechanisms and explaining
the methods of error reduction and mitigation. Hence the road
will be paved to enhance model design and device calibration
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
and end up with reliable high resolution parameter estimation
results.
II. RIMAX CHANNEL PARAMETER ESTIMATION FRAMEWORK
The ultimate goal of Channel Parameter Estimation is to
deduce a parametric model of the MIMO propagation channel
from the recorded data that is independent from the antennas
used during the measurement. This offers the possibility to
emulate the MIMO transfer properties of arbitrary antenna
arrays by reconstructing the output of the dedicated antenna
array from the estimated directional channel parameters which
will be weighted by the dedicated antenna’s radiation char-
acteristic [19]. The key techniques to estimate the individual
path parameters are HRPE and precise device calibration
which includes the antenna array response. From the variety
of estimation methods (see, e.g., [4] for an overview) mostly
subspace methods like ESPRIT and iterative ML procedures
like space alternating generalized expectation maximization
(SAGE) have been applied to multidimensional sounding data
[5], [6], [8], [20], [21]. We have developed a mixed SAGE and
gradient based iterative ML approach for parameter optimiza-
tion that clearly outperforms the pure SAGE based procedure
with respect to local convergence, especially in case of closely
spaced correlated paths ([10], [22]). In the following, we will
describe the resulting RIMAX ML estimation framework
and give details of the underlying hybrid propagation data
model that contains both resolved specular and non-resolved
stochastic components.
A. The Propagation Data Model
As indicated in the introduction the most widely accepted
data model for HRPE in propagation measurement consists of a
discrete multidimensional function. Essentially the same under-
lying model is used in electromagnetic wave propagation mod-
elling such as ray-tracing [23]–[25]. It is based on a ray op-
tical understanding of the propagation phenomena [26]. Con-
sequently, propagation paths are modeled by the sum of fre-
quency independent planar wave fronts in the equivalent base
band. Each path is described by nonlinear structural pa-
rameters: DoD , (azimuth and co-elevation), ToA ,
Doppler-shift , and DoA , , and 4 linear parameters
which are the complex polarimetric path weights: , ,
, . The - and -polarization are defined with respect
to the unit vectors and of the local spherical coordinate
system of the respective antenna array. Only in the equatorial
plane (co-elevation 90 ) the -polarization is identical to hori-
zontal-polarization and the -polarization is identical to vertical
polarization. For notational convenience the parameters of the
’th path are arranged in the parameter vector
(1)
In the discrete angular-delay-doppler parameter domain an in-
dividual path is described by an -dimensional Dirac function,
weighted by a 2 2 complex polarimetric path weight matrix
559
(2)
The Fourier transform reveals a set of multidimensional har-
monic functions in a spatial-frequency-temporal aperture do-
main which indicates that the estimation problem could be han-
dled as a harmonic retrieval procedure:
(3)
is hereby a function of the six parameters
with frequency being the
Fourier transform of the ToA and time being the Fourier
transform of the Doppler-shift . As far as the directional
part is concerned, a reasonable physical interpretation in the
spatial domain can only be achieved if an
additional geometric transformation similar to near/far-field
transformation is included that depends on the antenna array
structure. As will be shown later on, this is not explicitly
needed for the estimation since the RIMAX procedure relies
on measured (“calibrated”) far-field radiation patterns.
B. Extension to a Hybrid Propagation Data Model
Since specular reflection presumes wave interaction at plane
surfaces that are at least as large as some wavelengths it is well
known that the discrete-only (SC) approach is not sufficient to
describe the full reality of wave propagation. Although using a
vast number of weak discrete paths could help to approximate
the microscopic structure of wave interaction with a real propa-
gation environment, the situation in parameter estimation is dif-
ferent. Because of noise and with the limited calibration accu-
racy (as discussed already in the introduction) the resolution ca-
pability of any sounder is somewhat limited. Or in other words
there is not enough information available from the measurement
to resolve all those micro-paths and assign them to a determin-
istic parameter set . Overburdening the estimator would ob-
viously create unreliable results. There are some proposals to
model non-resolvable paths by an appropriate statistical distri-
bution. One approach considers a concentrated von-Mises dis-
tribution in the angular domain to model a plurality of non-re-
solvable paths gathered in a small angular region [27]. In [28]
the product of the von-Mises and the exponential distribution is
used to model the delay-angular distribution. We are following
the DMC approach as described in [29] which is motivated by
the request to describe the observed diffuse scattering in addi-
tion to the resolvable specular reflection.
While the electromagnetic background of diffuse scattering is
already well understood [30] and various attempts are made to
include diffuse scattering in ray tracing propagation modeling
[23]–[25], [31], its influence was widely neglected in HRPE
from sounding measurements. The first joint estimator for deter-
ministic specular and distributed diffuse multipath components
was published in [29]. Recent results have shown that the con-
tribution of the latter varies depending on the complexity of the
560 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

propagation environment [32]–[34]. Diffuse scattered compo- transmitter respectively. The radiation patterns of the re-
nents may be weak in macro-cell line of sight (LoS) scenarios spective antenna elements are in general not uniform for the
(around 20% of the total received power), but they are consid- individual antenna elements of the array. They are also dif-
erably stronger in densely build urban areas and can even dom- ferent for all of the four polarization components in (5). The
inate in complicated scenarios such as factory halls [35] (up to device frequency response can be presumed to be uniform if
80%). a back-to-back calibration is carried out. However, this data
For this reason the hybrid data model proposed in [29] com- model is still not directly applicable to parameter estimation,
prises one deterministic and two stochastic parts that are jointly since it still lacks the systems limited bandwidth and temporal
estimated by the RIMAX procedure. The total observed signal characteristics.1 Furthermore, in (5) the mapping of the angular
vector in (4) (also called snapshot or observation at a fixed parameters appears to be different from the mapping of the ToA
time stamp) contains the superposition of a limited and resolv- and Doppler shift. We can, however, restore a unified mapping
able number of deterministic components . This part cor- of all parameters and incorporate bandlimitation by using the
responds to (1) and is also called the structural part of the prop- EADF and introducing a concise vector/matrix notation
agation data model since it has a clear geometric interpreta-
tion. The second component represents the observed
dense multipath part (DMC) that cannot be resolved further and (6)
assigned reliably to SC components by the estimation proce-
dure. This part can be interpreted as the resulting global contri- The samples in the multidimensional aperture space are ar-
bution of diffuse scattering as observed at both the transmitter ranged in vectors as:
and receiver antennas if and only if model mismatch and esti-
mation errors can be excluded. The third part is the stationary (7)
noise related to the measurement equipment. Moreover, it turns
out that joint estimation of the stochastic parts together with where are sampled versions of the complex exponen-
the SC components will considerably enhance the estimation tials in (2) and (5) and are the normalized structural path
quality of SC part parameters, which are related to their physical counterparts by
unique projections , , ,
, , and a proper
normalization to the respective aperture size, e.g., frequency
bandwidth in case of . The linear projector matrices
(4)
(for transmit polarization and receive polarization
respectively) describe how the individual propagation paths are
C. Mapping of the Structural Propagation Data Model to perceived by the measurement device in the spatial, temporal,
Channel Observation and frequency aperture domain in terms of a multidimensional
linear transfer function. is the weighting of the
In Section II-A the data model of the radio channel was in- Doppler frequencies, is the frequency response,2
troduced using a multidimensional Dirac function or harmonic and and are the EADF matrices of
function respectively. While the two parameters and are di- the individual antenna array elements (related to Tx or Rx re-
rectly accessible by the channel sounder this is not true for the spectively). Each spatial response or is a transformed
angular parameters or their respective Fourier transform. There- two dimensional function of the complex steering vectors and is
fore, we need to explicitly define the mapping of the angular pa- of size . and de-
rameters to the individual antenna array elements of the sounder. note the number of samples of the EADF, see Sections III-B and
The contribution of any single specular propagation path to the III-C for a discussion about the size of the EADF.
total channel observation is described as follows: For simplicity it is assumed that the total projector matrix
is composed of the Kronecker products of to .
This is only valid if to are pairwise independent
which already implies some basic simplifications of the fol-
lowing data model.
1) Delay and Doppler are considered independent although
nonzero Doppler shift requires changing of delay.
2) Doppler is considered frequency independent although
Doppler in general causes a frequency dependent shift of
frequency.
3) In the Doppler domain reduces to an identity matrix and
(5) linear uniform movement and propagating wave directions
1The finite size of the antenna arrays is accounted for by the steering vectors.
The operator denotes the Kronecker product. The com- 2If back-to-back calibration is used will be the identity matrix. If the
plex harmonics in this equation directly correspond to (2), systems frequency response is to be included in the model is a diagonal
denote the steering vectors at the receiver and matrix with the systems frequency response as the main diagonal elements.
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
that do not change within the temporal observation interval
are assumed.
4) Antenna array responses are considered as frequency in-
dependent although spatial antenna separation transforms
to frequency dependent phase and radiation patterns will
have frequency responses which, in general, change with
direction.
5) or are considered independent on distance (re-
spectively time delay) which assumes plane wavefronts
and, hence, distant scatterers (far field).
6) Further separation of the spatial responses and
into Kronecker products of two vectors and
would imply independent azimuth and elevation charac-
teristics (“cuts”). In general this is not adequate as will be
discussed later.
Many of these simplifying assumptions will gradually lose their
justification with increasing bandwidth and array size.
D. Mapping of the Stochastic Part
As described above, the stochastic part of the observation
(4) is composed of two components. The part is attributed
to the stationary measurement noise. Therefore, it is not a part
of the Channel Impulse Response (CIR) since it does not con-
tribute to power transfer from transmitter (Tx) to receiver (Rx).
This part has to be interpreted separately and must be omitted
when a channel transfer matrix is resynthesized from the esti-
mated parameters. Nevertheless, it is of advantage to include it
into the propagation data model since it give us an estimate of
the noise level. Note that the measurement signal-to-noise-ratio
(SNR) is different and (preferably higher) than the SNR of some
transceiver system that might be simulated using the measured
CIR. Moreover, the estimation of the deterministic path param-
eters effectively results in some measurement noise reduction
because of correlation gain. The second stochastic part, on the
other hand, is clearly related to the impulse response since it is
an exponential decaying function which relates to the expected
power delay profile of the non-resolved DMC in
the delay domain. is the corresponding correlation func-
tion in the frequency domain. The describing parameter vector
of this simple model is composed of the parameters , propagation data model described in II-D the DMC influences
, , which are the coherence bandwidth, base delay and max-
imum power respectively:
(8)
In the same manner as the initial SC model in (5) this model also
implies infinite bandwidth whereas the data is observed within
the finite bandwidth of the measurement system. This is actually
a very important issue since it causes the DMC contribution
at any delay-bin to be a superposition of a reasonable number
of diffuse components which warrants its stochastic behavior.
That can be justified only by a finite bandwidth. The resulting
561
covariance matrix of the stochastic part in the frequency
domain is hereby given as
.. ..
. .
(9)
With being the number of samples in the frequency domain
and the spacing between the samples.
This model still does not represent any spatial or rather an-
gular distribution. That is because of the lack of knowledge
about the spatial and temporal structure of the observed global
diffuse scattering in complex environments. The reason is a
lack of measurements with reliable angular resolution of diffuse
components. First results from measurements are described in
[36]–[38].
E. The Parameter Estimation Process
With the unknown (hence random) parameters the observed
signal vector (4) shows a conditional probability density of
(10)
The related log-likelihood function is
(11)
The resulting estimation problem requires a joint solution for
two separate parameter sets, and respectively. Equa-
tion (11) contains the (total) covariance matrix of the
DMC. is assumed to be decomposable into the Kro-
necker-product of four individual matrices
(12)
With being the variance of the measurement noise. With the
the part only. The other covariance matrices are assumed
to be identity matrices which results in a uniform distribution
of the DMC in both spatial and Doppler domains. The Kro-
necker structure of the DMC part (although not really relevant
for this paper) seems to be valid if the correlation in the dif-
ferent domains is independent. Since the influence of the DMC
is not white in the delay domain according to (8), it acts as a
“whitening” function for estimating SC. The knowledge of the
covariance matrix will have considerable influence on the esti-
mation of SC and enhance its quality. However, the covariance
matrix depends on the estimated DMC. Therefore,
both problems are related and joint estimation is required for
the SC and DMC parameter.
Due to the Gaussian nature of the probability density, the
maximization of (11) with respect to the specular components
is essentially a non-linear weighted least squares problem.
Since an exhaustive search in the multidimensional parameter
space is not feasible, an iterative search framework is used
562 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
which is based on both sequential parameter update (SAGE)
and the gradient based Levenberg-Marquardt algorithm. The
latter enhances the convergence performance considerably in
case of coherent paths that are closely co-located (and, hence,
coupled in its influence to ) in the aperture limited multi-
dimensional parameter space. Although the search procedure
could work independently snapshot by snapshot, it is better
initialized with the parameters of the previous snapshot. Since
paths parameters are often persistent and slowly changing,
Kalman tracking can further reduce the search effort per snap-
shot [12]. The number of paths is adaptively controlled by a
birth and death process which discards disappearing paths and
searches snapshot per snapshot for emerging paths.
Paths are discarded if they appear as unreliable. This is indi-
cated by the variance of the estimated path-weights cal-
culated from the estimated Fisher Information Matrix (FIM)
which is available from the gradient estimation [10]. A path is
discarded if its relative variance is above a certain
threshold. In general can be used as the threshold which
means that the variance (or uncertainty) of the path-weight must
be less than halve of the power of the path. This procedure con-
tinuously adjusts the model order according to the evolution
of the propagation environment. Since the variance of the esti-
mates is related to the variance of the stochastic part and thus
to the DMC distribution, exaggerated deterministic approxima-
tion of dense components is avoided. This also avoids possible
path splitting that may result from model order overestimation.
Path splitting occurs when the estimator is forced to estimate
e.g., two paths where in fact only a single path is present. The
estimated paths will be very strong and closely spaced having
nearly equal magnitude and opposite phase. Although the two
estimated paths will approximate the single path with high ac-
curacy they are physically incorrect. This case, however, can be
detected due to the high variance of the parameters that will be
reduced to the usual value by discarding one path.
The parameter vectors and are estimated by an
alternating search procedure. The DMC model search uses a
Gauss-Newton procedure that performs parameter update using
estimates from previous snapshot. The update rate (step-size of
the Gauss-Newton algorithm) of the DMC model parameters
can be chosen smaller than that of the SC parameters since the
resulting diffuse scattering seems to change more slowly with
the “average environment.” The DMC model parameters are
estimated based on
(13)
that is equalized only with the frequency response calibration
vector retaining any influence of the antenna arrays to the
estimated DMC parameters.
F. Interpretation of the DMC-Part
Despite the deficiencies the DMC data model still has, it al-
ready considerably enhances the estimation performance of the
deterministic SC from the observed data by maximizing (11) as
will be demonstrated in Section V. The advantage comes from
the influence of the covariance matrix which causes
a “whitening” of the DMC in the delay domain which consider-
ably enhances the dynamic range of the SC. Without the covari-
ance whitening the SC estimator would not detect the weaker
components at larger delay. Moreover, without the knowledge
about the covariance matrix it is also not possible to validate
the quality of the estimates, thus, disturbing the control of the
model order as described in Section II-E Instead the estimator
would be trying to approximate the non-resolvable strong DMC
with unreliable SC which are better described by the stochastic
DMC distribution.
Furthermore, the DMC model allows the parameter estimator
to reallocate some error from the SC model mismatch to the
DMC. From the viewpoint of SC parameter estimation this is a
clear advantage since it helps to avoid model error approxima-
tion by SC. However, this might jeopardize the interpretation
of the estimated DMC as a result of distributed diffuse scat-
tering. Therefore, we will quantify the most important model
error mechanisms in Section III. It should be noticed that the
described advantages of the hybrid data model can only be re-
alized if both the deterministic and the stochastic part of the
model are estimated jointly. Just estimating the deterministic
part, removing it and then trying to interpret the remainder is
not enough.
III. ANTENNA ARRAY CALIBRATION AND EVALUATION OF
ESTIMATION RESULTS
A. System Calibration
In the following we solely assume that the frequency response
of the measurement device is calibrated. This is accomplished
by connecting the receiver and transmitter directly (without an-
tennas) and measuring the frequency response. However, this
already may include some potential pitfalls which are partly
related to the time domain switching principle of the sounder
system used [39], [40]. Time domain switching has the inherent
advantage of a common up- and down-converter channel re-
sponses for all antenna elements at the transmitter and receiver
side respectively. The limitations comes from the frequency
response of the antennas and antenna switches. This problem
is partly solved by calibrating the antenna array together with
the switching circuit. This makes sure that the different elec-
trical lengths of the cabling of the antenna ports to the switch
are included in the antenna calibration. A rigid construction of
the antenna module is thereby required in order to guarantee
perfect phase stability throughout the period of usage. Con-
cerning the antennas itself we assume frequency independent
responses as discussed already in Section II-C Consideration
of frequency dependent spatial response would require over-
the-air frequency response measurements which would make
antenna calibration even more challenging and would also in-
crease the complexity of the parameter estimator.
Another problem at the receiver side results from the Auto-
matic Gain Control (AGC) which is required to match the input
signal level to the ADC range. Especially in case of nonlinear
antenna arrays (e.g., circular) and directive elements, the AGC
attenuator will adjust the signal power when switching from
one antenna port to another since the received signal strength
changes according to the antenna orientation. So, in general,
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
the AGC switches during one snapshot which will cause phase
variation because of the changing electrical length of the AGC
circuit and, hence, severely distorting the DoA estimation. A
primitive solution is to fix the AGC within one snapshot but that
would be appropriate only in case of linear or planar antenna
arrays. In the RUSK sounder system the AGC induced phase
variation is compensated by using an AGC dependent calibra-
tion vector. That means for each of the different AGC settings
a calibration measurement is performed.
The time domain switching is also often suspected to be
highly sensitive to phase noise. Our investigations have shown
that for the results given here, phase noise was not the limiting
factor since its influence was smaller than the available model
accuracy [9]. Similar assessment is given in [41], [42]. Never-
theless, in chapter V we will include realistic correlated phase
noise in our simulations.
B. Antenna Array Calibration Using EADF
As explained in the previous section the spatial dimension of
the channel response of the AA is described by
and . Precise knowledge of the AA radiation pat-
terns in magnitude and phase is required in order to avoid es-
timation artifacts and to achieve the anticipated resolution and
accuracy. The AA calibration measurement is performed by far
field measurements in an anechoic chamber using a precise po-
larimetric reference horn antenna and the sounder system. As
reference we have used a dual slot coupled horn antenna with
polarization decoupling better than 40 dB in the far field in the
main-beam direction. This is adequate since the reference an-
tenna is always aligned towards the AA. AA calibration using
the broadband channel sounder system saves a lot of time com-
pared to vector network analyzer (VNA) but long term phase
drift can occur because of separate phase locked loop (PLL)s
at Tx and Rx even with 10 MHz reference synchronization.
This phase drift is in general not disturbing field measurements
with fast antenna switching since the required coherence time
interval is short (e.g., around a ms). But it may have a detri-
mental influence in calibration measurements which can last
over hours. Therefore calibration measurements must be cor-
rected either by recording the phase drift of an additional RF
reference channel or by estimating the phase drift using ML
methods as proposed in [43].
The AA element radiation patterns are described by their two
dimensional (2D) discrete Fourier transformation. For each an-
tenna element and for each polarimetric input/output combina-
tion in (6) we need a separate EADF. Since the transformation
of the far field radiation pattern can be interpreted as a distorted
inverse 2D near-field to far-field transform [44] we call it the
effective aperture distribution function EADF [1], [45]. Note
that we are using the EADF as an efficient radiation pattern de-
scriptor and we are not interested in the physically correct field
distribution. The same radiation pattern description (although
not named as EADF), was used in [46] for one-dimensional
(1D) and in [47] for 2D. Compared to the alternative vector
spherical harmonics (VSH) descriptor [48], the EADF is not ro-
tational invariant and the number of required aperture samples
563
may be larger and dependent on the alignment of the antenna
element. Nevertheless due to the use of fast Fourier transforma-
tion (FFT) algorithms the EADF is preferable since it is com-
putational more attractive [9] and rotational invariance is not of
importance.
Application of FFT to sampled radiation pattern is straight-
forward and does not give rise to any leakage error because the
patterns are naturally periodic with respect to the azimuth and
co-elevation angles. This is especially true since FFT processing
is also applied for radiation pattern interpolation and derivation
w.r.t. azimuth and elevation which are required by the gradient
based parameter search of the RIMAX algorithm. This way the
EADF descriptor can be directly applied to construct the linear
projector matrices at Tx and at Rx. Moreover, EADF
is an efficient means to calculate the CRLB of DoA/DoD esti-
mates in case of single and coherent multipath situations [45].
Since it can be determined based on measured as well as simu-
lated radiation patterns (e.g., by using full wave EM solvers) it
is a very powerful tool for evaluating the resolution potential of
real AAs.
C. Error Mitigation Using EADF
The EADF turns out to be not only an efficient tool to
handle the spatial array response. It can also be applied to
detect and minimize some calibration errors. Although AA
calibration measurement is conducted in an anechoic chamber,
parasitic reflections can still occur especially from scattering at
the positioning system which are expected to be attenuated by
10 dB to 40 dB w.r.t. the direct wave for frequencies between 4
GHz and 6 GHz depending on the angle of the incoming wave
and the absorbing material [49]. Since the EADF is related to
the electrical near-field distribution of a corresponding hypo-
thetical antenna element, it should be confined to a compact
support area in 2D space. The EADF will loose its spatial com-
pactness as a result of parasitic reflections from the positioning
device. So the EADF method indicates parasitic reflections
and to some extend also allows for enhancement of calibra-
tion response by spatial tapering which may be considered as
equivalent to temporal gating. Spatial tapering has to be han-
dled with sure instinct since EADF blurring also occurs as a
result of mutual antenna element coupling and if the radiation
pattern can not be measured over the full solid angle due
to the impact of the positioning system which causes tapering
in the radiation pattern domain and, hence EADF leakage.
Furthermore, the EADF will be somehow blurred if the radi-
ation pattern is not measured with respect to its phase center
which will always occur since we are rotating the whole array
during calibration. So it is important to use the optimum pivot
point in the center of the array which causes minimum EADF
blurring.
To demonstrate the effect of EADF distortion by parasitic re-
flections we have carried out simulations and measurements.
Both are related to a polarimetric uniform circular array (PUCA)
with 24 dual-polarized elements. The ideal dual polarimetric ra-
diation response was calculated using the simulation software
WIPL-D [50] with 7 adjacent antenna elements. The influence
564
Fig. 1. (a),(b) Normalized EADFs of a single antenna for vertical excitation
of a simulated and (c) measured PUCPA 24, magnitude in [dB]. [(a) Simulated
undistorted EADF, (b) simulated distorted EADF, (c) measured EADF].
of parasitic reflections from the positioning device was simu-
lated by a ray tracing approach. The strongest parasitic influ-
ence obviously results from a circular absorber plate which was
mounted below the foot of the AA to shield the positioner. De-
pending on the elevation angle of the impinging LOS reference
waveform the plate causes scattering, diffraction, and even LOS
shadowing. In the simulation 10 spatially distributed parasitic
waves with 25 dB attenuation w.r.t. the LOS wave and iden-
tical distributed polarization were assumed. Fig. 1 shows the
resulting EADFs of a single simulated antenna and matched po-
larization without and with distortion in comparison to a mea-
sured antenna element. From comparison of Fig. 1(a) and (b) it
can be concluded that EADF spreading occurs for spatial fre-
quencies and as a consequence of the
radiation pattern distortion due to parasitic reflections. The mea-
sured EADF shows similar behavior. Consequently, distortion
can be reduced by spatial truncation of the measured EADF in
both dimensions. Note that these reflections can not be removed
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 2. Estimated mean angular power distribution versus
deviation from true angle in azimuth as function of co-elevation applying (a)
CEADFs; deviation of angular estimates using CEADF and (b) TEADFs for
single path excitation and a simulated PUCPA with 24 elements (deviation of
angular estimates using TEADF).
by time gating since they are closer than the Rayleigh resolu-
tion of 2.5 m which results from the measurement bandwidth of
120 MHz.
In the following analysis, the performance of DoA parameter
estimation is investigated for two different AA models:
• The distorted and complete EADF (CEADF) with
and ;
• The distorted but truncated EADF (TEADF) with
and .
The parameter estimation results are shown in Fig. 2 for a sim-
ulated single-path excitation which acts as a known reference.
The direction of the impinging wave is tuned over the full
solid angle. In case of the undistorted antenna data model the pa-
rameter estimator RIMAX perfectly returns only one path from
the true directions (which is not shown in the picture). Contrary,
spurious paths appear in the distorted case since the model de-
viates from reality and the spurious are required to approximate
the model error. Up to 9 spurious paths show up for each sim-
ulated arrival direction with a considerable spread in azimuth.
Relative to the true paths the spurious are about
lower in power. Stronger artifacts are estimated closer to the
poles of the spherical coordinate system since in this region the
AA model shows a lower accuracy. This is related to the re-
duced magnitude of the radiation patterns toward these direc-
tions which gives artifacts a stronger relative influence. Spa-
tial truncation of the EADF reduces the level of the artifacts
by about 10 to 15 dB. The advantage of the TEADF model be-
comes most obvious for co-elevation angles of incidence around
where the remaining artifacts are almost negligible.
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
D. Measure of Antenna Array Calibration Accuracy
The mean square model accuracy is related to the Signal-to-
Remainder-Ratio (SRR) which is defined by (14) as the ratio
between the power of the estimated specular paths and the re-
maining signal power which is left after subtraction of the esti-
mated specular paths from the measured data :
(14)
where denotes the estimated noise power which is also re-
moved since by definition it is not a part of the data model. If we
force the parameter estimator to estimate only one single path
which we know is the true model order in the reference case,
the SRR is equal to the mean model accuracy of the data model.
However, if we do not fix the model order in this way (since in
general we do not know the true model order!) the contribution
of the spurious paths actually reduces the SRR thus pretending a
higher accuracy. Consequently, a better approximation of the re-
ceived waveform must not be interpreted as a better result if the
data model is wrong! In our example the resulting mean model
accuracy (estimated from the SRR with ) is around
25 to 33 dB in case of the TEADF which is approximately 8 dB
higher than with the CEADF AA model. This means that the
TEADF AA model is a better match to the “real” (undistorted)
model from which the reference data were deduced. Finally,
the model accuracy is lowest for directions close to the zeros of
the radiation patterns as discussed above.
IV. ESTIMATION ERRORS DUE TO THE APPLICATION OF
INCOMPLETE ANTENNA DATA MODELS
The major effect used in direction estimation ought to be the
phase response of the antenna elements which changes with
incident angle in azimuth and elevation depending on the an-
tenna array architecture. But any antenna element is somehow
directive in the full solid angle domain. This causes radiation
pattern magnitude to be an additional feature which must be
taken into account by the estimation procedure. There are also
many secondary effects that may result in distortion of the ra-
diation patterns. Reasons are manufacturing inaccuracy, influ-
ence of array mounting brackets, and, most of all, mutual el-
ement coupling. All these effects have influence to the mea-
sured antenna response in amplitude and phase. The measured
full solid angle polarimetric antenna response of the complete
array (the “calibration pattern” or the measured steering vector
in (5)) is directly exploited by the parameter estimator to match
the received array response for every path. However, there is a
common misuse of the calibration response which results from
an inappropriate simplification. The consequences of reducing
the model to 1D (e.g., by taking only azimuth cuts) is discussed
in Section IV-A In Section IV-B we discuss the adverse effect
of using only a single-polarimetric antenna data model instead
of a full, dual-polarimetric one.
A. Effect of Using Only 1D Antenna Data Models
For an ideal Uniform Linear Array (ULA), the phase response
along the array is inherently linear for each individual impinging
565
Fig. 3. DoA ambiguity ideal ULA and Phase over 16 element UCA for in-
cidence at different co-elevations . (a) Ambiguity-cone of an ULA, array ele-
ments are placed along -axis, (b) phase distribution of UCA for different co-el-
evation angles, the curve for shows highest variation, the curve for
shows no variation (const. phase for all elements (the curves in be-
tween are ordered according to the legend).
plane wave. Since the angle of incidence is defined with respect
to the array axis in end-fire direction, the phase response shows
a rotational symmetry around this axis. This means that any im-
pinging wave with solid angle of incidence on a cone around
a ULA end-fire direction will impose the same phase gradient
over the array, as shown in Fig. 3(a). There is no way of telling
from the array response which is the correct angle of incidence
in the azimuth plane. So we have to accept an unknown bias
in azimuth depending on co-elevation. More specifically, due to
the lack of aperture in the plane perpendicular to the line through
the array, estimation of the co-elevation angles of incidence with
linear arrays is not possible. So if we don’t have any idea about
the true co-elevation angle of the impinging wave we have no
other chance than using one azimuthal cut of the 3D radiation
pattern at an arbitrary co-elevation angle and we have to accept
an unknown bias error in azimuth. It should be noted that the
problem stated here is an inherent problem of ULAs that cannot
be solved. For practical array the situation turns out to be a little
bit more complicated as discussed below.
For an Uniform Circular Array (UCA) the situation is dif-
ferent. In this case, the phase changes with varying angles of
incidence along the array are not linear but sinusoidal (see
Fig. 3(b)) with phase deflection depending on co-elevation.
This constitutes additional information (compared to the ULA
case) that can be used to estimate the co-elevation of the
impinging wave. However, if we deliberately use only an
azimuthal cut (e.g., for 90 of co-elevation which corresponds
to the highest phase amplitude in Fig. 3(b)) of the antenna data
model, matching of a lower gradient phase distribution (e.g.,
566 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 4. Estimated angular power distribution versus the

azimuth deviation from the true path and co-elevation using the
1D data model of the PULA.

) by a single path is not possible. Therefore, the

parameter estimator will be forced to provide spurious paths
to match the incoming field with the applied AA azimuth-only
model and, hence, pretending angular spread.
1) Measurement Example for an ULA: The consequences
of the ULA’s ambiguity for direction estimation with a mea-
sured Polarimetric Uniform Linear Array (PULA) will be illus-
trated in the following example. In contrast to the ideal isotropic
ULA, the real PULA has directional polarized patch antennas
with nonuniform radiation patterns. Because of the secondary
effects mentioned above the phase slope is not exactly linear and
the magnitude response is similar but not uniform for a single
planar wave. Most important, this non-uniformity changes with
co-elevation. One could even try to deliberately exploit this fea-
ture to estimate the co-elevation angle. In general, for a reason-
able array design the available information is not sufficient to
achieve reliable elevation estimates but the same effect may be
strong enough to seriously degrade the dynamic range of the es-
timator if the co-elevation angle of the impinging wave differs
from that angle at which the azimuth pattern is measured during
calibration (typically: ). In Fig. 4 the results are shown
for a single -polarized wave (estimated with ).
The true angle of incidence is in azimuth with co-eleva-
tion ranging from 20 to 160 . The distribution in the figure is
clipped at amplitude which corresponds to the known
relative model accuracy (as discussed in Section III). The effect
of elevation on bias and variance of the azimuth angle estima-
tion is obvious (both are zero only for ). The spread at
other elevation angles results from the above mentioned pattern
mismatch that depends on elevation if only azimuth patterns for
elevation ) are used. This effect could only be avoided
if the true elevation angle would be available and used to select
the correct azimuth pattern.
2) Measurement Example for an UCA: In this example a 16
element UCA was used. Again, the array was illuminated by
a single wave (always matched in polarization) with the inci-
dence angle ranging over full azimuth and elevation. The es-
timation was performed using the reduced 1D ( , az-
imuth only) pattern. The results in terms of the mean model
accuracy (corresponding to SRR with in (14)) are
shown in Fig. 5(a). The mean model accuracy of the 1D data
model drastically decreases for co-elevation angles toward the
Fig. 5. (a) Mean model accuracy (SRR with ) and (b) estimated
angular power distributions for measured UCA, using 1D data
model.
poles of the spherical coordinate system whereas the full 2D
data model shows almost uniform accuracy resembling the ex-
pected behavior from the discussion in previous sections. The
variation of the curves is related to the circular non-uniformity
of the array in azimuth. In Fig. 5(b) the estimated angular power
spectrum with paths at each true azimuthal DoA is
depicted for the 1D data model. It shows that a single received
SC renders a number of estimated SC which are widely spread
around the true azimuthal angle of arrival except close to 90
co-elevation where the data model is correct. This detrimental
effect is generic for circular arrays and is caused by the funda-
mental dissimilarity between the antenna response measured in
the azimuthal plane and the antenna responses for
co-elevation incidence . For comparison the estimation
was also performed using the full 2D data model. The result (not
shown here) was almost perfect. So compared to the ULA case,
the circular array performs clearly worse if the elevation dimen-
sion was ignored. But on the other hand, the circular array offers
the desired degree of freedom to estimate co-elevation which
allows using elevation dependent azimuth patterns and, hence,
solving the problem.
B. Effect of Using Only Single Polarization Characteristics
Despite the fact that radio waves are of vectorial nature which
means they are composed of two orthogonal polarized compo-
nents perpendicular to the propagation direction, often only a
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
Fig. 6. Radiation pattern magnitude of the first element of the UCA for -po-
larized and -polarized excitation (at 90 co-elevation).
single polarization is considered. This simplification is moti-
vated by the assumption that an antenna element may be sen-
sitive to only one polarization. However, this is not true for a
physically realizable antenna especially if the full solid angle
of incidence is considered. So waves with the “wrong” polar-
ization orientation may deliver a considerable contribution to
the output. But the associate radiation response characteristics
will be widely different for the - and -polarized fields. That
causes a serious model mismatch and results in extensive es-
timation error if we try to estimate the DoA for one of the two
polarization components with data model for the other one. This
is again illustrated by the measured example of the 16 element
UCA.
Although the UCA array was designed from supposedly
-polarized antennas (monopoles), the cross polarization dis-
crimination (XPD), which is the power scaled ratio between
the radiation patterns for co-polarized and cross-polarized
excitation, is strongly varying with the angle of incident. Both
azimuthal radiation patterns of the first element of the UCA for
a constant co-elevation of 90 and - and -polarized excita-
tion is depicted in Fig. 6. The XPD is low (between 0 dB and
10 dB) except for the azimuthal directions . This
strong variation in azimuth maybe somewhat surprising for
vertical monopoles. But it seems that the XPD of the antenna
element is adversely influenced by mutual coupling in the array
configuration. Nevertheless, many publications neglect this
strong cross-polar sensitivity of real antenna elements when
estimating propagation directions. Simple single polarimetric
AA models are rather applied which may have catastrophic
influence to the quality of the parameter estimation results.
In the next example only the -polarized AA model was used.
That is, only the radiation pattern which was measured from
-excitation during calibration was applied for parameter esti-
mation. On the other hand, the incident field in the evaluation
experiment was chosen to be either a pure - or -polarized
single wave front with its angle of incidence ranging again over
full azimuth and elevation. In case of the -excitation (which
perfectly matches the antenna data model) a single path can be
perfectly estimated within the array model accuracy of 20 dB to
25 dB. But for the mismatching -excitation, the mean model
accuracy (SRR) drops to only 5 dB since the radiation pattern
for -polarization does not match. Fig. 7 shows the resulting
mean angular power distribution (averaged over azimuth)
relative to true azimuth direction only for the latter case. The
567
Fig. 7. Estimated angular power distribution for a reduced an-
tenna data model comprising only -polarized response in case of reception of
a single -polarized SC.
magnitude was clipped again at to suppress the spu-
rious components that may result from the limited calibration
accuracy. Displaying the mean in azimuth is reasonable as the
variation in estimation errors over the absolute azimuth angle is
small due to the rather homogeneous response of the UCA in
azimuth. The distribution in the Fig. 7 shows both a strong bias
in azimuth (at the true azimuthal angle almost no power is lo-
cated) and an asymmetric bias in co-elevation. This asymmetry
stems from the asymmetric mechanical structure of the UCA in
vertical direction. The apparent angular spread caused by using
the oversimplified data model is considerable (be reminded that
only a single discrete component was received in reality).
Although the particular shapes of the distributions in Fig. 7
are specific for the array, the problem of a significant loss of
accuracy in comparison with using a complete data model is
generic. One simply does not have control over the polariza-
tion direction of incoming waves during measurements and with
data models using only single linear polarization estimation of
components with the other polarization is bound to yield er-
rors of similar type as described here. So in general, a full po-
larimetric data model (consisting of the two radiation patterns
for both orthogonal polarized excitation) must be used also in
case of supposedly linear polarized (single port) antennas. Only
under very restrictive conditions such estimation errors could be
small. That happens e.g., when the array in fact only receives
a single linear polarization, either because the impinging field
has only a pure linear polarization (matching the calibration data
model) or because the XPD of the antenna elements is extremely
high. The former is not so likely in a real propagation environ-
ment [13] and the latter is only true for very few antenna designs
within a small range of angles of incidence. Otherwise large spu-
rious angular spread has to be expected.
The conclusion is that ULA or antennas arrays with otherwise
reduced calibration dimension can only be used if impinging
waves from those directions are avoided. That has to be guaran-
teed already during data recording. There is no way to correct
the resulting ambiguity and variance by a “better” estimator as
long as no additional calibration data information is available.
So, e.g., ULA could only be used in place of an elevated base
station site where we can assume that azimuthal angle range is
limited to a reasonable broadside range in azimuth and low ele-
vation spread. The latter also means that the ULA must be given
a down tilt to approximate (at least) the main elevation angle or
568 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
we use this azimuthal cross section from the calibration pattern
that corresponds to the expected elevation angle.
V. ESTIMATION QUALITY IN COMPLEX ENVIRONMENTS
This section deals with the overall influence of the real
measurement system including the antenna arrays on the HRPE
results in complicated multipath propagation environments.
Based upon simulations we will demonstrate the impact of im-
perfect calibration. We explain the beneficial influence of DMC
to the accuracy of the estimated SC if both components are
jointly estimated and we discuss the balance in interpretation
of the DMC as a result of non-resolved (“diffuse”) components
vs. approximation of the model error.
A. Simulation Setup
According to the discussion in Section IV we assume that the
complete (full solid angle polarimetric) 2D AA data model is
applied which already excludes the serious mistakes described
there. Here, we intend to include the influence of errors that
in essence cannot be avoided under practical circumstances.
These are remaining inaccuracies in calibration and also re-
ceiver noise. For the latter we consider additive independent
and identically distributed (i.i.d.) complex Gaussian noise and
phase noise. We will discuss the influence of correlated phase
noise because this is a critical issue if the sounder uses the
time division multiplexing (TDM) principle [39], [40]. TDM is
suspected to be error-prone in terms of phase variation during
the snapshot since the snapshot duration is extended in time by
sequential antenna switching. The total snapshot length is given
by the number of all CIRs that build the MIMO measurement
matrix times the length of a single CIR. At the same time
this mostly limits the time interval in which coherent signal
processing takes place. Therefore, as long as phase coherence
time is longer than the snapshot time the impact should be
small. In [41] it was shown that the effect of long-term phase
noise (random walk phase) can be neglected. The situation will
change if phase noise correlation distance becomes smaller.
The impact of correlated short-term phase noise was discussed
in [42], [51], [52]. Thus, we have included the TDM frame
into the simulations and the phase noise statistics used in the
simulation was estimated from the sounding device at hand.
In Sections III and IV we used a single path scenario as a ref-
erence to show the error effects. In a complicated multipath en-
vironment this comparison of measured responses to reference
seems not possible since the “true” channel parameters are never
known. Therefore, we follow the simulation approach. The CIR
raw data are generated by applying a state of the art three di-
mensional (3D) ray-tracer developed at the Karlsruhe [24]. This
gives us the desired access to the true angular distribution of the
rays. The 3D ray-tracing model consists of two major parts: a re-
alistic model of the propagation environment and a model to cal-
culate the multi-path wave propagation between Tx and Rx. The
propagation phenomena taken into account by the ray-tracer are
single reflections, combinations of multiple reflections and mul-
tiple diffractions (up to five interactions), and diffuse scattering
(single interaction). The power ratio between the strongest and
the weakest path is limited to 50 dB. This results in total number
of paths between 500 and 6000. This large number of paths is
related to the attempt of modeling diffuse scattering, which is a
superposition of a large number of non-resolvable weak paths
(weak in terms of power). The vast number of paths is necessary
to demonstrate the resolution limits of the parameter estimator
for the SC paths which will attribute the remaining part to DMC.
From our observation of measured results and also supported by
discussion in [23], [25] it becomes clear that this approach of
simulating diffuse scattering by ray-tracing still underestimates
the non-resolvable part. But, for the intended demonstration this
seems not to be of highest concern.
B. Impact of Distorted Antenna Array Calibration
Based upon the ray-tracing results used as a reference we
can directly compare the true path parameters to the
parameters which are estimated by the HRPE algorithm.
The simulated propagation environment was of a district of
Karlsruhe City, Germany, as shown in Fig. 8. A LoS simulation
route of 250 m length was chosen. The Tx array, a PULA with
8 polarimetric elements, was placed at a height of 38 m above
street level. The Rx array, a Stacked PUCA (SPUCA) with
24 polarimetric elements for each ring, is placed at a height
of 2 m above street level. The noise free channel realization
is generated based upon the true path param-
eters and weighted with the undistorted true AA data
model. For DoA estimation at the Rx simulated completely
undistorted and distorted, but improved by means of TEADF,
radiation patterns are applied as described in Section III. This
allows for comparison of the results of the HRPE algorithm in
case of ideal calibration (perfect knowledge of the radiation
patterns, indicated by the term NoDist) with the real case
with remaining calibration inaccuracy (indicated by the term
FullDist). The TDM switching scheme with an CIR length of
3.2 is applied for all the simulations.
• NoDist : In this ideal case infinite model accuracy is as-
sumed and only additive i.i.d. complex Gaussian noise
is added to the undistorted response ,
which results in the observation
(15)
• FullDist : In this case which resembles real measurements,
observations are generated that are affected both by ad-
ditive i.i.d. complex Gaussian noise and correlated short-
term phase noise
(16)
with the diagonal matrix embossing the phase noise
characteristic of a TDM sounder. The phase noise variance
and the phase noise covariance ma-
trix are estimated from the RUSK channel sounder avail-
able at Ilmenau University of Technology (IUT) and finally
applied to generate the correlated phase noise samples
in the diagonal elements of matrix . The dis-
torted AA data model that will be used for the DoA estima-
tion, is given in terms of the TEADF (which includes the
remaining effect of parasitic reflections during AA calibra-
tion) that cannot be further reduced. The resulting overall
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE 569

Fig. 8. Visualization of raytracer environment.

Fig. 9. Estimated of SC and DMC for the chosen LoS simulation
route in case of the estimation setups and
.
TABLE I
DATA MODELS USED FOR CHANNEL REALIZATION AND ESTIMATION

DMC peak power which is antenna dependent. As a con-

sequence, the following observations can be made.
1) In case of , of the estimated
DMC expresses the of unresolved SC since there
accuracy of the estimation model of the SC (respectively is no AA modeling error. If is close to 0 dB
) is thus around 25 to 30 dB, which is related to the estimated DMC represent mainly the mea-
the accuracy of the AA model and the impact of the phase surement noise, whereas for a higher than 0
noise. dB the estimated DMC can be considered a feature of the
The differences between the two models NoDist and FullDist radio channel.
are summarized in Table I. 2) If the in case of is
In the following two basic estimation approaches are applied higher than the of ,
to the two data models, NoDist and FullDist respectively, as de- then the estimated DMC of are
fined above. With joint SC and DMC estimation (indicated by partially related to model error. Consequently, they may
), we will show whether the estimated DMC re- not properly describe the propagation in the radio channel.
sult mainly from unresolved SC or from model error. By further Let us now analyze the illustrated in Fig. 9 for
comparing to the case of SC-only estimation we will demon- the SC and DMC in case of and
strate the advantage of incorporating the DMC estimation in the for the chosen simulation route.
HRPE algorithm. The legend can be read as follows: In front of the colon
To emphasize the relevance of estimating DMC in the pres- the component of interest is specified and after the colon
ence of model error let us reconsider (4). In the case of the ideal the simulation case NoDist or FullDist respectively is in-
undistorted data model for (and, hence, unbiased es- dicated. Apparently, the for both estimation
timation of ), we are able to subtract all resolvable paths setups are almost overlapping. The in case of
from the observation . This allows for attributing all the re- is always close to 0 dB with only a
maining power of the unresolved DMC part to diffuse prop- few exceptions. This means that almost all received SC are
agation. Using a distorted data model instead, the estimated resolved. For , the is ap-
allows less power to be subtracted with each path and proximately 5 dB to 20 dB higher. Thus, most of the estimated
more power is allocated to DMC. The power difference of the DMC power is related to model error only. Fig. 9 indicates
estimated DMC between the NoDist and the FullDist case in- some correlation between the and the
dicates how much of the DMC results from the measurement in case of the result. If the ratio
procedure or model error respectively. For this end, let us de- between the and is almost equal to
fine the parameter which in general is the ratio between the overall model accuracy of about 25 dB to 30 dB, we can
the mean peak power of an impulse response and the mean conclude that the DMC in case of are
noise power. and are defined as the mainly related to model error. This allows for an evaluation of
ratio between the mean peak power of the SC or the DMC part the estimated DMC in terms of mere model error approximation
in the estimated impulse responses relative to the mean esti- vs. true propagation contribution without the need to compare
mated noise power . The is calculated from the to a Reference which is not available in a practical situation
band-limited, but still antenna dependent response for all re- (see, also, analysis of measurements presented in [9] and [53]).
solved specular paths . This means the antenna response
is not de-embedded in order to allow comparison with the dif- C. Impact of DMC on Estimated SC
fuse part that cannot be de-embedded because of the missing Until now we have analyzed only the relevance of the
angular model. is calculated from the estimated estimated DMC, but we have got no evidence about the reli-
570
Fig. 10. Rx azimuth power spectra as a function of Rx position , for (a) the
ray-tracer (Reference), (b) the estimated spectra considering all unavoidable
model errors (FullDist) estimating , and (c)
ability of the estimated SC. For this purpose the Rx azimuth
power spectrum of the ray-tracing results—used as a Reference
(Fig. 10(a)) will be compared to the estimated spectra of the
(Fig. 10(b)) and
(Fig. 10(c)) cases that are deemed closest to the condition of
a measurement. The amplitudes are clipped 50 dB below the
maximum as the ray-tracing calculations were a-priori limited
to 50 dB dynamic range between the strongest and the weakest
path. Comparing the estimated angular power spectra of the two
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
simulation cases with the Reference spectrum, the estimated
artifacts around the “true” paths give information about the
quality of the discussed estimation approaches. With narrower
spreads around the “true” paths the estimation result is said
to be “better”. Analyzing Fig. 10 and taking the results of the
previous discussion about the reliability of the estimated DMC
into account, the following conclusions can be made:
• The artificial spreads that can be observed in case of
are caused by the array model error, phase
noise, and additive Gaussian noise. Due to the model
error, the estimated SC contribution is not sufficient to
describe the received response. The remaining difference
is approximated by additional specular paths that are
physically nonexistent.
• Joint estimation of SC and DMC can consider-
able improve the estimation result of the SC. For
the artifacts due to model
error contribute to the estimated DMC and, hence, do not
cause additional paths pretending resolvable SC.
The estimation results of shown in
Fig. 10(b) are the best results that we can achieve in a real mea-
surement. Again, if the ratio between the of the esti-
mated SC and DMC is higher or equal to the overall model ac-
curacy (here between 25 dB and 30 dB), the estimated DMC of
must be considered to be not a feature
of the radio channel and vice versa if the ratio is lower. It turns
out that the estimation of the SC becomes more reliable if the
DMC model is included in the estimation framework.
VI. CONCLUSION
Precise HRPE estimation of the structural parameters of mul-
tipath wave propagation (which has to include joint estimation
of the directions of propagation at both Tx and Rx and full
solid angle polarimetric processing) is the prerequisite for an-
tenna de-embedding and, hence, for flexible antenna-indepen-
dent channel characterization which is the ambitious aim of
channel sounding.
In this paper we have presented the decisive influence the
data model imposes to the quality of HRPE in multidimensional
channel sounding. The data model comprises both wave propa-
gation aspects as well as the influence of the measurement de-
vice. It is affected by simplifications and limited calibration ac-
curacy. Oversimplified data models that do not use the full dual
polarimetric solid angle antenna response will result in serious
estimation artifacts in complex multipath environments. This
further results in artificial spread of paths in the angular domain
only. which may be strong enough to render any subsequent propa-
gation analysis useless. This error mechanism is solely caused
by the data model applied and, hence can not be compensated
by a “better” estimator, although, of course, the specific param-
eter estimator chosen defines which data model can be applied.
Model mismatch, in essence, cannot be completely avoided but
somehow controlled by sophisticated device data models and
careful calibration procedures. This is especially true for the
AA influence. We have described the EADF method as a very
powerful tool to handle full solid angle polarimeric antenna cal-
ibration data that also allows some reduction of the remaining
calibration errors.
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
In this paper the RIMAX framework is applied for parameter
estimation. This iterative estimation framework includes a hy-
brid data model that consists of a deterministic part describing
the geometric structure of the SC components and a so called
DMC part which is stochastic in nature. This part shall describe
as much as possible the diffuse part of multipath propagation.
We have shown that this is only true if we use the full solid angle
dual polarimetric AA data model. Remaining model mismatch,
however, still increases the amount of power that is estimated
for the DMC. So for a reliable interpretation of the estimation
results we must have knowledge about the remaining calibra-
tion error to make sure that the channel model distortion is small
enough.
The DMC model part has another advantage for parameter
estimation since it can be used to approximate/compensate
the remaining calibration model error. This may considerably
enhance the quality of the SC estimation since it helps to avoid
model overestimation by unreliable (and hence wrong) SC.
Moreover, if one has precise knowledge about the calibration
data model error, the DMC part that has to be attributed to the
data model error can be removed from the estimation result
which allows for additional enhancement of estimation quality.
It must be emphasized, however, that most of this advantage be-
comes true only if the SC and DMC parts are jointly estimated.
Finally, it should be noted that it was not the goal of this
paper to refine the hybrid DMC propagation model. There might
be various approaches to achieve that [27], [36]–[38]. Rather it
was the goal to show how inaccurate and incomplete device and
propagation data models can severely affect estimation accuracy
and reliability which may have significant influence to the inter-
pretation of estimated model parameters. So any refinement of
the propagation data model must make sure that the pitfalls an-
alyzed in this paper are carefully avoided.
ACKNOWLEDGMENT
The authors would like to thank G. Sommerkorn for coopera-
tion and assistance during measurement campaigns and antenna
calibration as well as Dr. W. Kotterman for valuable discus-
sions. The cooperation with the University of Karlsruhe (IHE,
Prof. Wiesbeck) which allowed access to their ray tracing sim-
ulator is also gratefully acknowledged. Special thanks go to the
MEDAV company for excellent support and cooperation.
REFERENCES
[1] R. S. Thomä, M. Landmann, A. Richter, and U. Trautwein, “Multi-
dimensional high-resolution channel sounding,” in Smart Antennas in
Europe State of the Art, EURASIP Book Series on SP &C, T. Kaiser,
Ed. Cairo, Egypt: Hindawi Publishing, 2005, vol. 3.
[2] M. Landmann and R. S. Thomä, “Resolution limits and antenna array
calibration of real arrays,” in Mobile Broadband Multimedia Networks:
Techniques, Models and Tools for 4G Techniques, COST 273, L. M.
Correia, Ed. Waltham, MA: Academic Press, 2006, pp. 165–167.
[3] C. Schneider, G. Sommerkorn, M. Narandzic, M. Käske, A. Hong,
V. Algeier, W. Kotterman, R. S. Thomä, and C. Jandura, “Multi-user
MIMO channel reference data for channel modelling and system eval-
uation from measurements,” presented at the Int. IEEE Workshop on
Smart Antennas (WSA 2009), Berlin, Germany, Feb. 2009.
[4] M. Viberg and H. Krim, “Two decades of array signal processing re-
search: The parametric approach,” IEEE Signal Processing Mag., vol.
13, pp. 67–94, Jul. 1996.
571
[5] M. Kwakkernaat, Y. Jong, R. Bultitude, and M. Herben, “High-resolu-
tion angle-of-Arrival measurements on physically-nonstationary mo-
bile radio channels,” IEEE Trans. Antennas Propag., vol. 56, no. 8,
2008.
[6] B. H. Fleury, M. Tschudin, R. Heddergott, D. Dahlhaus, and P. K. In-
geman, “Channel parameter estimation in mobile radio environments
using the SAGE algorithm,” IEEE J. Sel. Areas Communi., vol. 17, pp.
434–450, 1999.
[7] M. Tschudin, C. Brunner, T. Kurpjuhn, M. Haardt, and J. A. Nossek,
“Comparison between unitary ESPRIT and SAGE for 3-D channel
sounding,” in Proc. IEEE 49th Vehicular Technology Conf., May
16–20, 1999, vol. 2, pp. 1324–1329.
[8] M. Haardt, R. S. Thomä, and A. Richter, “Multidimensional high-res-
olution parameter estimation with applications to channel sounding,”
in High-Resolution and Robust Signal Processing, Y. Hua, Ed. et al.
New York: Marcel Dekker, 2003, pp. 253–337.
[9] M. Landmann, “Limitations of Experimental Channel Characterisa-
tion,” Ph.D. dissertation, Technische Universität Ilmenau, Ilmenau,
Germany, 2008.
[10] A. Richter, “On the Estimation of Radio Channel Parameters: Models
and Algorithms (RIMAX),” Ph.D. dissertation, Technische Universität
Ilmenau, Ilmenau, Germany, 2005.
[11] J. Medbo, M. Riback, H. Asplund, and J. Berg, “MIMO channel charac-
teristics in a small macrocell measured at 5.25 GHz and 200 MHz band-
width,” in Proc. IEEE 62nd Vehicular Technology Conf., Sep. 2005,
vol. 1, pp. 372–376.
[12] J. Salmi, A. Richter, and V. Koivunen, “Detection and tracking of
MIMO propagation path parameters using state-space approach,” IEEE
Trans. Signal Processing,, vol. 57, no. 4, pp. 1538–1550, Apr. 2009.
[13] M. Landmann, K. Sivasondhivat, J.-I. Takada, I. Ida, and R. S. Thomä,
“Polarization behavior of discrete multipath and diffuse scattering in
urban environments at 4.5 GHz,” EURASIP J. Wireless Commun. Net-
work., Special Issue on Space-Time Channel Model. Wireless Commun.
Network., 2007.
[14] M. Landmann, W. Kotterman, and R. S. Thomä, “On the influence of
incomplete data models on estimated angular distributions in channel
characterisation,” presented at the 2nd Eur. Conf. on Antennas Propag.
(EuCAP 2007), 2007.
[15] A. Pal, C. M. Tan, and M. A. Beach, “Comparison of MIMO chan-
nels from multipath parameter extraction and direct channel measure-
ments,” in Proc. 15th IEEE Int. Symp. on Personal, Indoor and Mobile
Radio Communications, Sep. 2004, vol. 3, pp. 1574–1578.
[16] S. Wyne, N. Czink, J. Karedal, P. Almers, F. Tufvesson, and A. F.
Molisch, “A cluster-based analysis of outdoor-to-Indoor office MIMO
measurements at 5.2 GHz,” in Proc. IEEE 64th Vehicular Technology
Conf., Montreal, QC, Canada, Sep. 2006, pp. 1–5.
[17] N. Czink, X. Yin, H. Ozcelik, M. Herdin, E. Bonek, and B. H. Fleury,
“Cluster characteristics in a MIMO indoor propagation environment,”
IEEE Trans. Wireless Commun., vol. 6, pp. 1465–1475, Apr. 2007.
[18] A. Ferreol, P. Larzabal, and M. Viberg, “Performance prediction of
maximum-likelihood direction-of-Arrival estimation in the presence of
modeling errors,” IEEE Trans. Signal Processing, vol. 56, no. 10, pp.
4785–4793, 2008.
[19] R. S. Thomä, D. Hampicke, M. Landmann, A. Richter, and G. Som-
merkorn, “Measurement-based channel modelling (MBPCM),” pre-
sented at the Proc. Int. Conf. on Electromagnetics in Advanced Ap-
plications (ICEAA03), Torino, Italy, Sep. 2003.
[20] A. Richter, D. Hampicke, G. Sommerkorn, and R. S. Thomä, “MIMO
measurement and joint M-D parameter estimation of mobile radio
channels,” in Proc. IEEE 63rd Vehicular Technology Conf., May 2001,
vol. 1, pp. 214–218.
[21] I. Ziskind and M. Wax, “Maximum likelihood localization of multiple
sources by alternating projection,” IEEE Trans. Acoust. Speech Signal
Process., vol. 36, no. 10, pp. 1553–1560, Oct. 1988.
[22] A. Richter, M. Landmann, and R. S. Thomä, “A gradient based method
for maximum likelihood channel parameter estimation from multidi-
mensional channel sounding measurement,” presented at the XXVIIth
URSI General Assembly, Maastricht, NL, Aug. 2002.
[23] V. Degli-Esposti, D. Guiducci, A. de’Marsi, P. Azzi, and F. Fuschini,
“An advanced field prediction model including diffuse scattering,”
IEEE Trans. Antennas Propag., vol. 52, no. 7, pp. 1717–1728, Jul.
2004.
[24] T. Fügen, J. Maurer, T. Kayser, and W. Wiesbeck, “Capability of 3D
ray tracing for defining parameter sets for the specification of future
mobile communications systems,” IEEE Trans. Antennas Propag.,
Special Issue Wireless Commun., vol. 54, no. 11, pp. 3125–3137, Nov.
2006.
572 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[25] V. Degli-Esposti, F. Fuschini, E. Vitucci, and G. Falciasecca, “Mea-
surement and modelling of scattering from buildings,” IEEE Trans.
Antennas Propag., vol. 52, no. 1, pp. 143–153, Jan. 2007.
[26] M. F. Iskander and Z. Yun, “Propagation prediction models for wireless
communication systems,” IEEE Trans. Microwave Theory Tech., vol.
50, no. 3, pp. 662–673, 2002.
[27] X. Yin, L. Liu, D. K. Nielsen, T. Pedersen, and B. H. Fleury, “A SAGE
algorithm for estimation of the direction power spectrum of individual
path components,” in Proc. IEEE Global Telecommunications Conf.,
2007, pp. 3024–3028.
[28] C. B. Ribeiro, A. Richter, and V. Koivunen, “Joint angular-and delay-
domain MIMO propagation parameter estimation using approximate
ML method,” IEEE Trans. Signal Process., vol. 55, pp. 4775–4790,
Oct. 2007.
[29] R. S. Thomä, M. Landmann, and A. Richter, “RIMAX-a maximum
likelihood framework channel parameter estimation in multidimen-
sional channel sounding,” in Proc. Int. Symp. on Antennas and
Propag., Sendai, Japan, Aug. 2004, pp. 53–56.
[30] R. Vaughan and J. B. Andersen, “Channel propagation and antennas for
mobile communications,” Inst. Elect. Eng. Electromagn. Waves Series
London, no. 50, 2003.
[31] V. Degli-Esposti, F. Fuschini, and E. Vitucci, “A fast model for dis-
tributed scattering from buildings,” presented at the 3rd Eur. Conf. on
Antennas Propag., Mar. 2009.
[32] A. Richter and R. S. Thomä, “Joint maximum likelihood estimation
of specular paths and distributed diffuse scattering,” in Proc. IEEE
61st Vehicular Technology Conf., Stockholm, Sweden, May 30–Jun.
1 2005.
[33] N. Czink, A. Richter, E. Bonek, J.-P. Nuutinen, and J. Ylitalo, “In-
cluding diffuse multipath parameters in MIMO channel models,” in
Proc. IEEE 66th Vehicular Technology Conf., Baltimore, Sep. 30–Oct.
3 2007.
[34] J. Poutanen, J. Salmi, K. Haneda, V. Kolmonen, F. Tufvesson, and P.
Vainikainen, “Propagation characteristics of dense multipath compo-
nents,” IEEE Antennas Wireless Propag. Lett., vol. 9, pp. 791–794,
2010.
[35] W. Kotterman, M. Landmann, G. Sommerkorn, and R. S. Thomä, “On
diffuse and non-resolved multipath components in directional channel
characterisation,” presented at the XXVIIIth General Assembly URSI,
New Delhi, India, Oct. 2005.
[36] J. Salmi, J. Poutanen, K. Haneda, A. Richter, V.-M. Kolmonen, P.
Vainikainen, and A. F. Molisch, “Incorporating diffuse scattering in
geometry-based stochastic MIMO channel models,” in Proc. 4th Eur.
Conf. on Antennas Propag. (EuCAP), 2010, pp. 1–5.
[37] F. Quitin, C. Oestges, F. Horlin, and P. De Doncker, “Diffuse multipath
component characterization for indoor MIMO channels,” in Proc. 4th
Eur. Conf. on Antennas Propag. (EuCAP), 2010, pp. 1–5.
[38] M. Käske, M. Landmann, and R. S. Thomä, “Modelling and synthesis
of dense multipath propagation components in the angular domain,”
presented at the 3rd Eur. Conf. on Antennas Propag., Mar. 2009.
[39] R. S. Thomä, D. Hampicke, A. Richter, G. Sommerkorn, A. Schneider,
U. Trautwein, and W. Wiesbeck, “Identification of time-variant direc-
tional mobile radio channels,” IEEE Trans. Instrum. Meas., vol. 49, no.
2, pp. 357–364, 2000.
[40] R. S. Thomä, D. Hampicke, A. Richter, G. Sommerkorn, and U.
Trautwein, “MIMO vector channel sounder measurement for smart
antenna system evaluation,” Eur. Trans. Telecommun. ETT, Special
Issue Smart Antennas, vol. 12, no. 5, pp. 427–438, Sep./Oct. 2001.
[41] P. Almers, S. Wyne, F. Tufvesson, and A. F. Molisch, “Effect of
random walk phase noise on MIMO measurements,” in Proc. IEEE
61st Vehicular Technology Conf., Jun. 2005, vol. 1, pp. 141–145.
[42] A. Taparugssanagorn, X. Yin, J. Ylitalo, and B. H. Fleury, “Phase noise
mitigation in channel parameter estimation for TDM switched MIMO
channel sounding,” presented at the Asilomar Conference on Signals,
Systems and Computers, 2007.
[43] M. Landmann and R. S. Thomä, “Estimation of phase drift during cali-
bration measurements for efficient beam pattern modelling,” presented
at the NEWCOM-ACORN Workshop, Wien, Austria, Sep. 2006.
[44] C. A. Balanis, Near Field/Far Field Methods, ser. Numerical Anal-
ysis. New York: Wiley, 1997, pp. 852–858.
[45] M. Landmann, A. Richter, and R. S. Thomä, “DoA resolution limits
in MIMO channel sounding,” presented at the Int. Symp. on Antennas
Propag. and USNC/URSI National Radio Science Meeting, Monterey,
CA, Jun. 2004.
[46] M. A. Doron and E. Doron, “Wavefield modeling and array pro-
cessing—I: Spatial sampling,” IEEE Trans. Signal Process., vol. 42,
no. 10, pp. 2549–2559, 1994.
[47] J. E. Hansen, Spherical Near-Field Antenna Measurements, ser.
Numerical Analysis. London, U.K.: Peter Peregrinus, 1988, pp.
387–387.
[48] G. Del Galdo, J. Lotze, M. Landmann, and M. Haardt, “Modelling
and manipulation of polarimetric antenna beam patterns via spherical
harmonics,” presented at the 14th Eur. Signal Processing Conf. (EU-
SIPCO), Florence, Italy, Sep. 2006.
[49] Pyramiden-Absorber DATENBLATT 390-1, Noppen-Absorber
DATENBLATT 390-7 Emc-Technik & Consulting, 2007.
[50] Electromagnetic Modeling of Composite Metalic and Dielectric Struc-
tures [Online]. Available: http://www.wipl-d.com/.
[51] A. Taparugssanagorn, M. Alatossava, V. M. Holappa, and J. Ylitalo,
“Impact of channel sounder phase noise on directional channel estima-
tion by space-alternating generalised expectation maximisation,” IET
Microw. Antennas Propag., vol. 1, no. 3, pp. 803–808, 2007.
[52] A. Taparugssanagorn and J. Ylitalo, “Characteristics of short-term
phase noise of MIMO channel sounding and its effect on capacity
estimation,” IEEE Trans. Intrum. Meas., vol. 58, no. 1, pp. 196–201,
Jan. 2009.
[53] M. Käske , C. Schneider, G. Sommerkorn, and R. S. Thomä, PART
II: Reference campaign—quality check for channel sounding measure-
ments COST 2100 Temporary Document TD(09)777, 2009.
Markus Landmann was born in Zeitz, Germany,
in 1977. He received the Dipl.-Ing. (M.S.E.E.) and
Dr.-Ing. (Ph.D.E.E.) degrees in electrical engi-
neering and information technology from Ilmenau
University of Technology, Germany, in 2001 and
2008, respectively.
From 2001 to 2003, he was working as a Research
Assistant and Instructor at Ilmenau University of
Technology. In 2004, he was developing advanced
antenna array calibration methods and high resolu-
tion parameter estimation algorithm (RIMAX) for
propagation studies at MEDAV Company. In 2005, he was Visiting Researcher
and Instructor at Tokyo Institute of Technology (Takada Laboratory) in the field
of channel measurement and estimation techniques. From 2006 to 2008, he
was finalizing his doctoral thesis (Dr.-Ing) as a Research Associate at Ilmenau
University of Technology. He is currently with the Project Group Wireless
Distribution Systems/Digital Broadcasting, Fraunhofer Institute for Integrated
Circuits IIS Helmholtzplatz, Ilmenau, Germany. His current interests are
wireless propagation, channel modeling, and array signal processing.
Martin Käske received the Dipl.-Ing. (M.S.E.E.)
degree in electrical engineering (information tech-
nology) from Ilmenau University of Technology,
Germany, in 2008.
He is currently working as a Research Assistant at
the Electronic Research Measurement Lab, Ilmenau
University of Technology. He is focusing on diffuse
scattering—its modeling and estimation—as well as
advanced antenna array calibration methods.
Reiner S. Thomä (F’07) received the Dipl.-Ing.
(M.S.E.E.), Dr.-Ing. (Ph.D.E.E.), and the Dr.-Ing.
habil. degrees in electrical engineering and infor-
mation technology from Technische Hochschule
Ilmenau, Germany, in 1975, 1983, and 1989,
respectively.
From 1975 to 1988, he was a Research Associate
in the fields of electronic circuits, measurement
engineering, and digital signal processing at the
same university. From 1988 to 1990, he was a
Research Engineer at the Akademie der Wis-
senschaften der DDR (Zentrum für Wissenschaftlichen Gerätebau). During
this period, he was working in the field of radio surveillance. In 1991, he
spent a three-month sabbatical leave at the University of Erlangen-Nürnberg
(Lehrstuhl für Nachrichtentechnik). Since 1992, he has been a Professor of
electrical engineering (electronic measurement) at TU Ilmenau where he was
LANDMANN et al.: IMPACT OF INCOMPLETE AND INACCURATE DATA MODELS ON HRPE
the Director of the Institute of Communications and Measurement Engineering
from 1999 until 2005. Currently, he is Vice Speaker at the Intl. Graduate
School on Mobile Communication, TU Ilmenau and Speaker of the nationwide
DFG-focus project UKoLOS, ultrawideband radio technologies for communi-
cations, localization and sensor applications (SPP 1202). His research interests
include measurement and digital signal processing methods (correlation and
spectral analysis, system identification, array methods, time-frequency and
cyclostationary signal analysis), their application in mobile radio and radar
573
systems (multidimensional channel sounding, propagation measurement and
parameter estimation, ultrawideband radar), measurement-based performance
evaluation of MIMO transmission systems, and UWB sensor systems for object
detection and imaging.
Prof. Thomä has been serving as Chair of the IEEE-IM TC-13 on Measure-
ment in Wireless and Telecommunications since 1999. In 2007, he was awarded
the grade of IEEE Fellow for his “contributions to high-resolution multidimen-
sional channel sounding.”
574 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

A General Coupling-Based Model Framework for

Wideband MIMO Channels
Yan Zhang, Member, IEEE, Ove Edfors, Member, IEEE, Peter Hammarberg, Member, IEEE,
Tommy Hult, Member, IEEE, Xiang Chen, Member, IEEE, Shidong Zhou, Member, IEEE,
Limin Xiao, Member, IEEE, and Jing Wang, Member, IEEE

Abstract—A general coupling-based model framework for

wideband multiple-input multiple-output (MIMO) channels is
presented in this paper. Under this framework, the channel state
information (CSI) tensor can be expressed by the product of a
coupling tensor, a complex Gaussian tensor and three unitary ma-
trices. The unitary matrices can be either eigenbases or steering
matrices in different domains. The coupling tensor reflects the
relationship between the column vectors of these unitary matrices.
The complex Gaussian tensor is used to describe the small-scale
fading. Several realizations of this framework are introduced,
including the wideband Kronecker-based (WKB) model, the Fig. 1. The differences between (a) physical model and (b) analytical model.
wideband eigenvalue-decomposition-based (WEB) model, the The physical models are independent of the antenna arrays while the analytical
wideband virtual presentation (WVP) model and the wideband MIMO models consider the impact of the antenna arrays.
hybrid (WHY) model. To evaluate the performance of these
models, channel measurements were carried out in different in-
door scenarios both at Tsinghua University and Lund University.
The results show that these models have good agreement with the
In recent years, many MIMO channel models have been
measured data. Furthermore, we can see that the WHY model can developed [2], [3]. They can be classified into physical models
provide a tradeoff between complexity and accuracy in channel and analytical models [2], [4]. Physical models describe the
synthesis. electromagnetic wave propagation environment between the
Index Terms—Channel capacity, channel measurement, cou- transmitter (Tx) and receiver (Rx) locations. As illustrated
pling-based, multiple-input multiple-output (MIMO), wideband in Fig. 1(a), physical models are independent of the antenna
channel model. array configuration. Typical physical models include the geom-
etry-based stochastic models (GSCM) [5], the Zwick model [6]
and the Saleh–Valenzuela model [7]. Analytical models, on the
I. INTRODUCTION other hand, describe the channel impulse response (CIR) matrix
and are often used for channel synthesis and algorithm devel-
F OR next-generation wireless communication systems,
several new key technologies have been proposed to
support high transmit rates. Multiple-input multiple-output
opments. As Fig. 1(b) shows, the analytical MIMO models
consider the impact of the antenna arrays in both ends as parts
of the channel response.
(MIMO) systems, which can lead to large capacity gains [1],
Amongst the analytical models, the independent identically
have drawn considerable interest. Accurate MIMO channel
distributed (i.i.d.) channel model [1] is the simplest one. It is
models, which can mimic real wireless channels, is very im-
widely used for the calculation of ergodic MIMO capacity. The
portant for the MIMO system design and evaluation.
full-correlation model [4], which characterizes the channel by
the complex correlations between all channel matrix element
Manuscript received June 01, 2010; revised May 28, 2011; accepted
September 02, 2011. Date of publication October 26, 2011; date of current pairs, requires a vast number of parameters to be specified.
version February 03, 2012. This work was partially supported by National In order to simplify the full-correlation model, a narrowband
Basic Research Program of China (2012CB316002), Tsinghua University MIMO model based on the Kronecker product is presented
Initiative Scientific Research Program (20111081025), China’s Major Project
(2009ZX03007-003-02), China’s 863 Project (2009AA011501), PCSIRT,
in [8]. Weichselberger proposes an eigenvalue decomposition
Tsinghua-Qualcomm Joint Research Program, and China Postdoctoral Science (EVD)-based model in [9] and [10], and Sayeed develops a
Foundation. Support has also been received from the Swedish government virtual presentation model in [11]. The finite scatterer model
agency VINNOVA through the Sino-Swedish Strategic Cooperative Pro- [12] constitutes the coupling relationship through linking dif-
gramme on Next Generation Networks (2008-00962).
Y. Zhang is with the Department of Electronic Engineering, Tsinghua Uni- ferent scatterers in the propagation environment. The keyhole
versity, Beijing, 100084 China (e-mail: yan-z05@mails.tsinghua.edu.cn). and max-entropy models are established in [13] and [14],
O. Edfors, P. Hammarberg, and T. Hult are with the Department of Electrical respectively. In [15], a general unitary-independent-unitary
and Information Technology, LTH, Lund University, 221 00 Lund, Sweden.
X. Chen, S. Zhou, L. Xiao, and J. Wang are with the Department of Elec- (UIU) model is proposed to encompass most of the zero-mean
tronic Engineering, Research Institute of Information Technology, Tsinghua Na- channel models.
tional Laboratory for Information Science and Technology, Tsinghua Univer- Future wireless communication systems need wider fre-
sity, 100084 Beijing, China.
Color versions of one or more of the figures in this paper are available online
quency bands to achieve higher throughput. However, the
at http://ieeexplore.ieee.org. models mentioned above are all designed for the narrowband
Digital Object Identifier 10.1109/TAP.2011.2173436 channels. Therefore, it is necessary to investigate the wideband

0018-926X/$26.00 © 2011 IEEE

ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS 575

MIMO channel modeling. Until now, the basic modeling idea II. NARROWBAND ANALYTICAL MIMO MODEL
is to apply the narrowband models for each delay tap. For ex- In this section, we will review several popular analytical
ample, Yu extends the Kronecker model to the wideband case models of the narrowband MIMO channel. These models are
by calculating the channel correlation matrix tap by tap [16]. In the basis of our work and we will later show how these can be
[17], Costa establishes a novel structured model based on the extended to the wideband cases using the proposed wideband
EVDs of the delay, transmit, and receive correlation matrices. coupling-based framework.
This model describes the correlations between different taps
Now, Let us consider a narrowband MIMO system that em-
and is capable of generating the wideband CIR tensor.
ploys transmit antennas and receive antennas. The rela-
Our goal with this paper is to provide additional wideband
tionship between the transmitted and received signals can then
models to be used for channel synthesis and algorithm develop-
be expressed as
ments. A general coupling-based model framework is proposed,
and using this framework, we present several models for wide- (1)
band MIMO channels.
The specific contributions of this paper are as follows. where is the received signal, is the transmitted
• The wideband MIMO channel is described by a three-di- signal, and is the noise. The CSI matrix
mensional (3-D) channel state information (CSI) tensor
in the transmitter, receiver, and frequency domains. We .. .. ..
divide the wideband MIMO CSI tensor into a determin- . . . (2)
istic part and a stochastic part. The deterministic part cor-
responds to the line-of-sight (LOS) component. The sto-
is an matrix whose entry is the scalar valued fre-
chastic part, which is used to describe the diffuse com-
quency response between the th transmit antenna and the th
ponents, is decomposed into the product of a coupling
receive antenna.
tensor, a complex Gaussian tensor and three unitary ma-
For a Rician fading case, the channel matrix
trices. This decomposition is defined as the coupling-based
model framework.
• Under this framework, five wideband analytical models are (3)
presented, including the wideband i.i.d. model, the wide-
band Kronecker-based (WKB) model, the wideband eigen- can be divided into the deterministic part and the stochastic
value-decomposition-based (WEB) model, the wideband part [4]. Here, , a phase-shift-only MIMO matrix, reflects
virtual presentation (WVP) model, and the wideband hy- the contribution of the LOS component [4]. is a zero-mean
brid (WHY) model. The physical mechanisms and trans- stochastic part and is corresponding to the effect of the diffuse
form meanings of these models are also explained. components. is the Rician factor. which represents the ratio of
• Using the coupling-based framework, the wideband i.i.d., the LOS component power to the total power in the diffuse non
WKB, WEB, and WVP models can be viewed as the ex- line-of-sight (NLOS) components. In the following sections, we
tensions of the corresponding narrowband models. A novel only consider the stochastic part , i.e., . Then,
WHY model is proposed to provide a tradeoff between the and corresponds to a Rayleigh fading channel.
modeling complexity and accuracy.
• To evaluate the performance of these wideband cou- A. The i.i.d. Model
pling-based models, we carried out measurements both The i.i.d. model is the simplest model and is commonly
at Tsinghua University (THU) and at Lund University. used for information theoretical analysis. The matrix entries
The results show that the models derived from the cou- in this model obey zero-mean circularly symmetric complex
pling-based framework are in good agreement with the Gaussian (ZMCSCG) distribution. It assumes a rich scattering
measured channels. environment where numerous independent multipath compo-
The rest of this paper is organized as follows. In Section II, nents (MPCs) depart and arrive in all directions.
several popular narrowband analytical models are reviewed. The
coupling-based model framework and its several realizations are B. The Kronecker Model
given in Section III. In Section IV, we describe the measurements In the Kronecker model [8], the channel matrix is given by
carried out at THU and Lund University. In Section V, the per-
formance evaluation metrics are introduced. Measured data and
simulation results are shown in Section VI to evaluate the perfor- (4)
mance of the proposed wideband coupling-based models. Finally,
our conclusions are presented in Section VII. where
The following notation conventions will be adhered to:
Lower case letters are scalars and bold lower case (5)
letters are vectors. Bold upper case letters
are matrices, while upper case script letters are ten- are the transmit and receive correlation matrices. is an
sors in . is the trace of a matrix, is the expectation random matrix with i.i.d. ZMCSCG entries.
operator, stands for the Schur–Hadamard product and The Kronecker model relies on the assumption that direction
stands for the Kronecker product. of departures (DoDs) and direction of arrivals (DoAs) in the
576 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

MIMO channel can be separated. This means that each DoD be used to generate MIMO CSI tensors. Notation and relevant
is linked to all DoAs with the same pattern and that the joint tensor calculus in this paper follow those defined in [19] and
DoA-DoD spectrum can be expressed as the product of the DoA [20].
spectrum and the DoD spectrum. Obviously, the separability
condition is not met in general [4]. A. The Wideband MIMO CSI Tensor
Let us consider a wideband MIMO system with transmit
C. The Weichselberger Model antennas, receive antennas and a bandwidth of . The CSI
In the Weichselberger model [9], the EVDs are applied to matrix at frequency bin can be written as
the transmit and receive correlation matrices. Then, the channel
matrix can be synthesized as .. ..
.. (11)
. . .
(6)

Here, the unitary matrices and are eigenbases at the
receiver and transmitter domains and given by
(7)
In (7), the amplitude coupling matrix is used to de-
scribe the links between transmit eigenmodes and receive eigen-
modes. The Weichselberger model has been shown to agree
closely with measured data [4], [10]. Its disadvantages, as com-
pared to the Kronecker model, are the increased number of pa-
rameters that need to be specified and the increased complexity
in obtaining those parameters from the measured data [17].
Here, the entry is the complex frequency response be-
tween transmit antenna , receive antenna , and frequency
bin . The frequency bins are numbered and
divide the total bandwidth into equidistant intervals (sub-
carriers) of width .
Compared with the traditional tap-delay model, the wideband
MIMO channel response is modeled in frequency domain in-
stead of delay domain. We do not build the CIR matrix tap by
tap, but synthesize the CSI tensor in the frequency domain.
By changing , we can get a 3-D CSI tensor ,
which can be divided into two parts:

D. The Virtual Presentation Model (12)

The virtual presentation model [10] is designed for uniform
where still means the Rician factor, representing the ratio
linear arrays (ULAs). In this model, the channel matrix
of the LOS component power to the total diffuse components’
(8) power. The deterministic tensor denotes the
LOS component and the stochastic part represents the dif-
is modeled by describing the amplitude coupling between the fuse components. Again, we let , i.e., . Only the
virtual angles at transmitter and receiver. is the amplitude stochastic part will be considered in the following analysis.
coupling matrix and Based on the Raleigh-fading assumption, the entry of the
MIMO tensor satisfies the joint multivariate ZMCSCG distri-
bution. The wideband MIMO tensor can be fully characterized
by an full-correlation matrix

(9) (13)

Here, is the vector operator [17], [19]. character-

(10)
are the steering matrices into the virtual angles at transmitter
and receiver. Here is the normalized antenna
spacing and is the wavelength. is the th virtual DoA
and is the th virtual DoD, respectively.
III. A GENERAL COUPLING-BASED WIDEBAND
MIMO MODEL FRAMEWORK
The analytical models in Section II are all designed for the
narrowband MIMO channel. With an increase of transmission
bandwidths in future communication systems, it is important to
develop accurate models for the wideband MIMO channel. In
this section, a general coupling-based framework is proposed
for the wideband MIMO channel. Under this framework, sev-
eral wideband channel models are proposed. These models can
izes the correlations between all transmit antennas, receive an-
tennas and frequency bins. The MIMO CSI tensor can then be
synthesized by
(14)
where is an i.i.d. ZMCSCG random tensor.
This model can be defined as the full-correlation model of
the wideband MIMO channel [17]. Similar to the narrowband
case, the full-correlation model can entirely capture the channel
tensor’s spatial and delay property, so it can be used as a bench
mark to evaluate other models.
The drawback of this full-correlation model is its high com-
plexity. In order to synthesize , we need to specify with
complex parameters. In general the full-correlation
matrix size is too large for channel synthesis. In the following
sections, several simpler models will be presented.
ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS 577

B. Coupling-Based Model Framework any pair of transmit antennas should be independent of the re-
A coupling-based framework is proposed for the wideband ceive antenna index and the frequency bin index. Analogously,
MIMO channel. The wideband CSI tensor can be described by the frequency correlation between each pair of frequency bins
the following coupling-based structure should also be independent of the receive antenna index and
the transmit antenna index. The interpretation of the WKB
(15) model is as follows: The joint DoA-DoD-delay spectrum is the
Kronecker product of the delay spectrum, the DoD spectrum,
where is an amplitude coupling tensor and is and the DoA spectrum. All DoAs, all DoDs, and all delays are
used to describe the gain between the column vectors of , linked together in the same pattern. All delays show the same
and . , a complex Gaussian tensor, AoA/AoD spectrum. Similar to the narrowband Kronecker
reflects the small-scale fading in the propagation environment. model, these assumptions are in general too strict for realistic
are unitary matrices in the dif- propagation channel.
ferent domains. means multiplication in the th tensor di- This WKB model needs the knowledge of , and
mension [19]. Under this framework, several wideband models, to rebuild the CSI tensor. Hereby, there are
i.e., the wideband i.i.d., WKB, WEB, WVP, and WHY models, coefficients to be specified.
can be established based on different propagation assumptions. It should be mentioned that this WKB model is different from
1) Wideband i.i.d. Model: If , , the model in [16], which is also based on the Kronecker struc-
and , equals to . The ture. The WKB model does not build the channel response tap
coupling-based framework characterizes a simple wideband by tap as in [16], but considers the correlation in frequency do-
i.i.d. model, which can be used for analyzing the closed-form main together with the correlation in the spatial domain.
solution of wideband channel capacity. This wideband i.i.d. 3) Wideband EVD-Based Model: In this WEB model, the
channel model satisfies the rich-scattering assumption that a coupling relationship between the receive, transmit, and fre-
large number of independent multipath components superim- quency eigenmodes are considered.
pose at the receiver. The EVDs of receiver, transmitter, and frequency correlation
2) Wideband Kronecker-Based Model: If , matrices are given by
, and the coupling tensor (23)
(16) (24)
(25)
the coupling-based framework will describe a wideband Kro-
necker-based model: where , , and are the eigenbases in the receiver,
transmitter, and frequency domain, respectively. , , and
(17) are the diagonal eigenvalue matrices.
If , , and , the cou-
The receiver, transmitter and frequency correlation matrices are pling-based framework will describe the coupling relationship
defined as between the eigenvectors. The CSI tensor can be synthesized by
the following wideband EVD-based (WEB) model
(18)
(26)
(19)
Here, is an amplitude coupling tensor,
(20)
where the entries correspond to the coupling amplitude
between the th receiver eigenmode , the th transmitter
where means the th unfolding of [19]. The traces of
eigenmode , and the th frequency eigenmode . The
, and satisfy
coupling power can be given by
(21)

In this WKB model, the wideband full-correlation channel (27)

matrix
(22)
is constructed by the Kronecker product of the three correlation
matrices.
The WKB model implies the following assumptions. The re-
ceive antenna correlation between any pair of receive antennas
has to be independent of the transmit antenna index and the
frequency bin index. The transmit antenna correlation between
The WEB model can be viewed as an extension of the
narrowband Weichselberger model, where also the eigen-
modes in the frequency domain are taken account in the CSI
tensor modeling. Furthermore, , and are all
Karhunen–Loeve transform (KLT) matrices [15]. Thereby
the WEB CSI tensor and the tensor is
actually a 3-D KLT pair. In other words, the coupling tensor
can be transformed to the CSI tensor through three KLTs in the
receive, transmit and frequency domains.
578 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Furthermore, we can see that the WKB model is a special case
of this WEB model. If the amplitude coupling tensor
(28)
the WEB model will degrade to the WKB model.
In the WEB model, , , , and need to be
specified to generate , so there are total
unknown parameters.
4) Wideband Virtual Presentation Model: The coupling re-
lationships between the steering vectors in the receiver, trans-
mitter and frequency domains are considered in this model.
If steering matrices are used in (15), i.e., , lays, then the first path’s power will be “dispersed” to different
, and , the coupling-based frame-
work can extend the narrowband virtual presentation model to
the wideband case. The wideband MIMO CSI tensor can be
expressed as the coupling relationship between steering vectors
(29)
where , , and are the steering vectors in the re-
ceiver, transmitter and frequency domains, respectively.
and are defined in (12) and (13), and
(30)
is the steering matrix in the frequency domain. Similar to the
virtual angle definition in the narrowband virtual presentation
model, can be defined as the virtual delays.
divides the max delay range
uniformly.
In (29), is the amplitude coupling
tensor. Its entries represent the coupling amplitudes
between the th receive steering vector
the th transmit steering vector , and the
th frequency steering vector .
This model can be viewed as the extension of the narrow-
band virtual presentation model in [11]. Thereby, we call it the
wideband virtual presentation model. The physical meaning of
this WVP model can be explained as follows. The wideband
MIMO channel is modeled by using the predefined orthogonal
DoAs, DoDs and delays (virtual DoAs, DoDs, and delays). If
there exists a path at a unique pair of the virtual DoA, DoD, and
delay, the corresponding entry of is equal to the fading
gain of this path. For example, if there are two scatterers in the
propagation environment and only a single spectral reflection
occurs, as Fig. 2 shows, each scatterer causes an individual
path. The DoAs, DoDs, and delays of these two paths are
, and . The power coupling
coefficients and are equal to the power of the first
and the second path, respectively.
The WVP model can be viewed as “virtual” tap-delay models.
However the exact delay values are not required for the WVP
model. Similar to the virtual angles, these virtual delays are se-
lected to divide the delay domain uniformly due to the DFT
Fig. 2. Physical mechanism of virtual presentation model. The entry of
equals to the amplitude of the virtual path.
property. For example, in Fig. 2, if is not one of the virtual de-
virtual tap delays.
The spatial resolution of the WVP model depends on the an-
tenna array size. For example, receive (transmit) antennas
can only provide virtual DoAs (DoDs). The number of fre-
quency bins specifies the number of the virtual delays, i.e., the
delay resolution is determined by the bandwidth.
For this model, , , and are the discrete Fourier
transform (DFT) matrices. The WVP channel tensor and
the coupling tensor are therefore a 3-D DFT trans-
form pair, i.e., the coupling relationships between the virtual
DoAs, DoDs, and delays can be transferred to the frequency
response in the transmitter-receiver-frequency domain by these
DFT matrices.
The WVP model is only available for ULAs. In this model,
only the amplitude coupling matrix with parame-
ters needs to be specified.
5) Wideband Hybrid Model: In practical MIMO systems, the
number of antennas is often limited by cost, power restriction,
and terminal size. Thus, the spatial resolutions are limited at
both ends, especially at a mobile station. At the same time, the
system bandwidth (corresponding to the number of frequency
bins) is usually large enough to provide high delay resolution.
In this case, the complexity of the WEB model will be quite high
,
because of the increase in the size of the frequency correlation
matrix. The WVP model needs only a few parameters and no
EVD operation, while its spatial performance degrades quickly
if the array size is limited. Furthermore, the WVP model can
only be applied in the systems with ULAs.
Considering these limitations, we propose a novel wideband
hybrid coupling-based model, which combines the advantages
of the WEB and WVP models. If ,
and , the CSI tensor
(31)
can be modeled by the coupling relationships between the
transmit eigenmodes, the receive eigenmodes, and the fre-
quency steering vectors.
In this WHY model, the spatial characteristics are still
described in the eigendomains while the delay characteristics
are presented by the steering vectors. Then, the WHY model
can keep the performance in the spatial domain without losing
too much delay resolution. Furthermore, the computation com-
plexity in the frequency domain, which plays a major role in
the synthesis and simulation process, will be reduced.
ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS
TABLE I
THE DIFFERENCES BETWEEN PROPOSED COUPLING-BASED MODELS
In this WHY model, the coupling tensor, and the eigenbases
in the transmitter and receiver domain need to be specified. The
number of unknown parameters is therefore .
C. Discussion
The wideband i.i.d., WKB, WEB, WVP, and WHY models
are all realizations of the proposed coupling-based model frame-
work. As mentioned above, the wideband i.i.d. model and the
WKB model are based on the rich-scattering assumption and
separability assumption, respectively. In general, these assump-
tions are too strict for realistic channels. Therefore, only the
WEB, WVP and WHY model will be discussed in the following
sections. The differences between these three models are the
selection of the unitary matrices, as shown in Table I. In (15),
can be either selected as an eigen-
base or a steering matrix . The selection can be de-
termined by the antenna array configurations and the system
bandwidth. Using an eigenbase can provide higher resolution
but more complexity. Employing a steering matrix will lead to
less complexity. However, the performance of the model will
decrease if the array size, or the system bandwidth, is limited.
The complexity and accuracy requirement for the system simu-
lation should be considered when selecting the unitary matrices.
This coupling-based wideband model framework can be in-
terpreted in two ways. On one hand, the amplitude coupling
tensor reflects the coupling relationship between the receiver,
transmitter and frequency domains. Its entries are corresponding
to the coupling amplitudes between the columns of the unitary
matrices. On the other hand, the wideband MIMO CSI tensor
in the transmitter–receiver–frequency domain and the coupling
tensor in the DoA-DoD-delay domain are a 3-D transform pair.
Here, the transform can be either KLT or DFT.
Under this coupling-based framework, other narrowband
models could also be extended to the wideband case. Neverthe-
less, there are still several models that can not be derived from
this coupling-based framework, e.g., the wideband keyhole
channel model.
IV. MEASUREMENT DESCRIPTION
To evaluate the performance of the proposed coupling-based
models, we carried out measurement campaigns both at Ts-
inghua and Lund University.
A. Measurements at THU
At THU, the measurements were carried out by the THU
MIMO channel sounder [21]. Some important parameters were
shown in Table II. At the transmitter of the THU channel
sounder, a signal generator periodically outputed the test signal
at the center frequency of 3.52 GHz with 16 MHz bandwidth.
The transmitted RF signal was fed to each transmit antenna by
579
TABLE II
THE CONFIGURATION OF THU MIMO CHANNEL SOUNDER
a fast time-division multiplexed switching (TDMS) scheme.
Fast switching was also employed at the receiver. The test
signal length was 12.8 s. A guard interval of length
was inserted between adjacent transmissions. The snapshot
interval was 1254.4 s. The synchronization between the trans-
mitter and the receiver was achieved by using well-adjusted
Rubidium-based frequency references.
ULAs with half wavelength inter-element spacing were em-
ployed at both Tx and Rx. The maximum antenna configuration
was 7 7, while only the responses of 5 5 antenna pairs were
used in the following analysis. The antennas on the array edges
were viewed as dummy elements and their responses were dis-
carded. The Tx and Rx antenna arrays were placed at heights of
0.9 m and 1.3 m above the floor.
Campaigns were carried out in the Future Internet Tech-
nology (FIT) building, Tsinghua University, Beijing, China.
Two typical indoor scenarios were chosen, including a office
scenario and a corridor-to-office scenario. Because the receiver
was too large to move during the measurements, the position
of the receiver was fixed and the transmitter was moved to
different positions.
The office scenario was selected to be in a small lab. The area
of this office was 8.8 m 16.3 m. The receiver was fixed on a
desk in the center of the office, and the transmitter was mounted
on a trolley and moved to different positions. The main scat-
terers in this office were concrete walls, wooden tables, plastic
chairs and metal cabinets. There was a concrete pillar in the
center of the office, which could obstruct the LOS propagation
between the transmitter and receiver.
In the corridor-to-office scenario, the position of the receiver
position was the same as that in the office scenario. The trans-
mitter was moved to different positions in the neighboring cor-
ridor. Fig. 3(a) and (b) shows photographs of this scenario. The
corridor was 105 m long and 2.5 m wide. Fig. 4 illustrates the
site map of the measurement campaign. A total of 29 positions
were measured. The distance between adjacent positions was
3.6 m. There was no LOS propagation between Tx and Rx.
During the measurements, the doors of rooms 4–509 and 4–510
were kept open.
B. Lund Measurement Campaign
The commercial RUSK Lund channel sounder1 [22], was
used to collect the channel data at Lund University. Mea-
surements were performed at 2.6 GHz with a bandwidth of
200 MHz, using a transmit power of 27 dBm. The TDMS
1[Online] Available: http://www.channelsounder.de/
580
Fig. 3. Photos of the corridor-to-office scenario including (a) the transmitter
and (b) the receiver.
Fig. 4. Site map of the corridor-to-office scenario at THU. The receiver (red
arrow) was fixed in the office and the transmitter was placed in the different
positions in the corridor.
TABLE III
THE CONFIGURATION OF RUSK LUND CHANNEL SOUNDER
schemes were also employed to sequentially connect antenna
elements to a single RF chain at both the Tx and Rx side.
The length of the Tx signal was 1.6 s. The interval between
two consecutive snapshots was 3.4 ms. The main parameters of
the RUSK Lund channel sounder are shown in Table III.
The transmit antenna was a 2 8 rectangular array with
half-wavelength spacing. The receive antenna was a 16-element
uniform circular array (UCA) with an element spacing of . A. Joint DoA-DoD-Delay Power Spectrum
All elements are vertical polarized patch antennas. The heights
of Tx and Rx antenna arrays were 2.08 and 1.43 m, respectively.
The measurement site was the E-building, LTH, Lund
University. Three scenarios including corridor-to-office, cor-
ridor-to-corridor and lobby-to-corridor were measured. In
our analysis, only the positions without LOS propagation are
considered.
As shown in Fig. 5, for the corridor-to-office scenario the
transmitter was fixed at the end of the corridor and the receiver
was moved to 22 different positions in 20 offices. The directions
of arrows are the reference directions of the antenna arrays. The
reference direction of the Tx antenna array is the broadside of
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 5. Site map of the corridor-to-office scenario at Lund University. The
transmitter (the black arrow) was placed at the end of the corridor. The receiver
(the red circles) was placed in 22 positions in 20 different small offices. The
directions of arrows are the reference directions of the antenna arrays.
the rectangular array, while that of the Rx antenna array is the
direction between the first and sixteenth patches.
In the corridor-to-corridor scenario, the transmitter was still
fixed at the end of the corridor on the second floor. The receiver
was placed in 13 positions in the different corridors without
LOS propagations on the same floor.
In the lobby-to-corridor scenario, the transmitter was placed
in one side of the lobby on the first floor. The receiver was
moved to 15 positions in the different corridors on the same
floor.
V. PERFORMANCE EVALUATION METRICS
When we evaluate the performance of a wideband channel
model, both the modeling accuracy and complexity should be
taken into account. On one hand, the channel model should be
simple enough to synthesize. On the other hand, a good channel
model should characterize the propagation environment accu-
rately. Both the spatial property and the delay property should
be reproduced by the proposed channel model.
Two metrics, the joint DoA-DoD-delay power spectrum and
the channel capacity, will be employed to evaluate the accura-
cies of these analytical models [4].
The joint DoA-DoD-delay power spectrum can be expressed
using the Bartlett beamformer [10]:
(32)
with normalized steering vectors at delay ,
at DoD and at DoA . is the estimation
of the full-correlation matrix.
ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS
The computed DoA-DoD-delay power spectrum is an
tensor. It can reflect the spatial and delay characteristics
of the wideband MIMO channel. In the following analysis, the
joint DoA-DoD angular power spectrums (APS) of the strongest
tap is used to characterize the spatial structure. Furthermore, the
DoA-delay spectrum with the strongest departure direction will
be illustrated to describe the delay and receive angular charac-
teristics.
Because of Bartlett beamformer will lead a leakage between
tap delays, the power spectrum comparison may be unfairly in
favor of the WVP model. But considering the validity behind
using a Fourier base for the transform, the Bartlett beamformer
is sufficient for a quantitative comparison.
B. Capacity
Let us consider a channel unknown at the transmitter. The
capacity of a wideband MIMO channel with equally allocated
transmit powers can be calculated as [17]
(33)
where is an identify matrix, SNR is the average re-
ceive signal power to noise power ratio, and is the nor-
malized CSI matrix at frequency bin .
VI. RESULTS
The performance of the WEB, WVP, and WHY models will
be evaluated in the following section. Because the assumption of
the WKB model is obviously too strict for the realistic channel,
The WKB will fail completely when comparing the measure-
ment results. In this section the results of the WKB model are
not illustrated. For comparison, the traditional channel model
in [16], which builds the Kronecker structure tap by tap, is also
simulated for comparison. This model will be referred to as the
“Kronecker model” in the rest of the paper.
The results from the presented models will be shown in this
section. For the WEB, WVP, WHY, and Kronecker models,
CSI tensors are generated by Monte Carlo simulations. Firstly,
the model parameters are extracted from the measured data,
and using these the CSI tensors are generated. The generated
CSI tensors are then compared with the original measurement
data. For the Kronecker model, we first synthesize a narrow-
band channel matrix for each delay, then transform the CIR to
the frequency domain using a DFT.
A. Complexity
The number of parameters needed in the different models
are given in Table IV. An example is also provided when
. For the Kronecker model, refers to
the tap number, while is the number of frequency bins for the
proposed coupling-based models. The full-correlation model
can completely characterize the Rayleigh fading channel. How-
ever, in this model the full-correlation matrix with
parameters need to be specified. In three coupling-based
models, the WEB model needs the most parameters because
the coupling tensor and three correlation matrices need to be
581
TABLE IV
THE NUMBER OF PARAMETERS AND EVD OPERATIONS
NEEDED IN THE DIFFERENT MODELS
specified. Three EVD operations are also required in the WEB
model. With the increase in the number of the frequency bins,
the WEB model will become more difficult for synthesis. The
WVP model needs the fewest parameters to synthesize and
does not require any EVD operation. The WHY model omits
the correlation matrix computation and the EVD operation in
the frequency domain. Because the number of frequency bins is
usually larger than the number of antennas, the WHY model’s
complexity is close to the WVP model’s.
B. Joint DoA-DoD-Delay Power Spectrum
For the measurements at Lund University, the Tx and Rx
antenna patterns have not been measured yet. Thus, the DoA-
DoD-delay power spectrum cannot be estimated using the data
measured at Lund University. Therefore, only the THU mea-
surements will be presented.
1) THU Office Scenario: Figs. 6 and Fig. 7 show the results
from the measured data in the office scenario at THU. Fig. 6
illustrates the APSs of the strongest tap. In this scenario, the
transmitter and the receiver were placed in the same office
without LOS propagation. There are a lot of scatterers in
this position. It can be seen that the strongest path is close
to DoD DoA 22 23 and other paths appear at
50 27 , 10 32 , 21 78 , and 12 24 .
The Kronecker model forces the joint DoA-DoD spectrum to
be the product of two separate DoA and DoD spectrums. This
will lead to some artifact paths at the intersection of the DoD
and DoA peaks, e.g., the path at 22 30 . The WEB model
can reflect the spatial structure more accurately. Nevertheless,
the WEB model still fails to model some details of the cluster
arriving at about 27 . The WHY model performs identically
to the WEB model because they have the same spatial structure.
The performance of the WVP model is limited by the array size.
In this scenario, the WVP model performs well because most
real MPCs depart and arrive at the "virtual" angles.
In order to quantify the power spectrum differences, we use
the Kullback-Leibler divergence (KLD) to describe the differ-
ences. The KLD is defined as follows [25], [26]:
(34)
Here, is the nor-
malized . with equality iff .
In Fig. 6, let and represent DoA and DoD, respectively.
Then KLDs from the modeled APSs to the measured APS can be
582 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 6. MIMO APSs of the strongest tap from: (a) measured data, (b) WEB
model, (c) WVP model, (d) WHY model, and (e) Kronecker model in the office
scenario at THU.
Fig. 7. Joint DoA-Delay spectrum when DoD from: (a) measured
data, (b) WEB model, (c) WVP model, and (d) WHY model in the office sce-
nario at THU.
calculated. The results are shown in the first row of Table V. In
this scenario, it can be seen that the KLDs between the measured
TABLE V
THE KLDS BETWEEN MODELED AND MEASURED APSS
TABLE VI
THE KLDS BETWEEN MODELED AND MEASURED DOA-DELAY SPECTRUMS
APS and the WEB, WHY, and WVP models’ are very small.
The Kronecker model defined in [16] has larger KLDs because
it generates some artifact paths. Furthermore, the KLD between
the WEB and the WHY models’ APSs is 0.0149, near to zero.
That is because these two models have similar spatial structure.
Fig. 7 shows the joint DoA-delay spectrums at the DoD
22.2 . It can be seen that there are two main taps in the delay do-
main. The first tap has stronger power and larger angular spread
than the second one. The KLDs from the modeled DoA-delay
spectrums to the measure one are shown in the first row of
Table VI.
As shown in Table VI and Fig. 7, the performance of the
WEB model is still the best when it is used to characterize the
delay-DoA structure. The WHY and WVP models are also close
to the measured data in this scenario. For the WHY and WVP
models, two delay taps are overlapped, which may be caused by
the limited measurement bandwidth of the THU sounder. The
measurement bandwidth is 16 MHz, so the delay resolution is
around 0.0626 us. Larger measurement bandwidth will lead to
better performance for the WHY model.
2) THU Corridor-to-Office Scenario: Fig. 8 illustrates the
APSs of the position 1 in the corridor-to-office scenario. Sev-
eral paths arrive around 10 , which is the direction of the door
4–510. There is another path arriving at 67 , which is corre-
sponding to the door 4–509. In this scenario, because most paths
travel through doors opening, the joint APS can be considered
as separable. Thus, the performance of the Kronecker model is
acceptable while it still produces some artifact paths, e.g., at
31 24 and 24 64 . The KLD results from the modeled
APSs to the measured APS in this scenario are illustrated in the
second row of Table V. The WEB model and the WHY model
still performs better than the Kronecker and WVP models. The
WVP model mixes several paths together because of the lim-
ited spatial resolution. Thus, its KLD from the measured data is
large. Because the Kronecker structure is coincide with the re-
alistic channel, the KLD of the Kronecker model is small.
The DoA-delay spectrums when DoD 26.7 in this sce-
nario are plotted in Fig. 9. It is shown that the second tap is the
strongest tap in this scenario. The KLDs between modeled and
measured spectrums in Fig. 9 are shown in the second row of
Table VI. Again, the WEB model captures the spatial and delay
characteristics of this channel best. The WVP model cannot dis-
tinguish the paths arriving from different directions. The WHY
model provides acceptable resolutions both in delay and receive
ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS 583

TABLE VII
CAPACITY ERROR OF DIFFERENT WIDEBAND MODELS IN THE CORRIDOR-TO-OFFICE SCENARIO AT THU.
THE DIAGONAL DASHED LINE DENOTES NO MODEL ERROR

Fig. 9. Joint DoA-delay spectrum when DoD from: (a) measured

data, (b) WEB model, (c) WVP model, and (d) WHY model in the corridor-to-
office scenario at THU.

Fig. 8. MIMO APSs of the strongest tap from: (a) measured data, (b) WEB
model, (c) WVP model, (d) WHY model, and (e) Kronecker model in the cor-
ridor-to-office scenario at THU.
angular domains. Limited by the bandwidth, a small overlap-
ping in delay domain can be seen in the WHY model.
C. Capacity
The data from the corridor-to-office scenario at THU and
three scenarios from the Lund measurements are used for the
channel capacity analysis. The measured data from the THU of-
fice scenarios are not enough to get reliable statistical results, so
they are not illustrated here.
The evaluation SNR in (33) is selected as 20 dB and the ca-
pacity error of wideband models is defined as
Error (35)
Here, are the modeled capacity. is the measured
capacity. A positive error value means the overestimation of real
capacity, while a negative value means underestimation.
1) THU Corridor-to-Office Scenario: Table VII shows the
average capacity error of different models when
. The average capacity error in the last row is com-
puted by averaging absolute error values at all positions. It is il-
lustrated that the WEB model has the smallest estimation error
584
Fig. 10. Modeled capacity error using all data sets and all positions in the cor-
ridor-to-office scenario at THU.
Fig. 11. Capacity CDF for all data sets and all positions in the corridor-to-office
scenario at Lund University.
and the error of the WHY model is very close to that of WEB
model.
Fig. 10 illustrates the modeled versus synthesized capacity at
all 29 positions. The diagonal dashed line represents no model
error. Generally the WHY model beats the WVP model, but for
a few positions, the WVP model is better than the WVP model.
This may be caused by the propagation environment around
these positions. In general, if the number of scatterers is small
and the DoAs and DoDs are close to the virtual angles, the WVP
model will have good agreement with real channel. Otherwise
its performance may degrade quickly.
2) Lund Scenarios: Figs. 11, 12, and 13 show the capacity
results of the different scenarios at Lund University. Although
the receive cylindrical array response could be transformed to
a ULA response by parameter extraction and resynthesization,
the WVP capacity have not been calculated because the antenna
calibration files are unavailable. Thus, only the results of the
WEB, WHY and Kronecker models are illustrated. Similar to
the results in the THU scenario, the WEB and WHY models
outperform the Kronecker model.
The WHY model describes a continuous-time reality better
and better with increased sampling rate (bandwidth). Then
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 12. Capacity CDF for all data sets and all positions in the corridor-to-
corridor scenario at Lund University.
Fig. 13. Capacity CDF for all data sets and all positions in the lobby-to-corridor
scenario at Lund University.
with higher bandwidth, the WHY model will perform better
with significant complexity reductions. During the Lund mea-
surements, the measured bandwidth is 200 MHz which leads a
higher delay resolution. As a result, the WHY model’s capacity
estimation error is smaller in the scenarios at Lund University
than that at Tsinghua. For example, in Lund corridor-to-office
scenario, the averaging absolute capacity errors of the WEB,
WHY and Kronecker models are 2.67%, 2.45%, and 7.89%,
respectively. The WHY model’s performance approximately
equal to the WEB model’s. At the same time, the WHY model
requires much smaller computation and memory loads than the
WEB model.
D. Summary
Above all, the performance of these models is related to the
propagation environment. In general, the WEB model has the
best performance but also the highest complexity. The WVP
model decreases the computation load but it cannot describe
the spatial channel properties as well as the WEP model. The
WHY model can provide a tradeoff between the modeling com-
plexity and accuracy. Once again, we mention that the selection
ZHANG et al.: A GENERAL COUPLING-BASED MODEL FRAMEWORK FOR WIDEBAND MIMO CHANNELS
of models should be decided by the simulation requirements on
accuracy and complexity.
VII. CONCLUSION
In this paper, we presented a general coupling-based model
framework for wideband MIMO channels. We reviewed several
narrowband analytical models, and introduced the CSI tensor
expression and full-correlation model for the wideband MIMO
channel. A general coupling-based framework was proposed
by building the relationship between transmitter, receiver and
frequency domains. Several models were established under this
framework, including the wideband i.i.d. channel, the WKB
model, the WEB model, the WVP model, and the WHY model.
The WKB, WEB, and WVP models can be viewed as extensions
of the corresponding narrowband models. The novel WHY
model employed the steering matrix in the frequency domain
and eigenbases in the transmitter and receiver domains. In order
to evaluate the performance of these models, measurement
campaigns were carried out in different indoor scenarios both
at Tsinghua and Lund Universities. The joint DoA-DoD-delay
power spectrum and capacity were selected as the evaluation
metrics. Using data sets gathered in different scenarios, we com-
pared the performance of different models. The results showed
that the WEB model outperformed other models, but with the
highest complexity. The WHY model was shown to provide a
good tradeoff between the complexity and the performance.
REFERENCES
[1] E. Telatar, “Capacity of multi-antenna Gaussian channels,” Eur. Trans.
Telecommun., vol. 10, no. 6, pp. 585–595, Nov./Dec. 1999.
[2] P. Almers, E. Bonek, and A. Burr et al., “Survey of channel and radio
propagation models for wireless MIMO systems,” EURASIP J. Wire-
less Commun. Netw., vol. 2007, 2007, 10.1155/2007/19070, Article ID
19070.
[3] K. Yu and B. Ottersten, “Models for MIMO propagation channels—A
review,” Wireless Commun. Mobile Comput., vol. 2, no. 7, pp.
1211–1226, Nov. 2002.
[4] H. Özcelik, “Indoor MIMO channel models,” Ph.D. dissertation,
Institut für Nachrichtentechnik und Hochfrequenztechnik, Technische
Universität Wien, Vienna, Austria, Dec. 2004 [Online]. Available:
http://www.nt.tuwien.ac.at/mobile/theses
[5] P. Petrus, J. H. Reed, and T. S. Rappaport, “Geometrical-based statis-
tical mocrocell channel model for mobile environment,” IIEEE Trans.
Commun., vol. 50, no. 3, pp. 495–502, Mar. 2002.
[6] T. Zwick, C. Fischer, and W. Wiesbeck, “A stochastic multipath
channel model including path directions for indoor environments,”
IEEE J. Sel. Areas Commun., vol. 20, no. 6, pp. 1178–1192, Aug.
2002.
[7] A. A. M. Saleh and R. R. Valenzuela, “A statistical model for indoor
multipath propagation,” IEEE J. Sel. Areas Commun., vol. 52, no. 1,
pp. 128–137, Feb. 1987.
[8] J. P. Kermoal, L. Schumacher, K. I. Pedersen, and P. E. Mogensen, “A
stochastic MIMO radio channel model with experimental validation,”
IEEE J. Sel. Areas Commun., vol. 20, pp. 1211–1226, Aug. 2002.
[9] W. Weichselberger, M. Herdin, H. Özcelik, and E. Bonek, “A sto-
chastic MIMO channel model with joint correlation of both link ends,”
IEEE Trans. Wireless Commun., vol. 5, pp. 90–100, Jan. 2006.
[10] W. Weichselberger, “Spatial structure of multiple antenna radio chan-
nels,” Ph.D. dissertation, Vienna Univ. of Technology, Vienna, Austria,
2003 [Online]. Available: http://www.nt.tuwien.ac.at/mobile/theses
585
[11] A. M. Sayeed, “Deconstructing multiantenna fading channels,” IEEE
Trans. Signal Process., vol. 50, pp. 2563–2579, Oct. 2002.
[12] A. Burr, “Capacity bounds and estimates for the finite scatterers MIMO
wireless channels,” IEEE J. Sel. Areas Commun., vol. 21, no. 5, pp.
812–818, Jun. 2003.
[13] A. Sibile, “Keyholes and MIMO channel modeling,” Tech. Rep. COST
273 TD (01) 017, Bologna, Italy, 2001 [Online]. Available: http://www.
ensta-paristech.fr/~sibille/TD(01)017.pdf
[14] M. Debbah and R. R. Müller, “MIMO channel modeling and the prin-
ciple of maximum entropy,” IEEE Trans. Inf. Theory, vol. 51, no. 5,
pp. 1667–1690, May 2005.
[15] A. M. Tulino, A. Lozano, and S. Verdü, “Impact of antenna correlation
on the capacity of the multiantenna channels,” IEEE Trans. Inf. Theory,
vol. 51, no. 7, pp. 2491–2509, Jul. 2005.
[16] K. Yu, M. Bengtsson, B. Ottersten, D. McNamara, and P. Karlsson,
“Modeling of wideband MIMO radio channels based on NLoS indoor
measurements,” IEEE Trans. Veh. Technol., vol. 50, pp. 655–665, Jul.
2002.
[17] N. Costa and S. Haykin, “A novel wideband MIMO channel model and
experimental validation,” IEEE Trans. Antennas Propag., vol. 56, pp.
550–562, Feb. 2008.
[18] Y. Zhang, X. W. Hu, Y. Z. Jia, S. D. Zhou, X. Chen, and J. Wang, “A
novel coupling-based model for wideband MIMO channel,” in Proc.
IEEE Globalcom, Nov. 2009, pp. 1–6.
[19] M. Milojevic, G. Del Galdo, and M. Haardt, “Tensor-based frame-
work for the prediction of frequency-selective time-variant MIMO
channels,” in Proc. IEEE Int. ITG Workshop Smart Antennas, Feb.
2008, pp. 147–152.
[20] M. Weis, G. Del Galdo, and M. Haardt, “A correlation tensor Cbased
model for time variant frequency selective MIMO channels,” in Proc.
IEEE Int. ITG Workshop Smart Antennas, Feb. 2008, pp. 147–152.
[21] Y. H. Rui, Y. Zhang, S. J. Liu, and S. D. Zhou, “3.52-GHz MIMO
radio channel sounder,” in Proc. IEEE Int. Conf. Commun., Circuits,
Syst. (ICCCAS), May 2008, pp. 79–83.
[22] S. Wyne, A. F. Molisch, P. Almers, G. Eriksson, J. Karedal, and
F. Tufvesson, “Outdoor-to-indoor office MIMO measurements and
analysis at 5.2 GHz,” IEEE Trans. Veh. Technol., vol. 57, no. 3, pp.
1374–1386, May 2008.
[23] IST-WINNER II Deliverable D1.1.2, WINNER II Channel Models,
Sep. 2007 [Online]. Available: https://www.ist-winner.org/deliver-
ables.html
[24] P. Almers, F. Tufvesson, and A. F. Molisch, “Keyhole effect in MIMO
wireless channels: Measurement and theory,” IEEE Trans. Wireless
Commun., vol. 5, no. 12, pp. 3596–3604, Dec. 2006.
[25] G. Matz, “Characterization and analysis of doubly dispersive MIMO
channels,” in Proc. 40th Asilomar Conf. Signals, Syst., Comput., Pacific
Grove, CA, Oct. 2006, p. 946C950.
[26] T. T. Georgiou, “Distances and Riemannian metrics for spectral density
functions,” IEEE Trans. Signal Process., vol. 55, no. 8, pp. 3995–4003,
Aug. 2007.
[27] L. Bernadö, T. Zemen, A. Paier, G. Matz, J. Karedal, N. Czink, C. Du-
mard, F. Tufvesson, M. Hagenauer, A. F. Molisch, and C. F. Mecklen-
bräuker, “Non-WSSUS vehicular channel characterization in highway
and urban scenarios at 5.2 GHz using the local scattering function,” in
Proc. Int. ITG Workshop Smart Antennas, Darmstadt, Germany, Feb.
2008, pp. 9–15.
Yan Zhang (S’06–M’10) received the B.E. degree in
information engineering from the Beijing Institute of
Technology, Beijing, China, in 2005 and the Ph.D.
degree in communication and information systems
from Tsinghua University, China, in 2010.
He is currently an Assistant Researcher in the State
Key Laboratory on Microwave and Digital Commu-
nications, Department of Electronic Engineering, Ts-
inghua University, Beijing, China. His research in-
terests include the areas of wireless communications
and channel modeling.
586 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Ove Edfors (S’92–A’96–M’00) was born in Örn-
sköldsvik, Sweden, in 1966. He received the M.Sc.
degree in computer science and electrical engi-
neering and the Ph.D. degree in signal processing,
both from the Luleå University of Technology,
Sweden, in 1990 and 1996, respectively.
In spring 1997, he worked as a Researcher at the
Division of Signal Processing at the same university,
and in July 1997, he joined the staff at the Depart-
ment of Electrical and Information Technology, Lund
University, Sweden, where he is currently a Professor
of radio systems. His research interests include radio systems, statistical signal
processing, and low-complexity algorithms with applications in telecommuni-
cation.
Peter Hammarberg (S’07–M’11) received the M.S.
degree in engineering physics from Uppsala Univer-
sity, Sweden, in 2005 . He is currently working to-
wards the Ph.D. degree at the Department of Elec-
trical and Information Technology, Lund University,
Sweden.
His research interests include channel estima-
tion and receiver algorithm design, mainly for
MIMO-OFDM systems.
Tommy Hult (M’11) received the M.Sc. degree in electrical engineering with
emphasize on telecommunication and the Ph.D. degree, both from Blekinge In-
stitute of Technology, Sweden, in 2002 and 2008, respectively.
Since 2009, he has been working as a Research Fellow attached to the com-
munications group of the Department of Electrical and Information Technology
at Lund University, Sweden. His research interests are in space-time processing
(especially MIMO), radio channel modeling, radio channel measurement, and
radio wave propagation.
Xiang Chen (M’08) was born in Hunan, China, on March 24, 1980. He re-
ceived the B.E. and Ph.D. degrees, both from the Department of Electronic En-
gineering, Tsinghua University, Beijing, China, in 2002 and 2008, respectively.
He is currently an Assistant Researcher in the Wireless and Mobile Communi-
cation Technology R&D Center, Research Institute of Information Technology,
Tsinghua University, Beijing, China. His research interests include statistical
signal processing, digital signal processing, and wireless communications.
Shidong Zhou (M’98) received the B.S. and M.S. degrees in wireless commu-
nications from Southeast University, Nanjing, China, in 1991 and 1994, respec-
tively, and the Ph.D. degree in communication and information systems from
Tsinghua University, Beijing, China, in 1998.
He is currently a Professor at Tsinghua University. From 1999 to 2001, he was
in charge of several projects in China 3G Mobile Communication R&D Project.
He is now a member of the China FuTURE Project. His research interests are
in the area of wireless and mobile communications.
Limin Xiao (M’00) received the B.S. and M.S. degrees in wireless commu-
nication engineering from the Harbin Institute of Technology, Harbin, in 1992
and 1995, respectively, and the Ph.D. degree in communication and information
systems from Tsinghua University, Beijing, China, in 2000.
From 2000 to 2003, he was a Lecturer in the Department of Electronic Engi-
neering at Tsinghua University, China, where he is currently a Vice-Researcher.
Jing Wang (M’99) received the B.S. and M.S. degrees in electronic engineering
from Tsinghua University, Beijing, China, in 1983 and 1986, respectively.
Since 1986, he has been on the faculty of Tsinghua University, Beijing, China,
where he currently is a Professor and the Vice Dean of the Tsinghua National
Laboratory for Information Science and Technology. His research interests are
in the area of wireless digital communications, including modulation, channel
coding, multi-user detection, and 2-D RAKE receivers.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Multi-Link MIMO Channel Modeling Using
Geometry-Based Approach
Juho Poutanen, Fredrik Tufvesson, Senior Member, IEEE, Katsuyuki Haneda, Member, IEEE,
Veli-Matti Kolmonen, and Pertti Vainikainen
Abstract—Geometry-based stochastic channel models (GSCMs)
are extended to support multi-link simulations by applying the con-
cept of common clusters. This novel approach aims to control the
correlation between different links, inter-link correlation, by ad-
justing the amount of power simultaneously propagating via the
same clusters in the different links. The behavior of common clus-
ters is analyzed based on dual-link channel measurements, and a
multi-link GSCM is developed based on common clusters. In addi-
tion, the effects that the common clusters have on inter-link corre-
lation and on sum rate capacity are investigated based on simula-
tions. Finally, comparison between simulations and measurements
is done in order to indicate the validity of the proposed multi-link
GSCM.
Index Terms—Geometry-based channel modeling, multi-link
channel modeling, multi user MIMO, radio channel modeling.
I. INTRODUCTION
G EOMETRY-BASED stochastic channel models
(GSCMs) have attained much attention in MIMO
channel modeling during the past decade. This is due to their
inherent capability of modeling spatial and temporal correlation
properties in a straightforward manner. The basic idea behind
the GSCMs is to emulate the double-directional radio channel
by placing clusters in the simulation environment to act as
physical scattering objects. The clusters consist of groups of
closely located multipath propagation components (MPCs),
and the directions, delays, and complex amplitudes of each
MPC are directly computed based on the geometry of the
simulation environment. Examples of GSCMs, which all rely
on the cluster approach, are the COST 259 [1], COST 273 [2],
and WINNER [3] channel models.
Manuscript received May 27, 2010; revised September 13, 2010; accepted
October 09, 2010. Date of publication March 03, 2011; date of current ver-
sion February 03, 2012. This paper has been written within the framework of
WILATI+ which is a joint project between three Scandinavian universities and
a part of the NORDITE research program funded by the Finnish, Swedish and
Norwegian national research institutes Tekes, Vinnova and RCN, respectively.
The work was supported in part by the Jenny and Antti Wihuri Foundation and
the post-doctoral research project of the Academy of Finland, Helsinki, Finland.
J. Poutanen, K. Haneda, V.-M. Kolmonen, and P. Vainikainen are with the
Department of Radio Science and Engineering, Aalto University School of Sci-
ence and Technology, FI-00076 Aalto, Finland (e-mail: jpoutane@cc.hut.fi).
F. Tufvesson is with the Department of Electrical and Information Tech-
nology, Lund University, SE-221 00 Lund, Sweden.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2122296
0018-926X/$26.00 © 2011 IEEE
587
A. Motivation
Even though the state-of-the art GSCMs are extremely so-
phisticated, the capability of the current implementations to sim-
ulate multi-link scenarios has up to now remained an open ques-
tion. In principle, the cluster-based structure of the GSCMs sup-
ports multi-link simulations just by dropping multiple mobile
stations (MSs) and/or base stations (BSs) into the environment.
However, the major open question has so far been how to con-
trol the correlation between different links, the inter-link corre-
lation. Since clusters are generated randomly and independently
for each link, there is no guarantee that the different links result
in having proper correlation with respect to each other. Based
on previously reported work, however, it has been realized that
different links may encounter remarkable inter-link correlation
when the units are in close proximity to each other [4], but also
when they are largely separated in distance [5]–[8]. Two links
may also have very different inter-link correlation even if they
are in the same environment (e.g., in the same room) [9].
Those experimental findings indicate a true need for multi-
link MIMO channel models being able to reflect those proper-
ties. Furthermore, since the trend in novel radio communication
systems is going more and more towards applications that utilize
links between multiple nodes in the network for their operation,
the need for realistic multi-link channel model increases accord-
ingly. Examples of such systems where it would be very ben-
eficial to have realistic multi-link channel models include co-
operative communication systems, and indoor localization ap-
plications. Underestimating or neglecting the inter-link correla-
tion would usually lead to too optimistic performance results in
system simulations [4].
To the authors’ best knowledge, previous works on multi-
link MIMO channel modeling are restricted to the analytical
dual-link model proposed in [10]; no contributions on multi-link
GSCMs are available in the open literature.
B. Contributions
In this paper we propose an approach for extending a gen-
eral GSCM to fully support multi-link simulations by applying
the concept of common clusters (CCs). In short, the idea behind
the proposed method is to control the correlation between dif-
ferent links by allowing a certain proportion of the energy in
different links to propagate through the same clusters. The idea
to use CCs as a means to develop a multi-link GSCM is based
on observations made in measurements that significant amounts
of energy can indeed propagate through the same scatterers in
different MIMO links in certain types of environments [6].
588 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
At first, CCs are analyzed based on measured dual-link
MIMO data. Then, to reflect the characteristics of the measured
dual-link channels, a multi-link GSCM is implemented based
on the concept of CCs. The implemented simulation model is
used to study the effect of the CCs on channel characteristics
and system performance from two perspectives: first, it is
shown that CCs are a suitable way of adjusting the correlation
between different links, and, second, the effect of the CCs
on channel capacity is investigated. It is also shown that the
developed multi-link GSCM is able to accurately predict the
channel behavior in comparison with the measurement data.
Even though the GSCM developed in this paper is rather
simple, the modeling concepts are fully applicable to more
sophisticated GSCM implementations (such as [1]–[3]), as
well and it should be straightforward to extend those models to
handle the multi-link scenario.
C. Organization
The remainder of the paper is organized as follows. In
Section II, CCs are extracted from dual-link channel mea-
surement data. In Section III, the modeling philosophy of
the multi-link GSCMs is discussed, followed by a detailed
description of the model implemented in this work. Section IV
is dedicated to the analyses of the effect of the CCs on inter-link
correlation and dual-link capacity based on simulations.
In Section V, the relationships between the significance of
common clusters, inter-link correlation, and sum rate capacity
are studied based on measurement data. A comparison between
the developed model and measurement data in made in terms
of the dual-link channel capacity also in Section V. Finally,
Section VI concludes the work.
II. COMMON CLUSTERS IN MEASURED CHANNELS
For the development of multi-link GSCMs, it is essential to
understand the physical phenomena that increase correlation be-
tween different links. One of these phenomena is scatterers, or
clusters, that are common for two or more links. Hence, it is
vital to investigate how frequently the common clusters occur
in real multi-link propagation scenarios. To this end, we investi-
gate CCs based on dual-link channel measurements done in an
office corridor environment. First, the methodology to extract
CCs from measurement data is presented after which experi-
mental results on will be provided.
A. Extraction of Common Clusters From Measurement Data
1) Data Analysis: Physical scattering objects can be iden-
tified from the measurement data by combining the measured
radio propagation path parameter estimates with the geometry
information of a measurement environment. In this work,
the parameter estimates have been obtained by the extended
Kalman filter (EKF) [11]. Each MPC of the EKF parameter
estimates includes the DoD, DoA, delay, and polarimetric
path weights. Furthermore, each propagation path obtained by
the EKF has a lifetime over a certain number of consecutive
measurement samples, i.e., snapshots.
In order to identify the physical scatterers for each MPC, the
EKF estimates of the DoD, DoA and delay are used as inputs
Fig. 1. Extraction of common cluster from measurement data. The distance
between the scattering points of different links and the angular separation of
the scattering points seen from the MS determine if the cluster is considered as
common.
for a measurement-based ray tracer [12]. This ray tracer imple-
ments an algorithm that plots rays on top of a floor plan of the
environment according to the measured parameter estimates. It
enables the MPCs to be explicitly mapped to physical scatterers
in the environment. The identification of the scattering objects
can be performed simultaneously for multiple links, making it
possible to study if the same physical scatterer is common for
the different links, and thus forms a CC.
2) Definition of a Common Cluster: In situations where the
scatterer has a relatively large physical size, it is not always
meaningful to consider the whole scatterer as common for dif-
ferent links. In such cases, even if the scattering source is the
same physical object (e.g., a wall), the scattering points for dif-
ferent links might be separated by a large distance, and thereby
an MS or BS equipped with antenna arrays may be able to re-
solve the scattering points in the angular domain. On the other
hand, scattering points for the different links can be resolvable
even if the distance between them is small in a case where the
scattering object is very close to the antenna. Therefore, it is nec-
essary to establish conditions defining CCs based on 1) the dis-
tance between the scatterers of different links , and 2) the an-
gular separation of the scatterers seen from the MS , as shown
in Fig. 1.
It should be noted that multiple MPCs might be originating
from the same physical scatterer at the same time instant thus
forming a cluster. In such cases, the distance between the scat-
terers of the different links is calculated based on the cluster
center; the coordinates of the cluster center are calculated as
the power-weighted mean over the coordinates of the scattering
points of the individual MPCs belonging to the same cluster. In
Fig. 1, the small black dots correspond to the scattering points
of individual MPCs and the blue and red dots are the cluster
centers in different links. In this work, threshold values for the
and are selected to be 5 meters and 45 degrees, respectively.
The significance of common clusters quantifies
the amount of power that propagates through the CC. The
POUTANEN et al.: MULTI-LINK MIMO CHANNEL MODELING USING GEOMETRY-BASED APPROACH
is determined from the measurement data in the
following way. In a dual-link case, the significance of the th
CC is denoted as a function of a measurement time instant by
where is the significance of the th common scat-
terer with respect to the total power of the th link as
where . If the total number of scatterers that are common
for the different links is denoted by , the total
be expressed by the sum of the significances of the individual
CCs by
B. Significance of Common Clusters in an Office Corridor
Scenario
1) Measurement Setup: The analyzed measurement was car-
ried out in the corridor of the Department of Radio Science and
Engineering in Aalto University by using a dual-link channel
sounding system consisting of two channel sounders from Aalto
University School of Science and Technology (Aalto) and Lund
University (LU) operating at the frequency of 5.3 GHz [13]. The
system is capable of simultaneously measuring two links having
the MIMO matrix sizes of 30 30 and 30 32. A sample of a
full MIMO matrix, i.e., snapshot, is measured every 39.32 ms,
enabling measurements in dynamic environments. The mea-
surements involved dual-polarization, but in the present anal-
ysis the polarizations have not been treated separately but they
are summed up. In the measurement, the TX (from LU) was
moved along a continuous route acting as a MS, such as a smart
phone, whereas the two RXs (from LU and Aalto) were located
at fixed positions to emulate WLAN BSs of neighboring cells.
The floor plan of the measurement venue is shown in Fig. 2.
2) Results: The main propagation mechanism in the consid-
ered scenario is the waveguiding along the corridors in both
links. In addition, DoDs pointing towards walls B and C are
commonly seen on the MS side in both links. Interestingly, a
noticeable share of energy propagates directly through wall B
and the coffee room in the MS-BS1 link; this is possible since
wall B and the right wall in the coffee room are outdoor walls
having many windows meaning that the signal can penetrate the
building through windows or window frames.
In order to investigate in this scenario, the first scat-
tering point seen from the MS on wall A was calculated and
used to check the conditions for the CC. In addition, the points
on wall B where the signal either propagates through the wall
(MS-BS1 link) or reflects from it (MS-BS2 link) can be identi-
fied. In this scenario, other scatterers did not fulfill the criteria
of the CC. Even if waves reflected from wall C are also seen
589
(1)
(2)
Fig. 2. The floor plan of the measurement environment in the considered sce-
can
nario.
in parts of the measurement route, this propagation mechanism
was not active in both links simultaneously, and thus was not
(3) identified as a common scatterer.
The angular separation of the scattering points on wall A and
B are plotted in Fig. 3(a). In the case of wall A, the angular
separation is around 10 degrees in the beginning of the route,
but after snapshot 200 it starts to increase; at this location the
MS passes the corner of wall C and the scattering points in the
MS-BS2 link start to move further south. The threshold value
is exceeded approximately at snapshot 220. In the case of wall
B, the angular separation of the scattering points varies more
rapidly than in the case of wall A. Furthermore, wall B is an ac-
tive common scatterer only in parts of the route. The distance
between the scattering points on both walls A and B (Fig. 3(b))
follow the same trends as the angular separation. Also in this
case, the threshold value is exceeded in the end of the measure-
ment route.
Fig. 3(c) shows separately for scatterers A (blue
curve) and B (red curve) and the total (black curve). It
is seen that the waveguiding along the corridor (A) is a signifi-
cant propagation mechanism in both links, hence constituting a
significant CC. Also wall B forms a CC in parts of the route. The
total varies between 40% and 95%, but goes rapidly
to zero around snapshot 220 due to the fact that the threshold
values for the conditions of CC are exceeded.
III. MULTI-LINK GEOMETRY-BASED CHANNEL MODELING
When extending the current GSCMs to cover multi-link sce-
narios there are two main aspects to consider: 1) The singe-link
behavior should remain the same, while at the same time 2) the
correlation between links, i.e., interlink correlation, should be
represented in a realistic way.
In this section, a concept for extending the current GSCMs
to support multi-link simulations is presented. First, the con-
cept of common clusters, the proposed methodology in which
the correlation between links in GSCMs can be controlled, is
explained. Then, the implementation of the multi-link GSCM
used in this paper is detailed. Finally, relationships between the
multi-link GSCM implemented in this paper and the other ex-
isting GSMCs is briefly discussed.
590 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 4. Example of a common cluster (CC) in a scenario with one MS and two
BSs. The correlation between different links can be controlled by changing the
amount of power that propagates through the CC.

controlled by the cluster distributions and remains unchanged.

The approach is based on the observations made above and in
[6], that CCs can carry significant parts of the energy in realistic
multi-link environments.

B. Multi-Link GSCM Development

In the proposed multi-link GSCM, clusters are generated
so that each link is assigned with a set of uncommon clusters
(UCCs) and a certain number of CCs. The UCCs are generated
separately for each link and they contribute only to the impulse
response of the designated link, whereas CCs are shared by
the different links and contribute to both impulse responses,
see Fig. 4. In order to control the correlation between different
links, the amount of power that propagates through the CC
is set to a desired value by the following procedure. First,
the propagation path parameters, including the direction of
departure (DoD) and direction of arrival (DoA), delay, and
complex amplitude, are calculated for each MPC according to
the geometrical locations of the antennas and clusters. Then,
the power carried by the UCCs is scaled with a factor of so
that the condition

(4)

Fig. 3. (a) The angular separation of the scattering points belonging to the dif- is satisfied. In (4), i.e., the significance of common
ferent links seen from the MS. (b) The distance between the scattering points cluster, defines the ratio between powers carried by the CCs and
of different links. (c) The significance of common clusters as a function of the
measurement location in snapshots.
UCCs; and are the sum of powers carried by all the
CCs and UCCs, respectively.
Once the propagation path parameters with the desired
A. Common Clusters in Geometry-Based Models have been obtained, the MIMO channel matrices are
calculated for each link in the same way as for a conventional
A common cluster is a cluster that contributes to the channel GSCM as
between different links at the same time, as shown in Fig. 4.
Since the amount of power carried by the CCs can be set to a
desired value, it is possible to control the inter-link correlation
(5)
between different links; the larger is the amount of power that is
propagating via the CCs, the stronger is the correlation between
the different links. The intra-link correlation, i.e., the correla- where is the number of MPCs, is the amplitude of the
tion between antenna elements in a MIMO link, is as before th MPC, is the used radio frequency, is the propagation
POUTANEN et al.: MULTI-LINK MIMO CHANNEL MODELING USING GEOMETRY-BASED APPROACH 591

delay of the th MPC, and denotes the matrix transpose

operation. Furthermore, and are the array response
vectors calculated at the transmitter (TX) or receiver (RX) as

(6)

where is the wave number, is the unit direc-

tional vector consisting of the DoD and DoA of the th MPC,
and is the position vector of the th antenna element in the
array. The term represents the inner product between
vectors and . Finally, a random phase is added to
each MPC in order to increase the number of independent real-
izations of the channel matrices [14].

C. Relationship With Other GSCMs

The model developed in this work is based on clusters that

emulate physical scattering objects of real environments, in the
Fig. 5. Flow chart of the method to investigate the effect on common clusters
same way as in conventional GSCMs, and the MIMO channel on system performance based on simulations and measurement.
matrices are constructed as usual based on the propagation path
parameters by using (5). Regarding the single-link properties,
the implemented GSCM includes some simplifications with
desired , as shown in (4). The channel matrices for the
respect to the more comprehensive GSMCs, such as [1]–[3];
different links can then be calculated with the MPC parameters
in particular, only single-bounce clusters are considered, and
obtained from the geometry of the environment and with the de-
the concept of visibility regions is not included in the model
sired by (5). Finally, the channel matrices can be ap-
meaning that the activity of clusters as a function of location is
plied to the following equations in order to study the inter-link
not regarded. Furthermore, the full model parameterization (in-
correlation and sum rate capacity. In Fig. 5, the corresponding
cluding parameters such as angular and delay spreads, cluster
procedure for the measurement data is also shown; CMC and
shadow fading, etc.) based on measurement data has not yet
SRC are studied based on measurement data and compared with
been considered.
the simulation results in Section V.
The novelty of the GSCM of this work is that a part of the
clusters are shared by different links, which makes it possible to
control the inter-link correlation. Despite the above-mentioned
A. Definitions
simplifications, the developed GSCM can be used to investigate
the capability of the proposed modeling approach to reflect the
1) Inter-Link Correlation: The inter-link correlation was
essential features of multi-link scenarios, especially the ability
evaluated by calculating the correlation matrix collinearity
to control the inter-link correlation. Furthermore, the idea of
(CMC, or “collinearity”) as [15]
CCs can be applied in any other GSCMs as well, and, in fact, the
modeling approach developed in this work has also been applied
in the multi-link extension of the COST 2100 channel model
[15]. For a detailed description of the implementation principles (7)
of the multi-link extension of the COST 2100 channel model,
the reader is directed to [15]. where is the correlation matrix of the th link calculated as

IV. EFFECT OF COMMON CLUSTERS ON SYSTEM PERFORMANCE

(8)
In this section, the effect of the common clusters, or more pre-
cisely , on system characteristics is studied in terms of
1) the inter-link correlation and 2) the sum rate dual-link MIMO In (7) and (8), is the complex transpose of a matrix;
capacity by using the multi-link GSCM described in Section III. is the number of independent channel realizations; is the
Fig. 5 shows a flow chart of the procedure of how the effect of number of transmit antennas; and denotes the Frobenius
common clusters on CMC and SRC is evaluated. At first, the norm.
propagation parameters, i.e., the DoD, DoA, complex ampli- The CMC describes how similar the subspaces of the corre-
tude, and delay, are calculated based on the geometrical loca- lation matrices of the different links are, ranging between zero
tions of the clusters, BSs, and MSs. After that, the power car- (matrices are orthogonal to each other) and one (matrices are
ried by the uncommon clusters is scaled in order to obtain the similar).
592 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
2) Sum Rate Dual-Link Capacity: The capacity values were
calculated in the following way. First, the received power for
the th link was calculated as
(9)
where is the number of independent channel realizations;
is the number of receive antennas; is the number of
transmit antennas; is the channel matrix of the th link; and
denotes the Frobenius norm. In order to calculate the ca-
pacity, the channel matrices were normalized as
(10)
In the dual-link case, the received signal can be written as
[17]
(11)
where and are the normalized channel matrices of
the desired and interfering link, and and are the trans-
mitted signal vectors, respectively, and is an uncor-
related complex Gaussian noise vector. Since the channel ma-
trices and the noise variance are normalized, and represent
the signal-to-noise ratio (SNR) and interference-to-noise ratio
(INR). The capacity values were calculated as the ergodic ca-
pacity for different channel realizations as [17]
(12)
where, again, sub-indices “des” and “int” stand for the “desired”
and “interfering” links. In our case, we assume that there is no
channel state information at the transmitter (i.e., the transmit
signal covariance becomes an identity matrix) and a full knowl-
edge of the channel at the receiver.
Next, we study the capacity in two different cases, i.e., in the
single-link case (i.e., without interference) and in the dual-link
case (i.e., with interference). When calculating the capacity in
the single-link case, i.e., , the covariance matrix
in (12) was set to
(13)
and when calculating the capacity in the dual-link case, i.e.,
, the covariance matrix was set to
(14)
In the following, we analyze the relative sum rate capacity
(SRC, or “capacity”), which is denoted by the ratio between the
sum rate dual-link capacity
and the sum rate single-link capacity :
(15)
In (15), sub-indices “1” and “2” stand for links 1 and 2, meaning
that first link 1 was considered as the desired link and link 2 as
the interfering link, and then vice versa.
In the description below, SRC and “capacity” as well as CMC
and “collinearity” will be used interchangeably to improve read-
ability.
B. Simulation Studies
The effect of the CCs on inter-link correlation and sum rate
dual-link capacity was investigated by computer simulations in
three different scenarios. In the first scenario, the locations of
the clusters were changed in a controlled manner in order to
study the relationships between , CMC and SRC and
the influence of the environment through a simple example. In
the simulations of these controlled channels the environment
consisted of one UCC (per link) and one CC (shared by the
different links). Next, the simulation studies were continued by
placing the clusters in random locations. First, the random chan-
nels consisted of one UCC and one CC, and, after that, of five
UCCs and one CC, all being randomly positioned.
In each of the three scenarios, one MS and two BSs were lo-
cated at fixed positions. The clusters consisted of five MPCs
placed randomly within a diameter of one meter around the
center point of the cluster. Furthermore, a 4-element x-oriented
uniform linear antenna array (ULA) was used at the MS and
BSs in each case. The and were both fixed to 10 dB in all
the simulations. In the following, the presented CMC and SRC
values are the average values over 100 independent channel
realizations. Table I summarizes the three different simulation
scenarios.
As a general conclusion of the simulation studies, it can be
said that has a significant effect on both the inter-link
correlation and MIMO channel capacity; as increases,
the CMC increases and SRC decreases. This indicates the ability
of the multi-link GSCM based on CCs to manipulate the corre-
lation between different links. It also shows that it is important
not to neglect the effect of common clusters.
Next, the findings from the simulation studies are discussed
in detail separately for the controlled and random scenarios.
1) Controlled Channels: In the controlled scenario, clusters
were located so that the CC and the UCC for the link MS-BS2
(UCC2) were at fixed positions whereas the UCC for the link
MS-BS1 (UCC1) was moved on a circle with 45 degree steps
around the MS, as shown in Fig. 6(a). The simulations with the
different locations of UCC1 are marked with numbers 0–7 in
Fig. 6(a). At each location of the UCC1, the CMC and SRC
were simulated with ranging between 0% and 100%.
Figs. 6(b) and (c) show the collinearity and capacity as a func-
tion of for the different locations of the UCC1. It is
seen that the capacity is very high at low values of in
most of the locations of the UCC1, and gradually decreases to
approximately 10% of its original value as approaches
POUTANEN et al.: MULTI-LINK MIMO CHANNEL MODELING USING GEOMETRY-BASED APPROACH
TABLE I
SUMMARY OF THESIMULATION SCENARIOS
100%. The collinearity behaves in the opposite way, i.e., when-
ever the capacity is high, the collinearity is low, and vice versa.
However, at UCC1 locations 3 and 5 the capacity is low, and the
collinearity almost one, with all values of . At UCC1
location 5, both UCCs are located exactly at the same position,
meaning that the situation is equivalent to the case where all the
clusters are common among the different links. At UCC1 lo-
cation 3, there is a high correlation between the different links
since the MS cannot distinguish the waves coming from loca-
tions 3 and 5 due to its antenna array orientation.
2) Random Channels: As shown above, the locations of
the clusters may have a significant impact on the correlation
and capacity values observed at different values of . is 50%. This can be explained by the fact that as
In particular, it can be observed that with some combinations
of the cluster locations collinearity is high and capacity is low
even without the CC. In order to investigate the relationships
between , collinearity, capacity, and cluster locations
in a statistical manner, the simulation studies were continued
by placing the clusters in random locations.
The CDFs of the collinearity and capacity are shown for the
first random scenario (one UCC, one CC) for the
values of 0, 1, 10, 50, 90, 99, and 100% in Fig. 7(a) and (b). For
each value of , the CDF curve includes the values of
the collinearity and capacity from 1000 random channels. With
the values of 0, 1, and 10%, the matrix collinearity be-
haves quite similarly and is less than 0.2 in approximately 60%
of the cases. Even if for the most of the time the collinearity
is low and capacity high with low values of , also
highly correlated channels are observed: the collinearity is at
least 0.8 in about 15% of the cases even without the CC. With
the values of 0, 1, and 10%, the capacity gets high
values for the most of time, as could be expected. It was also
found that capacity is more sensitive than the collinearity to the
change of in the range of 0 to 10%: for instance, at
the CDF level of 0.5, the capacity decreased by approximately
25%-units when increased from 0 to 10%. This is an
important observation in the sense that even if only a small
amount of the power propagates through the common cluster,
the impact on the system performance is significant compared
to the case where common cluster is not considered at all.
When is 50%, both the collinearity and capacity are
almost uniformly distributed between the minimum and max-
imum values. With the values of 90, 99, and 100%,
593
we can see that the collinearity is very close to 1 the whole time.
However, variations between the respective capacity curves are
more clearly seen.
The CDFs of the collinearity and capacity are shown for the
second random scenario (5 UCCs, 1 CC) for the values
of 0, 1, 10, 50, 90, 99, and 100% in Fig. 7(c) and (d). Again, the
CDF curves include the values of the collinearity and capacity
from 1000 random channels for each value of .
In comparison with the first random scenario (1 UCC, 1
CC), the following observations can be made. First, with low
values of (0, 1, and 10%), the collinearity is generally
higher and therefore capacity lower; the same trends hold when
the number of clusters increases, it is harder for the antenna
array at the MS to separate them, and hence, the correlation
between links is likely to increase. Obviously, with larger
antenna arrays, this effect would be less significant due to better
capability of spatial filtering. With high values of (90,
99, and 100%), the curves are very similar to the first random
scenario, as could be expected, since anyway almost all the
power propagates through the common cluster.
V. COMPARISON BETWEEN MEASUREMENTS AND
SIMULATIONS
A. Capacity and Correlation in Measured Channels
Fig. 8(a) shows the collinearity and capacity along the mea-
surement route. The collinearity and capacity were calculated
by applying the measured propagation path parameters obtained
by the EKF to (5)–(15) and by using the same 4-element linear
array as in the simulations of Section IV (see Fig. 5 for the
flow chart of the procedure). Again, the collinearity and capacity
were calculated as the mean over 100 independent realizations
of the measured channels at each snapshot. Fig. 8(a) shows that
the collinearity gets very high values in the beginning of the
route but falls down to approximately 0.3 in the end of the route.
The capacity behaves the opposite way, as was the case also in
the simulation studies in Section IV.
By comparing Figs. 8(a) and 3(c), we can clearly see that
also in the measured channels, the inter-link correlation and sum
rate capacity are strongly related to the significance of common
clusters.
594 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 6. Effect of common cluster simulated in controlled channels. (a) Simu-
lation environment. (b) The correlation matrix collinearity (CMC), and (c) the
relative sum rate capacity (SRC) as a function of the significance of common
cluster. The legends show the position of UCC1.
B. Simulated vs. Measured Sum Rate Capacity
Finally, in order to investigate the validity of the multi-link
GSCM approach proposed in this paper, the simulations were
compared with the measurements in terms of capacity.
Two measurement locations were selected for the compar-
ison. The first location was at snapshot 120, where the
was as high as 85%, and the second at snapshot 270, where CCs
did not exist anymore. In the simulation, two UCCs (per link)
Fig. 7. Effect of common cluster on collinearity and capacity simulated in
the random scenarios. CDFs of the (a) collinearity and (b) capacity in the first
random scenario (one UCC, one CC). (c) Collinearity and (d) capacity in the
second random scenario (five UCCs, one CC). The CDFs include data from
1000 random combinations of cluster locations with each value of the .
Fig. 8. (a) Collinearity (black curve) and capacity (red curve) as a function
of the measurement location in snapshots. (b) Capacity in measurements and
simulations. The CDFs of the capacity at snapshots 120 and 270 and of three
independent simulation runs with corresponding significance of common cluster
are shown. In the simulations, 2 UCCs (per link) and 2 CCs were generated in
random locations.
and two CCs were randomly located around the simulation en-
vironment, whereas the BSs and the MS were at fixed positions.
Fig. 8(b) shows the CDFs of the capacity over 100 channel
realizations for the measurement (solid lines) and for three in-
dependent simulation runs (dashed and dotted lines). It is seen
that the simulation can always predict the dual-link channel be-
havior very accurately in terms of the capacity when
POUTANEN et al.: MULTI-LINK MIMO CHANNEL MODELING USING GEOMETRY-BASED APPROACH
is 85%. When is 0%, the model predicts the behavior
well for the most of the time; however, occasionally, due to
random cluster locations, the correlation between links happens
to be high in the simulation which results in underestimation of
the capacity.
VI. CONCLUSION
In this paper, an approach to extend geometry-based
stochastic channel models (GSCMs) to support multi-link
simulations has been presented. The proposed approach is
based on the concept of common clusters (CCs): the correlation
between different MIMO links, i.e., the inter-link correlation,
is controlled by adjusting the amount of power that simultane-
ously propagates via the same clusters in the different links.
A multi-link GSCM was implemented, and the effects that the
CCs have on inter-link correlation and on sum rate capacity
were investigated based on simulations. The existence of CCs
in real-world channels was confirmed by dual-link channel
measurements in an office corridor scenario. The measurement
results revealed that the significance of CCs can be as high as
95%.
In both simulations and measurements, clear relations were
found between the CC power, the inter-link correlation, and the
sum rate dual-link capacity. Generally, as the amount of power
carried by the CCs increases, the inter-link correlation increases
and at the same time the sum rate capacity decreases. In ad-
dition, comparison between simulations and measurements in
terms of the sum rate dual-link capacity revealed good agree-
ment, indicating the validity of the proposed multi-link GSCM.
The results of the paper indicate that the developed CC-based
multi-link GSCM provides a suitable means for controlling the
correlation between different MIMO links in a realistic manner.
Furthermore, it is evidently vital to include CCs in GSCMs since
neglecting them would lead to too optimistic performance re-
sults in system simulations.
ACKNOWLEDGMENT
The authors would like to thank Dr. P. Almers, Dr. T.
Abrudan, Dr. A. Richter, and J. Koivunen for their help in mea-
surements and data processing. Dr. J. Salmi is acknowledged
for providing the codes for calculating the parameter estimates;
Prof. A. Molisch for fruitful discussions on the modeling ap-
proach, and A. Palacios for his help in developing the channel
model implementation.
REFERENCES
[1] A. F. Molisch, H. Asplund, R. Heddergott, M. Steinbauer, and T.
Zwick, “The COST 259 directional channel model—Part I: Overview
and methodology,” IEEE Trans. Wireless Commun., vol. 5, no. 12,
Dec. 2006.
[2] , L. M. Correia, Ed., Mobile Broadband Multimedia Networks—Tech-
niques, Models Tools for 4G. Oxford, U.K.: Elsevier, 2006, p. 569.
[3] WINNER II deliverable D1.1.2 V1.1, Online, WINNER Channel
Models [Online]. Available: http://www.ist-winner.org/deliver-
ables.html
[4] F. Kaltenberger, D. Gesbert, R. Knopp, and M. Kontouris, “Correlation
and capacity of measured multi-user MIMO channels,” in Proc. IEEE
19th Int. Symp. on Personal, Indoor Mobile Radio Commun. (PIMRC
2008), France, Sep. 2008, pp. 1–5, Cannes.
595
[5] J. Poutanen, K. Haneda, J. Salmi, V.-M. Kolmonen, and P. Vainikainen,
“Analysis of correlated shadow fading in dual-link indoor radio wave
propagation,” IEEE Antennas Wireless Propag. Lett., vol. 8, pp.
1190–1193, 2009.
[6] J. Poutanen, K. Haneda, J. Salmi, V.-M. Kolmonen, T. Hult, F.
Tufvesson, and P. Vainikainen, “Significance of common scatterers
in multi-link radio wave propagation,” in Proc. 4th Eur. Conf. on
Antennas Propag. (EuCAP 2010), Barcelona, Spain, Apr. 2010, p.
1849081.
[7] V.-M. Kolmonen, K. Haneda, T. Hult, J. Poutanen, F. Tufvesson, and
P. Vainikainen, “Measurement-based evaluation of interlink correla-
tion for indoor multi-user MIMO channels,” IEEE Antennas Wireless
Propag. Lett., vol. 9, pp. 311–314, 2010.
[8] V.-M. Kolmonen, K. Haneda, F. Tufvesson, J. Poutanen, and P.
Vainikainen, “A dual-link capacity analysis of measured time-variant
indoor channel,” Electron. Lett., 2010, Accepted for publication in.
[9] N. Czink, B. Bandemer, G. Vazquez-Vilar, L. Jalloul, C. Oestges, and
A. Paulraj, “Spatial separation of multi-user MIMO channels,” in Proc.
IEEE 20th Int. Symp. on Personal, Indoor Mobile Radio Commun.
(PIMRC 2009), Tokyo, Japan, Sep. 2009, pp. 1059–1063.
[10] T. Hult, F. Tufvesson, V.-M. Kolmonen, J. Poutanen, and K. Haneda,
“Analytical dual-link MIMO channel model using correlated correla-
tion matrices,” presented at the 4th Eur. Conf. on Antennas Propag.
(EuCAP 2010), Barcelona, Spain, Apr. 2010.
[11] J. Salmi, A. Richter, and V. Koivunen, “Detection and tracking of
MIMO propagation path parameters using state-space approach,” IEEE
Trans. Signal Processing, vol. 57, no. 4, pp. 1538–1550, Apr. 2009.
[12] J. Poutanen, K. Haneda, J. Salmi, V.-M. Kolmonen, A. Richter, P.
Almers, and P. Vainikainen, “Development of measurement-based
ray tracer for multi-link double directional propagation parameters,”
in Proc. 3rd Eur. Conf. on Antennas Propag. (EuCAP 2009), Berlin,
Germany, Mar. 2009, pp. 2622–2626.
[13] V.-M. Kolmonen, P. Almers, J. Salmi, J. Koivunen, K. Haneda, A.
Richter, F. Tufvesson, A. F. Molisch, and P. Vainikainen, “A dynamic
dual-link wideband MIMO measurement system for 5.3 GHz,” IEEE
Trans. Instrum. Meas., vol. 59, no. 4, pp. 873–883, Mar. 2010.
[14] A. Molisch, M. Steinbauer, M. Toeltsch, E. Bonek, and R. Thoma, “Ca-
pacity of MIMO systems based on measured wireless channels,” IEEE
J. Sele. Areas Commun., vol. 20, no. 3, pp. 561–569, Apr. 2002.
[15] L. Liu, J. Poutanen, K. Haneda, P. Vainikainen, F. Tufvesson, and C.
Oestges, “A multi-link extension of the COST 2100 MIMO channel
model,” presented at the COST2100 11th Management Committee
Meeting, Aalborg, Denmark, Jun. 2–4, 2010, TD(11)11012.
[16] G. Golub and C. van Loan, Matrix Computations, 3rd ed. Baltimore,
MD: The Johns Hopkins Univ. Press, 1996.
[17] R. Blum, “MIMO capacity with interference,” IEEE J. Sel. Areas
Commun., vol. 21, no. 5, pp. 793–801, 2003.
Juho Poutanen was born in Helsinki, Finland, in
1983. He received the degree of Master of Science
in technology and Licentiate of Science in tech-
nology from Aalto University School of Science and
Technology, Helsinki, Finland, in 2007 and 2010,
respectively, where he is currently working towards
the Ph.D. degree.
Since 2005, he has worked as a Research Assistant
and Researcher in the Department of Radio Science
and Engineering, Aalto University School of Science
and Technology. His current research interests
include radio channel characterization and modeling.
Fredrik Tufvesson (SM’10) was born in Lund,
Sweden in 1970. He received the M.S. degree in
electrical engineering in 1994, the Licentiate degree
in 1998, and the Ph.D. degree in 2000, all from Lund
University, Lund, Sweden.
After almost two years at a startup company, Fiber-
less Society, he is now an Associate Professor of elec-
trical and information technology. His main research
interests are channel measurements and modeling for
wireless communication, including channels for both
MIMO and UWB systems. In addition, he also works
with his company ResQU on wireless search and rescue equipment as well as
research projects on OFDM and UWB system design.
596 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Katsuyuki Haneda (S’03–M’07) received the
Doctor of Engineering degree from the Tokyo
Institute of Technology, Tokyo, Japan, in 2007.
He has been a Postdoctoral Researcher at the
SMARAD Centre of Excellence, Aalto Univer-
sity (former Helsinki University of Technology),
Espoo, Finland, since 2007. His current interests
are radio wave propagation measurements and
modeling, ultrawideband radio, and multiple-input
multiple-output systems.
Dr. Haneda is a member of the Institute of Elec-
tronic, Information and Communication Engineers (IEICE), Japan. He was the
recipient of the Student Paper Award presented at the 7th International Sympo-
sium on Wireless Personal Multimedia Communications (WPMC’04).
Veli-Matti Kolmonen received the M.Sc. degree in
technology from Helsinki University of Technology,
Espoo, Finland, in 2004 and the D.Sc. degree in tech-
nology from Aalto University School of Science and
Technology, Espoo, in 2010.
Since 2003, he has been with the Department of
Radio Science and Engineering, Aalto University
School of Science and Technology, as a Research
Assistant, Researcher, and currently as a Postdoc-
toral Researcher. His current research interests
include radio channel measurements and modeling.
Pertti Vainikainen received the degree of Master of
Science in technology, Licentiate of Science in tech-
nology and Doctor of Science in technology from
Helsinki University of Technology (TKK), Espoo,
Finland, in 1982, 1989 and 1991, respectively.
From 1992 to 1993, he was Acting Professor of
Radio Engineering, since 1993 Associate Professor
of Radio Engineering and since 1998 Professor in
Radio Engineering, all in the Radio Laboratory (since
2008 Department of Radio Science and Engineering)
of TKK (since 2010 Aalto University). In 1993–97 he
was the Director of the Institute of Radio Communications (IRC) of TKK, and a
Visiting Professor in 2000 at Aalborg University, Denmark, and in 2006 at Uni-
versity of Nice in France. His main fields of interest are antennas and propaga-
tion in radio communications and industrial measurement applications of radio
waves. He is the author or coauthor of six books or book chapters and about
350 refereed international journal or conference publications and the holder of
11 patents.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Land Mobile Satellite Dual Polarized
MIMO Channel Along Roadside Trees:
Modeling and Performance Evaluation
Michael Cheffena, Fernando Pérez Fontán, Frédéric Lacoste, Erwan Corbel, Henri-Jose Mametsa, and
Guillaume Carrie
Abstract—We present a novel physical–statistical, generative
model for the land mobile satellite (LMS), dual polarized, mul-
tiple-input-multiple-output (MIMO) channel along tree-sided
roads. Said model is parameterized by means of a physical model
based on the multiple scattering theory (MST) which accounts for
the signal attenuation and scattering by trees. Moreover, finite-dif-
ference time-domain (FDTD) electromagnetic computations were
performed to characterize the scattering pattern of an isolated
tree, and to calculate the MIMO shadowing correlation matrix
required by the model, and not provided by MST. This modeling
framework also encompasses the single-input-multiple-output
(SIMO)/space diversity case. To illustrate the capabilities of the
developed model, time series were generated and used in system
performance calculations. The obtained results give an insight
into the advantages of dual polarized MIMO and SIMO/space
diversity techniques in these very frequent scenarios and may help
service providers in evaluating the technical feasibility of such
systems.
Index Terms—Diversity, LMS MIMO, signal fading, vegetation.
I. INTRODUCTION
T HE true broadcast nature of satellite-based mobile com-
munication systems offers great advantages for delivering
multicast and broadcast services to both populated and isolated
areas. However, due to the high path loss, limited satellite
power, and other impairments, current land mobile satellite
(LMS) systems show lower capacities and availabilities than
those of terrestrial systems [1]. Their performance can be
increased using multiple-input-multiple-output (MIMO) tech-
niques. To design such systems, a good understanding of the
propagation channel is required.
Manuscript received November 08, 2010; revised February 09, 2011; ac-
cepted July 02, 2011. Date of publication October 25, 2011; date of current
version February 03, 2012. This work is supported by the French Space Agency
(CNES), French Aerospace Lab (ONERA) and Thales Alenia Space-France.
M. Cheffena is with Gjøvik University College, N-2815 Gjøvik, Norway
(e-mail: Michael.Cheffena@hig.no).
F. Pérez Fontán is with the University of Vigo, E-36200 Vigo, Spain (e-mail:
fpfontan@tsc.uvigo.es).
F. Lacoste is with CNES, F-31401 Toulouse, France (e-mail: Frederic.La-
coste@cnes.fr).
E. Corbel is with Thales Alenia Space—France, F-31037 Toulouse, France
(e-mail: Erwan.Corbel@thalesaleniaspace.com).
H.-J. Mametsa and G. Carrie are with ONERA, F-31055 Toulouse, France
(e-mail: Henri-Jose.Mametsa@onera.fr; Guillaume.Carrie@onera.fr).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173447
0018-926X/$26.00 © 2011 IEEE
597
Signal propagation in the LMS channel is affected by dif-
ferent propagation impairments, among which are signal atten-
uation and fading caused by vegetation. The signal attenuation
due to vegetation depends on a range of factors such as tree
type, whether the trees are in-leaf or out-of-leaf, whether the
trees are dry or wet, the frequency band, the path length through
foliage, etc. [2]. A single-input-single-output (SISO), roadside
tree LMS channel model which takes into account the signal
fading caused by position-dependent tree scattered fields and by
swaying tree components due to wind was reported in [3] and
[4]. Fade mitigation techniques (FMTs) such as MIMO might
be used to counteract the signal fading caused by vegetation.
Availability of realistic channel time series and detailed knowl-
edge of their correlation properties are needed for performing
accurate simulations [5]. Several studies on terrestrial MIMO
systems can be found in the literature, while studies on LMS
MIMO are rather limited, among them are [1] and [6]–[9].
In this paper, we present a novel physical–statistical, gener-
ative channel model for the roadside tree, LMS, dual-polarized
MIMO channel, which includes as a special case the single-
input-multiple-output (SIMO) case. Fig. 1 shows the consid-
ered propagation scenario where a mobile terminal receives par-
tially correlated co- and cross-polar, right- and left-hand cir-
cular polarized (R/LHCP) tree scattered fields from both sides
of the road. The wanted model requires parameterization; this
is achieved by means of a physical model based on the multiple
scattering theory (MST) [10]–[12]. This technique is used to es-
timate the signal attenuation and scattering by a single tree. MST
assumes a uniform distribution of scatterers in both azimuth [0,
) and elevation (0, ) [3], [4], [13]. This assumption makes
it impossible to quantify spatial effects such as the cross-cor-
relation of the shadowing affecting the direct signal reaching
the receive antennas. To overcome this, finite-difference time-
domain (FDTD) electromagnetic computations were performed
to accurately calculate the MIMO shadowing correlation matrix.
In addition, calculated tree scattering patterns are also shown,
which coincide with results obtained using MST.
To show the potential of the proposed physical–statistical,
generative model, examples of performance evaluation for
SISO, SIMO, and MIMO configurations were carried out using
synthesized time series. The results give an insight into the
advantages of polarization MIMO and SIMO/space diversity
techniques and may be used in assessing the technical feasi-
bility of LMS systems in roadside tree scenarios.
598
Fig. 1. Propagation scenario for the LMS dual-polarized MIMO: a mobile ter-
minal receives correlated co- and cross-polar (R/LHCP) tree attenuated and scat-
tered fields from both sides of the road.
The paper begins in Section II by presenting the FDTD elec-
tromagnetic computation results. Section III discusses the basis
of the proposed physical–statistical model which, in general, is
oriented toward a MIMO configuration but also encompasses
both the SISO and SIMO configurations. In this section, the
MST-based SISO channel model used for parameterizing direct
signal attenuation and multipath is briefly discussed. Then, we
discuss the SIMO case. Finally, we go on to present the dual-po-
larized, physical–statistical, generative MIMO channel model.
In Section IV, we present examples performance calculations
using the synthetic time series. Such calculations include ca-
pacity and diversity studies. Finally, the conclusions are given
in Section V.
II. FDTD COMPUTATION
FDTD is a numerical technique for solving electromagnetic
problems where the computation space is divided into small
cubic cells in which the time-domain fields are solved using
an explicit finite-difference update scheme. Using FDTD, ac-
curate solutions of the electromagnetic interaction with arbi-
trary-shaped objects can be obtained [14]. In this section, the
attenuation and scattering characteristics of a tree are computed
using an FDTD tool. The attenuation time series are further used
to extract the correlation matrix of the shadowing effects in the
LMS dual-polarized MIMO channel of interest.
The input to FDTD is a three-dimensional cubically meshed
tree model having a dielectric constant of 28 [13] and conduc-
tivity of 0.02 S/m [15]. The dimensions of the tree in the
and directions are 2.74, 2.64, and 3.17 m, respectively. The
FDTD computation space and the tree model used are shown
in Fig. 2. The computational volume is divided into a scattering
field region and a total field region (which also includes the inci-
dent wave) by a virtual plane on which a plane wave is incident
into the total field region. The whole region is surrounded by
a perfectly matched layer (PML) implementing the absorbing
boundary conditions needed to prevent false echoes from being
produced at the boundaries of the computational volume. A cell
size of 1 cm was utilized, which resulted in a highest usable
frequency of 3 GHz. The number of cells in the , and di-
rections were 373, 363, and 417, respectively, which resulted in
a total of 56 461 383 cells.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 2. FDTD computation zones and boundary conditions. Cell size of 1 cm
and total cells of 56 461 383 are used.
Two scenarios were investigated: the far-field scattering pat-
tern and the total field in a straight line along the axis behind
the tree (see the dashed line in Fig. 2). Two runs of the FDTD
tool were performed, one with a vertical and the other with a
horizontal polarized incident wave. These were then combined
to yield circularly polarized signals. A Gaussian pulse was used
as the source field. A near-to-far-field transformation was per-
formed to calculate the far-field scattering pattern.
Figs. 3 and 4 show the normalized far-field scattering pat-
tern of the co- and cross-polar components of the RHCP and
LHCP signals at 2 GHz, respectively. The results are obtained
by rotating the observation point 360 around the tree as shown
in Fig. 2, where 0 represents the line-of-sight (LOS) path be-
tween the satellite and the mobile terminal (through the tree).
We can observe that the scattering patterns of the co-polar com-
ponents have a forward lobe with an isotropic background, the
same as with MST, as reported in [3] and [4], or the radiative en-
ergy transfer (RET), as reported in [16]. Fig. 5 shows the total
received field along a straight line parallel to the axis behind
, the tree for the co- and cross-polar components of the RHCP
and LHCP signals. The points at the start (before 0 m) and at
the end of the straight line (after 2.64 m) are not blocked (LOS
conditions) toward the source while the rest are shadowed by
the tree (the tree is located between 0 to 2.64 m along axis).
The fields along the straight line are passed through a running
average filter (with ten point window size) for calculating the
large-scale fading (shadowing) effects caused by the tree (see
Fig. 6).
As we move along the positive axis direction, we can ob-
serve from Fig. 6 the change of state due to tree shadowing,
i.e., LOS-shadowing-LOS. Table I shows the MIMO shadowing
(large scale fading) correlation matrix , calculated using the
mentioned filtered FDTD series at 2 GHz. In Table I, R/R refers
CHEFFENA et al.: LAND MOBILE SATELLITE DUAL POLARIZED MIMO CHANNEL ALONG ROADSIDE TREES 599

Fig. 3. Scattering patterns of the co- and cross-polar components of the RHCP Fig. 5. Time series of the co- and cross-polar components of the RHCP and
signal at 2 GHz. LHCP signal at 2 GHz.

Fig. 4. Scattering patterns of the co- and cross-polar components of the LHCP Fig. 6. Shadowing time series of the co- and cross-polar components of the
signal at 2 GHz. RHCP and LHCP signal at 2 GHz.

TABLE I
to the co-polar signal from an RHCP transmit antenna to an MIMO SHADOWING CORRELATION COEFFICIENTS
RHCP receive antenna, R/L refers to the cross-polar signal from OBTAINED FROM FDTD SIMULATIONS
an RHCP transmit antenna to an LHCP receive antenna, and so
on.
Note that the FDTD results presented in this section are only
applicable when the tree is in leaf. There exist many very dif-
ferent tree configurations, and the shadowing cross-correlation
may vary dramatically from tree to tree. This is an issue that
requires further investigation, to be carried out at a later time,
the main aim of this paper being the description of the phys-
SISO model and going on to present the SIMO and, finally, the
ical–statistical model discussed next and how such a model can
MIMO models.
be parameterized.
Physical–statistical models are deemed more accurate than
purely statistical ones as they take into account the geometry of
III. PHYSICAL–STATISTICAL, GENERATIVE ROADSIDE TREE the link(s), while they rely on physical (electromagnetic-based)
LMS DUAL-POLARIZED MIMO CHANNEL MODEL methods for calculating the needed model parameters. The un-
Here, a family of physical–statistical, generative models for derlying assumption made in this family of models is that the
LMS roadside tree environments is presented, starting with the received signal is composed of the direct component, which
600
Fig. 7. Tree model for the LMS channel with randomly distributed and oriented
leaves and branches. , , and are the orthonormal vectors of rectangular
coordinate system.
may be subjected to shadowing and accompanied by so-called
coherent component fluctuations, and the multipath component
due to diffuse scattering, i.e., a Ricean channel.
Thus, we implement a statistical random number generator
fitting a Rice distribution whose parameters are changing as the
terminal travels through a propagation environment described in
terms of simple, canonical volumetric forms representing trees.
Time variations are also taken into account, as mentioned ear-
lier, especially for stationary terminals.
The two Rice distribution parameters (direct signal ampli-
tude and multipath power) are calculated using MST: the di-
rect signal is affected by the specific attenuation of the tree
canopy provided by MST once the geometrical path through the
canopy is worked out for the different terminal route positions.
The second component, the multipath power, is also parame-
terized by means of MST, which provides the average coherent
and incoherent scattered powers; these are mapped to the rel-
evant Rice distribution parameter, also taking into account the
possible paths through and off the tree canopies on both sides
of the street.
At least two alternatives are possible for generating locally
Ricean time series: one is using a point-scatterer approach (used
in the current SISO and SIMO implementations), and the other
is by filtering a complex random Gaussian series. The latter op-
tion has been selected for implementing the MIMO channel.
A. LMS SISO Channel Model
A SISO model/simulator was reported by the authors in [3]
and [4]. The model uses MST as described in [10]–[13] with
modifications to account for slant paths and circular polariza-
tion. In the model, the tree canopy is modeled as a vertically ori-
ented cylindrical volume in a rectangular coordinate system
defined by the orthonormal vectors , , and . The canopy
volume contains randomly distributed and oriented leaves and
branches. Leaves are modeled as thin lossy dielectric disks,
branches as finite lossy dielectric cylinders, and the trunk as a
finite lossy dielectric cylinder (see Fig. 7).
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
The SISO model generates the position-dependent signal
fading by calculating the total coherent (the sum of coherently
scattered and free-space field) and incoherent tree scattered
fields. The phase variations in the scattered signal are accounted
for using a point-scatterer approach, where the contribution
from each tree is assumed to originate from a point in the
center of the canopy from which the distance-dependent phase
variations for the various route sampling points are calculated.
In addition, the signal fading due to wind swaying is ac-
counted for by modeling the tree components as masses attached
to springs. The position-dependent fading and the signal fading
due to tree swaying are then combined to give the overall signal
fading caused by vegetation. The model is applicable for dif-
ferent polarizations and elevation angles and was validated in
terms of first- and second-order statistics using measurements
at 2 GHz (see [3] and [4] for more details).
B. SIMO Receiver Space Diversity
The previously described SISO channel model can easily be
extended to the SIMO/space diversity case. The model is again
based on the physical–statistical tree model (see Fig. 7) reported
in [3] and [4] and MST. The considered propagation scenario is
shown in Fig. 8: two antennas with the same polarization sepa-
rated by a given distance are receiving tree attenuated and scat-
tered co-polar signals from each side of the road. The fades at
each antenna element are generated by taking into ac-
count the tree scattered fields from each side of the road as well
as the LOS field, given by
(1)
(2)
(3)
where and are the number of trees on each side of the
road, is the total number of scatterers in a single tree, and
is the wavenumber. Note here how the overall scattered signal
from one single tree has now been split into components, pro-
vided that the link budget of the scattered power and the MST
constraints are preserved. This shows a possible way toward a
wideband model. Parameters and are the am-
plitudes of the incoherent scattered components from side one
(satellite side) and side two (opposite side) of the road, respec-
tively. Parameters and are the amplitudes of
the coherent scattered field (side one) and the LOS field, respec-
tively.
The coherent and incoherent scattered fields can be estimated
from the physical–statistical tree model reported in [3] and [4].
Parameters and are the initial phase and path length of
each signal component, respectively. The coherent and inco-
herent scattered fields as well as the LOS field are received from
one side of the road (see (1)), while only the incoherent scattered
CHEFFENA et al.: LAND MOBILE SATELLITE DUAL POLARIZED MIMO CHANNEL ALONG ROADSIDE TREES 601

Fig. 8. Roadside tree LMS, SIMO/space diversity propagation scenario: two

antennas with the same polarization separated by a given distance are receiving
the tree attenuated and scattered co-polar signal from both sides of the road.

Fig. 9. Small-scale fading (absolute values) correlation coefficient versus an-

tenna separation distance (for two co-polar RHCP antennas) for the simulated
fields are received from the other side of the road (see (2)). These roadside tree LMS SIMO/space diversity (solid line) channel. Also shown is
are combined to produce the signal received at each antenna po- the theoretical curve of the square of the zeroth-order Bessel function of the
sition (see (3) and Fig. 8). first kind (dashed line).
The resulting simulated correlation coefficient versus antenna
separation distance for the small scale fading (absolute values)
is shown in Fig. 9 (for two co-polar RHCP antennas). As ex-
pected, we can observe that the correlation coefficient decreases
with increasing antenna separation distance. The resulting sim-
ulated correlation coefficient was also fitted to the square of
the zero-order Bessel function of the first kind, ,
where is the relative separation distance, and is the wave-
length (see the dashed line in Fig. 9). It can be seen that the fit
between the two curves is quite good. Thus, a model based on
the zeroth-order Bessel function of the first kind could be used
to model the correlation coefficient of the small scale fading in
the roadside tree LMS SIMO/space diversity channel. Fig. 10
shows examples of the simulated received signal time series on
the two antennas (RHCP co-polar antennas) for a separation dis-
tance of one wavelength. It can be seen that the time series have
similar large-scale fading characteristics but relatively decorre-
lated small-scale fading due to the separation distance between
the two antennas. Fig. 10. LMS SIMO receiver space diversity links at antenna separation dis-
tance of one wavelength (for two co-polar RHCP antennas).

C. LMS Dual-Polarized MIMO Channel Model

The SISO channel model discussed in Section III-A can be correlated processes that are the inputs to the shadowing filter,
extended to the MIMO case by taking into account the corre- see Fig. 11, are given by
lation between the co- and cross-polar channels. Table I gives
the MIMO correlation matrix for the large scale fading obtained (4)
using FDTD. The correlation matrix for the small scale fading
(multipath) can be found by assuming independent multi- where are partially correlated processes.
path fading on the MIMO channels as reported in [1] and [17]. are independent real-valued white
As previously indicated, the underlying assumption is a set of Gaussian processes with zero-mean and unit variance. is the
locally Ricean processes whose parameters are slowly variant Cholesky decomposition of .
along the traveled route. The slow-gading correlation proper- The complex white Gaussian processes (for
ties are then imposed on the direct signal component of the Rice ) are filtered (according to [18]) for Doppler
process, and the fast variation correlation properties are imposed spectrum shaping ( is the Doppler filter) (see Fig. 11).
on the multipath component. The resulting time series are then multiplied by the standard
The proposed model for the roadside tree LMS dual-polar- deviation of the noncoherent component of each channel
ized MIMO channel is presented in Fig. 11. In the model, the (estimated by means of MST [3] and [4]) to produce the small
602 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 11. LMS dual-polarized MIMO channel simulator for roadside trees. and (for ) are independent complex and real-valued white
Gaussian processes with zero mean and unit variance, respectively. is the Doppler filter. is the Cholesky factorization of . are partially correlated
processes. and are the standard deviation of the incoherent and the mean total coherent (which is the sum of coherently scattered and free-space field)
component, respectively. is the standard deviation of the total coherent component. and are the small and large scale fading, respectively.
is the fading due to swaying vegetation. is the signal phase corresponding to the traveled distance. is the Doppler shift of the direct signal component.
is a correlated complex signal envelope of the LMS dual-polarized MIMO channel.

scale fading. Similarly, the partially correlated processes, , parameter and then added to the attenuated direct signal
are low-pass filtered by a shadowing filter with transfer function (which are estimated using MST [3] and [4]). The time
given by [19] series are then converted to linear scale (i.e., to lognormal dis-
tributed time series) to obtain the large-scale fading. The small-
(5) and large-scale fading processes of each channel are summed
and weighted by (estimated from the physical–statistical
with tree model reported in [3] and [4]) to account for the signal
fading caused by wind swaying. Finally, the Doppler shift of the
(6) direct signal component is accounted for by multiplying each
channel by , where is the phase corresponding to
where is the sampling distance and is the correla- the traveled distance. The outputs of the simulator are partially
tion distance. Note that shadowing is only applied to the co- correlated complex signal envelopes of the dual-polarized 2 2
herent part [19]. can slightly differ between the co- and MIMO channel, which incorporate the total fading caused by
cross-polar shadowing channels [1]. Thus, the filtering step can vegetation. Note that the model is also applicable in the nomadic
modify the correlation coefficient between two channels. To case by assuming that wind swaying is the main cause of signal
deal with this effect, can be weighted by a correction co- fading.
efficient prior to taking its Cholesky factorization [17]. The cor- Fig. 12 shows examples of simulated partially correlated
rection coefficient is given by [17] roadside tree LMS dual-polarized MIMO channel time series.
The system parameters used in the simulations are given
(7) in Table II. Sizes, densities, and dielectric constants of tree
components reported in [13, Table I] were used. The periodic
where and are the filter coefficients for the co- and fading in the time series of the co-polar components indicate
cross-polar channels as defined in (6). The outputs of the shad- the presence of trees along the road. As expected, the simulated
owing filter are first weighted by the total coherent component cross-polar levels are lower than the co-polar components.
CHEFFENA et al.: LAND MOBILE SATELLITE DUAL POLARIZED MIMO CHANNEL ALONG ROADSIDE TREES 603

Fig. 12. Simulated partially correlated 2 2 dual-polarized MIMO channel Fig. 13. CDFs of the LMS channel capacities along the roadside tree scenario
time series. Simulation parameters are given in Table II. simulated both for SISO and dual polarized MIMO cases. The system simulation
parameters are given in Table II. A signal-to-noise ratio of 20 dB is used.

TABLE II
SYSTEM SIMULATION PARAMETERS
channel matrix, , and denotes the matrix determinant.
For the SISO case, the capacity is given by [20]

b/s/Hz (10)

where is the normalized complex channel gain relative to the

LOS.
Fig. 13 presents the cumulative distribution functions (CDFs)
of the LMS channel capacities for the dual-polarized MIMO
and SISO cases calculated using (9) and (10), respectively. The
channel time series are generated using the simulator shown in
Fig. 11 using the parameters listed in Table II. As expected,
we can observe the increase in capacity achieved with MIMO
compared to the SISO case.

B. Link Diversity
IV. PERFORMANCE EVALUATION Diversity improves link availability. The amount of diversity
gain/improvement obtained depends on the degree of correla-
A. Channel Capacity tion between the various parallel links which, in turn, depends
For a narrowband MIMO system with transmit and on the propagation environment. For the dual-polarized MIMO
receiver antennas, the receive signal is expressed as case, the instantaneous signal-to-noise ratio at the output of the
maximum-ratio-combiner (MRC) is given by [21]–[23]
(8)
(11)
where is the transmitted signal, is the noise, and is the
channel matrix, with being the channel response where is the largest eigenvalue of for (or
between the th transmit antenna and the th receive antenna. for ). For SIMO case
The capacity is given by [20]
(12)
b/s/Hz (9) where and are the complex envelopes of the SIMO
channel.
where is the identity matrix of size is the average Fig. 14 presents the CDFs of the received signal for the LMS
signal-to-noise ratio, is the complex transpose of the MIMO roadside tree scenario for SISO, SIMO/space diversity (for two
604
Fig. 14. CDFs of the LMS received signal along roadside tree areas for SISO,
SIMO receiver space diversity (for two co-polar RHCP antennas, one wave-
length apart) and dual-polarized MIMO diversity links. The system simulation
parameters are given in Table II.
co-polar RHCP antennas, one wavelength apart), and dual-po-
larized MIMO diversity with MRC. We can observe the sig-
nificant diversity gain achieved with dual-polarized MIMO and
SIMO/space diversity compared to the SISO case. It can also
be seen that the best performance is achieved with dual-po-
larized MIMO diversity. This is because, first, the cross-polar
components are not lost and contribute to increase the sum of
signal-to-noise ratios at the output of the MRC. Second, we can
observe from Fig. 12 that the cross-polar components are espe-
cially useful when the link is shadowed by the roadside trees.
At these positions, the levels of the co- and cross-polar compo-
nents are comparable (see Fig. 12), which results in significant
diversity gains.
V. CONCLUSION
In this paper, we studied the LMS dual-polarized MIMO
channel along roadside tree areas. A novel physical–statistical,
generative tree model based on multiple scattering theory
was proposed for calculating the attenuation and scattering
effects. FDTD electromagnetic computations were performed
to characterize the tree scattering pattern and to calculate
the shadowing cross-correlation matrix of the dual-polarized
MIMO channel. Generally, the use of FDTD has advantages
in obtaining an accurate solution of electromagnetic interac-
tion with arbitrary objects; however, it also has limitations in
terms of the size of the computation volume relative to the
wavelength, which is directly related to the required simulation
time. This means that performing FDTD simulations over large
computation areas/volumes may not be feasible, even using
parallel computing.
In addition, a channel model/simulator for the SIMO receiver
space diversity was also presented using the physical–statis-
tical tree model, thus extending the SISO model in [3] and [4].
The physical–statistical approaches discussed are more effec-
tive for simulating large scenarios compared to complex phys-
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
ical models, and more accurate than merely empirical or statis-
tical models.
In addition, performance analyses of the LMS SISO, dual-
polarized MIMO and SIMO/space diversity configurations were
carried out. Compared to the SISO case, significant increases in
performance were observed for the dual-polarized MIMO and
SIMO/space diversity cases. The results obtained can be used
to evaluate the technical feasibility of LMS systems in roadside
tree scenarios. They also give an insight into the kinds of system
performance analyses that could be carried out using the models
developed in this paper.
In general, the proposed channel models can be used to
carry out different system-level analyses such as capacity,
bit-error-rate, etc. Future work includes validating the devel-
oped LMS dual-polarized MIMO and SIMO/space diversity
channel models using measurements, as well as carrying out
further comparisons between MST and FDTD results.
ACKNOWLEDGMENT
The authors would like to thank the French Space Agency
(CNES), French Aerospace Lab (ONERA) and Thales Alenia
Space—France for supporting this work. The authors would
also like to thank X. Ferrieres and E. Bachelier of ONERA for
helping with the FDTD computations. In addition, the authors
would like to thank L. Castanet of ONERA for his useful com-
ments.
REFERENCES
[1] P. R. King and S. Stavrou, “Low elevation wideband land mobile satel-
lite MIMO channel characteristics,” IEEE Trans. Wireless Commun.,
vol. 6, no. 7, pp. 2712–2720, Jul. 2007.
[2] M. O. Al-Nuaimi and A. M. Hammoudeh, “Measurements and predic-
tions of attenuation and scatter of microwave signals by trees,” Proc.
Inst. Electr. Eng. Microw. Ant. Prop., vol. 141, no. 2, pp. 70–76, Apr.
1994.
[3] M. Cheffena and F. P. Fontan, “Land mobile satellite channel simulator
along roadside trees,” IEEE Antennas Wireless Propag. Lett., vol. 9, pp.
748–751, 2010.
[4] M. Cheffena and F. P. Fontan, “Channel simulator for land mobile
satellite channel along roadside trees,” IEEE Trans. Antennas Propag.,
vol. 59, no. 5, pp. 1699–1706, May 2011.
[5] G. J. Byers and F. Takawira, “Spatially and temporally correlated
MIMO channels: Modeling and capacity analysis,” IEEE Trans. Veh.
Technol., vol. 53, no. 3, pp. 634–643, May 2004.
[6] C. Oestges, “A stochastic geometrical vector model of macro- and
megacellular communication channels,” IEEE Trans. Veh. Technol.,
vol. 51, no. 6, pp. 1352–1360, Nov. 2002.
[7] M. Sellathurai, P. Guinand, and J. Lodge, “Space-time coding in mobile
satellite communications using dual-polarized channels,” IEEE Trans.
Veh. Technol., vol. 55, no. 2, pp. 188–199, Jan. 2006.
[8] G. Alfano and A. D. Maio, “A theoretical framework for LMS MIMO
communication systems performance analysis,” in Proc. 3rd Int. Wave-
form Diversity Design Conf., Jun. 4–8, 2007, pp. 18–22.
[9] K. P. Liolis, J. Gomez-Vilardebo, E. Casini, and A. Perez-Neira, “On
the statistical modelling of MIMO land mobile satellite channels: A
consolidated approach,” presented at the 27th AIAA Int. Commun.
Satellite Syst. Conf. (ICSSC), Edinburgh, U.K., Jun. 2009.
[10] L. L. Foldy, “The multiple scattering of waves,” Phys. Rev., vol. 67,
no. 3, pp. 107–119, 1945.
[11] M. X. Lax, “Multiple scattering of waves,” Rev. Mod. Phys., vol. 23,
no. 4, pp. 287–310, 1951.
[12] S. A. Torrico, H. L. Bertoni, and R. H. Lang, “Modeling tree effects on
path loss in a residential environment,” IEEE Trans. Antennas Propag.,
vol. 46, no. 6, pp. 872–880, Jun. 1998.
[13] Y. L. C. de Jong and M. H. A. J. Herben, “A tree-scattering model
for improved propagation prediction in urban microcells,” IEEE Trans.
Veh. Technol., vol. 53, no. 2, pp. 503–513, Mar. 2004.
CHEFFENA et al.: LAND MOBILE SATELLITE DUAL POLARIZED MIMO CHANNEL ALONG ROADSIDE TREES
[14] K. S. Yee, “Numerical solution of initial boundary value problems
involving Maxwells equations in isotropic media,” IEEE Trans. An-
tennas Propag., vol. AP-14, no. 3, pp. 302–307, May 1966.
[15] A. A. Chukhlantsev, A. M. Shutko, and S. P. Golovachev, “Conduc-
tivity of leaves and branches and its relation to the spectral dependence
of attenuation by forests in meter and decimeter band,” in Proc. IEEE
Int. Geosci. Remote Sens. Symp., Jul. 21–25, 2003, pp. 1103–1105.
[16] N. C. Rogers, A. Seville, J. Richter, D. Ndzi, N. Savage, R. Caldeir-
inha, A. K. Shukla, M. Al-Nuaimi, K. Craig, E. Vilar, and J. Austin,
“A generic model of 1–60 GHz radio propagation through vege-
tation—Final report,” in Radio Agency, U.K., May 2002 [Online].
Available: http://www.ofcom.org.uk/static/archive/ra/topics/research/
topics/propagation/vegetation/vegetation-finalreportv1_0.pdf
[17] G. Carrie, J. Lemorton, and F. Perez-Fontan, “State-of-the-art review
of LMS MIMO channel models at s and c band upgrade of the validity
domain of the available channel models,” ONERA, Toulouse, France,
Tech. Rep., Apr. 2010.
[18] T. Aulin, “A modified model for the fading signal at a mobile radio
channel,” IEEE Trans. Veh. Technol., vol. 28, no. 3, pp. 182–203, Aug.
1979.
[19] F. Perez-Fontan and P. M. Espineira, Modelling the Wireless Propa-
gation Channel: A Simulation Approach With MATLAB. Chichester,
U.K.: Wiley, 2008, ISBN: 978-0-470-72785-0.
[20] J. G. Foschini and J. M. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wireless Pres.
Commun., vol. 6, no. 3, pp. 311–335, 1998.
[21] T. K. Y. Lo, “Maximum ratio transmission,” IEEE Trans. Commun.,
vol. 47, no. 10, pp. 1458–1461, 1999.
[22] C. H. Tse, K. W. Yip, and T. S. Ng, “Performance tradeoffs between
maximum ratio transmission and switched-transmit diversity,” in Proc.
11th IEEE Int. Symp. Personal, Indoor, Mobile Radio Commun., 2000,
pp. 1485–1489.
[23] M. Kang and M. S. Alouini, “Largest egenvalue of complex Wishart
matrices and performance analysis of MIMO MRC systems,” IEEE J.
Sel. Commun., vol. 21, no. 3, pp. 418–431, Apr. 2003.
Michael Cheffena received the M.Sc. degree in elec-
tronics and computer technology from the University
of Oslo, Norway, in 2005 and the Ph.D. degree from
the Norwegian University of Science and Technology
(NTNU), Trondheim, Norway, in 2008.
For one year, he was a visiting researcher at the
Communications Research Centre (CRC), Canada.
From 2009 to 2010, he made a postdoctoral study
at the University Graduate Center (UNIK), Kjeller,
Norway. From 2010 to 2011, he was a Postdoc-
toral Fellow at the French Space Agency (CNES),
Toulouse, France. Currently, he is an Associate Professor at Gjøvik University
College, Norway. His research interests include modelling and prediction of
radio channels for both terrestrial and satellite links.
605
Fernando Pérez Fontán was born in Villagarcı́a
de Arosa, Spain. He received the Diploma degree
in telecommunications engineering and the Ph.D.
degree from the Technical University of Madrid,
Spain, in 1982 and 1992, respectively.
He is a Full Professor with the Telecommunica-
tions Engineering School, University of Vigo, Vigo,
Spain. He is the author of a number of international
magazine and conference papers. His main research
interest is in the field of mobile fixed radio commu-
nication propagation channel modeling.
Frédéric Lacoste received the Engineer degree from
Ecole Nationale Supérieur d’Ingénieurs en Construc-
tion Aéronautique et Spatiale (ENSICA), France, in
2002 and the Ph.D. degree in EM waves propagation
from Ecole Nationale Supérieure de l’Aéronautique
et de l’Espace (SUPAERO), France, in 2005.
In 2005, he joined the French Space Agency
(CNES), Toulouse, France, where he is currently
working on propagation channel modeling and
measurements for fixed and mobile Earth–Space
communication systems. He was involved in various
European and worldwide working groups dealing with Earth–Space propa-
gation modeling, such as SatNEx I and II, COST 280, and currently COST
IC0802, NoE-EWP, and ITU-R study group 3.
Henri-Jose Mametsa received the Diploma degree in telecommunication en-
gineering and the Ph.D. degree in electronics from the Ecole Nationale des
Telecommunications de Bretagne, France, in 1984 and 1986, respectively. He
co-directed Ph.D. students and received a postdoctoral degree (required to ad-
vise doctoral students or to oversee research) from the Université Paul Sabatier,
Toulouse, France.
He worked as an Engineer in the Antennas Design Department of
Thomson-CGR before joining the Electromagnetism and Radar Department
of the Office National d’Etudes et de Recherches Aerospatiales (ONERA),
Toulouse, France, where he has been working in different studies on electro-
magnetic scattering problems. His current research interests include modeling
electromagnetic scattering from man-made targets and natural media.
Guillaume Carrie was born on June 20, 1979. He
received the Engineer Diploma degree in aeronau-
tics and the D.E.A. degree in signal processing from
ENSICA, Toulouse, France, in 2003 and the Ph.D.
degree in array processing for GNSS receivers from
SUPAERO, Toulouse, France, in 2006.
From November 2006 to year-end 2007, he was
working in SILICOM as a Research Engineer on the
development of experimental receivers and on simu-
lations, both for Galileo systems. Since January 2008,
he has been working in the Radio Communication
and Propagation Research Unit of the Electromagnetism and Radar Department
of ONERA, Toulouse, France. His current research interests are mainly focused
on propagation channel modeling both for fixed and mobile satellite systems
and on channel adaptive GNSS receivers.
606 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Empirical-Stochastic LMS-MIMO Channel Model

Implementation and Validation
Peter R. King, Tim W. C. Brown, Member, IEEE, Argyrios Kyrgiazos, and Barry G. Evans, Senior Member, IEEE

Abstract—Land Mobile Satellite (LMS) networks, forming a key

component of future mobile Internet and broadcasting, can benefit
from Multiple-Input Multiple-Output (MIMO) techniques to im-
prove spectral efficiency and outage. LMS-MIMO networks can
be obtained using multiple satellites with single polarization an-
tennas with spatial multiplex channel coding, or by a single satel-
lite with dual polarization antennas providing polarization multi-
plex channel coding. In this paper, a guide is presented showing the
steps required to implement a simple empirical-stochastic dual cir-
cular polarized LMS-MIMO narrowband channel model with val-
idation both with and without a line of sight. The model is based
on an S-band tree-lined road measurement campaign using dual
circular polarizations at low elevations. Application of the model Fig. 1. Diagram of satellite MIMO channel structure.
is aimed at LMS-MIMO physical layer researchers and system de-
signers, who need an easy to implement and reliable model, repre-
sentative of typical LMS-MIMO channel conditions.
MIMO case, where its stochastic properties are derived from an
Index Terms—Channel model, Land Mobile Satellite (LMS), S-band tree-lined road measurement campaign using dual cir-
multiple-input multiple-output (MIMO), propagation, stochastic. cular polarizations at low elevations. Furthermore, this model
in particular compared to existing models to date [8], [9] con-
I. INTRODUCTION siders the interdependence between the small scale fading. The
authors have found the most appropriate means to accommodate
this interdependence is to use a Ricean fading model, where the

T HE benefit of applying multiple-input multiple-output

(MIMO) techniques to the Land Mobile Satellite (LMS)
channel, in terms of capacity gain and diversity gain, was shown
in [1] and [2] respectively. In recent years, the use of MIMO in
terrestrial wireless systems including the next generation wire-
less networks, IEEE 802.11n [3] as well as wide area networks,
IEEE 802.16m [4] and long term evolution of third generation
mobile (3GPP LTE) [5] has become widespread. More recently
the use of MIMO for LMS has gained interest with regards to
satellite based digital video broadcasting standards DVB-SH
and DVB-NGH [6]. However, before LMS-MIMO systems are
in widespread use, suitable and simple to implement channel
models that give a general model of the radio environment are
required. This paper defines an empirical-stochastic channel
model for such use.
As is typical for LMS channel models, as well as the model
presented in this paper, large scale fading comprises Markov
Chains to represent the “on/off” nature of the channel and fil-
tered log-normal simulation to represent the shadowing effects,
and Ricean simulation to represent the small scale fading ef-
fects [7]. However, this model extends the capabilities to the
Manuscript received November 29, 2010; revised September 23, 2011; ac-
cepted October 10, 2011. Date of publication October 25, 2011; date of current
version February 03, 2012. This work was supported in part by the EU FP6
project SatNex (Satellite Communications Network of Excellence).
The authors are with the Centre for Communication Systems Research,
University of Surrey, Surrey GU2 7XH, U.K. (e-mail: p.king@surrey.ac.uk;
t.brown@surrey.ac.uk; a.kyrgiazos@surrey.ac.uk; b.evans@surrey.ac.uk).
Digital Object Identifier 10.1109/TAP.2011.2173448
co-polar and cross-polar components are suitably correlated.
Section two of this paper describes the typical factors unique
to a dual circular polarization LMS-MIMO channel modeling,
including the need for a four-state Markov chain to form si-
multaneous dual polar shadowing models as well as small-scale
fading models. This is followed by sections detailing the mea-
surement campaign carried out to show the Markov chain be-
havior in the LMS-MIMO channel. A step by step guide is then
provided in generating the proposed empirical-stochastic model
informed by measurement data. The reader can use the infor-
mation presented to implement the necessary code for such a
model. Finally validation tests are shown to clarify the model’s
application in both line of sight and non line of sight regions.
II. LMS-MIMO CHANNEL MODEL CONSTRUCTION
The structure of an LMS-MIMO channel model is a 2 2
MIMO system whereby the two antennas at each end are dual
circular polarized with right hand and left hand circular polar-
izations (RHCP and LHCP) as illustrated in Fig. 1.
The model presented in this paper is considered to be suit-
able for the L-band and S-band frequency ranges such that the
ionospheric and tropospheric effects are considered negligible.
The largest impact would be Faraday rotation in the ionosphere
[7], which would be overcome using circular polarization. Ig-
noring any ionospheric and tropospheric effects, the main item
of interest for satellite MIMO is the multipath caused by local
scatterers near to the mobile. The scattering caused in this re-
gion will create some depolarization from RHCP to LHCP and
from LHCP to RHCP, which are represented in a 2 2 MIMO

0018-926X/$26.00 © 2011 IEEE

KING et al.: EMPIRICAL-STOCHASTIC LMS-MIMO CHANNEL MODEL IMPLEMENTATION AND VALIDATION
channel matrix, , where there are two co-polar (RHCP to
RHCP and LCHP to LHCP) and two cross-polar circularly po-
larized channels (RHCP to LHCP and LCHP to RHCP). These
four channels are conveniently represented as follows with sub-
scripts R and L
(1)
where the channel matrix is used as the multiplicative compo-
nent within the channel so that the output signal vector received
at the mobile, relates to the input signal vector by the
following equation:
(2)
where is a vector to represent additive white Gaussian noise
at the receiver. It is therefore of interest to model the channel
state at time , . The simplest means to model the LMS-
MIMO channel is to use a stochastic random process, which will
define the distribution of channel states (or first order statistics),
but it is also necessary to define how the channel evolves, by
considering what is known as the second order statistics.
For satellite MIMO, it is best to break the channel into
three components:
1) Free space path loss—This is defined theoretically by the
well known Friis’ formula [7] though for convenience of
being able to analyze the channel, it is normalized in this
case, since it is merely an offset value.
2) Shadowing or large scale fading—When the mobile is on
the ground, it may either have a direct line of sight link to
the satellite, or there may be a building, tree or other large
scattering object blocking the direct path. Thus the mobile
is within the shadow of the scatterer and will be subject
to extra path loss. In the case of satellite communications,
the mobile is constantly moving in and out of the shadow
regions as illustrated in Fig. 2. Here the mobile is a ve-
hicle moving along the road, it will enter regions between
scatterers on its left hand side where it will have a line of
sight link with the satellite. Therefore there is a need to
statistically model scenarios where there is both high and
low shadowing. Therefore it is necessary to model when
the mobile is switching between high and low shadowing,
which is best achieved by using a Markov chain [10].
3) Small scale fading—In the local area around the mobile
there will be several scattering objects, which will produce
reflected, refracted and diffracted signals. As the mobile
moves, these reflections, refractions and diffractions are
constantly changing and thus the received signal is con-
stantly changing. In some instances, the reflected signals
will add up constructively in phase, while in other cases
they will add up destructively out of phase and the received
signal will go into a deep fade.
It is therefore necessary to separately model the small scale
and large scale fading characteristics of the channel as it
changes over time with mobile movement.
For LMS-MIMO, there are a number of modeling challenges
not addressed in other LMS channel models that must be con-
sidered and which have led to the construction of a simple to
607
Fig. 2. Diagram illustrating an example of high and low shadowing regions.
implement stochastic model provided in this paper. The factors
that require consideration include the following:
• The MIMO branches are in the circular polarization do-
main, rather than the spatial domain. Therefore, channel
multiplexing occurs in this domain, which will give dif-
ferent characteristics in the eigen decomposition of the
channel, as will be seen from measurement data later in this
paper. Therefore different eigen characteristics [11] from
conventional MIMO channel models need to be modeled.
• The large scale fading changes constantly when switching
from high to low shadowing. This produces different
characteristics for the co-polar and cross-polar channel
branches and likewise should be modeled to be consis-
tent with simultaneous measurements of these channels.
Furthermore the shadowing characteristics should be com-
pared for both polarizations so that their interdependence
is maintained.
• The small scale fading channels for each of the four
MIMO paths are not necessarily independent, especially
when there is a line of sight link. Therefore in such cir-
cumstances, the interdependence must be appropriately
modeled so that both the model and real measured data
have comparable eigen decomposition.
Before developing models, the Markov chain, polarization
and interdependence characteristics need to be identified from
real measurement data, which the next section of this paper ad-
dresses. After describing the measurements, this paper describes
the stages involved in creating the large scale fading character-
istics, Markov chain and small scale fading.
III. MEASUREMENT SETUP
Extensive measurements were carried out on the edge of the
town of Guildford, UK, representative of a suburban/rural area
that would be applicable to LMS-MIMO during the summer
time (with trees in full foliage). An artificial terrestrially based
platform (acting as a satellite) was placed on top of a hill over-
looking a road, as illustrated in Fig. 3 containing directional
RHCP and LHCP antennas, spaced just under one wavelength
apart. Each antenna had a gain of 12 dBi and a 3 dB beamwidth
of 30 . A mobile van contained the receiver and its roof was
fitted with an omnidirectional RHCP and LHCP antenna spaced
four wavelengths apart. These two antennas had a beamwidth
of 70 in elevation. The satellite elevation angles ranged from
7 to 18 as the mobile moved along the tree-lined road. Al-
though many operational satellite elevations exceed these, some
geostationary and low Earth orbiting mobile satellite services
are required to work at low elevations. The present experiment
therefore represents these as worst case system scenarios, where
608 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 3. Diagram of measurement setup of artificial platform.

such elevations enable the highest degrees of multipath. Varying Fig. 4. Sample measurement showing the Markov chain characteristics of the
levels of Rice factor, as defined by the Ricean distribution, as LMS-MIMO channel.
well as channel correlations provide a usefully wide range of
empirical results.
An Elektrobit Propsound wideband MIMO channel sounder In order for these two sets of four shadowing models to
was configured for a carrier frequency of 2.45 GHz and a band- have interdependence, a 4 4 correlation matrix for large
width of 200 MHz. Each MIMO channel was obtained sequen- scale fading, , is then applied [7] to both high and low
tially by using fast switching at a rate of 152.7 Hz, which is over shadowing as analysis of measurement data has shown both
twice the maximum Doppler shift of 73 Hz for the vehicle speed, shadowing cases to follow the necessary Gaussian distribution.
thus meeting Nyquist criteria. This is assumed to be the case on This will accommodate the interdependence between the four
the tree-lined road where there were few vehicles in motion. MIMO branches so that correlated shadowing and
Within the data, suitable results were found whereby the is formed as follows:
channel could be captured within the channel sounder’s sensi-
tivity giving a signal to noise ratio that enabled the multipath
to be measured without error due to receiver noise. There were
many cases of interference from local wireless area networks
(3)
that had to be filtered out from the measurement data and
replaced with interpolated data. Data sampling being more than
where the 1/2 denotes the Cholesky factorization. An appro-
the required Nyquist criteria meant it was possible to achieve
priate set of values for are taken from the measurement
this.
data representing a typical tree lined road in a suburban envi-
An example of the measurement data obtained to illustrate the
ronment. To assist with understanding the matrix formation, the
Markov chain behavior of the high and low shadowing is shown
matrix has notations for each element where as an example, the
in Fig. 4 for all four MIMO branches. For clarity, the free space
correlation, gives the correlation of the shadowing
loss (FSL) is normalized out of the channel. Clearly it can be
between the right to right hand branch and the left to left hand
seen that there are different small scale fading characteristics in
branch. Therefore all sixteen possible permutations of this nota-
high and low shadowing regions, as well as Markov chain char-
tion are shown in the matrix as follows, from which values from
acteristics that are related for co-polar and cross-polar channels
the measured data are then given
though they are required to be generated separately. The next
section will explain in detail how the large scale fading, Markov
chain and small scale fading are generated in three separate steps
and then integrated in order to form a working channel matrix.

IV. MODEL GENERATION

A. Step 1—Generate Large Scale Fading for High and Low
Shadowing
For 2 2 MIMO, four simultaneous models of high shad-
owing that will vary over distance (i.e. in non line of sight re-
gion) are required to produce vector and four simultaneous
models of low shadowing (i.e. in line of sight region) are re-
quired to produce vector , in dB by using zero mean, unity
standard deviation Gaussian random noise signals.
(4)
The correlation values are high, as expected due to the close
proximity of the two transmit and two receive antennas and
other measurements have shown that high correlation is main-
tained in different channel scenarios. Using these correlations,
eight time-synchronized simulations are created that are defined
KING et al.: EMPIRICAL-STOCHASTIC LMS-MIMO CHANNEL MODEL IMPLEMENTATION AND VALIDATION
TABLE I
SHADOWING MODEL MEAN AND STANDARD DEVIATIONS
by the following process in order to form the second order sta-
tistics [12], [13]:
(5)
where the time variation is determined by a coherence distance,
for a given mobile speed, with sample time and
and are the correlated Gaussian distributed random vari-
ables with zero mean and unity standard deviation. Measure-
ments used in this paper have shown to be 25 m on average for
a tree-lined road environment. The range of values recorded
spanned from 23 m through to 29 m. It is assumed in this case
that each sample, , is taken for every meter. The shadowing
then requires normalization where a set of standard deviations to
the shadowing represented by vectors and and mean
values, represented by vectors and , all in dB, are ap-
plied, where denotes elementwise multiplication and the nor-
malized, filtered and correlated shadowing, and
are therefore:
(6)
Empirical values of standard deviations and
and mean values, and for co-polar and cross-polar
channels in the tree-lined road environment in dB are shown in
Table I, which are derived generically from all measurement
data taken as the highest and lowest values. Finally the data
must be reshaped to create two separate 2 2 channel matrices
and .
B. Step 2—Generate Markov Chain
Having generated data for high and low shadowing se-
quences, a Markov Chain [10] is used to select between the
regions of high and low shadowing for both co-polar and
cross-polar channels. This allows the sharp transitions that
occur as a mobile moves past buildings as illustrated in Fig. 2
to be suitably modeled. Therefore if two polarizations are
considered, there are four possible Markov states as illustrated
in Fig. 5. It is assumed that the behavior is the same whichever
of RHCP or LHCP is being transmitted. Once the polarization
is defined at the transmit end, these four possible states there-
fore consider whether the co-polar or cross-polar channels at
the receive end are both in a high or low state or in opposite
states, as can happen in certain instances. Given that there are
609
Fig. 5. Illustration of the four Markov states for the LMS-MIMO channel.
four possible states, there are therefore sixteen possible state
transitions as shown by the arrows. The Markov chain statistics
are extracted from the measurement data (once a threshold
is selected for high and low shadowing in the measurement
data) and the results are shown in Table II, which are derived
from analyzing the Markov chain of all measurement samples.
The Markov chain derived from the measurement data using
the chosen threshold was analysed to ensure that where state
transitions did occur, they were true cases of a real transitions.
There can be cases with extreme low probability where a high
state shadowing falls below the threshold while also a low state
threshold rises above the threshold. Inspection of the Markov
chain removed any of these remotely possible occurrences.
The columns of the state transitions represent the probability
of one state moving to another listed in the right hand column,
while each row represents the probability of moving to the state
shown on the right hand column from a previous state shown
on the bottom row. Thus the top right hand state transition of
0.1037 is the probability of moving from state “CP High, XP
High” to “CP Low, XP Low”, where CP is a co-polar channel
and XP is a cross-polar channel. In the majority of cases for
this measurement, both CP and XP are in a high shadowing
state, which reflects the measurement scenario being a road
lined with houses and trees with foliage thus providing dense
multipath and the satellite at a low elevation. From applying
the Markov chain analysis, sampled every meter, to the large
scale shadowing, a 2 2 channel matrix, can finally be
created.
After producing the Markov chain, the data must be up-sam-
pled to match the resolution of the small scale fading that is gen-
erated next. Up-sampling is required because the rate of change
of fading for the large scale fading is considerably low com-
pared to that of the small scale fading and therefore does not
require such a large scale of sampling (based upon the size of
) in the first instance. However, for the large scale fading to be
integrated with the small scale fading at a later stage, it is neces-
sary that the large scale fading matches the necessary sampling
rate for the small scale fading.
610 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE II
MARKOV STATE AND STATE TRANSITION TABLE
TABLE III
COMPARISON OF MEAN AND STANDARD DEVIATION VALUES OF COMPLEX
CORRELATION MAGNITUDE AGAINST RICE FACTOR
C. Step 3—Generate Small Scale Fast Fading
The small scale fading is modeled in this case by a Ricean
distribution, where the low shadowing region will have a higher
Rice factor than that of the high shadowing region, which will be
subject to denser multipath. The usage of values of Rice factors
based on measurements are discussed later in Table III.
Using the Rice factors, Ricean fading with appropriate second
order statistics can be generated for each MIMO branch by using
a ring scatterer model [14] (though other well known methods
to generate small scale fading such as Doppler filtering and
autocorrelation matrices [7] are equally acceptable). For each
sample , the small scale fading elements, denoted by subscripts
of are derived in this case as:
(7)
where is the sampling factor equal to the sampling frequency
divided by the maximum Doppler shift, due to mobile move-
ment. The Rayleigh (or scattered) part of the small scale fading
is normalized by so that its mean is unity. The compo-
nents of the Rayleigh part will arrive at angle and have a
random phase . The four elements are arranged into
the 2 2 matrix . (11)
In the high shadowing regions, there is in general a non line of
sight (NLOS) condition and the Rice factor is low. In the case of
the LMS, the scattering rich environment local to the mobile ter-
minal provides low correlation between the antenna branches,
while at the satellite there are directional antennas with highly
orthogonal circular polarizations, thus their correlation is also
low. At each end therefore, the correlation is controlled inde-
pendently and also remains low in a non line of sight scenario,
which therefore justifies that the channel is separable in order
to allow the well known Kronecker model to be applied [15] to
model the correlated small scale fading, as follows:
(8)
where vec is the vectorize function. Values for the small scale
fading correlation matrix are also taken from measurements,
where the values are derived as an average of correlation values
evaluated over the whole sampled measurement data. The
format of matrix follows the same pattern as that used
for in (4):
(9)
In the case where there is low shadowing and a line of sight
(LOS), the channel cannot be considered separable and thus the
Kronecker assumption does not hold. Thus the Kronecker model
has been extended here such that it is suited to a 2 2 satellite
MIMO system in a LOS environment with dual circular polar-
ization, which enables polarization multiplexing.
Firstly a co-polar correlation matrix, (within which a
complex correlation [7], of and is used) is defined
as the following 2 2 matrix:
(10)
The phase information in the matrix may be used, though as
the validation section later on will clarify, it is not essential to
include when generating a model. Secondly a cross polar cor-
relation matrix, is also defined in a similar way. The cor-
relation component, is derived by taking an average of the
complex correlation of and and the correlation of
and . Again the phase information is not essential and it is
assumed the two antennas at the transmit end have similar char-
acteristics in terms of gain patterns and polarization purity. The
same must also be true at the receive end, though the transmit
antennas do not have to be the same as the receive antennas.
The 2 2 correlation matrix for the cross-polar is defined as
follows:
The two matrices in (10) and (11) will influence the orthogo-
nality of the right hand and left hand polarizations, which will be
key to a MIMO channel. A 1 2 channel vector of the co-polar
components, , is then generated using Rice factors de-
termined from measurement using (7) and then the two co-po-
larizations are correlated to gain as follows:
(12)
KING et al.: EMPIRICAL-STOCHASTIC LMS-MIMO CHANNEL MODEL IMPLEMENTATION AND VALIDATION 611

TABLE IV TABLE V
COMPARISON OF MAXIMUM AND MINIMUM XPD AGAINST RICE FACTOR CORRELATION, XPD AND RICE FACTOR VALUES USED IN THE NLOS AND
LOS CASES FOR VALIDATION

Likewise the same can be done for the cross polar compo-
nents, which have a very different correlation and Rice factor
compared to the co-polar case:

(13)

The four elements once generated can easily be inserted into

a 2 2 matrix to generate the small scale fading, . After
applying the correlation, it is necessary to re-normalize the mean
values of all four MIMO branches. For the cross polar compo-
nents, and , the mean values must also be di-
vided by the square root of the cross polar ratio, . The
XPD is defined as the ratio of the mean co-polar power to the
mean cross-polar power. It is assumed XPD is the same whether
the co-polar component is left hand or right hand circularly po-
Fig. 6. Cumulative distribution plot of the modeled and measured channel for
larized. high shadowing (NLOS).
Measurements provided data showing Rice factors ranging
from 0 to 10 for co-polar data. The cross polar Rice factors
were also found to be wide ranging, though as a rule they are
always less than the co-polar Rice factor for a set of samples,
thus any Rice factors can be selected for a model that adhere
to the rule, though corresponding XPD and correlations have
to be used alongside given Rice factors. Table III and Table IV
present suitable corresponding values of correlation and XPD
respectively taken from analyzing measurement data available.
In the case of a high Rice factor, the co-polar correlation will
be inherently high, while for a low Rice factor the correlation is
lower and has a greater variance. XPD can be as high as 15 dB
for a high Rice factor, though on average it is closer to 10 dB.
For lower Rice factors, where the scattering causes significant
de-polarization the average XPD is closer to 0 dB.

D. Step 4—Integrate Steps 1, 2 and 3

Now that the large and small scale fading channels are cre-
ated, they can simply be multiplied together to form the final
channel model such that:
(14)
where denotes an elementwise multiplication of the two ma-
trices. It should be noted that the resultant large scale fading
is already normalized to the bulk mean free space path
loss and any other losses in the ionosphere or troposphere. How-
ever, the resultant small scale fading must be normalized
to unity mean power.
V. LMS-MIMO CHANNEL MODEL VALIDATION OF SMALL
SCALE FADING
An important validation for MIMO channel models is to en-
sure that the eigenvalue cumulative distributions produced by
Fig. 7. Cumulative distribution plot of the modeled and measured channel for
low shadowing (LOS).
the model are in good agreement with measured data. This will
not only ensure that the first and second order statistics of the
physical channel are suitably modeled but also that the inter-
dependence between them is suitably accounted for in order to
demonstrate the diversity and multiplexing capabilities of the
model.
To demonstrate the validation of the model, two LOS and
NLOS cases were chosen, which had values of Rice factor, XPD
and correlation shown in Table V. The values for Rice factor are
in linear form.
The following three subsections compare the first order,
second order and eigen analysis as a validation of the model
proposed using appropriate sections of the measurement data.
The narrowband measurement data used for validation has a
612 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 8. Comparison of the right to right hand and left to left hand polarized
Doppler spread in NLOS and LOS.
Fig. 10. Comparison of the model and measurement eigenvalues for the high
shadowing (NLOS) region.

Fig. 9. Validation of the second order statistics based on level crossing rate and
average fade duration.
Fig. 11. Comparison of the model and measurement eigenvalues for the low
shadowing (LOS) region.

sampling rate of more than twice the maximum Doppler shift

so as to meet Nyquist criteria.
Doppler spread shown verifies the suitability of the Ricean dis-
A. First Order Statistics tribution for small scale modeling based on a Classical bath tub
model with the addition of a delta function for the Rice compo-
Fig. 6 and Fig. 7 illustrate the cumulative distribution of
nent [7]. A high Rice component is identified in Fig. 8 for the
the small scale fading for all four branches of the NLOS and
LOS case while still a small Rice component is identified in the
LOS regions. In both cases there is good agreement between
NLOS case due to non uniform angle of arrival.
the measurement data and model, where in the LOS scenario,
A further validation of the second order statistics is presented
a wider gap can be identified between co-polar and cross-polar
in Fig. 9, which shows a good agreement between level crossing
branches. It is interesting to note that the mean value of the
rate and average fade duration when comparing the measured
cross-polar component for LHCP transmission is shifted by
narrowband data and the modeled narrowband data. Free space
over 6 dB when compared to RHCP, where the multipath was
loss and the maximum Doppler frequency are labeled FSL and
found to better combine constructively within the sampled
respectively.
time window. The model, however, has not accommodated this
difference, since the negligible contribution of the cross-polar
C. Eigen-Analysis
element to the MIMO channel has no real effect on the capacity
or eigen analysis. Fig. 10 presents results of the eigen analysis of the model and
measurement data as well as the modeled data for the NLOS
B. Second Order Statistics case. Clearly there is a good consistency, which verifies the Kro-
Fig. 8 shows the Doppler spread in both left hand to left necker model approach is sufficient for this scenario. In this
hand and right hand to right hand polarized cases, graph, the notation is denoted as an eigenvalue where is
where polarization makes little difference in LOS or NLOS. The 1 or 2 for a 2 2 MIMO system.
KING et al.: EMPIRICAL-STOCHASTIC LMS-MIMO CHANNEL MODEL IMPLEMENTATION AND VALIDATION
Fig. 11 on the other hand compares measured and modeled
eigenvalues in the LOS case using the new model approach.
Compared to the NLOS case, the LOS channel is clearly rich
in polarization multiplexing, as opposed to diversity because the
eigenvalue distributions are closer. This is expected due to fewer
scatterers. Results are in agreement, though it should be noted
that in this validation, the phase information was applied in the
correlation matrices. Were the phase information not applied,
the second eigenvalue would marginally change its gradient,
moving away from the measured data by less than 2 dB. Given
the negligible impact this would have on modeled channel ca-
pacity, use of the phase information is therefore not important.
A similar scenario occurs when modeling other LOS regions.
VI. CONCLUSION
The procedure for implementing a simple empirical-sto-
chastic based model for the dual circular polar 2 2
LMS-MIMO channel has been presented along with re-
sults to validate the model at low elevation, which is based
on switching between high and low shadowing regions with
different multipath conditions. The validation of the model at
such elevations will also be suited to higher elevation angles
where the multipath is reduced and the opportunity to imple-
ment polarization multiplexing is increased. The well known
Kronecker model is suitable for the non line of sight case, while
a new model has been presented to be applied to a polarization
multiplexing rich scenario in the line of sight case. Compar-
isons show good accuracy in both cases. Given the simplicity of
generating a Markov chain and correlated small scale and large
scale fading, it is highly appropriate for conformance testing for
satellite MIMO applications with the simplicity of controlling
the channel through altering Rice factors, correlation and XPD
values according to guidelines presented.
ACKNOWLEDGMENT
The help of U. Ekpe at the University of Surrey is acknowl-
edged for extracting the relevant narrowband measurement data.
REFERENCES
[1] P. R. King and S. Stavrou, “Capacity improvement for a land mobile
single satellite MIMO system,” IEEE Antennas Wireless Propag. Lett.,
vol. 5, no. 1, pp. 98–100, Dec. 2006.
[2] P. R. King, P. Horváth, F. Pérez-Fontán, I. Frigyes, and S. Stavrou,
“Satellite channel impairment mitigation by diversity techniques,” pre-
sented at the IST Mobile Summit, Jun. 2005.
[3] IEEE 802.11n Standard for Information Technology—Telecommu-
nications and Information Exchange Between Systems—Local and
Metropolitan Area Networks—Specific Requirements, IEEE 802.11n,
2009 [Online]. Available: http://standards.ieee.org
[4] IEEE 802.16 Standard for Local Metropolitan Area Networks, IEEE
802.16, 2009 [Online]. Available: http://standards.ieee.org
[5] Third Generation Partnership Project [Online]. Available: http://www.
3gpp.org
[6] Digital Video Broadcasting Standards [Online]. Available: http://www.
dvb.org
[7] S. R. Saunders and A. A. Aragón- Zavala, Antennas and Propagation
for Wireless Communications, 2nd ed. London, U.K.: Wiley, 2007.
[8] M. Sellathurai, P. Guinand, and J. Lodge, “Space-time coding in mobile
satellite communications using dual-polarized channels,” IEEE Trans.
Veh. Tech., vol. 55, no. 2, Jan. 2006.
613
[9] K. P. Liolis, J. Gomez-Vilardebo, E. Casini, and A. Perez-Neira, “On
the statistical modelling of MIMO land mobile satellite channels: A
consolidated approach,” in Proc. IET and AIAA Int. Communications
Satellite Systems Conf., Edinburgh, U.K., 2009, p. 422.
[10] F. P. Fontan and P. M. Espiñeira, Modelling the Wireless Propaga-
tion Channel: A Simulation Approach With MATLAB. London, U.K.:
Wiley,, 2008.
[11] R. G. Vaughan and J. B. Andersen, Channels, Propagation and An-
tennas for Mobile Communications. London, U.K.: IEE Press, 2002.
[12] M. Gudmundson, “Correlation model for shadow fading in mobile
radio systems,” IEE Electron. Lett., vol. 27, no. 23, pp. 2145–2146,
1991.
[13] M. J. Marsan, G. C. Hess, and S. S. Gilbert, “Shadowing variability in
an urban land mobile environment at 900 MHz,” IEE Electron. Lett.,
vol. 26, no. 10, pp. 646–648, 1990.
[14] M. Pätzold, Mobile Fading Channels. London, U.K.: Wiley, 2002.
[15] J. P. Kermoal, L. Schumacher, K. I. Pedersen, P. E. Mogensen, and
F. Frederiksen, “A stochastic MIMO radio channel model with exper-
imental validation,” IEEE J. Sel. Areas Commun., vol. 20, no. 6, pp.
1211–1226, Aug. 2002.
Peter R. King received the B.Sc. degree in electronic
engineering (telecommunications) from the Univer-
sity of Essex, U.K., in 1989, and the M.Sc. degree
in mobile & satellite communications and the Ph.D.
degree in mobile satellite radio propagation from the
University of Surrey, U.K., in 2002 and 2007 respec-
tively.
He has worked in radio frequency and radio propa-
gation research & development since 1990 for many
leading organizations such as Lucent Technologies,
Nortel, Roke Manor Research, Samsung, TTP Com-
munications, Airspan, Thales, ITT Defence and AceAxis. He now runs a small
RF consultancy business, providing solutions for mobile, satellite, aircraft and
wireless radio communication systems, circuits and products. In addition to his
consultancy work, he continues to research and teach at the University of Surrey
where he is a visiting academic. His research interests include terrestrial and
satellite radio propagation, and high-performance radio: power amplifiers, re-
ceivers, transmitters, frequency synthesis and power amplifier linearization.
Tim W. C. Brown (S’00–M’04) received the B.Eng.
degree in electronic engineering and the Ph.D. de-
gree in antenna diversity for mobile terminals from
the University of Surrey, U.K., in 1999 and 2004, re-
spectively.
Since completing his doctoral research, he has
continued his research interests in antennas, propa-
gation and radio frequency (RF) engineering. This
has included postdoctoral research from 2004–2006
at Aalborg University, Denmark and his present post
as a Lecturer in RF, antennas and propagation at the
Centre for Communication Systems Research (CCSR), University of Surrey.
His current research interests include mobile terminal antennas, satellite com-
munications, multiple input multiple output (MIMO), ultrawideband (UWB),
radar, radio frequency identification (RFID), near field communications (NFC)
and vehicular technologies.
Argyrios Kyrgiazos received the Dipl.-Eng degree
in electrical and computer engineering from the
National Technical University of Athens (NTUA),
Greece and the M.Sc. degree in mobile and satellite
communications from the University of Surrey,
U.K., in 2008 and 2010, respectively, where he is
currently working towards the Ph.D. degree.
In 2008, he joined the Greek army for his manda-
tory duty. His research interests include wireless ter-
restrial and satellite communications with emphasis
on physical layer and link layer issues, and satellite
radio propagation.
614 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Mr. Kyrgiazos is a member of the Institution of Engineering and Technology ESA and industry and leads the Surrey part of the UK-India network of excel-
(IET) and a member of the Technical Chamber of Greece (TEE). lence in next generation networks. He also leads the communications platform
in Surrey’s Knowledge Transfer Network. He is Editor of the International
Journal of Satellite Communications, an OFCOM spectrum advisory board
member, Chair of the steering committee of SatNEx, an EU/ESA network of
Barry G. Evans (M’86–SM’98) until 2009 was excellence, a member of the steering council of the Integral Satellite Initiative
Pro Vice Chancellor for Research and Enterprise at and Director of a spin off company Mulsys Ltd.
the University of Surrey and Director of the Centre Prof. Evans is a Fellow of the Royal Academy of Engineering.
for Communication Systems Research which is a
150 strong postgraduate research centre. He has a
personal research background in satellite and mobile
communications, researching in speech coding, radio
propagation, advanced physical layers and cognitive
radio with over 500 research publications and three
text books to his name. He now runs a number of
research projects in satellite communications with
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Effectiveness of Relay MIMO Transmission by
Measured Outdoor Channel State Information
Kentaro Nishimori, Naoki Honma, Tomoki Murakami, and Takefumi Hiraguri
Abstract—This paper investigates the effectiveness of the mul-
tiple input multiple output (MIMO) transmission with the aid of
relay nodes by using measured outdoor channel state information
(CSI) for the source, relay, and destination which are used in coop-
erative transmission. Single/multi-user MIMO (SU/MU-MIMO)
systems have attracted much attention as the technology enhances
the channel capacity with a limited frequency band. In this paper,
cooperative transmission using the relay station is applied to
SU/MU-MIMO systems: Relay MIMO transmission. Although
the performance of relay MIMO transmission is affected by the
actual propagation environment, the performance has not yet
been investigated using measured propagation channel data in
outdoor scenarios. In this paper, we demonstrate that the channel
capacity with the relay MIMO transmission can be improved com-
pared to the conventional MIMO transmission without the relay
nodes, especially in multi-user scenarios. Moreover, orthogonal
polarization is introduced and its effectiveness is shown in further
improving the channel capacity of relay MIMO transmission.
Index Terms—Cooperative transmission, heterogeneous
pathloss condition, orthogonal polarization, relay station,
single/multi-user MIMO (SU/MU-MIMO).
I. INTRODUCTION
D UE to the recent popularity of mobile phones and broad-
band wireless local area network (W-LAN), the mul-
tiple-input multiple-output (MIMO) technique is incorporated
into these broadband wireless systems using OFDM to achieve
higher transmission speeds without expanding the frequency
band [1], [2]. Moreover, multiuser MIMO (MU-MIMO) sys-
tems have recently attracted much attention as the technology
enhances the total system capacity by generating a virtual large
MIMO channel between a base station and multiple terminal
stations [3].
From the viewpoint of decreasing the transmission power
and expanding the service area, cooperative transmission using
relay stations was investigated [4]–[7]. In cooperative transmis-
sion, the source (S), destination (D), and relay (R) share their
resources in order to forward data, achieving a spatial diversity
gain against fading. In this paper, we incorporate MIMO or
MU-MIMO techniques into cooperative transmission. In other
words, MIMO transmission using the relay station, Relay
MIMO transmission is effective from the viewpoint of fre-
quency utilization and spatial diversity.
Manuscript received June 01, 2010; revised March 20, 2011; accepted May
11, 2011. Date of publication December 06, 2011; date of current version Feb-
ruary 03, 2012. This work was supported in part by KAKENHI, Grant-in-Aid
for Young Scientists (B) 22760272.
The authors are with Niigata University, Niigata 950-2181, Japan (e-mail:
nishimori@m.ieice.org).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173434
0018-926X/$26.00 © 2011 IEEE
615
Since cooperative transmission is a recently developed tech-
nique, detailed propagation characteristics have not yet been
thoroughly investigated in conventional studies, especially on
outdoor scenarios. For example, the path loss conditions are as-
sumed to be the same in the S-D, S-R, and R-D links in [6].
Moreover, R is assumed to be located between S and D. How-
ever, the conditions regarding antenna deployment for S, D,
and R are different in actual scenarios. The relay stations must
be deployed at high locations in order to ensure line of sight
(LOS) environment. The base stations should be basically lo-
cated on the rooftops of buildings. On the other hand, the an-
tenna heights of the terminal stations are usually low compared
to that of the surrounding buildings. Hence, the performance of
relay MIMO transmission is greatly affected by the actual prop-
agation environment due to the different conditions for S, R, and
D. Although there are a lot of measurement results regarding
MIMO/MU-MIMO transmission in outdoor environment (for
example, [8]–[11]), the performance of relay MIMO transmis-
sion has not been thoroughly studied using measured propaga-
tion channel data in outdoor scenarios.
In this paper, we evaluate the channel capacity of MIMO
transmission with a relay station considering actual measured
channel state information (CSI) for the S-D, S-R, and R-D links.
First, the channel capacity is derived using relay MIMO trans-
mission, and the influence of the relay MIMO transmission on
the channel capacity is clarified when heterogeneous path loss
conditions are given for the S-D, S-R, and R-D links. Then, we
compare the channel capacities between conventional and relay
MIMO transmissions using the measured CSI. Since MIMO
transmission with orthogonal polarization can reduce the spa-
tial correlation [9], [11], particularly between the relay and base
stations, the effectiveness using orthogonal polarization is clar-
ified.
The remainder of this paper is organized as follows. In
Section II, we derive the channel capacity when employing
the relay MIMO transmission and discuss the influence from
the heterogeneous path loss conditions. Section III describes
the employed measurement environment. In Section IV, the
effectiveness of the relay MIMO transmission in an actual
outdoor scenario is demonstrated. Moreover, the effect of
orthogonal polarization, which reduces the spatial correlation,
is investigated in Section IV.
II. CHANNEL CAPACITY IN THE RELAY MIMO TRANSMISSION
A. Derivation of Channel Capacity
Fig. 1 shows the transmission scheme for the relay MIMO.
We extend the scheme in [4] to MIMO transmission. In this
paper, R is assumed to decode and forward the signals trans-
mitted by S using two time slots. Let denote the index
616 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 2. Analysis model for Model A.

transmission rate in time slots 1 and 2, respectively, the sum

rate, , must satisfy the condition below.

(5)
(6)

where denotes the noise power. In order for D to decode the

signal without bit error, the achievable transmission rate in time
Fig. 1. Transmission scheme in the relay MIMO. (a) Slot 1, (b) Slot 2. slot , must satisfy

number of the time slot. As shown in Fig. 1, S transmits sig-

nals to R and D in time slot 1. , and denote the (7)
number of antennas at S, R, and D, respectively, as shown in
Fig. 1. When the signal transmitted by S in time slot 1 is rep- (8)
resented by , the received signals at D and R,
and , respectively, in time slot 1 Moreover, in order for R to decode the signal without bit error,
are given by the achievable transmission rate in time slot 1, , must satisfy
(1)
(9)
and
Since the sum rate, , must satisfy the capacity for the
(2) reception at R and D, the sum rate is constrained by

where denotes the average signal energy, when A (10)

and B denote the transmitter and receiver, respectively.
represents the channel matrix, where the (more details are given in [4]).
average power is equal to one. denotes the Gaussian noise Uplink transmission is considered in this paper. For the multi-
at receiver A. Next, in time slot 2, both S and R transmit signals user transmission, we assume that all the user terminals are
to D. Similar to time slot 1, the signal received by D in time regarded as S. The total number of antennas for the multiple
slot 2, is given by source stations are set to be in Fig. 1. Similarly, when con-
sidering the multiple relay stations, the total number of antennas
(3)
at all the relays are set to be in Fig. 1.
where we assume that the number of antennas at S and R sta-
B. Basic Characteristics Considering Heterogeneous Pathloss
tions, and are same. When using (1)–(3), the received
Conditions
signals at D in time slots 1 and 2, is ex-
pressed as In this subsection, the channel capacity for the relay MIMO
derived in Section II.A is evaluated by computer simulation.
Fig. 2 and 3 show the models (Model A and B) used in the
evaluation. The main parameters in Model A and B are shown
in Table I. Condition is assumed in Fig. 2 and
this condition is the same with that in [4]. R is located at the
(4) center between S and D in Model A. The path loss coefficient
: Path loss, : Distance between the transmitter
where denotes the transmit signals by S in and receiver) is the same and is set to 3.5 for the S-D, S-R, and
time slots 1 and 2. R-D links in Model A. is a typical value for mobile
In the following, the channel capacity for the relay MIMO communications.
is derived. The idea in [4] can be extended when considering It is clear that the channel capacity is maximized for condition
MIMO transmission. When and denote the achievable if the path loss conditions are the same for S-R and
NISHIMORI et al.: EFFECTIVENESS OF RELAY MIMO TRANSMISSION BY MEASURED OUTDOOR CHANNEL STATE INFORMATION
Fig. 3. Analysis model for Model B.
TABLE I
CONDITIONS FOR MODEL A AND B
TABLE II
PATH LOSS MODEL WITH HETEROGENEOUS PATH LOSS CONDITIONS
R-D when considering 2-hop transmission without S-D trans-
mission. However, R cannot be located at the center between
S and D when actual installation space regarding the antennas
is considered. Moreover, since the antenna heights are different
among S, R and D, heterogeneous path loss conditions [12], [13]
between the S-D, S-R, and R-D links must be considered in ac-
tual scenarios.
In order to evaluate the difference with the conventional
model such Model A, Model B is regarded as a realistic model
in a micro/macrocell environment as shown in Table I. The path
loss coefficient for S-D, in Model B is also set to 3.5. In
order to express the path loss coefficient for R-D, , Ichit-
subo model is adopted [12], [13]. Table II presents parameters
of Ichitsubo’s model. and [dB] are functions of the antenna
heights of the transmitter and receiver as shown in Table II.
This model was obtained considering many measurements,
which were carried out with 2 GHz band in urban areas in
Japan. Moreover, the path losses obtained by the measurements
and the equation in Table II agree well with each other, when
the transmissions of Tx 32 [m]—Rx 32 [m], Tx 32 [m]—Rx
15 [m], and Tx 32 [m]—Rx 2 [m] are considered [12], [13].
Here, Tx A [m]—Rx B [m] represent the antenna heights of
the transmitter and receiver at A and B [m], respectively.
becomes 3.0 when the antenna heights in Table I are used.
Since the path loss coefficient for S-R, is changed due
to path visibility which discriminates whether an environment
is LOS or Non-LOS (NLoS) [14], we assume that ranges
from two to six in Model B. For the sake of simplicity, the
channel responses for S-D, S-R, and R-D are assumed to be
617
Fig. 4. Channel capacity using the relay MIMO versus CNR. (a) Model A, (b)
Model B.
i.i.d., i.e., Rayleigh fading. are set to four, respec-
tively.
Fig. 4 shows the channel capacity versus the Carrier to Noise
power Ratio (CNR) at S-D link. The number of trials is 10 000
and the channel capacity in Fig. 4 is calculated by averaging
all the data. The channel capacity of single input single output
(SISO) is plotted for reference. Coop and S-D in Fig. 4 rep-
resent the channel capacity for the relay MIMO and transmis-
sion at S-D link without the aid of relay station, respectively.
Fig. 4 shows that the channel capacity for the relay MIMO using
Model B is higher than that using Model A. The improvements
in the channel capacity when using the relay MIMO transmis-
sion (Coop in Fig. 4) compared to that for the conventional
MIMO transmission without the relay station (S-D in in Fig. 4)
in Models A and B are 3.7 and 6.0 bits/s/Hz, respectively, when
ranges from 2 to 4: the effectiveness of the relay MIMO is
observed when considering actual propagation parameters. This
is because the path loss at R-D link in Model B is less than that in
Model A. On the other hand, the channel capacity using the relay
MIMO transmission is severely degraded when is greater
than or equal to five. Since the channel capacity using the relay
MIMO transmission is denoted by (10), the total channel ca-
pacity is affected by the channel capacity at S-R link, ,
when is small.
Fig. 5 shows the channel capacity versus the path loss co-
efficient at S-R link, . The figure shows that the channel
capacity is suddenly degraded when becomes greater than
4.3 in Coop of Model B. Moreover, the channel capacity for the
relay MIMO (Coop) is lower than that for conventional MIMO
(S-D) when is greater than 4.9. Here, equals 4.1, if the
618 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 5. Channel capacity versus pathloss coefficient between S and R
stations.
Ichitsubo model is used at S-R link: could be assumed to be
less than 4. Hence, we indicate the effectiveness using the relay
MIMO transmission if the pathloss co-efficient at S-R link is the
typical value in a realistic propagation environment. However,
how the pathloss conditions change depends on the propagation
environment. We show the results regarding CSI measurement
and clarify the effectiveness of the relay MIMO transmission in
an actual outdoor environment hereafter.
III. MEASUREMENT ENVIRONMENT
Fig. 6 shows the measurement environment. The measure-
ments were carried out at the central area of Yokkaichi city in
Japan, where is assumed to be a typical urban area. As shown in
Fig. 6, relay stations are located at six points (R1 to R6 in Fig. 6.
We assume uplink MIMO transmission using R1 to R6. The
transmitters and receivers are equipped at R, because R must
transmit the signal to D and receive the signal from S. The an-
tenna height on the R station is changed from 3 to 9 m. The trans-
mitters are equipped at S shown in Fig. 6. The antenna height
on the vehicle is 2.2 m. The receivers at D are equipped at the
site on the steel tower. The antenna height on the steel tower
is 50 m. Table III shows the distance of S-D and R-D links.
represents the index number for R and S in Fig. 6.
As shown in Fig. 6, S moved near R at the speed of 30 to
50 km/h. Hence, we assumed that S , R and D are a pair for
the evaluation when considering relay MIMO transmission. S
(Max.)/S (Min.) denotes the maximum/minimum distance of
S-D link, respectively, as shown in Table III.
The parameters for the measurement to obtain the CSI are
given in Table VI. The pathloss co-efficient in Section II.B is
not used for the measurement. The transmit power by S and R is
constant and its value is 2 [W]. Hence, not only the small fading
effect but also influence due to the path loss are included in the
measurement results. We measured the MIMO channel matrix
for the S-D, S-R, and R-D links. Data were obtained at the in-
terval of 50 ms in each measurement course. The location of R
is selected so that the propagation environment of R-D link is
LoS. For S-D links, the propagation environment is regarded as
NLoS except for the neighborhood around D because there are
15 to 40 m high buildings around the measurement course. The
CSI is obtained by transmitting the OFDM signals. Short and
long preamble signals are used for the timing synchronization
and CSI estimation, respectively. The channel capacity for the
relay MIMO is obtained using the CSI and equations derived in
Section II.
Fig. 6. Measurement environment.
TABLE III
DISTANCE OF S-D AND R-D LINKS [M]
The antenna configurations for S, R and D are represented
in Fig. 7. As shown in Fig. 7, the number of antennas at S, R
and D, , and , are set to 4, 4, and 8, respectively.
The array element spacing at S is 0.5 wavelengths. The array
element spacing at R and D is 1.0 wavelength. A circular array
is used for S, in order to reduce the dependency on the angle
from S. Linear arrays are used at R and D. Omni-directional
antennas are used at S and R.
The geometry of the dual polarized subarray at D is shown
in Fig. 8 [11]. Eight vertical and horizontal dipoles are arranged
vertically. Each polarization array has an individual RF feeder.
A reflector is placed at the backside of the element for sup-
pressing the backlobe. The actual gain of the subarray is 14.5
dBi for both polarizations. The 3 dB beamwidth of this antenna
is 90 degrees for both polarizations. The cross-polarization dis-
crimination (XPD) between vertically and horizontally polar-
ized antennas is 24.5 dB in the broadside direction.
In Sections IV.A and IV.B, we use sleeve antennas as verti-
cally polarized antennas. We compared the channel capacity due
to the difference in polarization in Section IV.C, i.e., vertical,
horizontal, and vertical/horizontal polarizations. A slot antenna
is used as a horizontally polarized antenna. Table IV shows the
combination of antennas which are used for S and R in Fig. 7.
The numbers of antennas which are used at D in Fig. 7 are shown
in Table V.
Fig. 9 shows the cumulative density function (CDF) of the
CNR when S-D, S-R and R-D transmissions are considered. We
created a no-signaling period with the interval of 320 sec to
NISHIMORI et al.: EFFECTIVENESS OF RELAY MIMO TRANSMISSION BY MEASURED OUTDOOR CHANNEL STATE INFORMATION
Fig. 7. Antenna configurations for S, R, and D.
Fig. 8. Geometry of antenna at D.
TABLE IV
COMBINATION OF ANTENNAS FOR S AND R IN FIG. 7
TABLE V
NUMBERS OF ANTENNAS USED AT D IN FIG. 7
measure the noise power. We used same receivers at R and D.
The measured noise power was dBm/Hz (
[dBm] at 20 MHz). The CNR is calculated as follows:
(11)
where [dBm] is the received power. The vertical polarization
is used and all the measured data are considered in Fig. 9. Fig. 9
shows that the CNRs for the R-D and S-R links are much higher
than that for the S-D link because there is basically a LOS for the
R-D and S-R links. Hence, we expect that the channel capacity
to be improved by using the relay MIMO transmission.
619
Fig. 9. CNR characteristics.
IV. EFFECTIVENESS OF THE RELAY MIMO TRANSMISSION
USING MEASURED CSI
A. Single-User Case
In this section, the effectiveness of MIMO transmission em-
ploying the relay station is demonstrated using measured CSIs
when considering a single user MIMO (SU-MIMO) scenario.
Fig. 10 denotes the channel capacity due to the difference in
antenna height of R. The values for , and are all
four. For comparison, the channel capacity of the conventional
MIMO transmission without the aid of relay station (S-D in
Fig. 10) is plotted in the figure. Fig. 10 shows that there is a
4.5 bits/s/Hz channel capacity improvement by using the relay
MIMO transmission (Coop in Fig. 10) compared to that when
using the conventional MIMO transmission (S-D). When we
focus on the difference due to the difference in the antenna
height of R, the channel capacity when m is slightly
degraded compared to that when or 9 m, because the
LOS environment is guaranteed at R-D link when or 9
m. Hence, the antenna height of R, is fixed at 9 m hereafter.
The channel capacity versus the transmit power is plotted in
Figs. 11 and 12. Since the transmit power at the terminal sta-
tion (S) is generally set to be lower than that at the base sta-
tion (D), it is very important to evaluate the effect of using R
when changing the transmit power. denotes the power
ratio between S and R in Figs. 11 and 12. Although the actual
transmit power by R is identical with that by S in the measure-
ment, is changed when the post processing is employed.
Figs. 11 and 12 take into account all the measured courses and
620
Fig. 10. Channel capacity due to difference in antenna height of R station (SU-
MIMO, ).
Fig. 11. Channel capacity versus transmit power (SU-MIMO,
).
Fig. 12. Channel capacity versus transmit power (SU-MIMO,
).
10% and 50% of the CDF are shown in these figures. In Fig. 10,
, and are 4, 4, and 4, and in Fig. 9 they are 2, 2, and
8, respectively.
Figs. 11 and 12 show that a higher channel capacity is ob-
tained using the relay MIMO transmission compared to that for
using the conventional MIMO transmission, regardless of the
transmission power. When ( , we ob-
serve a 4 bits/s/Hz and 1 bit/s/Hz
improvement in the channel capacity. Similarly, when
( , we observe a 1 bit/s/Hz
and 0.8 bits/s/Hz improvement in the
channel capacity. Hence, we find that the relay MIMO is the
most effective when the transmit powers between the R and S
stations are almost the same. Moreover, regarding the number
of antennas, the effect of the relay MIMO is maximized when
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 13. Measurement scenario for multi-user case (3-user case).
the number of transmit antennas is equal to that of the receiver
antennas.
B. Multi-User Case
Next, we focus on a multiuser scenario. Fig. 13 shows the
measurement scenario for multi-user case. 3-user case is de-
scribed in Fig. 13 as an example. For the evaluation on the
multiuser scenario, , and are set to 2, 2, and 8,
respectively. In other words, a four-user evaluation is possible
if the total number of transmit antennas at S or R is equal to
the number of antennas at D. The number of source stations
(users) is assumed to be equal to the number of relay stations.
As shown in Fig. 6, S moves near R , and S
and R are a pair in MU-MIMO relay transmission. Here, when
the number of users is , the source and relay stations with
is picked up from S (R in Fig. 6. 3 users
(sources) and relays are selected in Fig. 13. We assume mul-
tiple S access multiple R
(multiple Coop), and multiple S accesses D
(multiple S-D) at the same time in Fig. 6. We consider all pos-
sible combinations of multiple S (R ) and D for the number
of users. Hence, there are 15, 20 and 15 combinations when
is set to be two, three and four, respectively. The total number
of antennas on S and R for the channel capacity evaluation are
and , respectively. The different path loss
condition is considered for each user and CNRs for each user
are different, because the transmit power is constant (2W).
Fig. 14 shows the CDF for multi-access channel (MAC)
capacity. All the measured data were included in this figure.
Fig. 14 shows that the improvement in the channel capacity
when using the relay MIMO (Coop in Fig. 14) is higher com-
pared to that when using the conventional MIMO (S-D in
Fig. 14) when the number of users is increased. In particular,
an improvement of 8 bits/s/Hz in the channel capacity when
using the relay MIMO is obtained when , compared to
that when using the conventional MIMO.
Fig. 15 shows the MAC channel capacity versus the number
of users. The 10% and 50% values of the CDF are plotted in
this figure. Fig. 15 shows that the channel capacity using relay
stations (Coop in Fig. 15) can be improved compared to when
using the conventional MIMO transmission, when the number
of users is increased. Although the diversity effect exists for the
single user case in the conventional MIMO transmission (S-D
NISHIMORI et al.: EFFECTIVENESS OF RELAY MIMO TRANSMISSION BY MEASURED OUTDOOR CHANNEL STATE INFORMATION 621

Fig. 14. Channel capacity considering MU-MIMO transmission. Fig. 16. CNR characteristics between V-Pol. and H-Pol.

TABLE VI
MEASUREMENT CONDITIONS

Fig. 15. Channel capacity versus the number of users.

TABLE VII
CHANNEL CAPACITY COMPARISON ( , CDF=10%
in Fig. 15), the degree of improvement in the channel capacity is [BIT/S/HZ])
abated when the number of users increases. On the other hand,
the relay MIMO improves the channel capacity in proportion to
the number of users. As a result, we find that there is a 1.5 fold
improvement in the channel capacity when considering a 4-user
case.
TABLE VIII
C. Effectiveness Using Orthogonal Polarization CHANNEL CAPACITY COMPARISON ( , CDF=10%
[BIT/S/HZ])
In this subsection, we demonstrate the effectiveness when
using orthogonal polarization. S2(R2), S5(R5), and S6(R6) in
Fig. 6 are used as the locations for S and R when considering
a comparison among vertical, horizontal, and orthogonal (ver-
tical and horizontal) polarizations. Fig. 16 shows the CDF of
the CNR when considering S2(R2), S5(R5), and S6(R6). The
CNRs when employing vertical and horizontal polarization are TABLE IX
CHANNEL CAPACITY COMPARISON ( , CDF=10%
plotted in Fig. 16. The figures show that the CNR for vertical [BIT/S/HZ])
polarization is slightly higher than that for horizontal polariza-
tion for the S-D, S-R and R-D links.
Tables VII–IX show a channel capacity comparison among
vertical polarization (V.-Pol.), horizontal polarization (H.-Pol.)
and orthogonal polarization (V. and H.-Pols.) when considering
SU-MIMO transmission. The results for CDF=10% are given
in these tables. The numbers of antennas for S, R and D are capacity using V.H.-Pol is higher than those using V.-Pol. and
given in each table. Tables VII–IX show that the channel ca- H.-Pol., when considering the S-R and R-D links. In order to
pacity using V.-Pol is slightly higher than those using H.-Pol examine the reason, we present spatial correlation comparisons
and V.H.-Pol., when the conventional MIMO transmission (S-D among V.-Pol., H.-Pol., and V.H.-Pol. for the S-D, S-R, and R-D
in Tables VII–IX) is considered, because the received CNR with links, respectively in Fig. 17. The figure shows the spatial corre-
V.-Pol. is slightly higher than that with H.-Pol. lation is high when using V-Pol. and H-Pol. Particularly, the spa-
Next, we focus on the channel capacity for the S-R and R-D tial correlation for the R-D link when using V-Pol. and H-Pol.
links in Tables VII–IX. Tables VII–IX show that the channel exceeds 0.95 with a very high probability. On the other hand, the
622 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 18. MAC channel capacity comparison among V., H. and V.H.-Pols (S-D,
CDF=10%).

Fig. 17. Spatial correlation comparison.

Fig. 19. MAC channel capacity comparison among V., H. and V.H.-Pols
spatial correlation is reduced by using V.H.-Pol. Hence, orthog- (Coop, CDF=10%).
onal polarization is effective in environments where the spatial
correlation is very high when using a single polarization.
Finally, we focus on the channel capacity for relay MIMO
(Coop in Tables VII–IX). Tables VII–IX show that the channel
capacity of Coop is higher than that for the S-D link. Hence, the
relay MIMO transmission is effective regardless of the antenna
polarization. On the other hand, the channel capacity is almost
the same between V-Pol. and V.H.-Pol.
Fig. 18–20 shows the relationship between users and the
MAC channel capacity due to the difference of the polarization.
The conditions for evaluation are same with those in IV.C. The
results for % are shown in the figure. The figure
shows that the improvement in the channel capacity is higher
when using the relay MIMO (Coop in Fig. 19) compared to Fig. 20. MAC channel capacity comparison among V., H. and V.H.-Pols (S-R,
that when using conventional MIMO (S-D in Fig. 18), when R-D, CDF=10%).
the number of users is increased. In particular, a 4 to 6 bits/s/Hz
improvement in the channel capacity when using the relay
MIMO is obtained when , compared to that when using the S-D, S-R, and R-D links. First, we clarified that the differ-
conventional MIMO. ence in path loss conditions among the S-D, S-R, and R-D links
When considering conventional MIMO transmission (S-D), is very important in evaluating relay MIMO transmission. Then,
the V-Pol. can obtain the highest channel capacity in a mul- we evaluated the channel capacity in a single-user case using the
tiuser scenario. On the other hand, the channel capacity with measured CSI. The effect on the relay MIMO is shown to be
the V.H.-Pol. is higher than that with the V-Pol., when consid- maximized when the number of transmitting antennas is equal
ering the relay MIMO transmission (Coop). Fig. 20 shows that to that for the receiving antennas. Moreover, the effectiveness
the V.H.-Pol. can obtain the highest channel capacity for the S-R of the relay MIMO in a multiuser scenario is shown. We found
and R-D links. Hence, we confirmed that vertical and horizontal a 1.5 fold improvement in the channel capacity when using the
polarization can contribute to further improving the channel ca- relay MIMO compared to that for conventional MIMO without
pacity in the relay MIMO transmission. the aid of relay stations, when considering a 4-user case. Finally,
it is clarified that the use of orthogonal polarization contributes
to further improving the channel capacity in a multiuser sce-
V. CONCLUSION
nario compared to when using single polarization, because the
In this paper, we investigated the effectiveness of the relay spatial correlation is reduced by the relay MIMO transmission
MIMO transmission when considering actual measured CSI for with orthogonal polarization, especially between R-D links.
NISHIMORI et al.: EFFECTIVENESS OF RELAY MIMO TRANSMISSION BY MEASURED OUTDOOR CHANNEL STATE INFORMATION
ACKNOWLEDGMENT
The authors thank Y. Makise, M. Ida, and Dr. M. Yoshikawa
of NTT Advanced Technology Corporation for their help in con-
ducting the outdoor field measurements.
REFERENCES
[1] G. L. Stuber, J. R. Barry, S. W. Mclaughlin, Y. Li, M. A. Ingram, and T.
G. Pratt, “Broadband MIMO-OFDM wireless communications,” Proc.
IEEE, vol. 92, pp. 271–294, Feb. 2004.
[2] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wireless Per-
sonal Commun., vol. 6, pp. 311–335, 1998.
[3] Q. H. Spencer, C. B. Peel, A. L. Swindlehurst, and M. Haardt, “An
introduction to the multi-user MIMO downlink,” IEEE Commun. Mag.,
vol. 42, pp. 60–67, Oct. 2004.
[4] R. U. Nabar, H. Bolcskei, and F. W. Kneubuhler, “Fading relay chan-
nels: Performance limits and space-time signal design,” IEEE J-SAC,
vol. 22, no. 6, pp. 1099–1109, 2004.
[5] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diversity
in wireless networks: Efficient protocols and outage behavior,” IEEE
Trans. Info. Theory, vol. 50, no. 12, pp. 3062–3080, Dec. 2004.
[6] K. Yamamoto, H. Maruyama, T. Shimizu, H. Murata, and S. Yoshida,
“Spectral efficiency of fundamental cooperative relaying in interfer-
ence-limited environments,” IEICE Trans. Commun., vol. E91-B, no.
8, pp. 2674–2682, 2008.
[7] K. Hayashi, K. Shirai, T. Himsoon, W. P. Siriwongpairat, A. K. Sadek,
K. J. R. Liu, and H. Sakai, “Optimum relay position for differential
amplify-and-forward cooperative communications,” in Proc. Wireless
Personal Multimedia Communications (WPMC 2006), Sep. 2006, pp.
766–770.
[8] V. Erceg, P. Soma, D. S. Baum, and S. Catreux, “Multiple-input mul-
tiple-output fixed wireless radio channel measurements and modeling
using dual-polarized antennas at 2.5 GHz,” IEEE Trans. Wireless
Commun., vol. 3, no. 6, pp. 2288–2298, Nov. 2004.
[9] K. Nishimori, Y. Makise, M. Ida, R. Kudo, and K. Tsunekawa,
“Channel capacity measurement of 8 2 MIMO transmission by an-
tenna configurations in an actual cellular environment,” IEEE Trans.
Antennas Propag., vol. 54, no. 11, pp. 3285–3291, Nov. 2006.
[10] H. Taoka, K. Dai, K. Higuchi, and M. Sawahashi, “Field experiments
on ultimate frequency efficiency exceeding 30 Bit/Second/Hz using
MLD signal detection in MIMO-OFDM broadband packet radio ac-
cess,” in Proc. IEEE VTC2007-Spring, Apr. 2007, pp. 2129–2134.
[11] N. Honma, R. Kudo, K. Nishimori, Y. Takatori, A. Ohta, and S. Kubota,
“Antenna selection method for terminal antennas employing orthog-
onal polarizations and patterns in outdoor multiuser MIMO system,”
IEICE Trans. Commun., vol. E91-B, no. 6, pp. 1752–1759, Jun. 2008.
[12] K. Nishimori, R. D. Taranto, H. Yomo, P. Popovski, Y. Takatori, R.
Prasad, and S. Kubota, “Spatial opportunity for cognitive radio systems
with heterogeneous path loss conditions,” in Proc. IEEE 65th Vehicular
Technology Conf. (VTC2007-Spring), Apr. 2007, pp. 2631–2635.
[13] S. Ichitsubo, T. Furuno, T. Nagato, T. Taga, and R. Kawasaki, “2
GHz-band propagation loss prediction in urban areas; antenna heights
ranging from ground to building roof,” IEICE Tech. Rep., A., pp.
96–15, May 1996.
[14] Y. Ito, , Y. Hosoya, Ed., “Radiowave propagation handbook,” in The
Distribution of Height and Width Of Buildings. Japan: Realize Inc.,
1999, pp. 342–349.
623
Kentaro Nishimori received the B.E., M.E., and
Ph.D. degrees in electrical and computer engineering
form Nagoya Institute of Technology, Nagoya,
Japan, in 1994, 1996 and 2003, respectively.
In 1996, he joined the NTT Wireless Systems Lab-
oratories, Nippon Telegraph and Telephone Corpo-
ration (NTT), Japan. He was a Senior Research En-
gineer at NTT Network Innovation Laboratories. He
is now an Associate Professor in Niigata University.
He was a Visiting Researcher at the Center for Tele-
infrastructure (CTIF), Aalborg University, Aalborg,
Denmark, from Feb. 2006 to Jan. 2007. His main research interests are spatial
signal processing including MIMO systems and interference management tech-
niques in heterogeneous networks.
Prof. Nishimori is a senior member of IEICE. He received the Young En-
gineers Award from the IEICE of Japan in 2001, Young Engineer Award from
IEEE AP-S Japan Chapter in 2001, Best Paper Award of Software Radio Society
in 2007 Distinguished Service Award from the IEICE Communications Society
in 2005, 2008 and 2010, and the Best Paper Award of IEICE in 2011. He was an
Associate Editor for the Transactions on Communications for the IEICE Com-
munications Society from May 2007 to May 2010 and Assistant Secretary of
Technical Committee on Antennas and Propagation of IEICE from June 2008
to May 2010.
Naoki Honma received the B.E., M.E., and Ph.D. de-
grees in electrical engineering from Tohoku Univer-
sity, Sendai, Japan in 1996, 1998, and 2005, respec-
tively.
In 1998, he joined the NTT Radio Communication
Systems Laboratories, Nippon Telegraph and Tele-
phone Corporation (NTT), Japan. He is now an As-
sociate Professor in Iwate University. His current re-
search interest is planar antennas for high-speed wire-
less communication systems.
Prof. Honma s a member of IEICE. He received the
Young Engineers Award from the IEICE of Japan in 2003, the APMC Best Paper
Award in 2003, and the Best Paper Award of IEICE Communication Society in
2006, respectively.
Tomoki Murakami received the B.S. degree in elec-
trical, electronics and computer engineering and the
M.E. degree in electrical engineering and bioscience
from Waseda University, Tokyo, Japan, in 2006 and
2008, respectively.
In 2008, he joined NTT Network innovation Lab-
oratories, Nippon Telegraph and Telephone Corpo-
ration (NTT), Yokosuka, Japan. His current research
interests include multiuser MIMO systems and re-
source allocation.
Takefumi Hiraguri received the M.E. and Ph.D. de-
grees from the University of Tsukuba, Ibaraki, Japan,
in 1999 and 2008, respectively.
In 1999, he joined the NTT Access Network
Service Systems Laboratories, Nippon Telegraph
and Telephone (NTT) Corporation. He is now an As-
sociate Professor in Nippon Institute of Technology.
He has been engaged in research and development of
high speed and high communication quality wireless
LANs systems.
Prof. Hiraguri is a member of IEICE.
624 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Single and Multi-User Cooperative MIMO in a

Measured Urban Macrocellular Environment
Buon Kiong Lau, Senior Member, IEEE, Michael A. Jensen, Fellow, IEEE, Jonas Medbo, and Johan Furuskog

Abstract—We study the potential benefits of cooperative mul-
tiple-input multiple-output signaling from multiple coherent base
stations with one or more mobile stations in an urban macrocel-
lular environment at 2.66 GHz. The analysis uses fully-coherent
measurements of the channel from three base stations to a single
mobile station equipped with four antennas. The observed chan-
nels are used to explore the gains in capacity enabled by coopera-
tive base station signaling for point-to-point and multi-user com-
munications. The analysis shows that for point-to-point links, the
average capacity for cooperative signaling is 53% higher than that
achieved for a single base station. For downlink and uplink com-
munication with three mobile users, cooperative signaling yields
average sum rate increases of 91% and 63%, respectively.
Index Terms—Cooperative systems, multiple-input mul-
tiple-output (MIMO) systems, multiuser channels.
I. INTRODUCTION
W HILE multiple-input multiple-output (MIMO) tech-
nology has demonstrated the potential for offering
significant improvements in spectral efficiency for wireless
communication, realization of these gains depends on the
communication environment. For example, in cellular sys-
tems, compact device sizes at the mobile station (MS) limit
the number of antennas that can observe independent fading
[1]–[3]. At the base station (BS), the elevated position and
sectorized nature of the antennas lead to limited observed
angular spread, again limiting the benefit of multiple antennas
for reasonable inter-element spacing.
While work has been accomplished to limit coupling and im-
prove performance for compact element spacing at the MS [2],
less work has addressed the issue of limited angle spread at
the BS. One potential solution to this problem, however, in-
volves using multiple BS sites working cooperatively, a solution
that also potentially enables significant benefit in terms of inter-
ference control in multi-user signaling [4]–[6]. At its simplest
Manuscript received June 16, 2010; revised September 09, 2011; accepted
October 12, 2011. Date of publication October 25, 2011; date of current ver-
sion February 03, 2012. This work was supported in part by Telefonaktiebolaget
LM Ericssons Stiftelse för Främjande av Elektroteknisk Forskning, VINNOVA
under Grant 2008-00970 and in part by the U. S. Army Research Office under
the Multi-University Research Initiative (MURI) Grant # W911NF-07-1-0318.
B. K. Lau is with the Department of Electrical and Information Technology,
Lund University, Box 118, SE-221 00 Lund, Sweden (e-mail: bkl@eit.lth.se).
M. A. Jensen is with the Electrical and Computer Engineering Department,
Brigham Young University, Provo, UT 84602 USA (e-mail: jensen@byu.edu).
J. Medbo and J. Furuskog are with Ericsson Research, Ericsson AB,
SE-164 80 Stockholm, Sweden (e-mail: jonas.medbo@ericsson.com; johan.fu-
ruskog@ericsson.com).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173443
level, such coordination involves scheduling based on aware-
ness of interference created by multiple BS sites [7], although
more sophisticated cooperation is also possible. For example,
the benefit of cooperative BS communication has been studied
in the context of determining the channel and shadowing cor-
relation properties for multiple BS sites and a single MS [8],
[9], although [9] did not achieve coherence across the multiple
BSs and the equipment in [8] led to phase uncertainties in the
measurements. The pioneering work in [10], while using inco-
herent measurements from multiple BSs, uses a statistical anal-
ysis based on the low correlation between the channels from the
different BSs [11] to allow exploration of the point-to-point and
multi-user capacity assuming BS coherence in a campus envi-
ronment at 5.2 GHz. Also, recent time-synchronized multi-BS
and multi-MS measurements demonstrate the associated inter-
ference characteristics and are used to discuss some implica-
tions of cooperative MIMO signaling for multiple users in a
limited fashion [12], although here the results are for a micro-
cellular environment.
More recent work has focused on coherent channel measure-
ments either from multiple BS sectors or from multiple BSs
[13]. For example, the work in [14] develops cooperative com-
munication for MIMO downlink communication and demon-
strates its performance using coherent measured channels from
two sectors at the same BS site. Some experimental data from
multiple BSs in a cooperative environment are also reported in
[15], although the focus of this work is on comparing the be-
haviors of channel eigenvalues predicted using ray tracing with
those observed in the channel measurements. Since the initial
submission of the present paper, new results have appeared in
the literature showing the impact of BS cooperation in several
scenarios based on coherent multi-cell measurements [16], [17].
This paper reports on the analysis of fully-coherent measure-
ments from three BS sites to a single MS in a macrocellular
environment, measurements that complement those that have
been recently reported. The observed channels are first used to
explore the gains achieved with cooperative MIMO signaling to
a single user. This analysis shows that channel gain imbalance
plays a dominant role in determining the measured multi-BS ca-
pacity, consistent with conventional understanding, and that BS
cooperation leads to an increase in average capacity of 53% over
that achieved using a single BS. In places where all base stations
contribute nearly equal signal power to the MS, this increase in
average capacity can exceed 100%. We then turn our attention
to the performance of cooperative MIMO for multi-user com-
munications involving two and three MSs for both the down-
link or broadcast channel (BC) and the uplink or multiple access
channel (MAC) [5] based on different signaling strategies over
the observed channels. This analysis shows that cooperative

0018-926X/$26.00 © 2011 IEEE

LAU et al.: SINGLE AND MULTI-USER COOPERATIVE MIMO IN A MEASURED URBAN MACROCELLULAR ENVIRONMENT
Fig. 1. Birds eye view of measurement environment.
MIMO signaling can provide an average multi-user throughput
that is up to 91% higher than that achievable using more tradi-
tional multiple-access strategies under favorable channel condi-
tions.
One important area within this field of cooperative BS
communication is the development of models that can predict
the achievable performance. Some of the findings in this work
demonstrate that traditional channel models can explain the
key behaviors observed in cooperative BS channels provided
that the models include key properties of the link gains, which
is a useful observation for future channel model development.
However, a key challenge is knowing the typical distribution of
these link gains in practical implementation scenarios and val-
idating any developed models against measured observations.
The measurements reported here therefore provide critical
understanding that will assist in the development of models
appropriate for cooperative BS communication.
II. MEASUREMENT SETUP
The considered urban macrocell environment is the built up
area within Kista (also called “Mobile Valley”), Stockholm,
Sweden, depicted in Fig. 1. Three BS sites were selected that
emulate a realistic cellular deployment topology. At each BS, a
single antenna mounted a few meters above the average rooftop
level of approximately 25 m transmits a linearly-polarized (45
from vertical) signal. The main lobe of each antenna pattern
is pointed downwards between 6 and 8 from horizontal and
approximately towards the centroid of the triangle formed by
the three BS sites. The MS consists of two dipole and two loop
antennas mounted on the top of a measurement van as a square
array with an inter-element spacing of approximately 30 cm,
which is 2.6 wavelengths at the excitation frequency.
Measurement of the channel between all three BS and four
MS antennas is accomplished using the Ericsson channel
sounder that is based on a prototype for LTE [18] but with a
custom frame structure and rate. A single transmit unit gen-
erates orthogonal frequency-division multiplexing (OFDM)
channel sounding symbols that are distributed to the antennas
at the three BS sites using RF-over-fiber equipment. To avoid
problems with non-orthogonality of the OFDM symbols due
625
TABLE I
SPECIFICATIONS FOR THE ERICSSON CHANNEL SOUNDER
Fig. 2. Location of BSs and routes 1 and 2 – traveled by the MS. Dis-
tances in meters from the starting points are indicated by circles and diamonds
for routes 1 and 2, respectively.
to the MS mobility, the transmissions from these three BS
antennas are time multiplexed at the symbol level.
The MS uses four parallel receiver chains to simultaneously
down-convert the signals from the four receive antennas. Disci-
plined rubidium clocks (Stanford Research Systems, PRS10) at
the transmitter and receiver provide a highly accurate synchro-
nization (Allan standard deviation less than ) between the
BS and the MS. Based on this timing reference, error in the mea-
sured propagation distance over all routes is less than 1 m. The
resulting system generates a full 4 3 MIMO channel matrix
at a rate of 1500 observations per second (based on 0.667 ms
probing frames), but because of bandwidth limitations between
the system and the storage medium, the observations are stored
at a rate of 190 channels per second, providing high spatial res-
olution given the maximum van speed of 30 km/hr. All of the
parameters used in the measurements are provided in Table I.
The measurements consist of data from two different routes,
each requiring approximately 9 minutes of measurement time.
The routes include regions of line-of-sight (LOS), obstructed
line-of-sight (OLOS), and non-line-of-sight (NLOS) propaga-
tion. The position data for each channel sample is logged using
a GPS receiver. Fig. 2 shows the two routes along with markers
indicating the distance traveled along each route and the po-
sitions of the base stations. The observed signal-to-noise ratio
(SNR) computed by extracting the signal and noise powers from
626 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

the channel impulse response is 28 dB on average, with the SNR

for the strongest BS-to-MS link always above 22 dB and rarely
below 25 dB.
We emphasize that the unique feature of this data is the co-
herence between the measurements from the different BSs, a
feature that allows us to determine the impact of BS cooper-
ation. Throughout this discussion, several reference cases will
be presented where the network must match a BS to one or more
MSs, a pairing that requires some level of cooperation among
the BSs. However, when the term cooperative BS is used, it ex-
plicitly refers to the case where the BSs jointly and coherently
participate in the communication to the MS.

III. POINT-TO-POINT SIGNALING (MACRODIVERSITY)

We first study the communication between cooperative Fig. 3. Average capacity for the best BS-to-MS link and for cooperative ca-
BSs and a single MS with antennas. Let represent the pacity using all BS links for the MS on route 1.
measured multi-BS (MIMO) channel matrix at the
th frequency with th column representing the
single-input multiple-output (SIMO) link from the th BS. Each
channel matrix is normalized so that the average of the channel
power gains for the strongest BS-to-MS link is unity, or
where

(1)

represents the number of frequencies, and indicates a

Frobenius norm. If represents the total power transmitted and
if the additive noise is modeled as a zero-mean, unit-variance
complex Gaussian random process, then with this normaliza-
tion can be considered the average single-input single-output
(SISO) SNR observed on the strongest BS-to-MS link [19]. The
point-to-point capacity at each location averaged over the Fig. 4. Average channel gains for the three BSs and the dominant three
frequencies and assuming that the base stations do not possess eigenvalues of for the MS on route 1.
channel state information (CSI) is then given by

(2)
where is the identity matrix and indicates a conjugate
transpose. We perform capacity analysis assuming a reference
SNR of 10 dB, which is at least 10 dB below the SNR observed
in the measurements.
Fig. 3 plots the point-to-point capacity performance of the
best single BS-to-MS link and the cooperative communication
using all BS-to-MS links for route 1 shown in Fig. 2. To achieve
improved visual clarity in this plot, the results are smoothed in
the displacement variable with a moving average filter over a
window of 10 wavelengths and down-sampled. These results
demonstrate that BS cooperation provides significant potential
capacity performance gain, particularly in regions where the MS
lacks a clear view to a single (and therefore dominant) BS, such
as for positions between 300 and 700 m or 900 and 1200 m.
As a reference, the average capacity achieved when the channel
coefficients are modeled as independent identically distributed
(i.i.d.) zero-mean complex Gaussian random variables with 10
dB SISO SNR is indicated by the black triangle on the plot,
revealing that the capacity falls short of what is achievable under
ideal propagation conditions.
Fig. 4 shows the average channel gains for the three BSs as
well as averages of the largest three eigenvalues of
as a function of position along route 1. Comparison of these re-
sults with the capacity in Fig. 3 shows that the dominant three
eigenvalues and the capacity are largest when the channel gains
are nearly equal since all three BSs can effectively participate
in the communication. A similar yet less dramatic capacity in-
crease occurs when two BSs enjoy a strong link to the MS, such
as for MS positions between 1700 and 1800 m.
The point that cooperative communication provides gains
when the MS lacks a dominant link to a single BS can be more
emphatically demonstrated using a simple analysis involving
two quantities. The first is the ratio of the average cooperative
capacity to the average capacity of the best BS-to-MS link.
The second is the 3GPP geometry factor, which is defined as
the ratio of the power received on the best BS-to-MS link to
the sum of the noise and the powers received on the other links
[20], [21]. However, we are interested in using this metric to
quantify the relative channel gains from each BS to the MS, and
LAU et al.: SINGLE AND MULTI-USER COOPERATIVE MIMO IN A MEASURED URBAN MACROCELLULAR ENVIRONMENT
Fig. 5. Two dimensional pdf of the 3GPP geometry factor and increase in
the average cooperative capacity relative to the average capacity of the best
BS-to-MS link for all channels in routes 1 and 2.
Fig. 6. CDF of the capacity for the channels on routes 1 and 2 with and without
BS cooperation. The CDF for i.i.d. Gaussian random matrices scaled by the
average channel gains for each BS and MS are also shown for comparison.
therefore we exclude the noise in our analysis. Mathematically,
if (see (1)), then
(3)
Fig. 5 plots the joint probability density function (pdf) of
these two quantities for the combined data from the two routes.
This result clearly shows that when the geometry factor is near
0 dB indicating that all three BSs have similar gains to the MS,
the cooperative communication provides significant gain. How-
ever, when the geometry factor is large indicating a dominant
BS-to-MS link, cooperative communication provides little ca-
pacity benefit.
Fig. 6 plots the cumulative distribution function (CDF) of
the average capacities achieved for both the best BS-to-MS link
and for cooperative communication using the combined data for
both routes, a result that shows significant benefit of BS coop-
eration. In fact, for this combined data the average capacity re-
sulting from BS cooperation is 53% higher than that achieved
using the best BS-to-MS link.
627
To investigate the observation made in the analysis of Fig. 5
that gain imbalance in the three BS links plays a critical role
in determining the capacity, we scale the elements of the i.i.d
Gaussian channel matrix by the gains computed from the
observed channels. Specifically, we 1) scale the columns of
by the gain averaged over all receive antennas and frequencies
for each BS antenna and 2) scale the rows of by the gain
averaged over all transmit antennas and frequencies for each
MS antenna. Mathematically, we have ,
where and are diagonal matrices whose diagonal ele-
ments represent the average channel power gain for each MS
and BS antenna, respectively. Note that this is consistent with
the well-known Kronecker model for the channel spatial covari-
ance [22] where and are receive and transmit corre-
lation matrices, respectively. However, choosing these as di-
agonal matrices enforces zero correlation among the channel
transfer functions at the different antennas, a model previously
used for antenna arrays with inter-element separation exceeding
four wavelengths [23]. The CDF of the capacity resulting from
these scaled i.i.d. matrices is compared to that of the measured
channel in Fig. 6. As can be seen, scaling by the average gain
per BS antenna (assuming ) leads to an excellent match
to the measured capacity. Further comparison between the ca-
pacity achieved with the BS-scaled i.i.d. matrices and that ob-
tained using the measured channels shows that they differ by
less than 1% at each location. This highlights the dominant in-
fluence of channel gain imbalance on the capacity performance
for cooperative BS signaling and indicates that the Kronecker
correlation model with diagonal transmit and receive correla-
tion matrices works well in this type of cooperative BS envi-
ronment.
In contrast, the scaling by the average MS gains (assuming
) results in a capacity that is close to the average ca-
pacity predicted by i.i.d. channels without scaling (indicated by
the black triangle in Fig. 6). This is an indication that the av-
erage channel gains for different MS antennas are similar, de-
spite the fact that the MS antennas include both vertical (V)
and horizontal (H) polarizations. To investigate this further, for
transmission from the th BS we compute the ratios
, where we have dropped the frequency index no-
tation for convenience. Table II presents the average of several
values of , where we use the notation V/V (H/H) to indi-
cate that and are the indices associated with the two ver-
tically-polarized (horizontally-polarized) MS antennas and the
notation V/H to indicate the average over the four combinations
where and are associated respectively with vertically- and
horizontally-polarized MS antennas. While clearly the gain for
the vertical polarization is higher than that for the horizontal
polarization, the imbalance is relatively small, confirming our
postulation that the channel gains for the different MS antennas
are similar.
IV. MULTI-USER SIGNALING
The point-to-point analysis detailed in Section III reveals a
number of interesting aspects regarding the cooperative MIMO
channel. However, practical cellular systems support multiple
users, and it is therefore intriguing to explore the performance
628 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE II
AVERAGE CHANNEL POWER GAIN RATIOS FOR V- AND H-POLARIZED
ANTENNAS ON THE MS
impacts associated with multiple cooperative BSs communi-
cating with multiple MSs. We will first consider the downlink
channel (BC) and then focus our attention on the uplink channel
(MAC) [5].
Throughout this analysis, we focus on MSs on the mea-
surement routes. Since the data was obtained for a single mo-
bile node, this means that we use channels measured at different
times to obtain the required channel data from the BSs to the
spatially-displaced users. Naturally, the channel is not strictly
static over the time interval between these two different mea-
surements, and therefore the performance we obtain does not
generally represent the instantaneous performance that would
be obtained for simultaneous links. However, the focus of
this analysis is to explore statistical trends over an ensemble of
situations. Since the major scattering environment (buildings,
parked vehicles, etc.) does not change over the measurement
times, this study provides statistically representative multi-user
performance behavior.
We assume that each MS only receives or transmits a single
data stream, and therefore uses multiple antennas only for di-
versity or beamforming. While it is possible to also allow each
MS to receive or transmit multiple simultaneous streams (mul-
tiplexing), inclusion of this capability is of limited value in this
study since 1) the cooperative BSs only have three antennas in
total, so that even with only one MS can accommodate
multiple streams, 2) all of the various configurations used in the
comparative study can support a single stream (but not neces-
sarily multiple streams), and therefore this allows for fair com-
parisons, and 3) inclusion of multi-stream communication adds
complexity without providing detailed additional understanding
regarding the trends enabled by the cooperative communication.
The channel normalization is similar to that used for the
single-user analysis. However, in this case of multiple users,
we must preserve the differences in channel gain to each MS. If
the channel matrix from the BSs to the th MS at the
th frequency is with th column , we scale the
matrices for all MSs as where
(4)
Note that in the following we drop the frequency index for
notational simplicity, recognizing that all sum rate results rep-
resent averages over the frequencies.
A. BC Topologies
For the BC, the cooperative BSs apply a beamformer to
the signal for the th MS represented by the vector ,
where the th element of the vector is the complex weight ap-
plied to the signal from the th BS. The th MS then applies a
beamformer represented by the unit-norm vector to
the received signal. If the additive noise at each receiver is mod-
eled as a zero-mean, unit-variance complex Gaussian random
process, the sum rate experienced for this BC is [24], [25]
(5)
where
(6)
We assume that the total transmit power among all BS an-
tennas is constrained to be a constant such that [24]
(7)
With the normalization of (4) and given that we have assumed
Gaussian noise with unit variance, this total power can once
again be considered the SISO SNR to the strongest BS-to-MS
link. Also, implicit in this assumption is that the BSs use power
control and can change their power allocation up to a total of
. While we could adopt a per-antenna power constraint so
that all BSs transmit the same power [26]–[28], our observation
matches that of other studies [27] that while this per-antenna
constraint slightly reduces the capacity, the resulting trends
match those obtained with the sum power constraint. As a
result, we use the simpler sum power constraint of (7) in the
analysis.
1) MISO Signaling: We first assume that each MS receives
using only one of the vertically-polarized antennas in a mul-
tiple-input single-output (MISO) configuration. Assigning this
antenna to be antenna #1, we have for each
MS, where indicates a transpose.
Maximum Power Pairing: As a reference case, each mobile
user establishes a link with the BS for which the BS-to-MS gain
is maximum, even if multiple MSs share the same BS. There-
fore, each transmit beamformer has only a single non-zero
entry of value . However, this simple reference case is at
a strong disadvantage since it does not benefit from intelligent
spatial processing capabilities and therefore can create signifi-
cant interference at each MS. Motivated by this observation, for
this case only we also compute the sum rate when the MSs
equally divide the communication time (time division multiple
access or TDMA). In this case, each beamformer has a single
non-zero entry of value , the second term in the denominator
of (6) is zero, and we scale (5) as . The final reference
sum rate for this technique is given as the maximum of the rates
computed with and without TDMA.
Optimal Pairing: While pairing each MS to a BS based on
the maximum channel gain is simple, it will certainly not max-
imize the network sum rate. We therefore explore a second ref-
erence scenario in which we compute the sum rate for each pos-
sible BS-MS pairing and select the pairing that achieves the
largest sum rate. Once again, the transmit beamformer for each
MS has only a single non-zero entry of value . However,
LAU et al.: SINGLE AND MULTI-USER COOPERATIVE MIMO IN A MEASURED URBAN MACROCELLULAR ENVIRONMENT
this situation implies additional cooperation, since the network
must know the CSI from all BSs to each MS to determine this
optimal pairing.
RCI: Finally, we consider a true cooperative MISO BC where
the beamformer for each MS is selected to achieve an appro-
priate balance of high signal strength for the th MS and low
interference for the other MSs. To construct the beamforming
vectors, we use the iterative regularized channel inversion (RCI)
method that, assuming the correct initial conditions, achieves
the optimal sum rate under the constraint of linear processing
[24], [25], [29]. For our work, we initiate the iteration using the
transmit beamformers obtained for the optimal pairing. This is
not guaranteed to converge to the absolute optimum, but does
show the improvement possible using a practical beamforming
algorithm.
It should be noted that improved performance can be obtained
by using non-linear processing known as dirty paper coding
(DPC) [5], [30]. While DPC will achieve higher capacity, the
capacity difference in most cases is relatively minor, and the
trends with channel conditions for DPC and RCI beamforming
are similar [24], [31]. To avoid the complex encoding and de-
coding process associated with DPC and to be consistent with
the other linear processing assumed throughout this paper, we
choose to use the simpler RCI for our comparative study.
2) MIMO Signaling: Because we have multiple antennas at
each MS, we can also explore the BC when these antennas are
used. In this case, we use the same scenarios as outlined above
for the MISO case. However, we assume that the th MS knows
(through training) the effective channel for
and can therefore construct a minimum-mean-squared error
(MMSE) beamformer from [29]
(8)
where is chosen so that the vector has unit norm. For the iter-
ative RCI algorithm, a new MMSE beamformer for the receiver
is computed at each step of the iterative computation. Further-
more, to pair each MS with the BS for which the gain is highest,
we select the BS associated with the column of that has the
highest norm.
B. MAC Topologies
Optimal Pairing: The multiple access channel results when
multiple MSs transmit simultaneously and uncooperatively to
the cooperative BSs. As a reference case, we assume that each
MS transmits from antenna #1 and that the BSs cannot coher-
ently cooperate. For each possible pairing between the th MS
and the th BS, we compute the sum rate from (5) with
(9)
where is the th element of and each MS transmits
a power of . The pairing that achieves the best sum rate is
selected as the reference topology.
629
Cooperative BS: To formulate the capacity when BS cooper-
ation is considered, we let the vector represent the
th column of . If we continue to assume that each MS
transmits from antenna #1, we can construct the channel
matrix whose th column is . Since the MSs cannot co-
operate, the capacity for cooperative BSs is equivalent to the
point-to-point capacity for an uninformed transmitter based on
the composite channel matrix , or
(10)
Cooperative BS/Tx Diversity: Finally, to explore the ben-
efit of exploiting the multiple antennas at each MS, we apply
transmit selection diversity, where each MS selects the antenna
that achieves the maximum signal strength at the BSs, in addi-
tion to the cooperative BS processing. To accomplish this, for
each MS we select the value of that maximizes the norm of
the vector . The vector is used in place
of when constructing the channel matrix for use in (10).
C. Results
The computational results for these multi-user signaling
strategies based on the measured data use or
MSs and an average SISO SNR of 10 dB. The data exhibits a
coherence bandwidth, defined here as the frequency separation
at which the correlation coefficient of the data falls below
10%, of at least 1 MHz, and therefore to save computational
burden the sum rate is computed at 1 MHz intervals over the
full bandwidth and averaged in frequency.
Given the large set of possible combinations of locations for
the different MSs, we must select a multi-user scenario to use
in the analysis. Referring to Fig. 2, the first MS, designated as
, moves along the entirety of routes 1 and 2. The remaining
MSs are either at point 1 or point 2 which are respectively 700
m or 900 m from the starting point along route 2, locations that
allow these MSs to observe multiple BSs. Specifically, when
, simulations are run for the second MS at both
points, while when is at point 1 while is at
point 2.
1) BC Topologies: Before studying composite sum rate
statistics for the presented scenario, we investigate the sum rate
as moves along route 1 between the displacements of 200
m and 1000 m with at point 1 to allow discussion of the
impact of different propagation characteristics on the sum rate.
The top plot in Fig. 7 shows the sum rate achieved assuming
BC MISO signaling for the three topologies discussed in
Section IV.A1, where the data has been smoothed as discussed
in Section III. We first observe that the maximum power pairing
works well compared to the optimal BS-MS pairing when
is on the main roads (e.g., between displacements of 750
and 900 m) and enjoys nearly LOS (or strong urban canyon)
propagation and therefore a dominant link with a single BS.
However, when deviates into a small “inlet” (e.g., between
displacements of 250 and 550 m), the maximum power pairing
increases the multi-user interference, and therefore a different
pairing that reduces interference is beneficial. We emphasize
that in these interference-limited scenarios, the maximum
power pairing would suffer significant additional degradation
630 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 8. CDF of the sum rate achieved for different MISO and MIMO BC sig-
Fig. 7. Sum rates computed for different MISO BC and MAC signaling ap-
naling approaches when travels along the entirety of routes 1 and 2 and
proaches when travels along a portion of route 1 and is at point 2.
is either at point 1 or point 2. The SISO SNR is 10 dB.
The SISO SNR is 10 dB.

were it not for the ability to switch to TDMA. But we also

note that in a few cases (not shown), the TDMA capability
allows the maximum power pairing to outperform the optimal
pairing. Finally, since the link gain for two or more BSs to a
single MS is similar in these regions, allowing the multiple BSs
to collaborate to control interference and maximize link gains
through application of the RCI beamforming weights provides
significant additional sum rate capability.
Fig. 8 plots the cumulative distribution function (CDF)
of the data using both MISO BC and MIMO BC signaling
(Sections IV.A1 and IV.A2) for MSs, with the statistics
computed by concatenating the simulations for at points
1 and 2. It is intuitive that having multiple antennas at the
receivers (MIMO BC) enables an increase in the average sum
rate. However, these results also show that for MIMO BC Fig. 9. CDF of the sum rate achieved for different MISO BC signaling ap-
signaling, the optimal pairing provides a sum rate approaching proaches when travels along the entirety of routes 1 and 2 and and
that achieved using the full RCI beamforming. This situa- are at two different points along route 2. The SISO SNR is 10 dB.
tion occurs because each MS uses knowledge of the transmit
beamforming weights to construct its own MMSE receive
beamforming weights and therefore is able to suppress the bulk of 37% for the two-user scenario considered in Fig. 8. Further-
of the interference. This is in contrast to the case of MISO more, because of the increased interference created by simul-
BC signaling, where the receiver is unable to use array signal taneous transmissions to three different MSs, using the optimal
processing for interference suppression. As a result, the benefit pairing provides some benefit over the maximum power pairing,
of cooperative transmission is generally less significant for although it still naturally falls far short of what is achievable
MIMO BC signaling with this optimal pairing, although it using full BS cooperation.
should be again emphasized that such optimal pairing requires 2) MAC Topologies: The bottom plot in Fig. 7 provides
significant network cooperation and that the multi-antenna the sum rate for the uplink (MAC) scenarios detailed in
reception requires additional complexity at each MS. In most Section IV.B as moves along route 1 between the dis-
networks, achieving this level of complexity is more feasible in placements of 200 m and 1000 m with at point 1. Once
the infrastructure (BS) rather than in the MSs. Therefore, these again, we observe significant performance benefit when the
results reveal that BS cooperation is an effective technique for base stations can cooperate in the multi-user reception. Inter-
dramatically improving the performance. estingly, when is between the displacements of 200 and
Fig. 9 shows the CDF obtained when MSs for MISO 700 m, which is where cooperation tends to give the most
BC signaling. Because , communicating with three MSs benefit, using transmit selection diversity offers little additional
uses all of the spatial resources available from the cooperative benefit. This occurs because the signal processing achieved by
BSs. In this case, the average sum rate achieved with coopera- the cooperative receivers (BSs) already leverages the diversity
tive transmission (RCI) is 91% higher than that achieved with in the channels. However, over the region from 750 to 900 m,
maximum power pairing. This compares with a relative increase where cooperation is less beneficial because each MS has a
LAU et al.: SINGLE AND MULTI-USER COOPERATIVE MIMO IN A MEASURED URBAN MACROCELLULAR ENVIRONMENT
Fig. 10. CDF of the sum rate achieved for different MAC signaling approaches
when travels along the entirety of routes 1 and 2 and and
applicable) are at one of two different points along route 2. The SISO SNR is
10 dB.
dominant link to a different BS, the use of transmit diversity to
overcome fast fading is highly effective.
Finally, Fig. 10 shows the CDF of the sum rate achieved for
the uplink scenarios for both and MSs. These
results reveal increases in the average sum rate of 44% and
63% achieved for 2 and 3 MSs, respectively, assuming only a
single antenna used at each MS (no transmit diversity). Perhaps
the most striking observation is that the average rate achieved
for MSs is only slightly higher than that achieved for
MSs. This occurs because the impact of the increased
total power transmitted is partially offset by the increased inter-
ference when , particularly when only one or two BSs ex-
perience a strong link to the MSs. The effect is particularly dom-
inant when using optimal pairing since in this technique the BSs
cannot cooperate to reduce the multi-user interference, and as a
result the sum rate for MSs is generally lower than that
for MSs. When all three BSs experience a strong link to
the MSs and coherent cooperation is allowed, then the sum rate
achieved for is much larger than that for , as evi-
denced by the differences in the two curves at the high sum-rate
levels (upper portion of the curves). Unfortunately, such situa-
tions are rare, which is why they occur at relatively low proba-
bility.
V. CONCLUSION
This paper uses fully-coherent measurements from three BS
sites to a single MS in a macrocellular environment to explore
the potential gains achievable with cooperative BS communi-
cation for single-user and multi-user scenarios. Specifically,
computations with the data for point-to-point links demonstrate
that the average capacity increases by 53% as a result of co-
operative BS communication. The analysis further shows that
the capacity behavior follows that achieved with i.i.d. Gaussian
random channel matrices whose columns are properly scaled to
achieve the observed BS-to-MS gains. Evaluation of the data
with practical BC signaling strategies shows that cooperation
between the BSs can increase the average multi-user sum rate
by 37% and 91% for two and three MSs, respectively, when
631
each MS has a single antenna. For MAC signaling, the average
improvements are 44% and 63% for two and three single-an-
tenna MSs, respectively, with even more gains achievable
using multiple antennas at each MS. Such dramatic capacity
improvement motivates further study of coherent cooperative
communications for macrocellular settings.
REFERENCES
[1] J. W. Wallace and M. A. Jensen, “Mutual coupling in MIMO wireless
systems: A rigorous network theory analysis,” IEEE Trans. Wireless
Commun., vol. 3, pp. 1317–1325, July 2004.
[2] B. K. Lau, “Multiple antenna terminals,” in MIMO: From Theory to Im-
plementation, C. Oestges, A. Sibille, and A. Zanella, Eds. San Diego:
Academic Press, 2011, pp. 267–298.
[3] M. A. Jensen and J. W. Wallace, “A review of antennas and propagation
for MIMO wireless communications,” IEEE Trans. Antennas Propag.,
vol. 52, pp. 2810–2824, Nov. 2004.
[4] D. Gesbert, S. Hanly, H. Huang, S. S. Shitz, O. Simeone, and W. Yu,
(if “Multi-cell MIMO cooperative networks: A new look at interference,”
IEEE J. Selected Areas Commun., vol. 28, pp. 1380–1408, Dec. 2010.
[5] A. Goldsmith, S. A. Jafar, N. Jindal, and S. Vishwanath, “Capacity
limits of MIMO channels,” IEEE Trans. Inf. Theory, vol. 21, pp.
684–702, Jun. 2003.
[6] S. Zhou, M. Zhao, X. Xu, J. Wang, and Y. Yao, “Distributed wireless
communication system: A new architecture for future public wireless
access,” IEEE Commun. Mag., vol. 41, pp. 108–113, Mar. 2003.
[7] V. Jungnickel, M. Schellmann, L. Thiele, T. Wirth, T. Haustein, O.
Koch, W. Zirwas, and E. Schulz, “Interference aware scheduling in the
multiuser MIMO-OFDM downlink,” IEEE Commun. Mag., vol. 47,
pp. 56–66, Jun. 2009.
[8] N. Jaldén, P. Zetterberg, B. Ottersten, and L. Garcia, “Inter- and intra-
site correlations of large-scale parameters from microcellular measure-
ments at 1800 MHz,” EURASIP J. Wireless Commun. Netw., 2007.
[9] M. Alatossava, A. Taparugssanagorn, and V. Holappa, “Measurement
based capacity of distributed MIMO antenna system in urban microcel-
lular environment at 5.25 GHz,” in Proc. IEEE Vehicular Technology
Conf. Spring, Singapore, May 11–14, 2008, pp. 430–434.
[10] V. Jungnickel, S. Jaeckel, L. Jiang, U. Krüger, A. Brylka, and C. von
Helmolt, “Capacity measurements in a cooperative MIMO network,”
IEEE Trans. Veh. Technol., vol. 58, no. 5, pp. 2392–2405, Jun. 2009.
[11] S. Jaeckel, L. Thiele, A. Brylka, L. Jiang, V. Jungnickel, C. Jandura,
and J. Heft, “Intercell interference measured in urban areas,” in Proc.
2009 IEEE Int. Conf. Commun., Dresden, Germany, Jun. 14–18, 2009,
pp. 1–6.
[12] T. W. C. Brown, P. C. F. Eggers, and K. Olesen, “Simultaneous 5
GHz co-channel multiple-input-multiple-output links at microcellular
boundaries: Interference or cooperation?,” IET Proc. Microw. An-
tennas Propag., vol. 1, no. 6, pp. 1152–1159, Dec. 2007.
[13] R. Irmer, H.-P. Mayer, A. Weber, V. Braun, M. Schmidt, M. Ohm, N.
Ahr, A. Zoch, C. Jandura, P. Marsch, and G. Fettweis, “Multisite field
trial for LTE and advanced concepts,” IEEE Commun. Mag., vol. 47,
pp. 92–98, Feb. 2009.
[14] V. Jungnickel, L. Thiele, T. Wirth, T. Haustein, S. Schiffermüller, A.
Forck, S. Wahls, S. Jaeckel, S. Schubert, S. Gäbler, C. Juchems, F.
Luhn, R. Zavrtak, H. Droste, G. Kadel, W. Kreher, J. Mueller, W.
Stoermer, and G. Wannemacher, “Coordinated multipoint trials in the
downlink,” in Proc. 5th IEEE Broadband Wireless Access Workshop
(BWAWS)/IEEE GLOBECOM Workshops, Honolulu, HI, 30 Nov.–4
Dec. 2009, pp. 1–6.
[15] R. Fritzsche, J. Voigt, C. Jandura, and G. P. Fettweis, “Verifying
ray tracing based CoMP-MIMO predictions with channel sounding
measurements,” in Proc. Int. ITG/IEEE Workshop on Smart Antennas
(WSA’10), Bremen, Germany, Feb. 23–24, 2010, pp. 161–168.
[16] R. Irmer, H. Droste, P. Marsch, M. Grieger, G. Fettweis, S. Brueck,
H.-P. Mayer, L. Thiele, and V. Jungnickel, “Coordinated multipoint:
Concepts, performance and field trial results,” IEEE Commun. Mag.,
vol. 49, pp. 102–111, Feb. 2011.
[17] P. Marsch, M. Grieger, and G. Fettweis, “Field trial results on different
uplink coordinated multi-point (CoMP) concepts in cellular systems,”
in Proc. IEEE Global Telecomm. Conf., Miami, FL, Dec. 6–10, 2010,
pp. 1–6.
[18] Y. Selen and H. Asplund, “3G LTE simulations using measured MIMO
channels,” in Proc. IEEE Global Telecomm. Conf., New Orleans, LA,
30 Nov.–4 dec. 2008, pp. 1–5.
632 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[19] J. W. Wallace, M. A. Jensen, A. L. Swindlehurst, and B. D. Jeffs, “Ex-
perimental characterization of the MIMO wireless channel: Data ac-
quisition and analysis,” IEEE Trans. Wireless Commun., vol. 2, pp.
335–343, Mar. 2003.
[20] S. Plass, X. G. Doukopoulos, and R. Legouable, “Investigations on
link-level inter-cell interference in OFDMA systems,” in Proc. Symp.
on Communications and Vehicular Technology, Liege, Belgium, Nov.
23, 2006, pp. 49–52.
[21] “Analysis for Simulation Scenario Definition to Interference Mitiga-
tion Studies,” Tech. Rep. Document R4-060117, Feb. 2006, 3GPP TSG
RAN WG4.
[22] J. P. Kermoal, L. Schumacher, K. I. Pedersen, P. E. Mogensen, and
F. Frederiksen, “A stochastic MIMO radio channel model with ex-
perimental validation,” IEEE J. Selected Areas Commun., vol. 20, pp.
1211–1226, Aug. 2002.
[23] V. Pohl, V. Jungnickel, T. Haustein, and C. von Helmolt, “The effect
of path loss variations on the capacity of MIMO systems,” in Proc.
5th Eur. Personal Mobile Communications Conf., Glasgow, U.K., Apr.
22–25, 2003, pp. 458–462.
[24] M. Stojnic, H. Vikalo, and B. Hassibi, “Rate maximization in multi-
antenna broadcast channels with linear preprocessing,” IEEE Trans.
Wireless Commun., vol. 5, pp. 2338–2342, Sep. 2006.
[25] A. L. Anderson, J. R. Zeidler, and M. A. Jensen, “Reduced-feed-
back linear precoding with stable performance for the time-varying
MIMO broadcast channel,” IEEE J. Sel. Areas Commun., vol. 26, pp.
1483–1493, 2008.
[26] S. Shi, M. Schubert, N. Vucic, and H. Boche, “MMSE optimization
with per-base-station power constraints for network MIMO systems,”
in Proc. 2008 IEEE Int. Conf. Commun., Beijing, China, May 19–23,
2008, vol. 55, pp. 4106–4110.
[27] F. Boccardi and H. Huang, “A near-optimum technique using linear
precoding for the MIMO broadcast channel,” in Proc. IEEE Int. Conf.
Acoustics, Speech, and Signal Processing, Honolulu, HI, Apr. 15–20,
2007, vol. 3, pp. 17–20.
[28] W. Yu and T. Lan, “Transmitter optimization for the multi-antenna
downlink with per-antenna power constraints,” IEEE Trans. Signal
Processing, vol. 55, pp. 2646–2660, Jun. 2007.
[29] Q. H. Spencer, J. W. Wallace, C. B. Peel, T. Svantesson, A. L. Swindle-
hurst, and A. Gummalla, “Performance of multi-user spatial multi-
plexing with measured channel data,” in MIMO System Technology and
Wireless Communications. Boca Raton, FL: CRC Press, 2006.
[30] G. Caire and S. Shamai, “On the achievable throughput of a multi-
antenna Gaussian broadcast channel,” IEEE Trans. Inf. Theory, vol.
49, pp. 1691–1706, July 2003.
[31] J. Lee and N. Jindal, “High SNR analysis for MIMO broadcast chan-
nels: Dirty paper coding versus linear precoding,” IEEE Trans. Inf.
Theory, vol. 53, pp. 4787–4792, Dec. 2007.
Buon Kiong Lau (S’00–M’03–SM’07) received the
B.E. degree (with honors) from the University of
Western Australia, Perth, Australia and the Ph.D.
degree from Curtin University of Technology, Perth,
in 1998 and 2003, respectively, both in electrical
engineering.
During 2000 to 2001, he worked as a Research En-
gineer with Ericsson Research, Kista, Sweden. From
2003 to 2004, he was a Guest Research Fellow at
the Department of Signal Processing, Blekinge Insti-
tute of Technology, Sweden. Since 2004, he has been
at the Department of Electrical and Information Technology, Lund University,
where he is now an Associate Professor. He has been a Visiting Researcher at
the Department of Applied Mathematics, Hong Kong Polytechnic University,
China, Laboratory for Information and Decision Systems, Massachusetts In-
stitute of Technology, and Takada Laboratory, Tokyo Institute of Technology,
Japan. His primary research interests are in various aspects of multiple antenna
systems, particularly the interplay between antennas, propagation channels and
signal processing.
Dr. Lau is an Associate Editor for the IEEE TRANSACTIONS ON ANTENNAS
AND PROPAGATION and a Guest Editor of the 2012 Special Issue on MIMO
Technology for the same journal. From 2007 to 2010, he was a Co-Chair of
Subworking Group 2.2 on “Compact Antenna Systems for Terminals” (CAST)
within EU COST Action 2100. Since 2011, he has been a Swedish national del-
egate and the Chair of Subworking Group 1.1 on “Antenna System Aspects”
within COST IC1004.
Michael A. Jensen (S’93–M’95–SM’01–F’08)
received the B.S. and M.S. degrees from Brigham
Young University (BYU), Provo, UT, in 1990 and
1991, respectively, and the Ph.D. degree from the
University of California, Los Angeles, in 1994, all
in electrical engineering.
Since 1994, he has been at the Electrical and
Computer Engineering Department, BYU, where he
is currently a Professor and Department Chair. His
research interests include antennas and propagation
for communications, microwave circuit design, and
multi-antenna signal processing.
Dr. Jensen is currently the Editor-in-Chief of the IEEE TRANSACTIONS ON
ANTENNAS AND PROPAGATION. Previously, he was an Associate Editor for
the same journal and for the IEEE ANTENNAS AND WIRELESS PROPAGATION
LETTERS. He has been a member and Chair of the Joint Meetings Committee
for the IEEE Antennas and Propagation Society, a member of the society
AdCom, and Co-Chair or Technical Program Chair for six society-sponsored
symposia. In 2002, he received the Harold A. Wheeler Applications Prize
Paper Award in the IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION in
recognition of his research on multi-antenna communication.
Jonas Medbo received the Högskoleexamen på
Fysikerlinjen degree (B.S.) in physics from Stock-
holm University, Sweden, in 1986 and the Ph.D.
degree in particle physics from Uppsala University,
Sweden, in 1997.
From 1986 to 1989, he was with Ericsson Telecom
AB developing software for protection and mon-
itoring of telecommunication transmission lines.
Since 1997, he has been with the Propagation Group
at Ericsson Research. His main research interests are
measurements and modeling of radio channel and
propagation as well as positioning of user equipment in wireless communica-
tions. He has contributed to the Hiperlan/2, IEEE 802.11 TGn, COST 273 and
3GPP SCM channel models.
Johan Furuskog received the M.Sc. degree in en-
gineering physics from Uppsala University, Uppsala,
Sweden, in 2007.
After graduating, he joined Ericsson Research,
Kista, Sweden, as a researcher where much of his
work has concerned testbed development and field
trials with focus on MIMO channel characteristics
and LTE multi-antenna performance. He is currently
involved in testbed projects targeting design and
deployment of future radio access technologies and
works with concept development and evaluation of
new components for the LTE standard.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
User Influence on MIMO Channel Capacity for
Handsets in Data Mode Operation
Jesper Ødum Nielsen, Boyan Yanakiev, Ivan B. Bonev, Morten Christensen, and Gert Frølund Pedersen
Abstract— The current paper concerns realistic evaluation of the
capacity of the MIMO channel between a BS and handheld device,
such as a PDA or smartphone, held in front of the user’s body (data
mode). The work is based on measurements of the MIMO channel
between two widely separated BSs in a micro-cellular setup, and six
handsets located in an indoor environment. The measurements are
done simultaneously in both the 773.5–778.5 MHz and 2250–2350
MHz bands, and from the two BSs. The handsets are realistic types
and were measured both in free space and with twelve different
users, using both one and two hands. The random capacities of the
channels are evaluated in terms of outage capacity. For an SNR
of 10 dB, median capacities in free space of about 4.4–4.7 bit/s/Hz
for the low band and about 3.3–3.8 bit/s/Hz for the high band were
found. The mean decrease in outage capacity due to the user was
found to be up to about 2.2 bit/s/Hz, depending on the band and
handset. More results are presented in the paper.
Index Terms—Channel capacity, dual-band propagation,
MIMO channels, optical link, propagation measurements,
user-interaction.
I. INTRODUCTION
D URING the last 10–15 years it has been known that the
power transmitted and received from a mobile handset (or
cellular phone) may vary significantly. The importance of this
has often been reported with differences of several dB’s found
between handsets [1], and in some cases variations of more than
10 dB were found for different users of the same handset [2]–[5].
The large variations stress the importance of including the user
in the design and testing of future handsets, since this has an
impact on network performance, battery lifetime, and general
user experience.
For a long time handsets have typically been used in talk
mode, i.e., the situation where the handset is held by the user
next to the head for voice usage. A current market evolution is
from voice-centric devices into devices where data and appli-
cations are equally or more important, such as for smart mobile
platforms or smartphones, collectively referred to as “handsets”
Manuscript received September 17, 2010; revised June 02, 2011; accepted
June 16, 2011. Date of publication October 25, 2011; date of current version
February 03, 2012. The work of J. Ø. Nielsen and B. Yanakiev was supported
in part by the Danish National Advanced Technology Foundation via the Con-
verged Advanced Mobile Media Platforms (CAMMP) project. The results and
conclusions presented by the authors in this article are not necessarily supported
by the other partners of the CAMMP project.
J. Ø. Nielsen, I. B. Bonev, and G. F. Pedersen are with the Antennas, Propaga-
tion and Radio Networking Section, Department of Electronic Systems, Faculty
of Engineering and Science, Aalborg University, DK-9220 Aalborg, Denmark.
B. Yanakiev and M. Christensen are with Molex Antenna Business Unit,
DK-9220 Aalborg, Denmark.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173435
0018-926X/$26.00 © 2011 IEEE
633
in the following. With this trend, data mode operation becomes
more important where the handset is in front of the user and
held with one or two hands. The locations of the user’s hands
and fingers on the handset may be different from those used in
talk mode [6]. It is known that the user’s hand is the single most
important issue when considering the variation in performance
in terms of power obtained with different users [7]. Therefore
large performance variations may also be expected in data mode
operation, since the user’s fingers still may interact with the
antennas.
Along with the trend towards data mode operation comes a
demand for higher data rates. Given the scarcity of radio spec-
trum, a promising way to achieve higher data rates is to employ
multiple-input multiple-output (MIMO) techniques utilizing
several antennas on both the transmitter (Tx) and receiver (Rx)
side. For example the upcoming long-term evolution (LTE)
standard has MIMO capabilities [8].
Today’s mobile handsets are densely packed with battery,
electronics, and are often equipped with several antennas for
different systems. Since small handsets are generally preferred
by the users, adding more antennas for MIMO will be difficult
and require compromises to be made between the performance
and the design and location of the antennas on the handset. The
influence of the user’s hand on the MIMO performance will be
crucial.
It is well known that the performance of a MIMO system is
highly dependent on the properties of the radio channel between
the Tx and Rx [9], and thus must be included in evaluation.
Given that the user interacts with the handset antennas in the
near-field, possibly in a dynamic way, it is difficult to include all
aspects of both the mobile environment and the user influence
without actually including both in a performance measurement.
The work in [10] reports on some of the first results on MIMO
performance for handheld devices based on propagation mea-
surements with a handset and several live users.
It may be possible to simplify the evaluation, e.g., by using
radiation pattern measurements including users, similar to
what has been done in the context of single-input single-output
(SISO) handset performance evaluation, see [11], [12]. Eval-
uation of diversity systems, i.e., single-input multiple-output
(SIMO), in handsets have also been carried out in this way
including phantoms of the user head and hand for talk mode
scenarios in [13] and data mode in [14].
Another approach to performance evaluation is presented in
[15]. Here a combination of the radiation pattern measurements,
including user phantoms, and models of the propagation channel
is used, where the model describes all individual plane waves in
the channel. Assuming far-field conditions, this method allows
a practical separation of antenna measurements and propagation
634 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

measurements. The work in [15] considers only talk mode. Data

mode operation results are given in [16] and [17], where a sig-
nificant user influence on the capacity was found.
For data mode, the work in [18] studies the influence of the
user’s hand on the capacity, based on simulations of the channel.
Based on simulations of both the channel and the radiation pat-
terns, [19] considers the influence of the user’s body when the
device is carried in a pocket. In the latter two references a signif-
icant reduction of (ergodic) capacity was found due to the user.
Other related work includes [20] where the capacity of
handsets in data mode is studied with special focus on
cross-polarization difference (XPR), based on anechoic room Fig. 1. View from the antenna location of BS2.
measurements, and [21] where methods for MIMO antenna
evaluation are studied, utilizing channel measurements, but
focusing on methods rather than practical devices. The early
work in [22] studies the performance of different principal
antenna types based on propagation measurements, but without
user influence. From simulations and measurements in a setup
with dipoles in a reverberation chamber, including a simple
user phantom and assuming uncorrelated Rx branches, the
work in [23] provides a parametric study of how the capacity is
influenced by the reduced efficiency and signal blocking, that
may be introduced from a nearby user.
Much of the earlier work employs phantoms to mimic the in-
fluence of the user, but issues like dynamic behavior and varia-
tions in the MIMO performance among the users are difficult to
include with phantoms. Furthermore, results on different types Fig. 2. View towards BS1 from the measurement site.
of handsets antennas used in data mode are scarce.
The main topic of the current work is the performance
evaluation, in terms of capacity, of different realistic MIMO TABLE I
OVERVIEW OF THE TWO BASE STATIONS
handsets. Focus is on both achievable capacity as well as the
influence of the users of the handsets. The investigations are
based on an extensive radio channel measurement campaign in
a micro-cellular setup. Simultaneous measurements were car-
ried out from two base stations (BSs) in both 773.5–778.5 MHz
and 2250–2350 MHz bands. Six realistic handsets of different
types, all equipped with two antennas, are all measured in an
indoor environment both in free space and with twelve users in
of both high capacity and coverage. In an attempt to create a re-
data mode.
alistic scenario for the measurements used in the current work,
The next section describes the measurements in more detail,
a setup with two base stations (BSs) was used. BS1 was envi-
including the developed handsets equipped with optical units
sioned to result in high capacity channels, being located some
ensuring correct data acquisition. Section III describes the pro-
150 m away from the measurement building with partial line of
cessing of the raw measurement data, while Section IV concerns
sight (LOS) and the antennas near rooftop height of surrounding
the obtained results on mean effective gain (MEG), capacity and
buildings. In contrast, BS2 was located about 500 m away on
the user influence. Section V concludes the paper.
top of a tall building overlooking the surrounding buildings. An
overview of the base stations is given in Table I and Figs. 1–2.
II. MEASUREMENT SETUP Both indoor and outdoor measurements were made, where
the current paper focuses on the indoor part. The measurements
A. Scenario
took place inside a 3rd floor room with windows towards BS1,
Successful use of spatial multiplexing modes in a MIMO where the LOS was partly blocked by buildings. In the room a 4
system requires a rich scattering environment with a wide an- m by 4 m square was marked on the floor. During the first 5 s of
gular spread of scatterers both near the Tx and the Rx. This gen- a measurement the user walked from a corner forward along one
erally results in a high rank channel matrix with low correlation side of the square to the next corner; the next 5 s the user walked
among the elements, which in turn results in a high channel ca- backwards towards the first corner. This was then repeated re-
pacity [24]. For a cellular network a BS should preferably be sulting in a total measurement time of 20 s in which the user
near rooftop level or below and not in a highly elevated loca- kept the same orientation. Four handsets (described below) were
tion that might be preferred from a network coverage point of measured simultaneously, held by four test users each walking
view. Clearly a successful network has to provide a compromise along one of the four sides of the square.
NIELSEN et al.: USER INFLUENCE ON MIMO CHANNEL CAPACITY FOR HANDSETS IN DATA MODE OPERATION 635

B. Frequency Bands TABLE II

OVERVIEW OF HANDSETS USED
Two bands were measured simultaneously. An effective
sounding bandwidth of about 5 MHz was used at the center
frequency of 776 MHz. This band is subsequently referred to as
the low band (LB). The high band (HB) was centered at 2300
MHz where an effective sounding bandwidth of about 100 MHz
was used. The two bands were chosen to resemble the LTE bands
in the 700–800 MHz and 2.3–2.6 GHz ranges, respectively
[25]. In practice, both the center frequencies and the bandwidths
are compromises given the available equipment and unused
frequency spectrum, resulting in the unequal bandwidths.

C. Handsets
The six handsets used in this work are special mock-up hand-
sets which are realistic with respect to the antennas, electro-
magnetic properties, shape and handling, and at the same time
allows for connection to the channel sounding equipment. A
straightforward approach would be to connect the antennas in
the handsets to the sounder using conventional coaxial cables,
but this is an undesirable solution. It is well known that the use
of conducting cables on small devices changes the electromag-
netic properties, because the cable becomes part of the antenna
[26]–[28]. Coaxial cables for measurements need to be low-loss
and phase-stable, and are typically of the order 1 cm in diam-
eter, somewhat inflexible and heavy. Attaching such a cable to
a small device often makes its handling difficult and hence un-
natural, where it is noted that a stiff choke may be needed on
Fig. 3. Handset grips, one hand (OH) for H2 (left) and two hand (TH) for H1
the cable, in order minimize the cable influence. In addition, the (right).
cable may have to be lead out at an awkward location on the
device with respect to easy handling.
An attractive way to avoid the above mentioned problems is on the handsets and held the handset in front of the body at an
to use an optical fiber between the handset and the sounding angle of about 45 . The two grips are shown in Fig. 3.
equipment. By modulating a laser diode with the RF signals As mentioned above, variation in performance is expected
received by the antennas it is possible to transfer the signals to among the users and therefore more users are involved to allow
the sounder using a flexible plastic fiber. The main difficulty is in averaging. Since no a priori information exist on the capacity
designing optical units that are small enough to fit into a typical distribution, measurements with twelve users were carried out
handset. As described in detail in [29], this has been done for based on the experience with measurements of MEG [30]. All
the current work. combinations of the four square sides, two grips and twelve
The six handsets used in this work all have integrated optical users were measured twice. Firstly with the handsets H1, H2,
units and all have two antennas, single or dual-band. All the H3, H4, and secondly with the handsets H1, H5, H6, H7.
handsets were placed in a plastic casing from PC-ABS material In addition all handsets were measured in free space where
made in a rapid prototyping printer. The material has , the handsets were mounted at an angle of 45 using Styrofoam
which is comparable to most plastics found in today’s phones. on top of a table with wheels. The table was then pushed by a
The reason for this is to mimic the user handling as closely as person (bending down) to be measured in the same way as with
possible. The plastic covers provide natural feeling and prevents the users. These measurements were made twice.
the user from directly touching the PCB and disturbing the cur-
rents and fields in an abnormal way. Finally, grip markings were E. Sounder Setup
embedded on the covers for better grip control. An overview of The measurements were carried out using a MIMO channel
the six handsets is given in Table II. Note that ‘H6’ is missing sounder [31], allowing truly simultaneous measurement of the
from the list. This handset was part of the measurement cam- channels from all seven (three LB and four HB) Tx antennas on
paign, but broke during the campaign and therefore the data was the base stations, to the four dual-band Rx antennas. These four
discarded. In all cases the handsets are designed for 50 MHz and Rx antennas are located, one in each, in the four different hand-
100 MHz bandwidth in the LB and HB, respectively. sets that are measured at the same time. As described above,
each handset has two antennas which are connected via a multi-
D. User Grips and Repetitions
plexing switch. Hence, eight dual-band Rx antennas are mea-
Two grips were used, one hand (OH) and two hand (TH). In sured, so that in total a 7 16 MIMO (Tx Rx) wide band
each case the users placed their fingers in predefined markings channel matrix was measured at a rate of 60 Hz to cope with
636 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
channel changes due to the movements of the users and other
changes in the channel.
III. MEASUREMENT STATISTICS
Given the measurements described in the previous section,
different MIMO constellations can be studied, i.e., which fre-
quency band and how many Tx and Rx antennas are used. The
following are considered in this work,
BS1,Lo The two LB Tx antennas from BS1 are used
to form a compact 2 2 MIMO setup for
each handset.
BS1,Hi1-2 Similarly, two of the HB Tx antennas from
BS1 are used to form a compact 2 2
MIMO setup.
BS1,Hi1-4 All four HB Tx antennas from BS1 are used,
resulting in a compact 4 2 MIMO setup.
BS1 2,Lo This is a 3 2 distributed MIMO setup
where the two Tx antennas from BS1 are
used in addition to the single Tx antenna on
BS2. Via the normalization described below
it is assumed that the Tx power is adjusted
so that the average Rx power is the same
from BS1 and BS2.
The MIMO channel is described by the matrix con-
sisting of the elements where indices denote the
-th square side, the -th Tx antenna, -th Rx antenna, and -th
time index. The MIMO constellations and the choice of handset
defines the channels used, and is indicated by the -index. The
scalar is the complex gain of the narrow-band channel
between the Tx and Rx antennas, obtained via discrete Fourier
transforms of the measured impulse responses (IRs).
To ensure a fair comparison the channels are normalized to
the mean power of all handsets in free space. The mean is com-
puted independently for every Tx antenna, mainly to remove
path loss differences in the distributed MIMO case and between
bands. The free space average power gain for the -th Tx an-
tenna is computed as
(1)
where is the number of sides of the square, is the
number of Rx antennas of the handsets, and is the
number of IR samples along each side. The averaging is done
over handsets. The normalized channel matrix has
the elements
(2)
Assuming no knowledge at the Tx about the channel state,
the instantaneous channel capacity is given by [9]
(3)
where the signal to noise ratio (SNR) is is the -th eigen-
value of the matrix and .
The number of Tx antennas for the constellation is given by
. The channel capacity is random, and hence a sta-
tistical approach is needed. A useful measure is the outage ca-
pacity (OC) [32], which is the value such that the probability
, where is the probability level in per-
cent. Thus, the term OC is another name for capacity percentile.
This work focus on , and , i.e., OC at the 10%, 50%,
and 90% levels, respectively. The percentiles are found from
the empirical cumulative distribution functions (CDFs) by com-
bining all instantaneous capacities from all four square sides,
i.e., for all values of and .
The capacity results presented in this work are assuming an
SNR ratio of 10 dB for the average handset, obtained via the
normalization described above. This is equivalent to fixing the
Tx power and is aimed at creating a fair comparison among the
handsets. For example some handsets may have antennas with
higher efficiency than the average and as a result effectively
have a higher average SNR.
The issue of normalization and hence SNR ratio is related
to the update rate of the power control in the cellular system.
The normalization chosen in the current work is based on the
average over the complete route (four sides of the square path).
Hence both slow and fast fading is preserved and the SNR will
vary locally along the route, depending on the handset antennas
and the channel. This ensures a fair comparison of the handsets,
which would be difficult if, e.g., the slow fading was estimated
and removed individually for the handsets, approximating fast
power control.
With the aim of understanding capacity results it is useful
to study the SISO channels comprising the MIMO channels in
terms of MEG. The MEG was originally defined as the ratio
of the average power obtained with an antenna under test to
the average power obtained with a reference antenna, where the
averaging is over measurements carried out in the same realistic
environment [33].
Denoting by a complex sample of the IR at
time-index , delay-index , for the -th Tx antenna, -th Rx
antenna, and measured in the -th side of the square in the room,
the average total power gain is computed as
(4)
where is the number of delay samples, is the
number of Tx antennas for the considered band and base. The
meaning of and are defined as for (1). The value of
may be viewed as the MEG for the -th antenna of handset/
band , where the reference antenna is a hypothetical antenna
collecting all the transmitted power in both polarizations. Note
that is computed in (4) using the wideband data since the
measurements are calibrated for equal Tx power in the LB and
HB, having different bandwidths.
The body loss (BL) for the -th Rx antenna is defined as
the ratio of average total power gains with and without a user [5]
(5)
NIELSEN et al.: USER INFLUENCE ON MIMO CHANNEL CAPACITY FOR HANDSETS IN DATA MODE OPERATION
Fig. 4. The MEG in free space conditions. The -axis indicates the handsets.
The different lines in the plot indicates combinations of base, band, and Rx an-
tenna element. The measured points are connected by lines only to ease reading.
where is the average total power gain in free space
conditions, and is the gain when a user is present. The
body loss not only includes signal power absorbed in the user’s
body, but also indirect changes in the received power due to the
user, such as de-tuning of the antenna and load-pull of power
amplifiers in case of uplink transmission. In the following all
MEG statistics are based on the logarithms of the mean channel
gain . ther OH (one-hand) or TH (two-hand). The different lines in the plot indicates
The presence of a user is expected to result in lower OC
compared to the free space case [17], [34]. With the purpose
of studying this influence, the term capacity loss (CL) is intro-
duced. In analogy with the BL, the CL is the difference in OC
obtained with and without the user when handset is operated in
the same environment. More precisely, the CL is defined as
(6)
where is the OC at the % level, for the -th user, and
is the similar OC obtained from the -th measurement
in free space conditions.
In order to also quantify the variation of the OC among the
users, the capacity variation (CV) is defined as the sample stan-
dard deviation
(7)
where is the mean OC among the users.
IV. RESULTS
A. Free Space MEG
The MEG for the free space case is shown in Fig. 4, where
the handsets are given on the -axis and all combinations of
the two base stations, the two bands, and the two Rx antennas
are shown using different lines. First of all it is evident that the
gains for the channels originating in BS2 are much smaller than
those from BS1. This is due to the longer distance and hence
path-loss. Furthermore, for BS1 the HB channel has a higher
loss than the corresponding LB channel. Table III lists the path
loss averaged over the handsets and Rx antennas, and here the
637
TABLE III
PATH LOSS OBTAINED FROM FREE SPACE MEASUREMENTS. AVERAGE OVER
ALL HANDSETS AND RX ANTENNAS
Fig. 5. The mean of the body loss obtained with 12 different users. The -axis
labels are in the form Hn/Grip, where “Hn” is the handset and “Grip” is ei-
combinations of base, band, and Rx antenna element. The measured points are
connected by lines only to ease reading.
LB to HB difference is found to be 10.4 dB. Using Friis’ power
transmission equation and assuming, for a moment, free space
propagation conditions and identical gains in both the Tx and
Rx antennas, the change in frequency alone results in about 9.4
dB power difference. Although these assumptions are dubious
it illustrates the importance of the frequency dependence of the
channel gain.
The MEG depends on the joint properties of the channel in
terms of power distribution versus angle, and the properties of
the handset in terms of radiation patterns, including polarization
and efficiency. The performance may be analyzed using these
terms, see e.g., [12], but here it is simply noted from Fig. 4
that there may be several dB’s of difference between the two
Rx antennas of the same handsets, especially for H1, H2, and
H7, as well as among the handsets.
B. Body Loss
The mean BLs of all combinations of handset, grip, base,
band, and Rx antenna are shown in Fig. 5. From the plot both
very high values of about 15 dB are found and also very low
found, down to about dB.
The negative BL of about dB for H2 is for the Rx2
antenna which is located at the top of the handset, and therefore
may be affected only slightly by the users, as evidenced by the
generally small BL values for this handset. Although a negative
BL is possible theoretically, the observed negative BL may also
be the result of a small BL and measurement inaccuracy (see
later in Section IV.E). Also for H7, the negative BL is obtained
for Rx2 which is located at the top of the handset. The very high
13–15 dB BL found for H7, LB, Rx1 has been identified to be
638
Fig. 6. The outage capacity (OC) for the different handsets in free space conditions. The four plots represent different MIMO constellations. Top-left: BS1, low
band. Top-Right: BS1 2, low band. Bottom-left: BS1, high band, Tx1-2. Bottom-Right: BS1, high band, Tx1-4. The measured points are connected by lines only
to ease reading.
caused by severe de-tuning. This antenna is furthermore located
at the bottom of the handset and hence likely to be affected by
the users.
1) Top/Bottom Differences: Some of the handsets have both
an antenna mounted at the top as well as the bottom of the
handset, where the user is much more likely to influence the
antenna performance. For these handsets the mean difference in
BL for the bottom and the top mounted antenna is about 5.5 dB.
For all these handsets the bottom antenna has a higher BL than
the top antenna, but the difference is varying from about 0.4 dB
for H1, BS1, HB to about 14 dB for H7, BS2, LB.
2) OH/TH Differences: When the TH grip is used the BL is
about 1.5 dB larger on average compared to the BL when the
OH grip is used. Again, the differences vary depending on the
specific combination, but in all cases the TH grip results in the
largest BL, ranging from about 0.1 dB for H2, BS2, LB, Rx2,
to about 4 dB for H1, BS1, HB, Rx1.
3) Left/Right Differences: Regarding the handsets where
both the antennas are top mounted, the two antennas may also
have a difference in the BL. For H3 the right antenna has a BL
4–5 dB larger than the left antenna. For H4 the difference is
smaller and less clear, and which antenna has the largest body
loss depends on the grip. The BL for the left antenna of H5 is
about 1.1 dB larger than for the right. Thus, there is no clear
tendency and this probably depends on the particular design.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Finally, it is noted that the BL obtained with a given antenna
is very similar for BS1 and BS2, as expected.
C. Free Space Capacity
Fig. 6 shows the OCs at the 10%-, 50%-, and 90%-level for
free space conditions. The results are computed as described
in Section III, using the four different MIMO constellations. It
should be noted that H3 does not have LB antennas and H5 does
not have HB antennas.
The expression in (3) shows that in general the capacity de-
pends on both the eigenvalues and the SNR. However, it is well
known that capacity is strongly dependent on the SNR with a
weaker dependence on the eigenvalues [35]. Therefore it is not
surprising that the results of Fig. 6 agree well with the MEG
shown in Fig. 4.
H4 only has a single antenna in the LB which explains the
generally lower capacity of this handset compared to, e.g., H1.
Although H7 is a two antenna handset in both the LB and HB
the performance in terms of power is rather poor (Fig. 4). In
both bands it is essentially a single antenna Rx, resulting in a
generally low capacity.
With the above comments on power for the LB, 2 2 MIMO
can be formed effectively for H1, H2, and H5. Comparing
these handsets, the OCs are found to be fairly similar, with ,
NIELSEN et al.: USER INFLUENCE ON MIMO CHANNEL CAPACITY FOR HANDSETS IN DATA MODE OPERATION 639

Fig. 7. The mean reduction in the OC when the user is present compared to free space. The -axis labels are in the form Hn/Grip, where ‘Hn’ is the handset and
‘Grip’ is either one hand (OH) or two hand (TH). The four plots represent different MIMO constellations. Top-left: BS1, low band. Top-Right: BS1 , low band.
Bottom-left: BS1, high band, Tx1-2. Bottom-Right: BS1, high band, Tx1-4. The measured points are connected by lines only to ease reading.

, and in the ranges (2.7–3.1, 4.4–4.7, 5.9–6.1) bit/s/Hz,
respectively.
For the HB, 2 2 MIMO can be formed effectively with
H1, H2, H3, H4. Here the OCs are similar for H2-4, (1.9–2.4,
3.3–3.8, 4.7–5.3) bit/s/Hz for the 10%-, 50%, and 90%-levels,
respectively, but significantly higher for H1, (3.2, 4.8, 6.8)
bit/s/Hz for the three levels. This can be attributed to a higher
signal to noise ratio due to a larger MEG for this handset
compared to the rest.
1) LB/HB Differences: Comparing the obtained OCs for the
LB and HB no clear tendency is apparent. For H1, the HB has
the higher OCs 0.4–1.0 bit/s/Hz, whereas for H2 the LB has
higher OCs by 1.2 bit/s/Hz. For H4 the HB OCs are larger by
0.9–1.1 bit/s/Hz, while for H7 they are about the same with
differences of to 0.2 bit/s/Hz. It should be recalled from
Section IV.A that compared to the LB, the HB requires about 10
dB higher Tx power to obtain the same SNR.
2) Extra Tx Antennas, Same BS: Comparing the OCs for
the two MIMO constellations BS1,Hi1-2 and BS1,Hi1-4 it is
found that the two extra Tx antennas do increase the OC, at
least for H1-4. Thus, the extra antennas provide more diversity,
although the improvement is marginal. H1 benefits the most,
0.3–0.5 bit/s/Hz, while for H2-4 the OC generally increase by
about 0.2 bit/s/Hz.
3) Extra BS: Introducing extra diversity by means of an
extra BS may also be beneficial. Comparing the results for the
BS1,Lo constellation with those of the BS1 2,Lo constellation
reveals that for H2 it is improved 1–1.3 bit/s/Hz for the three
OC levels whereas for H1, H4, H5, H7 mainly the values
are improved by 0.6–0.9 bit/s/Hz, followed by the values by
0.3–0.6 bit/s/Hz, while the values for H1, H4, and H7 are
changed by bit/s/Hz. For H5 is larger by 0.6 bit/s/Hz,
but the overall tendency for H1, H4, H5, H7 is that the is
improved the most and the main effect of the extra BS is to in-
crease the diversity in the channel.
D. User Influence on Capacity
The CL defined in (6) is shown in Fig. 7 for the handsets
measured in the current work. Below these results are analyzed
from different viewpoints.
1) Handset Differences in CL: Comparing the CL observed
for the different handsets, it is immediately apparent that H1 has
the highest CL, about 2.2 bit/s/Hz in mean over all grips, levels,
and constellations. H1 is of the PDA type and has one of the
antennas at the bottom where it may be affected by the users, as
evidenced by a high body loss (see Fig. 5).
H3 and H5 have medium CL of about 1.3 bit/s/Hz in mean,
despite being relatively small bar types of handsets. The reason
640 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE IV
INCREASE IN OUTAGE CAPACITY (OC) OBTAINED BY ADDING MORE TX
ANTENNAS, EITHER ON AN EXTRA BASE (BS2), OR ON THE SAME BASE (BS1).
SHOWN VERSUS HANDSET AND OC LEVELS AND COMPUTED AS MEAN
OVER GRIPS
may be that both antennas are located at the top of the handsets,
and thus somewhat protected from user influence.
Handsets H2, H4, and H7 have low CL. H2 is a relatively long
(when open) clamshell type that seems to protect the antennas
from the influence of the users, with a CL of about 0.8 bit/s/Hz in
mean. Again, Fig. 5 shows that this handset also has a relatively
small BL.
H4 is of the PDA type, but unlike H1 with both antennas at
the top where the users are unlikely to touch. For H4 the mean
CL is about 0.8 bit/s/Hz.
H7 is effectively a single antenna handset, where only the top
mounted antenna is receiving significant power. This may ex-
plain why this handset in the mean has a CL of only 0.5 bit/s/Hz,
the lowest of the handsets.
2) Dependence of CL on Level: Comparing the CL for the
different OC levels it appears that sometimes and are
changed more than the corresponding mainly for H1 in all
constellations, but also, e.g., for H4 and H5 in the LB. Thus, in
these cases there is a tendency that high instantaneous capacity
values are reduced more than low values.
3) Frequency Dependence of CL: Regarding dependence of
the CL on the frequency band, H1, H2, H4, H7 are interesting
since they are dual band. Comparing results for BS1,Lo and
BS1,Hi1-2, there is a tendency that the CL is higher for the HB
than for the LB, in mean by about 0.5 bit/s/Hz.
4) CL for Extra Tx Antennas on the Same Base: Comparing
the CLs for BS1,Hi1-2 and BS1,Hi1-4, i.e., when using two or
four Tx antennas for the HB, it seen that the CL is generally
larger for the BS1,Hi1-4 constellation. The overall mean differ-
ence is about 0.16 bit/s/Hz, but for H1 they are generally larger,
about 0.3 bit/s/Hz.
The overall increase in OC by adding the two extra Tx an-
tennas is shown in the right half of Table IV, where the CLs due
to the users are included. From the table it is clear that the OC
improve marginally.
5) CL for Extra Base Station: Similarly, the CLs for the con-
stellations with or without the extra BS is compared, i.e., results
for BS1,Lo and BS1 2,Lo. The general tendency is that the CL
is larger for the BS1 2,Lo constellations with an overall av-
erage of about 0.25 bit/s/Hz. The overall gain by using the extra
BS2 transmitter is shown as the left part of Table IV, where it
is clear that there is a gain. The question is obviously whether
this gain of maximally 0.7 bit/s/Hz justifies the extra cost and
difficulties associated with implementing a distributed MIMO
system.
The capacity variation (CV) is defined above in (7) and Fig. 8
shows the computed values for the different combinations of
handsets, probability levels, and MIMO constellations.
6) Dependence of CV on Level: A first observation is that
there is a clear tendency that , i.e., large
capacity values are more sensitive to the variations that the users
introduce.
7) Handset Differences in CV: On average
, and bit/s/Hz but there are some variations
around these mean values depending on both the handset and
band.
On the LB the CV for H2 is in general larger than for H1,
perhaps explained by the smaller size of H2 in the part of the
clamshell with the user grip. Both H1 and H2 have an antenna at
the bottom, and the smaller size could allow for more variation
in the grip style. Also H4, with only top antennas, has roughly
the same or less variation as does H1.
On the other hand, H2 has significantly less variation on the
HB, whereas H1 and H4 have roughly the same variation as
on the LB. Thus, although size and location of antennas could
explain some of the CV, specific design of the antennas seems
to be important too.
Comparing the results for BS1,Lo and BS1,Hi1-2, the CV tend
to be a bit lower for the HB than for the LB, about 0.1 bit/s/Hz
in the mean. A possible explanation for this is that the whole
handset tends to act as antenna for the LB, where for the HB the
radiating parts are more confined to the antenna element area.
The difference is more pronounced for H2 and H7 than for H1,
since the latter has one antenna at the bottom where the user
holds and the antennas on H2 and H7 are located where they
allow more freedom for variations in the influence of the user.
8) CV for Extra Tx Antennas: The CV values obtained
for the two MIMO constellations at the LB, i.e., BS1,Lo and
BS1 2,Lo, are roughly the same. Also the CV values for
BS1,Hi1-2 and BS1,Hi1-4 are roughly the same.
E. Repeatability
In the preceding sections the performance of the mobile hand-
sets is studied in terms of OC obtained from the measurements.
In order to reach conclusions, it is important to address the re-
peatability of the combined measurement and processing. In
principle a repeated measurement with the same user should
yield the same OC, but in practice this will not be the case for
several reasons, including the following.
• Noise and other errors in the measurement system.
• Differences in the handling of the handset, such as exact
location of the user’s fingers. Even if the user is instructed
to use the same grip, small changes are inevitable.
• Similarly, minor changes in, e.g., the user’s route, orienta-
tion, and walking speed must be expected.
• Changes in the surrounding environment.
The repeated measurements allow to investigate the repeata-
bility of the derived channel capacity statistics. Every combi-
nation of MIMO constellation, user, and grip results in repeated
samples of OC. Based on these values (in total 96), percentiles
NIELSEN et al.: USER INFLUENCE ON MIMO CHANNEL CAPACITY FOR HANDSETS IN DATA MODE OPERATION 641

Fig. 8. The STD of the OC when the user is present. The -axis labels are in the form Hn/Grip, where “Hn” is the handset and “Grip” is either one hand (OH)
or two hand (TH). The four plots represent different MIMO constellations. Top-left: BS1, low band. Top-Right: BS1 2, low band. Bottom-left: BS1, high band,
Tx1-2. Bottom-Right: BS1, high band, Tx1-4. The measured points are connected by lines only to ease reading.

TABLE V TABLE VI
PERCENTILES OF DEVIATIONS FROM MEAN IN REPEATED MEASUREMENTS. PERCENTILES OF DEVIATIONS FROM MEAN IN REPEATED MEG
THE ROWS OF THE TABLE REPRESENTS THE OUTAGE CAPACITY (OC). THE MEASUREMENTS. ALL VALUES ARE IN DB
COLUMNS SHOW THE PERCENTILES OF THE DEVIATIONS FROM THE MEAN OF
THE REPEATED OC. THE VALUES ARE IN BIT/S/HZ

table it is noticed that 90% of the observations are within about

were computed to obtain an overview of the repeatability. Sim-
ilar to the measurements with users, statistics were computed
from the in total 64 combinations in free space. The percentiles
regarding accuracy are shown in Table V for both free space and
H1. It is noticed that 90% of the observed differences are 0.26
bit/s/Hz or below, and that the deviations tends to increase with
the OC level.
Similarly, the repeatability of the measured MEG was
studied. Every combination of base, band, Rx antenna, grip,
and person resulted in repeated samples of power, in total 144
samples. For the free space case in total 48 combinations are
available. Table VI shows the percentiles of the absolute differ-
ences for both the free space and user measurements. From the
dB and dB of the mean value in the free space and
user cases, respectively. Furthermore, in all cases the free space
percentiles are smaller than those for the user cases, indicating,
as expected, that the user introduces extra variability in the
measurements. However, the largest part of the variation is due
to other sources.
V. CONCLUSION
The user influence on the channel power gain was investi-
gated in terms of the body loss (BL). Similar to previous find-
ings for talk mode, the BL in data mode was found to depend
highly on the design of the handset and the usage, with approx-
imate mean values ranging from 0 dB to 10 dB.
642 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
In free space the outage capacity (OC) is generally similar
for the handsets, but a high MEG also results in a higher OC,
as this effectively gives a higher SNR. Measured values of the
50% OC were 3.3–4.7 bit/s/Hz for an SNR of 10 dB, depending
on handset and frequency band. The path loss is about 10 dB
higher on the high band (HB) than on the low band (LB).
As expected, the OC is reduced the most when users are likely
to touch areas near the antennas. A reduction of up to about
2.2 bit/s/Hz was found for a handset with an antenna at the
bottom, compared to about 0.8 bit/s/Hz for a relatively large
handset with top mounted antennas. In mean the OC reduction
is 0.5 bit/s/Hz higher on the HB than on the LB.
Using more Tx antennas on the base than on the mobile may
in free space introduce extra diversity, increasing the 10% OC
about 0.2 bit/s/Hz, but when users are introduced the increase is
only marginal. Using an extra BS (distributed MIMO) can pro-
vide some extra diversity in the free space case, with a 10% OC
increase of up to about 0.9 bit/s/Hz, which is reduced to a max-
imum of about 0.7 bit/s/Hz when the user is present. Note that
these numbers are for the best case with no path loss differences
between the BSs.
ACKNOWLEDGMENT
The authors would like to thank Prof. C. Luxey and this col-
leagues at the University of Nice Sophia-Antipolis, France for
developing the special decoupling technique and designing the
antennas used in two of the handsets.
REFERENCES
[1] L. M. Correia, Wireless Flexible Personalised Communications. COST
259: European Co-Operation in Mobile Radio Research. New York:
Wiley, 2001.
[2] M. Murase, Y. Tanaka, and H. Arai, “Propagation and antenna mea-
surements using antenna switching and random field measurements,”
IEEE Trans. Veh. Technol., vol. 43, no. 3, pp. 537–541, Aug. 1994.
[3] G. F. Pedersen, J. Ø. Nielsen, K. Olesen, and I. Z. Kovacs, “Measured
variation in performance of handheld antennas for a large number of
test persons,” in Proc. VTC, May 1998, pp. 505–509, IEEE.
[4] K. Boyle, “Mobile phone antenna performance in the presence of
people and phantoms,” in Technical Seminar: Antenna Measurement
and SAR, May 2002, pp. 8/1–8/4, IEE Antennas and Propagation
Professional Network.
[5] J. Ø. Nielsen and G. F. Pedersen, “In-network performance of hand-
held mobile terminals,” IEEE Trans. Veh. Technol., vol. 55, no. 3, pp.
903–916, 2006.
[6] M. Pelosi, O. Franek, M. B. Knudsen, M. Christensen, and G. F. Ped-
ersen, “A grip study for talk and data modes in mobile phones,” IEEE
Trans. Antennas Propag., vol. 57, no. 4, pp. 856–865, 2009.
[7] D. A. Sánchez-Hernández, Multiband Integrated Antennas for 4G Ter-
minals. Norwood, MA: Artech House, Inc., 2008.
[8] D. Astély, E. Dahlman, A. Furuskär, Y. Jading, M. Lindström, and S.
Parkvall, “LTE: The evolution of mobile broadband,” IEEE Commun.
Mag., vol. 47, no. 4, pp. 44–51, Apr. 2009.
[9] D. Gesbert, M. Shafi, D. Shan Shiu, P. J. Smith, and A. Naguib, “From
theory to practice: An overview of MIMO space-time coded wireless
systems,” IEEE J. Sel. Areas Commun., vol. 21, no. 3, pp. 281–302,
Apr. 2003.
[10] W. Kotterman, G. Pedersen, and K. Olesen, “Capacity of the mobile
MIMO channel for a small wireless handset and user influence,”
in Proc. 13th IEEE Int. Symp. Personal, Indoor and Mobile Radio
Commun., Sep. 2002, vol. 4, pp. 1937–1941.
[11] K. Sulonen and P. Vainikainen, “Performance of mobile phone antennas
including effect of environment using two methods,” IEEE Trans. In-
strum. Meas., vol. 52, no. 6, pp. 1859–1864, Dec. 2003.
[12] J. Ø. Nielsen and G. F. Pedersen, “Mobile handset performance eval-
uation using radiation pattern measurements,” IEEE Trans. Antennas
Propag., vol. 54, no. 7, pp. 2154–2165, 2006.
[13] V. Plicanic, B. K. Lau, A. Derneryd, and Z. Ying, “Actual diversity
performance of a multiband diversity antenna with hand and head ef-
fects,” IEEE Trans. Antennas Propag., vol. 57, no. 5, pp. 1547–1556,
May 2009.
[14] V. Plicanic, B. K. Lau, and Z. Ying, “Performance of a multiband di-
versity antenna with hand effects,” in Proc. iWAT, 2008, pp. 534–537.
[15] P. Suvikunnas, J. Villanen, K. Sulonen, C. Icheln, J. Ollikainen, and P.
Vainikainen, “Evaluation of the performance of multiantenna terminals
using a new approach,” IEEE Trans. Instrum. Meas., vol. 55, no. 5, pp.
1804–1813, 2006.
[16] F. Harrysson, J. Medbo, A. Molisch, A. Johansson, and F. Tufvesson,
“Efficient experimental evaluation of a MIMO handset with user influ-
ence,” IEEE Trans. Wireless Commun., vol. 9, no. 2, pp. 853–863, Feb.
2010.
[17] F. Harrysson, A. Derneryd, and F. Tufvesson, “Evaluation of user hand
and body impact on multiple antenna handset performance,” in Proc.
APSURSI, Jul. 2010, pp. 1–4.
[18] T. Zervos, K. Peppas, F. Lazarakis, A. Alexandridis, K. Dangakis, and
C. Soras, “Channel capacity evaluation for a multiple-input-multiple-
output terminal in the presence of user’s hand,” IET Microw. Antennas
Propag., vol. 1, no. 6, pp. 1137–1144, Dec. 2007.
[19] A. Michalopoulou, T. Zervos, A. Alexandridis, K. Peppas, F.
Lazarakis, K. Dangakis, and D. Kaklamani, “The impact of the user’s
body on the performance of a MIMO terminal in “pocket position”,”
in Proc. EuCAP, Nov. 2007, pp. 1–7.
[20] Y. Okano and K. Cho, “Dependency of MIMO channel capacity on
XPR around mobile terminals for multi-band multi-antenna,” in Proc.
EuCAP, 2007.
[21] P. Suvikunnas, J. Salo, L. Vuokko, J. Kivinen, K. Sulonen, and P.
Vainikainen, “Comparison of MIMO antenna configurations: Methods
and experimental results,” IEEE Trans. Veh. Technol., vol. 2, no. 57,
pp. 1021–1031, Mar. 2008.
[22] K. Sulonen, P. Suvikunnas, L. Vuokko, J. Kivinen, and P. Vainikainen,
“Comparison of MIMO antenna configurations in picocell and micro-
cell environments,” IEEE J. Sel. Areas Commun., vol. 21, no. 5, pp.
703–712, Jun. 2003.
[23] J. Valenzuela-Valdés, M. García-Fernández, A. Martínez-González,
and D. Sánchez-Hernández, “The influence of efficiency on receive
diversity and MIMO capacity for Rayleigh-fading channels,” IEEE
Trans. Antennas Propag., vol. 56, no. 5, pp. 1444–1450, May 2008.
[24] D. Gesbert, H. Bolcskei, D. Gore, and A. Paulraj, “Outdoor MIMO
wireless channels: Models and performance prediction,” IEEE Trans.
Commun., vol. 50, no. 12, pp. 1926–1934, Dec. 2002.
[25] 3GPP TS 36.101 [Online]. Available: http://www.etsi.org version
8.11.0 release 8
[26] W. A. T. Kotterman, G. F. Pedersen, and P. Eggers, “Cable-less mea-
surement set-up for wireless handheld terminals,” in Proc. PIMRC,
Sep. 2001, pp. B112–B116.
[27] C. Icheln, J. Ollikainen, and P. Vainikainen, “Reducing the influence of
feed cables on small antenna measurements,” Electron. Lett., vol. 35,
no. 15, pp. 1212–1214, Jul. 1999.
[28] C. Icheln, J. Krogerus, and P. Vainikainen, “Use of balun chokes in
small-antenna radiation measurements,” IEEE Trans. Instrum. Meas.,
vol. 53, no. 2, pp. 498–506, 2004.
[29] B. Yanakiev, J. Ø. Nielsen, and G. F. Pedersen, “On small antenna mea-
surements in a realistic MIMO scenario,” in Proc. EuCAP, Apr. 2010,
pp. 1–5.
[30] J. Ø. Nielsen, G. F. Pedersen, K. Olesen, and I. Z. Kovács, “Statis-
tics of measured body loss for mobile phones,” IEEE Trans. Antennas
Propag., vol. 49, no. 9, pp. 1351–1353, Sep. 2001.
[31] J. Ø. Nielsen, J. B. Andersen, P. C. F. Eggers, G. F. Pedersen, K.
Olesen, E. H. Sørensen, and H. Suda, “Measurements of indoor
16 32 wideband MIMO channels at 5.8 GHz,” in Proc. ISSSTA,
2004, pp. 864–868.
[32] L. Ozarow, S. Shamai, and A. Wyner, “Information theoretic consid-
erations for cellular mobile radio,” IEEE Trans. Veh. Technol., vol. 43,
no. 2, pp. 359–378, May 1994.
[33] J. B. Andersen and F. Hansen, “Antennas for VHF/UHF personal
radio: A theoretical and experimental study of characteristics and
performance,” IEEE Trans. Veh. Technol., vol. 26, no. 4, pp. 349–357,
Nov. 1977.
[34] C. Waldschmidt, C. Kuhnert, M. Pauli, and W. Wiesbeck, “Integration
of MIMO antenna arrays into hand-helds,” in Proc. 5th IEE Int. Conf.
3G Mobile Commun. Technol., 2004, pp. 16–23.
[35] J. Ø. Nielsen, J. B. Andersen, G. Bauch, and M. Herdin, “Relationship
between capacity and pathloss for indoor MIMO channels,” in Proc.
PIMRC, 2006, pp. 1–5.
NIELSEN et al.: USER INFLUENCE ON MIMO CHANNEL CAPACITY FOR HANDSETS IN DATA MODE OPERATION
Jesper Ødum Nielsen received the M.S. degree in
electronics engineering in 1994 and the Ph.D. degree
in 1997, both from Aalborg University, Denmark.
He is currently employed at Department of Elec-
tronic Systems, Aalborg University, where his main
areas of interests are experimental investigation
of the mobile radio channel and the influence on
the channel by mobile device users. He has been
involved in channel sounding and modeling, as
well as measurements using the live GSM network.
In addition he has been working with handset
performance evaluation based on spherical measurements of handset radiation
patterns and power distribution in the mobile environment.
Boyan Yanakiev received the B.S. degree in physics
from Sofia University, Bulgaria, in 2006 and the M.S.
degree in wireless communication from Aalborg Uni-
versity, Denmark, in 2008 where he is currently pur-
suing the Ph.D. degree.
His current position is as an industrial Ph.D.
student in cooperation with Molex Antenna Business
Unit. His primary interests are in the area of small
integrated mobile antennas, optical antenna measure-
ment techniques and radio channel measurements.
He has been involved in the design and development
of multiple RF-to-optical convertors, for onboard handset measurements.
Ivan B. Bonev was born in Yambol, Bulgaria, in
April 1980. He received the B.Sc. and M.Sc. degrees
(hons.) in telecommunications from the Technical
University of Varna, Varna, Bulgaria, in 2002 and
2004, respectively, where he also received the
M.Sc. degree in electrical engineering in 2007. He
is currently pursuing the Ph.D. degree in wireless
communications at Aalborg University, Aalborg,
Denmark.
He has been a Visiting Researcher at Ecole Po-
litechnique de luniversite de Nantes, Nantes, France;
Tampere University of Technology, Tampere, Finland; and Atlantic Cape Com-
munity College, Cape May Court House, NJ. In 2007, he joined the Antennas,
Propagation, and Radio-Networking (APNET) Group at Aalborg University.
His current research interests include small antennas, antenna interactions with
a human body, computational electromagnetics, numerical techniques, hearing
aid compatibility of mobile phones, specific absorption-rate evaluation issues,
performance evaluation of multiple-antenna terminals.
Mr. Bonev is a member of Sigma Xi and the Hearing Loss Association of
America. He is the recipient of the URSI young scientist award for 2011.
643
Morten Christensen was born in 1973. He received
the M.Sc. degree in electrical engineering from Aal-
borg University, Denmark, in 1998.
In 1998, he joined Bosch Telecom A/S, Pandrup,
Denmark (acquired by Siemens Mobile Phones in
2000) where he designed integrated antennas for
mobile terminals. In 2006 he joined Motorola A/S,
Mobile Devices Aalborg where he was heading
the EMC and Antenna Department. He is now
with Molex Antenna Business Unit responsible for
the RF Research activities. His areas of interests
includes handset antenna design, performance evaluation methods and radio
propagation models.
Gert Frølund Pedersen was born in 1965. He re-
ceived the B.Sc.E.E. degree (hons) in electrical en-
gineering from the College of Technology, Dublin,
Ireland, and the M.Sc.E.E. and Ph.D. degrees from
Aalborg University, Aalborg, Denmark, in 1993 and
2003, respectively.
He has been employed by Aalborg University
since 1993 where he is now a Full Professor heading
the Antenna, Propagation and Networking Group
and is also the Head of the Doctoral School on Wire-
less which some 100 Ph.D. students enrolled. His
research has focused on radio communication for mobile terminals especially
small antennas, diversity systems, propagation and biological effects and he
has published more than 75 peer reviewed papers and holds 20 patents. He
has also worked as consultant for developments of more than 100 antennas
for mobile terminals including the first internal antenna for mobile phones in
1994 with lowest SAR, first internal triple-band antenna in 1998 with low SAR
and high TRP and TIS, and lately various multi antenna systems rated as the
most efficient on the market. He has been one of the pioneers in establishing
over-the-air measurement systems. The measurement technique is now well
established for mobile terminals with single antennas and he was chairing the
COST2100 SWG2.2 group with liaison to 3GPP for over-the-air test of MIMO
terminals.
644 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Exposure Compliance Methodologies for Multiple
Input Multiple Output (MIMO) Enabled Networks
and Terminals
Nicholas Perentos, Member, IEEE, Steve Iskra, Antonio Faraone, Senior Member, IEEE, Ray J. McKenzie,
Giorgi Bit-Babik, Member, IEEE, and Vitas Anderson, Member, IEEE
Abstract—Multiple input multiple output (MIMO) enabled
handsets and base-stations feature antenna systems that generate
electromagnetic fields for which relevant exposure standards and
guidelines do not explicitly define compliance testing methodolo-
gies. Here, through computational modeling, we explore several
field summation schemes for evaluating such exposures and
propose compliance testing methodologies that limit the degree of
exposure under/over-estimation for both base stations and hand-
sets. The methodologies rely on scalar field probe measurements
thus avoiding significant equipment upgrades and are applicable
to cases where access to signals from each MIMO antenna element
can be arranged.
Index Terms—Dosimetry, electromagnetic analysis, multiple
input multiple output (MIMO) systems, Mobile antennas, safety.
I. INTRODUCTION
M ULTIPLE INPUT multiple output (MIMO) refers to the
use of multiple antennas at the radio transmitter and re-
ceiver to improve the performance of a wireless communica-
tion system. Recent technology standards have defined MIMO-
enabled networks including the IEEE 802.11n (WiFi), IEEE
802.16 (WiMAX), 3GPP Rel 7 (HSPA+) and 3GPP Rel 8 (LTE)
[1]–[4].
As is the case with other wireless equipment of similar
transmit power levels, MIMO enabled devices are required to
comply with standards [5] or internationally recognized guide-
lines [6] for limiting human exposure to radio-frequency (RF)
fields. A pertinent feature of MIMO technologies in this regard
is the possibility of correlated signal transmissions at the same
frequencies from multiple antennas leading to constructive or
destructive interference effects in the RF exposure field of the
antenna array. The RF exposure assessment guidance in current
safety guidelines and product compliance standards may not be
suitable for assessing such exposures.
Manuscript received July 05, 2011; revised September 29, 2011; accepted
September 30, 2011. Date of publication October 25, 2011; date of current ver-
sion February 03, 2012. This work was supported by the GSM Association and
the Mobile Manufacturers Forum under Grant BS123456.
N. Perentos, S. Iskra, and R. J. McKenzie are with Swinburne University of
Technology, Hawthorn, VIC 3181, Australia.
A. Faraone and G. Bit-Babik are with Motorola Solutions, Inc., Fort Laud-
erdale, FL 33309 USA.
V. Anderson is with Swinburne University of Technology, Melbourne, VIC
3122, Australia (e-mail: vitas@ieee.org).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173453
0018-926X/$26.00 © 2011 IEEE
This paper explores the issues that may arise when as-
sessing exposures from MIMO enabled networks from both
the perspective of the user equipment as well as the base
station. Methodologies to deal with the identified issues are
proposed, and procedures for exposure compliance assessment
are presented in the form of flowcharts suitable for inclusion in
appropriate standards.
II. MIMO BACKGROUND
A. MIMO Technology
Multiple antennas at both the transmitter and receiver can
improve system performance by improving coverage, system
capacity, or improved service provisioning [7]. There are three
ways in which multiple antennas can be utilized to achieve these
objectives.
Spatial Multiplexing: Here, the spatial dimension is reused
to achieve higher overall data rates by transmitting distinct data
streams from each antenna. It is also possible to continually
optimize signal transmission for specific users based on feed-
back from the receiver. In this case, different MIMO antenna
elements transmit uncorrelated waveforms since they carry un-
correlated data streams.
Transmit Diversity/Cyclic Delay Diversity: Copies of the
same data stream are transmitted from each antenna element
(sometimes time delayed) leading to signal diversity at the
receiver. The diversity provides an improved signal to noise
ratio which in turn allows for higher data rates to be used. In
this case, different MIMO antenna elements transmit correlated
time-domain waveforms since they feature the same informa-
tion content.
Classical Beamforming: Here energy focusing in a subsector
of a cell achieves an improved signal to noise ratio. To achieve
classical beamforming different antenna arrangements are re-
quired to those typically used in MIMO enabled transceivers.
For instance, co-polarized, regularly spaced column arrays are
frequently required to achieve the desired pattern shape and
angular scan agility. Closely spaced antenna elements expe-
rience high mutual fading correlation and thus are unable to
provide the necessary signal diversity that is commonly de-
sirable to achieve the performance improvements of MIMO.
For this reason classical beamforming arrays are sometimes not
included among MIMO transmitters, although classical beam-
forming is another case of correlated time-domain waveforms
emitted by each antenna element, thus leading to interference
patterns in the corresponding spatial field distribution as for
PERENTOS et al.: EXPOSURE COMPLIANCE METHODOLOGIES FOR MIMO ENABLED NETWORKS AND TERMINALS
MIMO. Compliance assessment for classical beamforming ar-
rangements has been previously considered [7]. A different kind
of beamforming is achievable with low mutual correlation an-
tennas, which is sometimes called pre-coding [7].
B. MIMO Enabled Base Stations
Typically when a MIMO antenna system is to be imple-
mented at a base station, it is required that independent fading
conditions are experienced by each antenna element. The mu-
tual fading correlation can be used to assess this, where a low
value implies that the sources experience independent channels
while a high value implies that the sources experience similar
channels. One way of achieving low mutual fading correlation
is to place the antenna elements sufficiently far apart. In typical
urban environments a distance of a few wavelengths apart
is sufficient. Another way is to co-locate antenna elements but
use a different polarization for each.
C. MIMO Enabled User Equipment
Contrary to the case for base station, user equipment typi-
cally operates in close vicinity to multiple scatterers, including
the human body, which can provide good multipath fading con-
ditions even with relatively closely spaced antenna elements
.
III. APPLICABILITY OF CURRENT COMPLIANCE
METHODOLOGIES TO MIMO ANTENNAS
MIMO antenna elements used in diversity or beamforming,
thus transmitting time-domain correlated waveforms, belong to
the broader category of correlated sources. Current assessment
methodologies adequately cover multiple uncorrelated sources
(spatial multiplexing MIMO) since the resulting exposure can
be calculated by the summation (sometimes linearly scaled) of
the individual exposures [5], [6]. However the same does not
apply for correlated sources.
A. Total RF Exposure From Multiple Correlated Sources
Consider two correlated sources in close vicinity to each other
(i.e. a few apart or closer) transmitting signals and in the
same frequency band. According to the principle of superpo-
sition the resulting combined field distribution at any
point in space and time will be the phasor vectorial combination
of the two fields, so that the magnitude may be calculated as
(1)
Since signals are expected to combine constructively or de-
structively at different spatial locations, the spatial distribution
of RF energy will be dictated by the resulting interference pat-
terns. The dynamic relation between MIMO elements will lead
to as many distinct interference patterns as amplitude/phase
combinations between elements. It follows that if field as-
sessments through probe measurements are to be performed,
at any spatial location, an equal number of measurements as
interference patterns will need to be performed in order to
guarantee capturing the maximum achievable exposure level.
Alternatively it can be argued that adequately long field level
645
monitoring periods will eventually capture the true maximum
exposure level. Such approaches are resource and time con-
suming so can be impractical.
B. Total RF Exposure From Multiple Uncorrelated Sources
Because they protect from excessive temperature rise that
may be produced by RF exposure, safety standards and guide-
lines [5], [6] define RF exposure limits on the basis of time-
averaged metrics which depend on RMS fields. Consequently,
the superposition of exposures from time-domain uncorrelated
fields is readily performed by straight summation of the indi-
vidual exposures. Considering two uncorrelated field sources
transmitting signals and in the same frequency band, the
combined exposure is the sum of the respective squared mag-
nitudes
(2)
Should summation according to (2) be applied to correlated
fields, the result is expected to sometimes overestimate and
sometimes underestimate exposure depending on the spatial
location under consideration.
IV. METHODS
A. Field Summation Schemes
Assuming that access to identifiable, correlated signals
from each MIMO element is possible, several field summation
schemes are proposed in the following, which provide conser-
vative estimates of the true exposure assessed according to (1).
Whereas the application of (1) would entail the use of vector
field probes capable of measuring magnitude and phase for
each field component, the summation schemes that we propose
can be implemented with scalar field probes since they do not
require knowledge of any phase information. This is a sig-
nificant advantage since scalar probes and associated readout
instrumentation are less complex, broadband in frequency
response, easier to operate, and less expensive. For correlated
source signals and , the proposed summation schemes are as
follows.
1) summing the squares of the sums of local field component
magnitudes
(3)
2) the square of the sum of local field magnitudes
(4)
Summations according to (3) and (4) are mathematically
shown [9] to overestimate the true field level provided by (1).
Our aim is to quantify the degree of typical over- and under-es-
timation associated with each summation scheme, including the
case when (2) is (improperly) used to estimate the combined
exposure from correlated sources.
646
Fig. 1. Base station antenna setup. The panels are separated by distance
and each has five dipoles. The elevation from the ground is
B. Modeling Strategy
In this study, the field summation schemes described above
were applied to the field outputs of computational models rep-
resenting a sample of MIMO enabled base stations and portable
devices. Comparisons of the actual combined exposure level
given by (1) with that of (2), (3) and (4) provide indications
on the typical degree of overestimation and underestimation of
each summation scheme. We consider the case of MIMO en-
abled base stations and the cases of MIMO enabled handset de-
vices and laptop devices. It is important to consider these three
categories of devices since standards define distinct method-
ologies of exposure compliance assessment for each. Base sta-
tion compliance is usually demonstrated versus electric field
(E) reference levels while handheld and portable devices are
tested against the specific absorption rate (SAR) basic restric-
tions through E-field scans in the dielectric media of head and
flat phantom models [8], [10], [11].
C. Base Station Modeling
A typical urban scenario was modeled using the EMSS FEKO
Electromagnetic Software Analysis Tool (v5.4) where two panel
antennas, each featuring five dipoles fed by impedance matched
75 sources, are placed at a distance apart, 6 m above
ground and excited with a 900 MHz signal (Fig. 1). Three simu-
lations were performed for the same arrangement with the phase
difference between the two elements set at 0 , 90 , and 180 ,
illustrating the dependence of the radiation and interference pat-
tern on the phase difference.
Two more simulations were performed where the antenna
panels were excited one at a time in separate simulations.
The two models were otherwise identical so as to allow for
field combination based on the superposition principle. For
this reason all active and inactive antennas were loaded with
passive resistive loads of 75 . The two E-field vector outputs
were then processed and combined and the power density
at a distance of 100 m from the antenna was calculated as
. The percentage deviation of the calculated
power density for the (2), (3) and (4) scalar combinations was
compared to the benchmark vector phasor combination (1).
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
.
Fig. 2. An LFMA handset against a sphere volume representing the head.
D. Modeling of Portable Devices
XFDTD version 6.5 modeling software (REMCOM) was
used for simulations described in this section. This computa-
tional tool allows easy access and manipulation of output files
and for SAR calculations to be performed independent of the
FDTD computation therefore allowing SAR calculations after
superposition of fields.
The MIMO antenna elements were excited one at a time in
separate simulations. The same approach described for the base
station antennas, where one simulation per antenna element was
performed while loading the others with their source imped-
ances, was followed to compute the fields.
The electric field output files of the FDTD solutions were
combined to obtain according to (1)through to (4) while
the phase difference between the MIMO elements was varied
from 0 to 180 in 15 steps (Appendix A describes how the
field components were combined numerically). From the com-
bined electric field values SAR was then calculated. SAR values
were subsequently imported into XFDTD where averaged SAR
calculations (peak point SAR, 1 g and 10 g SAR) were per-
formed. Percent deviation from (1) for the other summation
schemes ((2)–4) was then calculated.
Handset Models: Two handset models were simulated
against a sphere with tissue equivalent
dielectric medium and specific mass density properties ac-
cording to the IEEE 1528 standard ( , ,
) [10]. We opted for this simple geometric
structure instead of the more complex SAM phantom [10] or
any inhomogeneous models so as to more clearly seethe influ-
ence of the interference between MIMO elements on the SAR
value without the added clutter of additional field scattering that
may result from more complex geometrical shapes or dielectric
boundaries.
The first model was a pair of parallel dipoles
( , ,
) placed 10 mm away and
tangentially to the surface of the sphere. The second model
(Fig. 2) was a multimode antenna presented by Kim et al.
[12] called a Low Profile Folded Monopole Antenna (LFMA)
operating at 2 GHz with a ground plane of dimensions
PERENTOS et al.: EXPOSURE COMPLIANCE METHODOLOGIES FOR MIMO ENABLED NETWORKS AND TERMINALS 647

Fig. 3. The two simulated orientations of the laptop model against the standard flat phantom are depicted. Distance d represents the separation between the two
MIMO elements. Distance d is varied as 87 186 and 286 mm.

comparable to mobile phones. Antenna and ground plane

dimensions were identical to those in [8]. SAR was calculated
in the section of the sphere closest to the active elements
of the antennas in a volume of and
, for the dipole and LFMA antenna models
respectively.
Laptop Models: According to the IEC 62209-2 standard [11],
handheld and body mounted devices should be tested against a
standard flat phantom representing the torso of a human body.
This standard also describes an elliptical flat phantom and dis-
tinct testing orientations of devices in relation to it. Accordingly,
a laptop model was simulated against this flat phantom with ori- Fig. 4. Electric field (normalized to unity) 100 meters away from two-panel
antenna (see Fig. 1). Panels separated by a distance of . The
entations as shown in Fig. 3. The laptop model was made of two shift in peaks and troughs is clearly seen among the three patterns. Observation
ground planes representing the conventional clamshell structure angle 0 corresponds to the direction of the x-axis (Fig. 1). ; 0 , ; 90 and
of laptops. Planar Inverted F-type antennas (PIFA’s) were situ- ; 180 .
ated on the top part of the lid as shown in Fig. 3(a). The sepa-
ration distance between the PIFA antennas was set at 33,87,186
and 286 mm for orientation 1 and at 286 mm for orientation 2.
The selected antenna separation distances cover a representative
range of possible antenna configurations on laptops. For orien-
tation 1, the maximum point, 1 g and 10 g SAR was calculated
in the high SAR region that was observed in the phantom nearby
to the antennas and the region in between. For orientation 2, the
maximum SAR was calculated in the high SAR region nearby to
the front edge of the laptop where it was abutting the phantom.

V. COMPARISON OF SUMMATION SCHEMES

A. Base Station Transmitting Correlated Waveforms
The dependence of the MIMO base station radiation pattern
on the phase difference between the antenna elements is seen
in Fig. 4. Some lobes are seen to shift by around 10 which
corresponds to a distance of 17 m at a radius of 100 m away
from the center point of the two panels.
The true value of the combined normalized power densities
according to (1) are compared to the alternate field summation
schemes (2) (3) (4) 100 m away from the base station and 1
m above the ground in Fig. 5 with phase difference between
MIMO elements set to 0 . When using (2) the exposure level is
overestimated or underestimated depending on the fields phase
relationship at the evaluation point. For (3) and (4) results are
almost identical with the exposure never being underestimated.
Fig. 5. Power density (normalized to unity) at a radius of 100 m from the center
point of the antenna arrangement of Fig. 1 versus observation angle. Observation
angle 0 corresponds to the direction of the x-axis. ; true value, i.e. summation
according to (1), ; summation according to (2), ; summation according to (3)
and ; summation according to (4).
B. Portable Devices Transmitting Correlated Waveforms
Handset Models: Table I summarizes the findings for the
dipole and LFMA handset placed against the sphere phantom.
For the dipoles model, at worst there is a 3% overestimation of
maximum SAR using (4) and underestimation using (2) asso-
ciated with the three scalar summation schemes. Significantly
different values are recorded for the LFMA MIMO handset. An
underestimation range of 16–33% is observed using (2) and an
overestimation range of 32–67% using (3) and (4).
648 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE I
OVER OR UNDER ESTIMATIONS COMPARED TO (1): HANDSET MODELS
TABLE II
OVER OR UNDER ESTIMATIONS COMPARED TO (1): LAPTOP MODELS
Antenna separation distance
Laptop Models: Table II summarizes how the ((2)–(4)) scalar
summation schemes differ from the true value (1) for SAR esti-
mates in the laptop models for both device orientations against
the elliptical flat phantom (Fig. 3). For each summation scheme,
the trends for maximum point, 1 g and 10 g SAR were very
similar. For 10 g SAR, using (2) led to under estimations in the
range of 11–42%, using (3) led to overestimations in the range of
8–20% and using (4) led to similar overestimations of 11–23%.
An example of an SAR plot versus phase shift between the PIFA
antennas is shown in Fig. 6 for an antenna separation of 286 mm
and the laptop device oriented as in Fig. 3(a).
For orientation 1, the maximum SAR occurred nearby to each
of the two PIFA antenna elements, and so the interference in-
teraction between them generally decreased as the separation
distance, , increased as each had progressively less field con-
tribution to the other. For orientation 2, the maximum SAR oc-
curred in the phantom region adjacent to the front edge of the
laptop, due to currents induced in the laptop base from both an-
tenna elements. In this case there is a more balanced contribu-
tion of field strength from each antenna for the maximum SAR,
since both are roughly equidistant from the laptop front edge.
This is reflected in the higher underestimation percentages for
Fig. 6. The 10 g averaged SAR normalized to the vector phasor addition max-
imum versus the phase variation is displayed here for the PIFA laptop with an-
tenna separation of 300 mm: true value, i.e. summation according to (1);
summation according to (2); summation according to (3), and; summation
according to (4). It is seen that (2) can overestimate or underestimate, while (3)
and (4) always overestimate exposure.
summation scheme (2) for the 286 mm antenna separation in
orientation 2.
VI. COMPLIANCE ASSESSMENT METHODOLOGIES
A. Field Summation for Base Stations
In Figs. 5 and 6 it can be seen that the scalar summation
according to (2) tends towards the average of the vector
phasor summation (1). Actual time-averaged exposure will
tend towards that predicted using (2) if the variation in phase,
and consequently the interference pattern, is rapidly and ran-
domly varying. The IEEE C95.1 standard [5] and the ICNIRP
Guidelines [6] specify that the average power density over
any 6-minute exposure interval should be used to demonstrate
compliance to reference levels. In this context, a rapid time
frame is any time interval substantially shorter than 6 minutes.
However, the assumption of rapid and random phase change
may not always hold. For example in a static environment
with no moving scatterers, small number of fixed users and no
spatial multiplexing or diversity coding implemented, a fairly
constant radiation pattern maybe expected, whereas in highly
variable environments with moving users and scatterers where
a base station switches between beam forming to diversity
transmission schemes, a variable exposure may be observed.
Fig. 5 shows that field summation by (3) and (4) leads to an
average overestimation of approximately 70% compared to the
true value (1). The summation results of (3) and (4) are very
similar to each other because the electric field produced by the
simulated MIMO antenna is predominantly z-polarized due to
the vertical orientation of the dipoles. In such cases it is easily
seen that (3) can be approximated by (4) since:
If the MIMO transmitter had significant field components in
more than one Cartesian direction, as would be the case with
two cross-polarized dipole antennas or multi-reflection environ-
ments, distinctly different results would be observed.
The summation scheme defined by (3) is observed to be sen-
sitive to the local frame of reference used for the measurement
of the field components (see Appendix B). In practical terms
PERENTOS et al.: EXPOSURE COMPLIANCE METHODOLOGIES FOR MIMO ENABLED NETWORKS AND TERMINALS
Fig. 7. Flow chart for the assessment of MIMO base stations.
this implies that when fields are recorded by an isotropicprobe
and combined according to (3) the outcome will depend on the
orientation of the probe in relation to the MIMO antenna. This
introduces a source of uncertainty on the amount of overestima-
tion in the exposure evaluation using (3).
B. Field Summation for Portable Devices
LFMA and Dipole Handsets: Table I lists the range of re-
sults for handheld devices. The results are consistent with ex-
pectations where summation according to (2) tends towards the
average (see also Fig. 6) whereas (3) and (4) overestimate expo-
sure with (4) more so than (3). First we note that the simplistic
model of the twin dipoles results in small over and under es-
timations. However this canonical model does not share many
common features with probable implementations of MIMO an-
tenna systems in real handsets. One basic feature that is missing
is the common ground that the antenna elements are likely to
share on handsets where as this is adequately addressed in the
LFMA model handset. Significantly different results are ob-
served for the LFMA handset where over estimation was up to
67%. The summation scheme defined by (3) leads to no signifi-
cant reduction in overestimation for all determined SAR quan-
tities ( 1%).
An interesting trend observed in Table I is that the level of
SAR increases (i.e. under-estimation is reduced or over-estima-
tion is increased) with an increasing averaging mass for sum-
649
mation schemes (2–4) relative to (1). Our explanation for this
is that scalar summations according to (2), (3) and (4) remove
any phase information and as a result assume constructive in-
terference at all spatial locations within the averaging mass. For
a10 g SAR averaging cube, the cube dimensions
are comparable to the wavelength inside the
tissue at 2 GHz . Accordingly, each 10 g cube
should always contain areas of destructive interference, but for
summation schemes ((2)–4) these areas are summed construc-
tively. As a result the mass averaged SAR is inflated leading to
an increase in the degree of over estimation as the averaging
volume is increased.
Laptops: Large underestimations are observed for the sum-
mation scheme of (2) for both laptop orientations (see Table II).
Overestimations resulting from the summation techniques given
by (3) and (4) do not exceed 23%. It is observed that overesti-
mation decreases when antenna separation is increased. This is
so because the contribution of the one element in the vicinity of
the other is minimized due to increased field decay. There is an
exception however for the separation distance of 186 mm which
may be a result of standing wave patterns being formed on the
ground plane of the laptop. Consistent with the respective ob-
servation in the handset modeling, an increasing SAR level is
observed for increasing averaging volume. Summation scheme
(3) appears to provide a benefit of a few percent decreased over-
estimation when compared to the results of (4). Although this
650 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

improvement may appear marginal (less than 10%), where ex-

posure approaches the compliance limit this summation scheme
can be of benefit. However the sensitivity of this scheme to the
E-field probe local reference frame will need to be assessed fur-
ther to quantify the likely benefit.

C. Exposure Assessment Protocols for Base Stations

The proposed summation schemes rely on access to distin-
guishable field recordings from each MIMO element. In the case
of base station assessments this could be achieved by switching
off all MIMO antenna elements but one and performing field
measurements according to established methodologies for ex-
ample as defined in the IEC 62232 standard [8]. The same pro-
cedure is then repeated for all MIMO elements and finally fields
are combined according to the chosen summation scheme. This
approach is inconvenient since the normal operation of the base
station will need to be disrupted to perform the measurement.
There are alternatives however. For example consider the case
of an LTE MIMO enabled base station. Each MIMO antenna el-
ement carries its own unique pilot signal. Pilot signals of each
MIMO element occur in distinct resource blocks (a unique com-
bination of an orthogonal frequency sub carrier and time slot)
and are therefore uncorrelated (see [7, pp. 324–330]). It is there-
fore feasible that through the use of suitable protocol analyzers,
one can measure all pilot signal levels ‘simultaneously’, scale
the measured field level according to the maximum transmitter
output bandwidth and power and then sum the field contribu-
tions according to the preferred summation scheme. A possible
approach to measurement for MIMO base stations is detailed
in the form of a flow chart in Fig. 7. Owing to the limited ben-
efit in terms of reduced overestimation and the increased com-
plexity and potential uncertainty of using summation scheme Fig. 8. A flow chart of a possible routine that can be followed for SAR assess-
ments of MIMO enabled devices. Significant increases in scanning times should
(3), scheme (4) has been recommended in the proposed pro- be expected as the number of antenna elements increase and as the physical di-
tocol as the best compromise while maintaining a conservative mensions of the device increase. The terms area scan and zoom scan are defined
approach. in [10].

D. Exposure Assessment Protocols for Portable Devices

In the case of portable terminal devices simpler procedures
can be envisaged where special software testing routines are in-
corporated into the MIMO enabled device where one element
is switched on and all others are off while the entire area of in-
terest is being scanned with a SAR probe. In such protocols it
would be necessary to ensure that the physical dimensions of the
device are scanned in entirety to ensure that the field combina-
tion can be estimated at all probable SAR peak locations. This
is necessary because antenna locations or accurate prediction of
possible SAR peak locations may not be possible. Field combi-
nation according to (3) can be problematic since the SAR probe
orientation may change with respect to the local reference frame
while a measurement is performed. The issue may be avoided if
during SAR compliance assessments the orientation of the elec-
tric field probe is kept the same between and within SAR scans.
This may be difficult to achieve and can lead to results that are
not easily reproducible between laboratories. It may be simpler
to account for the expected sensitivity through the incorporation
of an additional factor into the uncertainty budget. Alternatively
it may be determined through physical measurements that the
associated uncertainty is simply small enough to be negligible.
A possible compliance testing procedure for handheld devices
is detailed in the form of a flow chart in Fig. 8. As for base sta-
tions, we have recommended summation scheme (4) as the best
compromise which always provides a conservative estimate of
exposure.
E. Future Trends and Opportunities
In the case of LTE-enabled portable devices, such as lap-
tops, tablets and PDA-like smartphones, it would be desirable
to conduct a SAR assessment using the approach of discrimi-
nating concurrent signals with protocol analyzers as described
for base stations in Section VI-C. In this way, the device under
test could be operated while transmitting from all antennas si-
multaneously, thereby allowing the completion of a SAR assess-
ment in a single test. However, this approach is not feasible at
the present time since SAR probes currently feature low-pass
filters and diode-detectors that do not allow discrimination of
different resource blocks. Nonetheless, ongoing efforts to de-
velop vector SAR probes that could capture the LTE waveform
PERENTOS et al.: EXPOSURE COMPLIANCE METHODOLOGIES FOR MIMO ENABLED NETWORKS AND TERMINALS 651

Fig. 10. Two distinct cases of cross-polarized antenna setups. In the first, the
orientation of the antennas coincides with the local reference frame defined by
the chosen x-y coordinates ( is parallel with the -axis and is parallel
to the -axis). In the second case on the right, the cross-polarized antennas are
rotated by clockwise.

Fig. 9. The twelve electric field components (four for each Cartesian coordi-
nate) that form a valid tissue voxel in the FDTD implementation.
Considering the 12 components that form a valid tissue voxel
in the FDTD environment

[13], [14] make it possible to envisage the use of the aforemen-

tioned approach in commercial SAR measurement systems once
those efforts come to fruition.
(6)
VII. CONCLUSIONS
We have investigated three alternative summation schemes where n represents the four vertices that make up any of the x,
for combining fields from MIMO antenna elements to produce y or z components of the electric field at the centre of the voxel.
estimates of RF exposure levels for the purposes of safety com- Sum of the squares of the sums of local field magnitudes:
pliance assessments. The summation schemes rely on scalar
field probe measurements thus avoiding significant equipment (7)
upgrades. They are applicable to cases where access to uncorre-
lated signals from each MIMO element source can be arranged, Square of the sum of the local field magnitudes
for example, through software control of the RF transmitter
function of a device. Two of the summation schemes have been
shown to overestimate the true SAR in all cases, and one of these
has been recommended because of its relative simplicity. Using
the preferred scheme based on the square of the sum of the local
field magnitudes, an average over-estimation of 70% compared
to the true value may be expected when assessing MIMO en- (8)
abled base stations, while an overestimation of up to 67% can be
expected when assessing SAR compliance of portable devices Taking into consideration the 12 components that form the voxel
such as handsets and laptops. The compliance assessment of live
traffic-carrying base stations requires ‘simultaneous’ measure-
ment of the fields from all pilot channels and the results com-
bined and scaled according to the preferred summation scheme.
Further study is required to identify the practicalities and addi-
tional uncertainties associated with the proposed methodologies
when applied to live MIMO enabled networks and devices. (9)

APPENDIX The above equation is only applicable for homogeneous

A. Summation Formulae for Scalar Combinations
In the FDTD computational environment a tissue voxel is
formed by twelve Cartesian components, or twelve complex
electric fields as shown in Fig. 9, see also [15].
Therefore the scalar superposition of two or more electric
fields according to (2), (3) and (4) can be implemented in the
following respective ways:
Sum of the squared magnitudes
(5)
tissue models. Complications arise in the case of inhomoge-
neous models due to different media properties in adjacent
FDTD voxels. However inhomogeneous models are not appli-
cable to assessments in the homogeneous phantoms scenarios
specified in SAR compliance standards [5], [6].
B. Sensitivity of (3) to the Orientation of the Local Reference
Frame
Consider the case of two sets of cross-polarized antennas as
depicted in Fig. 10(a) and 10(b).
Assuming that the fields are linearly polarized and orthog-
onal with complex amplitudes and the
652 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
combined field according to (3), using the probe arrangement in
Fig. 10(a), is given by
(10)
since . However if we consider the case
depicted by Fig. 10(b) we arrive at a different result, , [11] Human Exposure to Radio Frequency Fields From Hand-Held and
through the use of a rotational transformation matrix, , where
(11)
After rotation we find
(12)
The square of the sum of the local field magnitudes is calcu-
lated as
(13)
When comparing (13) with (10) it is clear that
except when sin , thereby indicating variable
results for field summation according to (3) depending on the
orientation of the local reference frame.
ACKNOWLEDGMENT
The authors thank J. Cambell and Prof. Q Balzano for useful
discussions.
REFERENCES
[1] IEEE Standard for Information Technology—Telecommunications and
Information Exchange Between Systems—Local and Metropolitan
Area Networks—Specific Requirements Part 11: Wireless LAN
Medium Access Control (MAC) and Physical Layer (PHY) Spec-
ifications Amendment 5: Enhancements for Higher Throughput,
802.11n-2009 IEEE Standard, Jun. 12, 2007.
[2] IEEE Standard for Local and Metropolitan Area Networks—Part
16: Air Interface for Broadband Wireless Access Systems, New York,
802.16-2009, May 29, 2009.
[3] 3rd Generation Partnership Project; Technical Specifications and Tech-
nical Reports for a UTRAN-Based 3GPP System, 3GPP TS 21.101
V7.5.0 (2009-12) 2009.
[4] 3rd Generation Partnership Project; Technical Specifications and Tech-
nical Reports for a UTRAN-Based 3GPP System, 3GPP TS 21.101
V8.2.0 (2009-12) 2009.
[5] IEEE Standard for Safety Levels With Respect to Human Exposure to
Radio Frequency Electromagnetic Fields, 3 kHz to 300 GHz, IEEE
Std C95.1, The Institute of Electrical and Electronics Engineers, Inter-
national Committee on Electromagnetic Safety, New York, April 19,
2005.
[6] ICNIRP, “Guidelines for limiting exposure to time-varying electric,
magnetic, and electromagnetic fields (up to 300 GHz),” Health Phys.,
vol. 74, pp. 494–522, Apr. 1998.
[7] E. Dahlman, S. Parkval, J. Sköld, and P. Beming, 3G Evolution: HSPA
and LTE for Mobile Broadband, 2nd ed. Burlington: Academic Press,
2008.
[8] IEC 62232 Ed. 1 Methods for the Assessment of Electric, Magnetic
and Electromagnetic Fields Associated With Human Exposure, IEC/
TC106, The International Electrotechnical Commission, 2009.
[9] IEC 62630 Ed.1: Guidance for Evaluating Exposure From Multiple
EM Sources, , The International Electrotechnical Commission, 2010.
[10] IEEE Recommended Practice for Determining the Peak Spatial-Av-
erage Specific Absorption Rate (SAR) in the Human Head From
Wireless Communications Devices: Measurement Techniques, IEEE
Std 1528-2003, The Institute of Electrical and Electronic Engineers,
New York, 2003, IEEE SCC34.
Body-Mounted Wireless Communication Devices—Human Models, In-
strumentation, and Procedures Part 2: Procedure to Determine the
Specific Absorption Rate (SAR) in the Head and Body for 30 MHz to 6
GHz Handheld and Body-Mounted Devices Used in Close Proximity to
the Body, IEC62209-2 Ed 1, The International Electrotechnical Com-
mission, 2010.
[12] Y. Kim, T. Hayashi, Y. Koyanagi, and H. Morishita, “Compact built-in
handset MIMO antenna using L-shaped folded monopole antennas,”
IEICE Trans. Comm., vol. E91-B, pp. 1743–1751, 2008.
[13] T. Onishi, H. Togo, N. Shimizu, K. Kiminami, S. Uebayashi, and T.
Nagatsuma, “SAR measurement employing electro-optic (EO) probe
without using metal,” presented at the Bioelectromagnetic Society An-
nual Meeting, Dublin, June 2005, P-C-21.
[14] H. Togo, N. Shimizu, and T. Nagatsuma, “Near-field mapping system
using fiber-based electro-optic probe for specific absorption rate
measurement,” IEICE Trans., vol. E90-C, no. 2, pp. 436–442, Feb.
2007.
[15] K. S. Yee, “Numerical solution of initial boundary value problems in-
volving Maxwell’a equations in isotropic media,” IEEE Trans. An-
tennas Propag., vol. 14, pp. 302–307, May 1966.
Nicholas Perentos (M’09) received the Bachelor de-
gree and the Ph.D. degree in engineering telecommu-
nications from RMIT University, Melbourne, Aus-
tralia, in 2005 and 2008, respectively.
In 2009, he joined the Australian Centre for
Radiofrequency Bioeffects Research (ACRBR) as a
Postdoctoral Research Fellow and in 2010 he joined
the University of Cambridge, Cambridge, U.K., as a
Research Fellow. He is interested in the development
of medical implants as well as interference issues of
electromagnetic exposures to such devices.
Steve Iskra received the B.E. (Hon.) degree in elec-
trical engineering from the University of Melbourne,
Australia, in 1982.
In 1982, he joined the Electromagnetic Compati-
bility (EMC) Group, Telstra Research Laboratories
(TRL). In 2001, he moved to TRL’s Electromagnetic
Energy Safety Research Group and has been involved
in the measurement of electromagnetic fields, com-
putational electromagnetics, and RF interference ef-
fects on medical devices. He has published in peer
reviewed journals in the areas of EMC, human expo-
sure to electromagnetic fields and RF personal dosimetry.
Mr. Iskra is a Past Chairperson of Standards Australia (SA) Committee TE/3
(EMC), is a member of committees SA TE/7/2 and TC106 WG4 of the Interna-
tional Electrotechnical Commission (both in the area of human exposure to RF
fields).
PERENTOS et al.: EXPOSURE COMPLIANCE METHODOLOGIES FOR MIMO ENABLED NETWORKS AND TERMINALS
Antonio Faraone (M’97–SM’05) was born in Rome,
Italy, in 1966. He received the Ph.D. degree in applied
electromagnetics from the University of Rome “La
Sapienza,” in 1997.
He then joined the Motorola (now Motorola So-
lutions Inc.) Corporate EME Research Laboratory,
Fort Lauderdale, FL, engaging in mobile antenna
technology and RF dosimetry research. He is a Mo-
torola Scientific Advisory Board Associate member,
Motorola Distinguished Innovator, and. Fellow of
the Technical Staff within the Chief Technology
Office. He holds 18 patents in antenna technology and has coauthored more
than 30 refereed journal publications.
Dr. Faraone is actively involved in IEEE and IEC standards related to human
exposure to RF energy and the Convener of the IEC Technical Committee 106
(Methods for the assessment of electric, magnetic and electromagnetic fields
associated with human exposure), Working Group 4.
Ray J. McKenzie received the B.S. degree in
applied science (with first class honours) in physics
from RMIT University in Melbourne, Australia, in
1996.
He was the Project Leader of the EME Safety
Research Group, Telstra Research Laboratories
(TRL), Clayton, Australia, until 2005, joining Tel-
stra’s Chief Technology Office in 2006. He was also
the Research Director, Dosimetry, at the Australian
Centre for RF Bioeffects Research (ACRBR) until
its closure in 2011. He specializes in electromagnetic
propagation and physical interactions, in particular, the dosimetry and mea-
surement of ambient fields and SAR, as well as computational electromagnetic
modeling. He has spent 23 years in the area of electromagnetic energy and
has coauthored over 50 publications and conference papers in RF dosimetry,
radiation protection, and electromagnetic interference.
Mr. McKenzie is active on Standards Australia, IEEE and IEC RF exposure
standards committees, and is a Technical Assessor for the National Association
of Testing Authorities (NATA) in the area of RF measurements. He is a member
of the Applied Computational Electromagnetics Society and the Bioelectromag-
netics Society.
653
Giorgi Bit-Babik (M’97) was born in Tbilisi,
Georgia, in 1972. He received the M.Sc. and Ph.D.
degrees in radio physics and electronics from Tbilisi
State University (TSU), Tbilisi, Georgia, in 1994
and 1998, respectively.
Until 2000, he was an Associate Professor at
TSU, where he was engaged in research on com-
putational electromagnetics. From 2001 to 2008,
he was with Motorola Corporate Electromagnetic
Energy Research Laboratory, Fort Lauderdale, FL.
He is currently a Distinguished Member of the
Technical Staff within the Chief Technology Office at Motorola Solutions, Inc.
His research interests in applied and computational electromagnetics include
antenna technology, numerical techniques and RF exposure and dosimetry. He
is actively involved in the IEEE and IEC standards related to human exposure
to RF energy and is chairing the working group developing the computational
methods for evaluation of exposure from mobile radio antennas. He has
coauthored 16 peer reviewed journal publications and over 70 conference
papers and holds nine patents in antenna technology.
Vitas Anderson (M’00) was born in Melbourne,
Australia, in 1960. After studying medicine for three
years, he changed to engineering and completed
his mechanical engineering degree (Hons) at the
University of Melbourne, in 1985 and received
the Ph.D. degree in biophysics from Swinburne
University of Technology (SUT), Australia, in 2001.
His main professional area of interest has been in
bioelectromagnetics (especially electromagnetic and
thermal dosimetry) which he began in 1989 at the Tel-
stra Research Laboratories in Melbourne, and con-
tinued as a Research Engineer until 2000. From 2001 to 2008, he was mainly
engaged as a consultant in electromagnetic safety policy, training and research.
In 2008, he was appointed as an Associate Professor of bioelectromagnetics at
SUT, and returned to consulting in late 2011. Since 1995, he has been an active
contributor to radiofrequency (RF) safety and exposure assessment standards
for the ICES/IEEE, Standards Australia and the International Electrotechnical
Commission (IEC). He has coauthored 28 peer reviewed journal publications
or book chapters and 53 conference papers.
654 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
MIMO Transmission Using a Single RF Source:
Theory and Antenna Design
Osama N. Alrabadi, Member, IEEE, Julien Perruisseau-Carrier, Member, IEEE, and
Antonis Kalis, Senior Member, IEEE
Abstract—An approach for transmitting multiple signals using a
single switched parasitic antenna (SPA) has been recently reported.
The idea there is to map the signals to be transmitted onto a set of
basis functions that serve as “virtual antennas” in the beamspace
(i.e., wavevector) domain. In this paper, we generalize the deriva-
tion of the antenna pattern basis functions regarding a three-el-
ement SPA of arbitrary radiating elements, within a symmetric
array topology, for multiplexing signals in the wavevector domain
(using different beampatterns) rather than in the hardware an-
tenna domain with multiple feeding ports. A fully operational an-
tenna system example is modeled, optimized regarding its return
loss and the power imbalance between the basis functions, and fi-
nally realized. The measurements of the SPA show good agreement
with the simulated target values, revealing an accurate design ap-
proach to be adopted as a fast SPA prototyping methodology. The
SPA has been successfully employed for multiplexing two binary
phase-shift-keying (BPSK) datastreams over-the-air, thus paving
the way for practically compact and highly efficient MIMO trans-
ceiver designs.
Index Terms—Basis functions, MIMO, reconfigurable antenna,
switched parasitic antenna.
I. INTRODUCTION
M ULTI-INPUT MULTI-OUTPUT (MIMO) communica-
tion has gained lots of attention over the last decade as
it enhances the spectral efficiency by exploiting the precious
spatial resource dimension [1], [2]. Since the emergence of this
technology, the classical approach has been assuming a trans-
mitter with a number of transmit RF chains in order to inde-
pendently map a set of signals onto a corresponding set of an-
tennas. The receiver on the other hand performs some complex
signal processing so as to decode the linear mixture of the sig-
nals and extract the useful data. However, having multiple RF
Manuscript received May 14, 2010; revised August 29, 2011; accepted
September 02, 2011. Date of publication October 25, 2011; date of current
version February 03, 2012. The work of J. Perruisseau-Carrier was supported
by the Swiss National Science Foundation (SNSF) through its Professorship
program.
O. N. Alrabadi is with the Antennas, Propagation and Radio Networking
(APNet) Group, Department of Electronic Systems, Aalborg University,
DK-9220 Aalborg, Denmark (e-mail: ona@es.aau.dk).
J. Perruisseau-Carrier is with the Group for Adaptive MicroNano Wave
Systems, LEMA/Nanolab, Ecole Polytechnique Fédérale de Lausanne (EPFL),
Lausanne CH-1015, Switzerland (e-mail: julien.perruisseau-carrier@epfl.ch).
A. Kalis is with the Broadband Wireless and Sensor Networks (BWiSE)
Group, Athens Information Technology (AIT), GR-19002, Athens, Greece
(e-mail: akal@ait.edu.gr).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173429
0018-926X/$26.00 © 2011 IEEE
chains at the user mobile terminal is rather costly. For example,
the LTE–Release 8 standard supports a single antenna for the up-
link transmission and two antennas for the downlink reception
[3], [4]. The asymmetry in the number of antennas is mainly in-
tended for avoiding the costly power amplifiers in the transmit
RF chains. Although antenna selection is a terminal option, it
requires instantaneous channel state information from the re-
ceiver back to the transmitter, which is a burden on the wireless
communication system. Consequently, classical MIMO trans-
mission especially in uplink scenarios may not be supported due
to the practical limitations of the portable RF units.
To overcome these challenges, the authors in [5] describe how
a half rate space-time (ST) code is transmitted with a single
radio. In fact, a simple time-switched ST code [6] will outper-
form the approach in [5] regarding both performance and com-
plexity. In [7], the authors propose an antenna system of two
RF sources and four antenna elements. The proposed antenna
system is capable of changing its polarization state (at the mod-
ulation rate), and thus transmitting the 4 4 Jafarkhani code.
However, having two transmit RF chains may still be costly for
low-end terminals.
On the other hand, the authors in [8] proposed a MIMO-like
system using a switched parasitic antenna (SPA) with a single
RF source. The SPA was shown to have a throughput potential
comparable to that of conventional MIMO systems by switching
the SPA far-field at the modulation rate, however no specific
multiplexing techniques were proposed. In fact, parasitic an-
tenna systems have been proposed over the past as a promising
solution for addressing the problems associated with the dif-
ficulty of integrating multiple RF chains in compact portable
units [9]. Such antenna systems comprise a single RF branch
and multiple antenna elements loaded by variable reactive im-
pedances. By controlling the reactance via a dc control, basic
antenna properties, like the beampattern, can be reconfigured.
Parasitic antennas have been widely used for providing receive
angular (or pattern) diversity (examples are given in [10] and
[11]) and have recently been proposed for analogue beam and
null steering [12].
The use of a compact-sized SPA for emulating open-loop
MIMO transmission has been first proposed in the work of Kalis
et al. in [13] followed by work of Alrabadi et al. [14]. The idea
of using an SPA as a MIMO terminal is to drive the central active
antenna with a high frequency RF signal modulated by the first
datastream, while simultaneously driving a set of parasitic ele-
ments (PEs) strongly coupled to the active one with a baseband
(low frequency) control signal as shown in Fig. 1. The baseband
control signal has information about the other datastreams to be
ALRABADI et al.: MIMO TRANSMISSION USING A SINGLE RF SOURCE: THEORY AND ANTENNA DESIGN 655

complex conjugate transpose (Hermitian) operators, respec-

tively. The notation indicates an identity matrix of size
. The operator returns a square matrix with the
elements of the vector laid across the main diagonal of the
matrix. Moreover, we consider a classical uniform three-di-
mensional angular power spectrum seen by the transmitter (the
mobile terminal), which is approximately the case when the
mobile unit is surrounded by many scatterers.
The rest of the paper is organized as follows. In Section II, we
describe a technique for transmitting two BPSK signals simulta-
neously via a single RF front end. Section III expresses the basis
functions of a three-element SPA based on full-wave electro-
magnetic modeling and optimizes the SPA for BPSK signaling
regarding the average rate of transmission. Section IV describes
an SPA example of printed dipoles and explains its design im-
plementation. Section V shows both simulation and measure-
ment results, and finally Section VI concludes the paper.

II. MIMO TRANSMISSION WITH A SINGLE RF SOURCE

In this section, we first prove the existence of an orthog-
onal basis whenever an MIPP can be formed. Based on this, a
technique for transmitting two BPSK signals using an arbitrary
Fig. 1. Schematic diagram of the proposed technique where the first bitstream
single radio based antenna system capable of forming an MIPP
is modulated, up converted and fed into the central active element whereas the is described.
second bitstream is XORed with the first one. The output control signal is used
for swapping the loads of the PE.
A. Orthogonal Bases Using an MIPP
The correlation between two arbitrary beampatterns
transmitted over the air. By this way, it has been shown that the and is given by
input datastreams are mapped onto an orthogonal set of basis
functions in the wavevector domain via a single radio and com-
pact array dimensions. (1)
In this paper, we focus on binary phase-shift-keying (BPSK)
signaling format (the extension to all phase-shift keying (PSK)
is straightforward by following the approach in [14]) where where
we first generalize the derivation of the bases from mirror
image pattern pairs (MIPPs) i.e., when one beampattern is a
mirrored version of the other, regardless of how the MIPPs are
expressed. We therefore extend previous findings by decou-
(2)
pling the wavevector domain [15] from the antenna domain
and thus enabling MIMO functionality through any antenna
system capable of creating MIPPs. At the receiver side, we are the spatial integration of the power beampatterns of
prove that the receive antenna response to a beampattern that and over the space, respectively. Whenever ,
is a linear mixture of basis functions is nothing more than the two beampatterns are called “balanced.”
the linear combination of the receive antenna responses to the Lemma 1: For an MIPP and , the set of the
different basis functions. By this way, the receiver decodes angular functions defined as
the transmitted data symbols by estimating the basis responses
using classical training techniques.
A practical antenna system example of printed dipoles is pro-
posed, modeled, optimized regarding the average rate of trans-
mission, and finally designed and demonstrated. The measured
(3)
return loss and radiation patterns are in good agreement with
the target parameters, revealing a fast and accurate designing
methodology. form an orthogonal basis.
Throughout the paper, a bold small letter designates a Proof: For two beampatterns that form an MIPP, we have
vector, and a bold big letter designates a matrix. The operators since one beampattern is just a mirrored version of the
and designate complex conjugate, transpose, and other. Moreover, the correlation between the two beams is real
656 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

(see the proof in the Appendix), and thus . Based on TABLE I

these observations, the proof is straightforward and is given as TWO BPSK SIGNALS COMBINATIONS
follows:

compactness. Letting the sole RF port be fed by the signal ,

the antenna beampattern in the far-field becomes either

(4) State 1

Corollary 1: A balanced basis is obtained by designing the

two beampatterns and described in Lemma 1 or State 2
to be orthogonal to each other, i.e., if and are
orthonormal, and are orthonormal too.1
Proof: Letting , the proof is straight-
forward, as shown in (6). In (6), and are the spatial inte- or generally as
gration of the power beampatterns of and ,
(7a)
over the space, given, respectively, by
(7b)

where is the antenna system state such that is State

(5)
1 within which the antenna system transmits over and
is State 2 within which the antenna system transmits over
. From (7b), it is obvious how the two BPSK signals:
, which is modulated in the baseband, upconverted, and fed
into the input RF port and , which is spatially
modulated on the antenna far-field by controlling the antenna
state , are mapped onto the space of and ,
respectively. In general, for any PSK modulation of order
is a set of complex numbers evenly distributed over the
unit circle, as discussed in Section IV of [12]. Table I shows the
state required for transmitting according to the value of ,
where is input vector of bits modulated into .
(6) Fig. 1 shows a schematic diagram of the proposed technique,
where the XORing of the two bitstreams gives the required ,
i.e., giving 0 and 1, which correspond to
and , respectively.
B. Transmission Technique Description
In this part, we show that an arbitrary antenna system that has C. System Training
a single RF input but has the capability of creating an MIPP will The two BPSK signals that are transmitted in the beam-space
be capable of transmitting two BPSK signals and , simulta- domain and received using a classical uniform linear array
neously. The two BPSK signals are mapped onto an orthogonal (ULA) of antenna elements ( -element ULA), can be
set of basis functions; thus, independent fading between the two decoded by first estimating the receive antenna responses to the
signals is almost always guaranteed regardless of the transceiver proposed basis.
1The reason that we acquire an orthonormal basis and Proposition 1: A beampattern comprising a linear mixture of
from an MIPP is that the MIPP by itself represents a linear combination (desired basis functions (at the transmitter side) triggers a linear combi-
multiplexing relation) of the basis onto which the signals are mapped. The di-
versity action of the system directly depends on the transmit covariance of the nation of the individual channel responses to the different basis
basis (proportional to the identity matrix when the basis is orthonormal). functions (at the receiver side).
ALRABADI et al.: MIMO TRANSMISSION USING A SINGLE RF SOURCE: THEORY AND ANTENNA DESIGN
Proof: This directly stems from the principle of superposi-
tion in linear systems. To have a deeper insight, we first define
2 1 column vectors , and
, where the first and the second elements of every
column vector represent the and polarizations of the cor-
responding pattern, respectively. We also define as
the vector of the polarization components of the th receiver an-
tenna pattern . As in [16], we assume that the propa-
gation channel between the transmitter and the receiver consists
of a set of plane waves, with the th wave characterized by a
complex voltage gain , angle of departure , and
angle of arrival . We also assume that each plane
wave undergoes a polarization transformation due to scattering
that can be expressed as the unitary matrix
(8)
The response of the th receive antenna when il-
luminated by the beampattern is the complex channel
gain representing the ratio of the received voltage signal to the
transmitted voltage signal and may be written as shown in (9),
where is a constant that depends on the receiver and the
transmitter active gains and impedances [17], and
are the responses of the th receive antenna to and
, respectively. By applying the same analysis, the re-
sponse of the th receive antenna when illuminated by
becomes . by the superposition of the retarded currents induced on the
(9)
Based on this, the receiver can decode the two BPSK signals by
estimating the channel responses of the basis as
(10a)
(10b)
By constructing the matrix of the receive antennas’ responses,
the receiver can zero-force the received signal by inverting the
channel matrix (or using any other reception techniques) for de-
coding and .
657
III. ANTENNA MODEL AND OPTIMIZATION
In this paper, we adopt the antenna topology proposed in [14],
i.e., a symmetrical three-element SPA, where the central ele-
ment is the active one, while the other two are passive. The
two parasitic elements are loaded with pure imaginary loads
as the real part of a complex load degrades the ef-
ficiency of the antenna system. Obviously, the antenna system
can create an MIPP (the plane in Fig. 1) by simply permuting
the reactive loads of the PE as , based
on image theory. In other words, having the first beampattern
at , the beampattern
is obtained at . Consequently, by feeding the central
active element with the first BPSK datastream and permuting
the loads according to the second datastream, the two streams
are simultaneously transmitted out of a single radio and mapped
onto an orthogonal basis according to Lemma 1, irrespective of
and . Having the two loads and as a degree of
freedom when considering BPSK signaling, we can optimize the
loads according to a specific criterion as shown in Section III-C.
A. Generalized Derivation of Antenna Basis Functions
Although the beampattern of thin electrical dipoles (or
monopoles) can be practically approximated as an array factor
wire antenna elements such as in [14, eq. (6)], this is not true
when considering general2 radiating elements, e.g., flat or
fractal dipoles, slot antennas, etc. To overcome this problem,
we implement full wave electromagnetic modeling based
on the SPA scattering parameters (S-parameters) denoted by
, as well as the 3D complex active port
patterns3 of the antenna elements 0, 1, and 2 shown in Fig. 1,
denoted by and , respectively. An
expression of the electric far-field beampattern of a three-el-
ement SPA based on the aforementioned quantities and the
variable antenna loading has been derived in [19] using Mason’s
rule. From [19] and after correcting the equations to properly
adhere to Mason’s rule, the two basis functions obtained when
swapping the imaginary loads of the two parasitic elements
become
(11)
2Again, we emphasize that the arbitrariness of the elements is limited to the
center element being a self mirror image, and the outer two being respective
mirror images of each other, both about a vertical plane that divides the left and
right sides of the SPA structure.
3The active port pattern is defined as the beampattern obtained when driving
the corresponding port (whether being active or passive) with a unit excitation
voltage signal while terminating the other ports with reference impedances [18].
658 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

where vector , the transmit covariance

matrix5 is obtained as

(16)

which is simply the power distribution across the basis functions

(12) since according to Lemma 1. Notice that
, which is easily obtained from (2) and the
basis definition in (3), where is the average transmit power.
and such that Defining the power imbalance ratio between the basis functions
as , we can write as , where is the
normalized power distribution across the basis functions such
that . can be written as such that
(13) and . From the above, the received
signal model becomes
(14)
(17)

where we assumed by having the source impedance at where is the power into the transmitter (input power) and
the central driven port equal to the reference impedance is the efficiency of the transmit antenna system being
. The basis coefficients in (12) are derived with respect to equal to , where is the SPA return loss derived
a general scattering matrix. Swapping the two reactive loads in [19]. Finally, is the vector of the modulated
as , swaps the coefficients BPSK signals (see Table I), and is a vector representing the
in (11), thus phase-shifting by white Gaussian noise, with zero mean and variance.
180 without affecting . By this way, the factor
in (7a) is obtained. The two functions and C. Optimization Criterion
are the basis functions that are used to transmit two PSK signals
In this paper, we define the optimal SPA loads as the ones that
of any modulation order [14].
maximize the average rate of transmission. However, in MIMO
communications, average rate computation often demands tack-
B. Received Signal Model ling calculations of expectations with respect to random ma-
trices rather than random scalar variables. For this reason, we
Considering a narrowband, flat-fading, point-to-point com- derive an upperbound on the average rate and deploy it as an
munication link where the two BPSK symbols are transmitted in optimization criterion. We assume open-loop operation where
the beam-space domain over two basis functions (equivalent to the channel is known to the receiver but unknown to the trans-
two uncorrelated virtual antennas) and received using an -el- mitter. The ergodic capacity of a MIMO random channel, de-
ement ULA of uncorrelated and uncoupled antenna elements. noted by , is the ensemble average of the information rate
Assuming independent fading statistics at the transmitter and over the distribution of the elements of the channel matrix
the receiver, the Kronecker product [20] can be assumed and . By using the formula [22], the
thus the channel transfer function can be written as4 upper bound that comes from the Jensen’s inequality and the
concavity of ,6 we get
(15)

where the elements of the matrix are inde-

pendent and identically distributed (i.i.d.) complex Gaussian
random variables with zero mean and unit variance. The cor-
relation at the receiver side is ignored by the aforementioned
assumptions regarding the receiving ULA. Defining the row
4In [21], the correlation based channel model accounts for the mutual cou-
pling by explicitly incorporating the coupling matrices. However, in (15), the
mutual coupling is implicitly taken into consideration within the calculation of
the basis functions in (11).
5Since the basis functions are imbalanced, the transmit covariance matrix
rather than the transmit correlation matrix is considered.
6The is concave over positive semi-definite matrices [23]. Since
is positive semi-definite, the term is positive semi-
definite too, as it is a one-to-one mapping of , thus preserving the positive
definiteness.
ALRABADI et al.: MIMO TRANSMISSION USING A SINGLE RF SOURCE: THEORY AND ANTENNA DESIGN 659

(18)

In (18), the average transmitted power is not divided by the

number of the basis functions (the number of the virtual an-
tennas), since the trace of is normalized to a unity rather than
to the number of the basis functions (both forms are equivalent).
The optimal loading is defined as the one that maximizes the av-
erage throughput upperbound in (18), i.e.,

Fig. 2. Schematic diagram of the SPA initially proposed in [24].

(19)
In (19), is made part of the optimization criterion by con-
straining rather than as the SPA efficiency is a key de-
sign parameter when considering portable RF units with limited
storage batteries.

IV. ANTENNA SYSTEM DESIGN

In this section, we consider the three-element SPA shown in
Fig. 2, where the radiating elements are thin printed dipoles. The
planar topology of the SPA makes it better fit in compactness-
constrained mobile units as compared to the majority of the wire
parasitic antennas already proposed in the literature. The current
SPA was proposed earlier in [24]; however, in this paper, we
complete the work by describing the implementation and the
measurements of the prototype. Fig. 3. An optimization contour map regarding the upperbound on with
respect to and .

A. Design Parameters and Optimal Loading

The first design steps consist of making some initial choices
on the antenna materials and the basic topology. We consider a
three-element SPA of flat dipoles as radiating elements as shown
in Fig. 2, designed on an 1.5-mm-thick substrate of relative
permittivity . The dipole lengths and spacing are 48.3
and 11 mm, respectively. The spacing is at the desired
operational frequency of 2.6 GHz. The SPA was simulated using
HFSS, with ports at the locations of the variable loads. The re-
sulting scattering matrix is given by
(20)
where the matrix is symmetric by the reciprocity theorem i.e.,
by the usual assumption of employing antennas with electrically
reciprocal materials, thus . Moreover, the symmetric
topology of the SPA shown in Fig. 1 ensures that and
. The antenna system is lossy as when
compared to the lossless four-port network (expressed by ) of Fig. 4 so that the overall impedance in each state
in [19] as by the energy conservation principle
when including the radiated beams in the network structure. Fur-
ther, the diagonal elements of are nonvanishing as we aim at
diminishing the return loss of the central active element rather
than . The resulting three-port S-parameters and the complex
3D active port patterns were exported to Matlab, where a com-
puter routine scans the realizable range of the reactance space
searching for given by (19). Fig. 3 shows an op-
timization contour plot of at a transmit signal to
noise ratio7 (SNR) 10 dB. The figure shows that is
maximized at . At such loading,
the upperbound on is 5 b/s/Hz, the power imbalance be-
tween the two basis functions is 0.56 dB, and the SPA efficiency
is 97%.
B. Reconfigurable Impedance Implementation
The design of the variable load, explained in more detail in
the earlier partial work of [24], consists of the following steps:
First, an adequate layout for the reconfigurable load area, to be
controlled using a PIN diode, is selected [“Reconfigurable Load
Area” in Fig. 4(a)]. The parasitic capacitance ( , and
) between the different pads are extracted from full-wave
simulations. Here the inductive effect in the pads can be ne-
glected in the design. Subsequently, the surface-mounted ele-
ments to implement and are deduced from the circuit
and
7In fact, the transmit SNR is commonly used in the literature when
evaluating the system performance. However, as the SPA loading will affect
the transmit SNR through the matching efficiency, it seems more reasonable
to use , which is simply the transmit SNR before the mismatch effect
represented by . On the other hand, the way of calculating the receive SNR
is different and is shown later in (20).
660 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 4. Reconfigurable dipole load impedance: (a) Layout and elements view,
including biasing network; (b) target two-state variable impedance; and (c) de-
tailed implementation circuit, including layout parasitic capacitances.

[see Fig. 4(b)] match the target values deduced in the previous
section, namely and
. Finally, a dc biasing network was designed using
large RF-block inductors and a resistor to precisely
control the diode biasing current. As can be seen in Fig. 6, the
dc paths are then driven to the other side of the substrate by vias,
where they can conveniently be connected to the dc voltage ref-
erences in the antenna environment (see Section V). The PIN
diode (Aeroflex Metelics MPN7310A-0805) serves as a low Fig. 5. Measured load impedance in each diode state of the PIN diode, ex-
tracted from the S-parameter measurements on a dedicated microstrip TRL cal-
capacitance fast switch, with a negligible transient switching ibration kit, from [24]. The OFF and ON diode states correspond to a reversed
time (orders of nanoseconds). and are capacitors of 0 V and forward ( 9 mA) bias, respectively.
0.5 and 0.8 pF, respectively. The biasing network elements are
22 nH and 910 .
In order to experimentally validate the reconfigurable load
design prior to its insertion in each of the SPA parasitic dipoles,
it was fabricated and measured using a thru-reflect-line (TRL)
calibration kit, which allows placing the measurement reference
planes at the desired locations, as required here. It is then pos-
sible to extract the desired impedances and from the mea-
sured S-parameters and microstrip line impedance, as shown
in Fig. 5 for each of the diode states. The imaginary parts of
the measured impedance and at the design frequency of
2.6 GHz are 38 and 108 in the ON and the OFF states,
Fig. 6. Photograph of the fully operational SPA, optimized for the proposed
respectively. These values are close to the target reactances of aerial MIMO approach.
27 and 100 , considering the tolerances of the SMD el-
ements and the impact of the biasing network. The real parts of
and are not exactly zero due the diode and SMD com- A. Antenna Demonstration
ponents finite resistances, which were neglected in the design
procedure (their measured average values are only 5 and A photograph of the fully operational fabricated antenna is
3 in the OFF and the ON states, respectively). The target basis shown in Fig. 6. It was observed that a good balanced excita-
functions (at ) and the achieved tion of the active dipole is simply obtained by connecting the
ones are compared in Fig. 7, central and the outer conductors of a coaxial connector to each
showing very good agreement. of the dipole arms. The variable load designed and character-
ized in Section IV-B was introduced in each parasitic dipole of
the SPA, including the dc biasing network. The dc ground pad
V. SIMULATION AND MEASUREMENT RESULTS of each variable load on the backside of the substrate is con-
The current section presents the measurements of the dif- nected by a printed line to the coaxial connector outer conductor
ferent SPA parameters and compares them to the corresponding (which thus serves as a dc ground), whereas each actuation pad
parameters obtained by computer simulations. (shown as “ ” in Fig. 4) is connected by a thin wire to the
ALRABADI et al.: MIMO TRANSMISSION USING A SINGLE RF SOURCE: THEORY AND ANTENNA DESIGN
Fig. 7. The magnitude of the H-plane co- polarized basis functions at the target
loads of and at the practically achieved loads of .
The two basis functions resemble the omni and the angular sine functions, which
are orthogonal to each other.
Fig. 8. Setup of the antenna for reconfigurable radiation pattern measurements,
with a 9-V battery places behind the absorber cone in a direction of the low field
intensity.
bias voltages for controlling the states of the diodes. In order to
improve the antenna performance and provide pure measured
patterns, the dc wires are driven along the coaxial feed, which
is oriented toward the minimum radiation of the SPA (i.e., par-
allel to the dipoles, see Fig. 6). A standard 9-V battery is used as
a dc source in the radiation pattern measurements. The battery
is placed behind a piece of an absorber (located in the direction
of minimum radiated power density), as can be seen in Fig. 8.
Therefore the antenna states were simply selected by connecting
each of the two dc wires to the 0 or 9 V references. The impact
of the biasing voltages on the antenna performance was investi-
gated, showing similar responses for 10 to 0 V as the OFF (or
“reverse-biased”) state, while 3 V to 10 V are acceptable for
the ON (or “forward-biased”) state.
B. Return Loss
Fig. 9 shows the simulated and measured return loss of the
SPA around the design frequency of 2.6 GHz. The graph only
shows the response in the operational states of the antenna,
namely when it is loaded by the reactance load pairs
and . As explained earlier in Section IV, the return
loss is the same for both states due to the SPA symmetry, which
is confirmed here by the similarity between the two measured
661
Fig. 9. Return loss (dB) of the SPA for both loading states, i.e., and
.
curves in Fig. 9. The SPA was found to have poor matching
in the two (unused) states and ), which
are not shown here. The agreement between simulations and
the measurements is moderate, since the measured bandwidth
is larger than the one obtained by simulation and is not exactly
centered around the design frequency of 2.6 GHz. Neverthe-
less the measurements show good return loss at 2.6 GHz. The
10-dB measured bandwidth is 5.6% and 7.1% for a reference
of 10 dB, for and , respectively.
C. Radiation Patterns
Fig. 10 shows the H-plane co- and cross- polarized far
fields in the first operational antenna state . Note
that the maximum of the co-polarized beampattern, located at
90 , corresponds to the direction of the load in the OFF
state. The simulated and measured co- and cross- polarized
beampatterns are in good agreement, as shown in Fig. 10.
Because of the SPA symmetrical structure and the reactance
pair antisymmetry, the other antenna beampattern should
simply be a mirror image of the first beampattern around the
0 180 axis, which is well verified by the measured
prototype, as can be seen in Fig. 11.
D. Experimental Results
The proposed antenna prototype has been successfully used
for spatially multiplexing two BPSK datastreams over the air
at 2.6 GHz. The experiments constitute to the best of the au-
thors’ knowledge the first MIMO transmission with a single RF
source yet to be proposed. The first train was modulated into
a BPSK symbol stream (using a raised-cosine waveform with
0.3 roll-off factor) and upconverted to 2.6 GHz. The high fre-
quency signal was modulated to the central element within a
modulation bandwidth of 533 kHz. The second binary train was
XORed with the first binary train in the baseband domain and
the output baseband control signal was amplified and used for
662 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 10. Simulated and measured co- and cross- polarization components of the
beampattern in the H-plane, i.e., , at 2.6 GHz.
Fig. 11. Measured co- and cross- polarization components of the beampatterns
and at 2.6 GHz. Notice that
, resulting in an MIPP.
switching the SPA loads. A simple zero-forcing decoding was
implemented by the receiver which was equipped with two dis-
tant omnidirectional monopole antennas separated from each
other by 23 cm or , and both are located several wavelengths
from the SPA (the receiver is located in the broadside direction
of the SPA, but completely blocked from the transmitter in the
sense that no line-of-sight between the transmitter and the re-
ceiver exists). The receiver first estimates the receive antennas’
responses to the two beampatterns and using
classical training, then the response to the basis is obtained from
(10a)–(10b). Finally, the 2 2 complex channel matrix is in-
verted and used for equalizing the received signal. A total bit
rate of 820 kb/s was obtained with arbitrarily low error; thus, a
spectral efficiency of 1.54 b/s/Hz can be claimed. Although this
seems far from the target upperbound of 5 b/s/Hz, it is well justi-
fied by the fact of using real signaling with uniform distribution
rather than complex signaling with Gaussian distribution. The
details of the experiments’ setup are detailed in [25].
Fig. 12 shows the received signal constellations after equal-
ization (spatial separation), onto which the transmitted signals
Fig. 12. Scatter plot of received signal constellation after equalization.
Fig. 13. Probability of error versus the transmit SNR (per bit).
(red dots) are also projected, for comparison reasons. Every
demodulated signal comprises of two noisy clouds such that
and . The receive SNR of the th
cloud is calculated as
SNR (21)
where returns the sample mean of the operand, and
returns the sample variance of the operand. The four
clouds have almost the same SNR, and the mean of the four
SNRs is finally considered. The bit SNR referred to as
is calculated by adding to the average SNR (in
decibels), where is the number of samples per one symbol,
whereas the 0.5 factor is due to using real signaling. In this ex-
periment, was set to five samples per symbol such that each
transmission has 410 symbols or equivalently 2048 samples.
On the other hand, Fig. 13 shows the bit probability of error
versus obtained by measurements as well as the
performance of a 2 2 BPSK-MIMO with a Rayleigh channel
of independent and identically distributed coefficients, and
ALRABADI et al.: MIMO TRANSMISSION USING A SINGLE RF SOURCE: THEORY AND ANTENNA DESIGN
zero-forcing decoding.8 The figure shows that the performance
of the beamspace MIMO is comparable to the conventional
one, thus validating the importance of such a new approach for
realizing single radio compact-sized MIMO transceivers.
VI. CONCLUSION
The paper generalized a previously reported approach for
transmitting multiple signals using a single RF source. The idea
is to obtain an orthogonal or orthonormal basis out of MIPPs.
The paper also provided design steps for an example of a three-
element SPA, capable of forming an MIPP that are mirror im-
ages of each other. The SPA was optimized for BPSK signaling
by deriving a criterion that maximizes the SPA efficiency and
minimizes the power imbalance between the basis functions, si-
multaneously. A reconfigurable impedance was designed and a
fully operational SPA for single radio MIMO transmission was
demonstrated for the first time. The measured SPA parameters
are in good agreement with the target values, regarding the SPA
return loss and the radiation patterns in the different SPA states.
Finally, the SPA has been successfully used for multiplexing
two BPSK datastreams with a total bit rate of 820 kb/s.
APPENDIX
In this part, we prove that the cross-correlation of an MIPP in
a uniform field is real.
Proof: Assuming an MIPP over the angular domain,
where is the azimuth polar system of coordinates with a refer-
ence axis taken from the MIPP axis of symmetry (the can be
dropped for simplicity). The MIPP can generally be written as
and . The can be further written
as , where and are the real and
imaginary parts of . The cross-correlation of the MIPP
becomes
ACKNOWLEDGMENT
The authors like to thank P. Pardo for fabricating and mea-
suring the variable load circuit, and P. Miskovsky for his help in
8Theoretically, of a 2 2 BPSK-MIMO under Rayleigh fading and a zero-
forcing receiver is unsurprisingly identical to the performance of 1 1 BPSK-
SISO i.e., [26].
663
the antenna measurement, both at Centre Tecnològic de Teleco-
municacions de Catalunya (CTTC), Barcelona, Spain. Finally,
the authors express their deepest gratitude to the editor and to the
anonymous reviewers for their helpful and constructive com-
ments and suggestions.
REFERENCES
[1] J. H. Winters, “On the capacity of radio communication systems
with diversity in a Rayleigh fading environment,” IEEE J. Sel. Areas
Commun., vol. SAC-5, pp. 871–878, Jun. 1987.
[2] G. J. Foschini and M. J. Gans, “Limits of wireless communication in
a fading environment when using multiple antennas,” Wireless Pers.
Commun., vol. 6, no. 3, pp. 311–335, 1998.
[3] Overview of 3GPP Release 8 V0.0.3, Nov. 2008 [Online]. Available:
http://www.3gpp.org/Release-8, available online at
[4] J. Kotecha, “LTE:MIMO techniques in 3GPP-LTE,” Freescale Semi-
conductor, Inc., Jun. 2008, available online.
[5] M. Okoniewski, S. V. Hum, A. Sutinjo, and G. G. Messier, “A space-
time coding scheme utilizing phase shifting antennas at RF frequen-
cies,” IEEE Antennas Wireless Propag. Lett., vol. 4, pp. 369–372, 2005.
[6] J. Yuan and B. Vucetic, Space-Time Coding. New York: Wiley, 2003.
[7] A. Grau, J. Romeu, M. Lee, S. Blanch, L. Jofre, and F. De Flaviis, “A
dual-linearly-polarized MEMS-reconfigurable antenna for narrowband
MIMO communication systems,” IIEEE Trans. Antennas Propag., vol.
58, no. 1, pp. 4–17, Jan. 2010.
[8] M. Wennstorm and T. Savantesson, “An antenna solution for MIMO
channels: The switched parasitic antenna,” in Proc. 12th IEEE Int.
Symp. Pers., Indoor, Mobile Radio Commun., Sep. 2001, vol. 1, pp.
159–163.
[9] R. Vaughan, “Switched parasitic elements for antenna diversity,”
IIEEE Trans. Antennas Propag., vol. 47, no. 2, pp. 399–405, Feb.
1999.
[10] T. Sawaya, K. Iigusa, M. Taromaru, and T. Ohira, “Reactance diver-
sity: Proof-of-concept experiments in an indoor multipath-fading en-
vironment with a 5-GHz prototype planar ESPAR antenna,” in Proc.
Consumer Commun. Netw. Conf., Jan. 5–8, 2004, pp. 678–680.
[11] M. Yamamoto, M. Taromaru, H. Sadamichi, and A. Shimizu, “Per-
formance of angle switch diversity using ESPAR antenna for mobile
reception of terrestrial digital TV,” in Proc. IEEE 64th Veh. Technol.
Conf. (VTC–Fall 2006), 2006, pp. 1–5.
[12] C. Sun, A. Hirata, T. Ohira, and N. C. Karmakar, “Fast beamforming
of electronically steerable parasitic array radiator antennas: Theory
and experiment,” IIEEE Trans. Antennas Propag., vol. 52, no. 7, pp.
1819–1832, Jul. 2004.
[13] A. Kalis, A. G. Kanatas, and C. B. Papadias, “A novel approach to
MIMO transmission using a single RF front end,” IEEE J. Sel. Areas
Commun., vol. 26, no. 6, pp. 972–980, Aug. 2008.
[14] O. N. Alrabadi, C. B. Papadias, A. Kalis, and R. Prasad, “A universal
encoding scheme for MIMO transmission using a single active element
for PSK modulation schemes,” IEEE Trans. Wireless Commun., vol. 8,
no. 10, pp. 5133–5142, Oct. 2009.
[15] A. S. Y. Poon, R. W. Brodersen, and D. N. C. Tse, “Degrees of freedom
in multiple-antenna channels: A signal space approach,” IEEE Trans.
Inf. Theory, vol. 51, no. 2, pp. 523–536, Feb. 2005.
[16] M. L. Morris and M. A. Jensen, “Network model for MIMO systems
with coupled antennas and noisy amplifiers,” IEEE Trans. Antennas
Propag., vol. 53, no. 1, pp. 545–552, Jan. 2005.
[17] C. Waldschmidt, S. Schulteis, and W. Wiesbeck, “Complete RF
system model for analysis of compact MIMO arrays,” IEEE Trans.
Veh. Technol., vol. 53, no. 3, pp. 579–586, May 2004.
[18] D. M. Pozar, “The active element pattern,” IEEE Trans. Antennas
Propag., vol. 42, no. 8, pp. 1176–1178, Aug. 1994.
[19] L. Petit, L. Dussopt, and J. Laheurte, “MEMS-switched parasitic-an-
tenna array for radiation pattern diversity,” IIEEE Trans. Antennas
Propag., vol. 54, no. 9, pp. 2624–2631, Sep. 2006.
[20] D. S. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, “Fading corre-
lation and its effect on the capacity of multielement antenna systems,”
IEEE Trans. Commun., vol. 48, no. 3, pp. 502–513, Mar. 2000.
[21] Y. Fei, Y. Fan, B. K. Lau, and J. S. Thompson, “Optimal single-port
matching impedance for capacity maximization in compact MIMO ar-
rays,” IEEE Trans. Antennas Propag., vol. 56, no. 11, pp. 3566–3575,
Nov. 2008.
664 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[22] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
Communications. Cambridge, U.K.: Cambridge Univ. Press, 2003.
[23] R. A. Horn, Matrix Analysis. Cambridge, U.K.: Cambridge Univ.
Press, 1996.
[24] J. Perruisseau-Carrier, O. N. Alrabadi, and A. Kalis, “Implementation
of a reconfigurable parasitic antenna for beam-space BPSK transmis-
sions,” in Proc. Eur. Microw. Conf. (EuMA), Paris, France, Sep. 2010,
pp. 644–647.
[25] O. N. Alrabadi et al., “Spatial multiplexing with a single radio:
Proof-of-concept experiments in an indoor environment with a 2.6
GHz prototype,” IEEE Commun. Lett., vol. 15, no. 2, pp. 178–180,
Dec. 17, 2010.
[26] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
Cambridge, U.K.: Cambridge Univ. Press, 2005.
Osama N. Alrabadi (M’10) was born in Jordan
in 1979. He received the Electrical Engineering
Diploma from the University of Jordan (JU),
Amman, in 2002, the Master’s degree in information
and telecommunications (M.S.I.T.T.) (with highest
hons.) from Athens Information Technology (AIT),
Greece, in 2007, and the Ph.D. degree from Aalborg
University (AAU), Denmark, in 2011.
In 2001, he joined Orange Mobile (previously
MobileCom) for undergraduate training in the
Microwave and GSM Network Maintenance De-
partment. From 2002 to 2006, he was working at the National Electrical Power
Company (NEPCO) and the Jordan Electrical Power Company (JEPCO) in
remote control, metering, and communications. He is currently a Postdoctoral
Fellow at the Antennas, Propagation and Radio Networking (APNet) group at
AAU. He is currently working on the Smart Antenna Frontend (SAFE) project,
in cooperation with the antenna company Molex, the tuner company Wispry,
Irvine, CA, and Intel Mobile Communications. His research interests include
design and modeling of small antenna arrays, space–time signal processing, as
well as MIMO and cognitive radio transceiver architectures.
Dr. Alrabadi has been a member of the Jordan Engineering Association since
2002.
Julien Perruisseau-Carrier (S’07–M’09) was born
in Lausanne, Switzerland, in 1979. He received
the M.Sc. and Ph.D. degrees from the Ecole Poly-
technique Fédérale de Lausanne (EPFL), Lausanne,
Switzerland, in 2003 and 2007, respectively.
In 2003, he was with the University of Birm-
ingham, U.K., first as a visiting EPFL M.Sc. student
and then as a short-term researcher. From 2004 to
2007, he was with the Laboratory of Electromag-
netics and Acoustics (LEMA), EPFL, where he
completed his Ph.D. while working on various EU
funded projects. From 2007 to 2011, he was an Associate Researcher with the
Centre Tecnològic de Telecomunicacions de Catalunya (CTTC), Barcelona,
Spain. Since June 2011, he has been a Professor at EPFL funded by the Swiss
National Science Foundation (SNSF), where he leads the group for Adaptive
MicroNano Wave Systems. He has led various projects and work packages
at the National, European Space Agency, European, and industrial levels.
He has authored more than 50 conference papers and more than 20 journal
papers. His main research interest concerns interdisciplinary topics related
to electromagnetic micro- and millimeter-waves: dynamic reconfiguration,
application of micro/nanotechnology, metamaterials, and joint antenna-coding
techniques.
Prof. Perruisseau-Carrier was the recipient of the Young Scientist Award of
the URSI-EMTS 2007 conference (Ottawa, Canada), of a Torres Quevedo Grant
awarded by the Spanish government, and of the Raj Mittra Travel Grant 2010
presented by the IEEE Antennas and Propagation Society.
Antonis Kalis (A’01–M’03–SM’10) received the
Electrical Engineering Diploma degree from the
Electric Engineering Department of the University
of Patras, Greece, in 1997 and the Ph.D. degree from
the University of Patras in 2002.
In 1997, he joined the Laboratory of Electromag-
netics at the University of Patras, participating in var-
ious R&D projects for the Greek Government and the
European Union, as a Member of Research Staff. In
2000, he worked as a Research Engineer and an As-
sistant Research Unit Manager at the Computer Tech-
nology Institute. He joined Athens Information Technology (AIT) in January
2003. During the summer semester of the same year, he taught the course Net-
work Design and Evaluation at Carnegie Mellon University, Pittsburgh, PA. In
2004, he supervised the work of D. Leang in Pittsburgh, PA, which received the
Kennametal Fellowship Award. Since 2005, he has been an Adjunct Professor
at Carnegie Mellon University. He is also currently a Professor at AIT. From
June to November 2009, he spent a semester in the Centre Tecnològic de Tele-
comunicacions de Catalunya (CTTC), Barcelona, Spain, as a Visiting Scientist.
Since 2003, he has also worked as a consultant for a number of small enter-
prises in Cyprus and the United States. His research interests are in the areas
of radio communications, antenna design, and wireless networks. He has nu-
merous journal and conference publications and a U.S. patent.
Dr. Kalis received the Chester Sall Memorial award of the Consumer Elec-
tronics Society in the areas of radio communications, antenna design, and wire-
less networks in 2000. His research output has drawn the interest of the sci-
entific community, underlined by a significant number of invited talks (e.g.,
in Bell Labs, Princeton University, Carnegie Mellon University, University of
Trento, University of Piraeus, etc.) and one IEEE tutorial in PIMRC 2007, en-
titled “Developments in ESPAR Antennas.” A paper that he coauthored ap-
peared in the October 2010 issue of the IEEE TRANSACTIONS ON WIRELESS
COMMUNICATIONS, and was nominated for the Marconi paper award. Other
recognitions include numerous invitations to act as an organization chair, ses-
sion chair, and member of the technical program committee in major IEEE con-
ferences in the field of communications. He has been a member of the Technical
Chamber of Greece since 1998.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
MIMO Capacity Enhancement Using Parasitic
Reconfigurable Aperture Antennas (RECAPs)
Rashid Mehmood, Student Member, IEEE, and Jon W. Wallace, Member, IEEE
Abstract—The capacity of multiple-input multiple-output
(MIMO) systems employing reconfigurable apertures (RECAPs)
is carefully analyzed with a realistic thermal noise model for three
different power constraints: average receive signal-to-noise ratio
(SNR), maximum effective isotropic radiated power (EIRP), and
average transmit power. Performance is studied not only for a
noise-limited single link, but also in the presence of interference
and multiple RECAP-equipped users. The impact of loss and finite
bandwidth on the operation of the RECAP is also considered. For
the practical EIRP constraint, results show that a compact MIMO
RECAP provides 30%–50% capacity improvement for
a single link. It is also found that RECAPs are even more beneficial
in interference-limited and multiuser scenarios, where capacity
is increased by 50% to 800% depending on the severity of the
interference, indicating that RECAPs are an attractive solution
for future wireless systems employing aggressive spectral reuse.
Index Terms—Information rates, interference suppression, mul-
tiple-input multiple-output (MIMO) systems, reconfigurable an-
tennas.
I. INTRODUCTION
R ELIABLE and high performance transmission con-
tinues to be a major goal of wireless communication
systems, which is significantly enhanced by arrays employing
beamforming and diversity techniques. Multiple-input mul-
tiple-output (MIMO) wireless technology emerged in the 1980s
and has gained increasing attention due to the significant gains
in channel capacity [1], [2], possible by exploiting channel
multi-path with spatial multiplexing. In a communication
system, the channel matrix includes effects of the physical
propagation environment and antenna radiation and reception
characteristics. Antennas can be viewed as transmit and re-
ceive filters that are ideally matched to the physical channel,
enhancing signals of interest and mitigating noise and inter-
ference to maximize capacity [3]. Although for a single fixed
antenna, no adaptation of spatial filtering is possible, multiple
fixed antennas connected to multiple radio frequency (RF) and
digital signal processing (DSP) chains can employ the “smart
Manuscript received July 21, 2010; revised November 30, 2010; accepted
January 15, 2011. Date of publication October 25, 2011; date of current version
February 03, 2012.
The authors are with the School of Engineering and Science, Jacobs Uni-
versity Bremen, Bremen 28759, Germany (e-mail: r.mehmood@ieee.org;
wall@ieee.org).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173445
0018-926X/$26.00 © 2011 IEEE
665
antenna” concept [4] to provide dynamic spatial filtering. How-
ever, the increased RF and DSP resources may be prohibitive
for many applications.
The term reconfigurable aperture (RECAP) antenna [5], [6]
refers to a large array of analog reconfigurable elements (REs),
which can be manipulated in order to support beam-steering,
signal-to-noise (SNR) maximization, interference suppression,
and dynamic matching. In contrast to smart antennas, RECAPs
adapt directly in the analog radio-frequency (RF) domain and
require only a single RF chain and modest DSP resources, po-
tentially providing lower cost. RECAPs are also interesting for
MIMO systems, where the optimal antenna array exploits the
multi-path to provide peak capacity while using as few active
RF chains as possible. Also, for multi-user systems RECAPs
can adapt patterns to dynamically partition spatial reuse of spec-
tral resources.
Optimal antenna selection for MIMO has been considered
(e.g., [7]), where only a few antennas out of a set of antennas
are chosen for capacity maximization with lower complexity.
The improvement in the channel capacity using a reconfigurable
antenna is presented in [8], where moderate sized switched par-
asitic arrays with relatively few REs are used. A practical an-
tenna solution providing multiple patterns with a single fixed
antenna is presented in [9], exhibiting improved performance
compared to spatially separated dipoles. Capacity maximization
using planar RECAPs at transmit and receive is investigated in
[10], where each antenna acts as a single RECAP. A reconfig-
urable MIMO array consisting of two dipole elements is intro-
duced in [11], where by adaptively changing the length of the
dipoles, modest increases in single-user capacity are possible.
The important study in [12] shows that MIMO systems with re-
configurable antennas have a maximum diversity order equal
to the product of the number of transmit antennas, receive an-
tennas, and the reconfigurable states. This idea is expanded in
[13], where not only practical space-time coding methods that
code over the antenna state to maximize diversity are developed,
but also practical aspects like antenna switching time are con-
sidered.
This previous work on reconfigurable MIMO systems has
some limitations. First, only simple termination-independent re-
ceiver noise has been considered, which is known to be in-
accurate for analyzing MIMO systems with variable termina-
tion [14]. Second, limited reconfigurability has been considered,
which may be insufficient to exploit the degrees of freedom of
the occupied aperture. Third, the role of the power constraint
has not been studied in detail, where typically only an average
transmit power constraint has been assumed. Finally, capacity
maximization for a single link limited by thermal noise has been
666 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

considered, but multi-user systems with interference are more

realistic for today’s wireless scenarios.
This work provides a more comprehensive analysis of ca-
pacity enhancement possible with reconfigurable antennas by
addressing these previous shortcomings. To this end, we study
a RECAP consisting of a 9 9 parasitic array having sufficient
complexity to exploit a compact aperture as studied
in [15]. In contrast to [10], the complete aperture is exploited
rather than using separate RECAPs for each MIMO antenna. A
realistic noise model is considered in order to take into account
the effect of matching on amplifier noise. Three realistic but
distinct power constraints are also considered, indicating where
RECAPs are most effective: 1) average signal-to-noise (SNR),
where transmit and receive power are normalized and the focus
of optimization is on channel orthogonality and multi-path en-
hancement; 2) effective isotropic radiated power (EIRP), which
allows power enhancement at receiver but not at transmitter,
which is more practical for many of today’s communication sys-
tems; and 3) average transmit power, which is a commonly as-
sumed constraint allowing power enhancement at both transmit Fig. 1. Configurations for non-RECAP and RECAP arrays: (a) perspective
view of the parasitic RECAP with 2 feeds. (b) Top view of RECAP configu-
and receive. In addition to considering a single link limited by rations, where red boxes show RE positions, blue stars and circles show feed
thermal noise, we also consider fixed interference and multiple locations for 2 2 and 4 4 MIMO respectively. (c) Top view of antenna
RECAP-equipped links. positions for non-RECAP for 2 2 MIMO (stars) and 4 4 MIMO (circles),
where boxes are empty locations.
The remainder of the paper is organized as follows: Section II
explains the simulation method that was used to study RECAPs,
followed by Section III that defines MIMO capacity with power standards. Thus, switching at the beginning of a transmission
constraints. Section IV studies the MIMO channel capacity frame should incur negligible overhead. On the other hand,
using RECAP antennas in comparison with non-RECAP arrays, if tunable MEMS devices are used, switching times could be
and Section V concludes the paper. higher, requiring longer block lengths to mitigate overhead as
explained in [13]. Although a detailed analysis of switching
II. ANALYSIS OF PARASITIC RECAP STRUCTURE
time for dynamic adaption in time-varying channels is beyond
The structure considered in this study is depicted in Fig. 1(a), the scope of this paper, this remains an important consideration
which is a 9 9 dipole array consisting of -oriented half-wave for practical systems. Thus, this present work demonstrates
dipoles constrained to an area of in the plane. Each the benefit of the RECAP concept for MIMO systems, and
dipole is either an active “feed” (connected to a transmit or a re- practical aspects are to be treated in future work.
ceive chain) or terminated with a reconfigurable element (RE),
each of which has 8 possible reconfigurable states (RSs) to en- A. Efficient Simulation of RECAP Structure
sure sufficient control over the aperture [15]. The top view of the Efficient simulation of the RECAP is accomplished by com-
structure is shown in Fig. 1(b), where REs and feeds are indi- bining full-wave simulation of the array with network analysis
cated by squares and circles, respectively. In this work, we con- for RE loading. The structure is first analyzed using the Nu-
sider propagation in the plane, where the two-dimensional merical Electromagnetics Code (NEC), which yields the ad-
array can generate patterns with both endfire and broadside char- mittance matrix and short circuit embedded radiation pat-
acteristics. terns of the arrays, where gives the radiation
REs are assumed to be variable capacitances, such that the re- into polarization for unit voltage excitation on port , when
flection coefficient presented at the th port is , where the other ports are short-circuited. S-parameter analysis is more
. We have assumed that is uniformly dis- convenient for this study, where the S-parameter matrix and
tributed on as presented in [15]. Although planar matched ( -terminated) patterns are computed as
RECAPs with realistic RE biasing are arguably more practical,
this study employs this simple structure, since it can be simu- (1)
lated efficiently and its performance is not constrained by prac-
tical limitations of existing switch technologies, biasing, sub- and
strate losses, etc.
Time required for RE switching for a dynamic reconfigurable (2)
antenna was treated in [13] for MEMS switches, indicating that
this overhead can significantly impact system throughput. For respectively, where is the real scalar normalizing impedance,
our analyzed structure employing variable capacitances (i.e. and is the identity matrix. Note that the matched circuit pat-
varactor diodes), the switching time is expected to be a few terns in (2) are embedded patterns, where the entry
ns, which is on the order of a symbol for existing modulation gives the pattern radiated into polarization when port is
MEHMOOD AND WALLACE: MIMO CAPACITY ENHANCEMENT USING PARASITIC (RECAPs) 667

Fig. 2. MIMO system model and noise matching for LNAs. (a) System model; (b) LNA matching.

driven with a unit incident wave and the other ports are termi- source voltage with internal impedance . The in-
nated in loads . cident traveling waves on the feed ports are simply
Next, network analysis is used to find radiation patterns and , and radiated far fields are given by (7). The RE-ter-
the input reflection matrix of the feeds for arbitrary RE termina- minated receive array is modeled in a similar manner, expect
tion. Defining and as the vectors of incident and that due to external incident field, a source wave term must
reflected waves on the feed ports and and as the corre- be included, such that . Assuming a plane
sponding vectors on the RE ports wave arriving at angle and reciprocity

(3)
(8)

where has been partitioned according to feeds and REs. Ter- where gives the polarization and complex amplitude of the
minating RE ports with loads having reflection matrix , we plane wave.
have , where is a diagonal matrix with A multi-path model is assumed consisting of clusters and
. Combined with (3) paths (or rays) within the th cluster, where the th path in
the th cluster has angle of departure , angle of arrival
(4) , complex amplitude , and time of arrival , or
(5) , where is frequency.
Although depolarization of the paths is not considered in this
work, this could be included by making a matrix. Superim-
where is the input reflection coefficient matrix looking posing the waves due to all paths
into the feed ports for the given termination at the REs. In this
work, we choose to be closely matched to a single
half-wave dipole. The field radiation pattern of the RECAP with
RE termination is

(6) (9)

C. Noise Modeling
where and represent matched patterns cor- In order to consider a realistic system, where noise from the
responding to feeds and REs respectively. Substituting (4) into low-noise amplifier (LNA) at the receiver depends on the feed
(6) yields reflection, we employ an LNA model equivalent to [14], where
equivalent forward and reverse traveling noise waves
at the LNA input are needed to properly model real transistors.
The receiver consists of a matching network, forward and re-
(7) verse noise sources, and LNA as shown in Fig. 2(a). The mul-
where represents matched patterns of feed ports with tiport LNA is assumed to consist of multiple uncoupled LNAs
RE port termination . with optimal reflection coefficient , normalized equivalent
noise resistance , and minimum noise figure , available
B. System Model
from a standard LNA data sheet.
Next, we consider using RECAPs at transmitter (Tx) and re- Straightforward analysis at the connection of array and
ceiver (Rx) to form a complete system, as depicted in Fig. 2(a). matching network reveals
Note that unprimed and primed RECAP quantities denote those
at Tx and Rx, respectively. At Tx the th feed is connected to (10)
668 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

where where , is the equivalent-noise resistance,

is the noise voltage covariance, , is
(11) Bolzmann’s constant, is reference temperature, and is
bandwidth. Since is the same for the reference and RECAP
(12)
systems and SNR of the reference system is fixed, has no
effect on capacity and is set to 1.
The incident wave into the amplifier is found to be,
Note that since a data sheet typically assumes ,
which is different from the value used in our analysis, the trans-
(13)
formation

In this work, a fixed (non-RECAP) array is used as a refer- (21)

ence case for gauging performance improvement, and an iden-
tical uncoupled matching network is assumed on each of the is required, where and are optimal reflection and ref-
receive feeds such that for the reference, where erence impedance from the data sheet.
is the source reflection coefficient that provides optimal Summarizing, the MIMO input-output relationship is given
noise performance. The same uncoupled matching network is by (16), where RE-dependent noise covariance is computed
employed for the RECAP, and noise coupling from one feed to from (17)–(20), where parameters , , and are avail-
the next and deviation of can lead to reduced SNR. able from a standard LNA specification. Although computation
Fig. 2(b) shows the uncoupled matching arrangement on each of the noise covariance in this way seems cumbersome com-
branch that transforms to using pared to MIMO analyses that directly specify , the added
a reciprocal lossless matching network, such that , complexity is necessary to capture noise coupling of the active
where is a 2 2 matrix. The required conditions are satisfied ports and variable input impedance of the receive array, which
with both affect capacity when thermal noise is significant compared
to interference.
(14)
(15) D. LNA Specification
This analysis uses the MAXIM MAX2656 LNA, having
is a 2 2 block matrix, where the th block is equal to , noise-equivalent resistance ,
from (14) and (15), and is an identity matrix. and optimum reflection coefficient (at
Note that the fixed matching network can also be lumped into 1960 MHz and ) [17]. Since the specific LNA may
the LNA to form the effective LNA shown in Fig. 2(b), with affect the simulations and conclusions, we briefly consider the
new optimal reflection coefficient and equivalent noise re- impact of the LNA choice.
sistance . Analysis of the RECAP mainly depends on how sensitive the
Plugging (9) into (12) and the result into (13) yields noise figure of the LNA is to the reflection coefficient presented
by the RECAP. Noise figure for uncoupled LNAs is [16]

(22)

where is the reflection coefficient looking into the output of

one of the matching networks. Assuming a lossless matching
(16) network shown in Fig. 2(b) that transforms the source reflection
to of the LNA, it can be shown
where is the channel matrix, is from (9), is noise, that for ,
and and are input and output signals. The linear term
applied to both signal and noise does not change capacity and is
omitted. The noise covariance is (23)

(17)

where , , and are [16] where is the the equivalent LNA noise resistance referenced
back to the input of the matching network where .
(18) Fig. 3 shows noise figure degradation in dB for dif-
ferent values of and using (23), where .
(19) As increases, the penalty of mismatch can increase dramati-
cally. However, our chosen amplifier (as indicated in the figure)
has moderate sensitivity to mismatch, making it a good candi-
(20)
date for this initial study. Amplifiers with much higher would
MEHMOOD AND WALLACE: MIMO CAPACITY ENHANCEMENT USING PARASITIC (RECAPs) 669

Although RECAPs that can adapt to each instantaneous value

of , and are optimal, this rate of adaptation may be
unrealistic for practical implementation. Thus we also consider
a system that adapts average RECAP performance when the
and are fixed, but only the are random. We refer to the
former and latter cases as instantaneous and average RECAP
optimization, where average performance is computed in the
latter case using 10 realizations of .

F. Genetic Algorithm
Due to the large number of RE combinations, obtaining the
optimal solution with an exhaustive search is not feasible, and a
genetic algorithm (GA) is employed. The GA employed in this
work is basically equivalent to that described in [15], except
that REs at both transmit and receive end are jointly optimized
to maximize the capacity.
Fig. 3. Effects of and on .
Note that the purpose of using genetic algorithms in this work
is only to explore the peak potential of RECAP-enabled MIMO,
simply increase mismatch penalty, which would more strongly since such algorithms may be too expensive for in-situ optimiza-
constrain the set of useful RECAP states, possibly resulting in tion due to the extensive training overhead and computation
reduced capacity. time required. The development of direct RECAP optimization
methods more suitable for real-time optimization is a long-term
E. MIMO Channel Modeling goal of this research, to be treated in later work.
The channel matrix is given by (16), where is found
according to the path-based model in (9). In this work we con- III. MIMO CAPACITY WITH CONSTRAINTS
sider models where attention is restricted to the azimuthal plane For the analysis of MIMO channel capacity, we have used the
( and are 90 ). Two propagation models are assumed for the RECAP structure explained in Section II, for both Tx and Rx,
paths between Tx and Rx: forming a MIMO system. In order to properly scale power and
1) Uniform Model: In this simple model, a single cluster is assess RECAP capacity gain, a reference non-RECAP antenna
assumed with rays having arrival times array is considered, having the same number of feeds and area
. Angles of arrival and departure are uniformly as the RECAP, but consisting of matched dipoles. Although an-
distributed on and has a unit variance complex tennas were placed as far apart as possible for the reference case,
normal distribution. some initial experiments were required to find the best place-
2) SVA Model: The more realistic Saleh-Valenzuela angular ment of feeds for the RECAP to give peak capacity. Having
(SVA) model [18] assumes clusters, where the arrival of the feeds too close to the aperture center or edge reduced the ca-
th cluster has the conditional pdf pacity of the RECAP, and a balanced arrangement gave the best
performance.
Channel capacity is computed from
(24)
where is the arrival rate of the clusters. Relative arrival time (27)
of the th ray within the th cluster has the pdf
and for equal power allocation , where is
the total Tx power, and is the number of Tx feeds. Lumping
(25) noise covariance into the channel matrix results in
where is the arrival rate of rays.
The complex amplitude of the th ray in the th cluster (28)
is complex gaussian, where the variance decays exponentially
with arrival time according to where is computed using (17). Plugging (28) into (27)

(26) (29)
and and are the cluster and ray decay time constant, re-
which can be interpreted as the capacity of an effective channel
spectively. The azimuthal angle of the th cluster at transmit
for i.i.d. noise (unit variance) and transmit power .
and receive is and , respectively, which are uniformly
A convenient way of enforcing the different power con-
distributed on . The relative transmit and receive an-
straints in this study is to first define the average single-input
gles of the th ray in the th cluster are and
single-output (SISO) gain of a given system as
, which follow a double-sided Laplacian distri-
bution with pdf , where is
(30)
the angular spread.
670 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
where is Frobenius norm and is the number of feeds at
Rx, which indicates the average power gain provided by channel
matrix with respect to all active ports. The desired SNR of a
system can then be fixed by normalizing that system by its SISO
gain according to and setting transmit
power as , resulting in the equivalent ca-
pacity expression
(31)
where represents the normalized quantity. Below we explain
how this normalization can be used to implement power con-
straints for three realistic cases.
Fixed SNR Constraint (Case 1): The total amount of trans-
mitted/collected power is the same for both the non-RECAP and
RECAP structures. This constraint ensures that the RECAP can
only increase capacity by improving channel orthogonality or
conditioning. For Case 1, channel matrices corresponding to a
non-RECAP reference (REF) and the RECAP are normalized
as
(32)
(33)
Normalizing each system individually by its own SISO gain
forces the RECAP and reference case to both have average SISO
SNR when computing capacity with (31).
Max EIRP Constraint (Case 2): Here we constrain the
EIRP of the RECAP to be no larger than that of the reference
(non-RECAP) system. This is accomplished by setting transmit
power such that a prescribed SNR is obtained for the
reference system, and this same transmit power is also used for
the RECAP system. Maximum EIRP of the RECAP system is
then limited to be equal to or lower than that of the reference
system by scaling the embedded RECAP radiation patterns
according to
(34)
where refers to the radiation pattern of the th feed at
Tx, refers to the one corresponding to the th feed of
the reference (non-RECAP) Tx, and is used in place of
when computing in (9) when is less than or
equal to 1 (i.e. when a RECAP feed provides higher maximum
gain than a non-RECAP feed). The non-RECAP and RECAP
channels are normalized respectively with (32) and
(35)
Note that although the advantage of transmit beamforming by
the RECAP is removed due to the maximum EIRP normaliza-
tion, both channels are normalized by the SISO gain of the ref-
erence system, preserving possible enhanced power collection
with receive RECAP beamforming.
Fig. 4. Instantaneous and average MIMO channel capacity for a simple non-
RECAP array (dashed lines) and RECAP (solid lines): (a) 2 2 MIMO with
, (b) 4 4 MIMO with .
Average Transmit Power Constraint (Case 3): In this case,
only average transmit power is constrained such that a pre-
scribed SNR is obtained for the reference system, and no
constraint is placed on directional gain of Tx or Rx antennas.
Specifically, channel matrix normalization is done using (32)
and (35), allowing the RECAP to obtain a power advantage
through both transmit and receive beamforming.
IV. MIMO CAPACITY ANALYSIS
Since the advantage of RECAP may depend on the number
of antennas, we consider both 2 2 and 4 4 MIMO systems.
Fig. 1 shows the top view of transmit/receive antennas for the
RECAP and non-RECAP structures for the analyzed 2 2 and
4 4 MIMO systems. Results are for the uniform path-based
model and reference SNR , unless otherwise noted.
A. Single User MIMO Capacity
Channel capacity for single user communication is computed
using (31) where is computed using the cases in Section III.
1) 2 2 MIMO System: Fig. 4(a) plots the capacity for the
RECAP and non-RECAP structures for average and instanta-
neous optimization. Note that capacity for the non-RECAP does
not change with constraint type, since the reference has con-
stant SNR, and the slight difference with respect to optimization
type is due to different Monte Carlo realizations. For fixed SNR
(Case 1), RECAP capacity is only marginally better than that
of the non-RECAP, indicating that two channels of sufficient
quality are obtained without reconfigurability, and the RECAP
cannot significantly improve this.
The main advantage of RECAP is power, with significant im-
provements seen when moving to the EIRP constraint (Case 2)
and the transmit power constraint (Case 3). It is also apparent
that average optimization is only slightly worse than instan-
taneous optimization, which is reasonable for power enhance-
ment, since multi-path directions are mainly important, not the
phases of signals sent in those directions.
MEHMOOD AND WALLACE: MIMO CAPACITY ENHANCEMENT USING PARASITIC (RECAPs) 671

2) 4 4 MIMO System: Fig. 4(b) shows that RECAP perfor-

mance is more flat with respect to the constraint type for the 4
4 MIMO system, suggesting that power advantage is less impor-
tant and more opportunity for improving channel conditioning
exists. Also a more significant gap is seen between average and
instantaneous optimization, indicating not only mutli-path di-
rections but also phases are important to attain peak capacity.
Finally it is interesting that the transmit power constraint for the
2 2 RECAP system gives almost the same performance as the
4 4 system with the EIRP constraint.

B. Single User MIMO Under Interference Constraint

Most practical systems for personal wireless communications
are interference limited, and we therefore study the effect of
interference on single-link capacity in detail. In order to model
the effect of interference, we extend to be the covariance
matrix of noise and interference, or Fig. 5. Channel capacity for single user 2 2 MIMO with fixed interference
for a simple non-RECAP array (dashed lines) and RECAP (solid lines): (a) uni-
form multi-path , (b) SVA model.
(36)

where represents the channel matrix between the interferer

and receiving antennas, other quantities are for interferer, anal-
ogous to those at Tx, and we assume . Plugging (36)
into (27) and simplifying yields

(37)
where is interference-to-noise ratio. Since
depends on proximity of the interferer, values of between
0 to 20 dB are considered. We assume that the interfering node
employs a non-RECAP structure and is computed as

(38)
1) 2 2 MIMO System: Fig. 5(a) plots the capacity of a
2 2 system with multi-path for both non-RECAP and
RECAP. The case for , is similar to no interference. As
increases, non-RECAP capacity steadily drops towards zero,
since for interference with rank , it is not possible for
the non-RECAP to null the effect. Although capacity for both
the non-RECAP and RECAP is falling with increasing , closer
inspection reveals that the capacity gain of using the RECAP
over the non-RECAP actually increases with increasing . Per-
formance degradation is much smaller for the RECAP since REs
can be used to null interference, suggesting the possibility of ag-
gressive spectral reuse.
2) 4 4 MIMO System: The results for varying are shown
in Fig. 6(a) for . The overall effect is same as that of 2
2 system, but the curves are flatter with respect to the power
constraint, indicating that power advantage is less important for
more feeds for fixed interference as well. However, improve-
ment relative to the non-RECAP is still very significant, espe-
cially for severe interference.
Fig. 6. Channel capacity for single user 4 4 MIMO with fixed interference
for a simple non-RECAP array (dashed lines) and RECAP (solid lines): (a) uni-
form multi-path with , (b) SVA model.
C. Multi-User MIMO
Building on the idea of employing aggressive spectral reuse,
we next consider a true multi-user scenario where users opti-
mize their RECAPs to maximize sum capacity. This is different
from the case of fixed interference, since the role of the transmit
RECAP becomes more important to reduce interference to the
other user. Although for fixed interference, the user does not
have control over interference, he also does not care about how
much interference he causes. For the multi-user case, interfer-
ence can be controlled but users also impact each other. Ca-
pacity degradation due to interference depends on proximity,
and between 0 and 20 dB is again considered.
Two links are considered, where each receiving user expe-
riences interference from the transmitter of the other link and a
joint optimization is done in order to maximize the sum capacity
672 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 7. Channel capacity for two-user 2 2 MIMO for a simple non-RECAP
array (dashed lines) and RECAP (solid lines): (a) uniform mutli-path with
, (b) SVA model.
Fig. 8. Channel capacity for two-user 4 4 MIMO for a simple non-RECAP
array (dashed lines) and RECAP (solid lines): (a) uniform mutli-path with
, (b) SVA model.
of both links. Individual capacity of each link is calculated using
(37), except now in (38), the Tx RECAP patterns of the other
link are employed instead of for a non-RECAP.
Fig. 7(a) shows MIMO channel capacity per user for the 2 2
multi-user system with increasing , exhibiting similar RECAP
capacity gain as the fixed interference case. The relative gain in
moving from Case 1 to Case 2 is higher for multi-user compared
the single user case with fixed interference, likely due to the fact
that Tx RECAPs can now be controlled to avoid interference.
Fig. 8(a) shows the results for the 4 4 MIMO system. By in-
creasing , the improvement with respect to constraint type be-
comes even flatter than the 4 4 case for fixed interference. Sur-
prisingly, RECAP capacity per user is now lower than that for
fixed interference, indicating that jointly suppressing incoming
interference and avoiding outgoing interference becomes more
difficult for more active feeds.
Fig. 9. Effect of losses for a simple non-RECAP array (dashed lines) and
RECAP (solid lines): (a) 2 2 MIMO with uniform multi-path and , (b)
4 4 MIMO with uniform multi-path and .
D. SVA Propagation Model
The simple path-based model is convenient, but possibly
over-simplistic to represent true propagation scenarios, and
hence we also consider the Saleh Valenzuela Angular (SVA)
model [18]. The parameters of the model are assumed to be
, , , and
, taken from [18]. The model makes use of a threshold
value after which it stops looking for multi-path, which we have
assumed to be 10 dB, generating 50 multi-path on average.
Results for SVA channel simulations have been plotted
next to the respective plots for the simple path-based model
in Figs. 5–8. There is not a dramatic impact compared to the
simple channel model. For the non-RECAP case, capacity is
slightly more degraded in some results with increasing due to
more paths. For the RECAP case the curve trends are similar,
but curves are shifted up slightly in some results showing
that RECAP is more advantageous with increased multi-path.
However, in general changes are only marginal.
E. Effects of Losses and Bandwidth on Channel Capacity
Bandwidth limitations and RE loss are important considera-
tions in practical RECAP structures, and in this section we study
these two effects. First we consider loss by including a series re-
sistance with each RE ranging from 0–10 . Fig. 9(a) shows the
results corresponding to the 2 2 single user system without
any interference. There is no impact of loss for Case 1 since
power differences are removed. Moving to Cases 2 and 3, the
impact of loss becomes increasingly prominent, resulting from
reduced gain of the RECAP, which decreases the channel ca-
pacity. More performance loss is observed for the 4 4 MIMO
system as shown in Fig. 9(b).
Another important aspect is finite bandwidth, and in order to
study its effect a two sided bandwidth of 20 MHz is assumed
at a center frequency of 3 GHz. Capacity is computed as the
average capacity at the center frequency and two band edges
for a single fixed RECAP structure. Channels are normalized as
MEHMOOD AND WALLACE: MIMO CAPACITY ENHANCEMENT USING PARASITIC (RECAPs)
Fig. 10. RECAP capacity for 20 MHz bandwidth and the SVA model for a
simple non-RECAP array (dashed lines) and RECAP (solid lines): (a) 2 2,
(b) 4 4.
before, except now the largest (worst case) normalization factor
of the three frequencies is used.
Fig. 10(a) indicates that finite bandwidth results in a small
capacity reduction for the 2 2 non-RECAP. Although RECAP
performance is minimally impacted in Case 1, for Cases 2 and 3
some reduction is seen, comparable to the difference of average
versus instantaneous optimization. Results for the 4 4 system
in Fig. 10(b) are similar with a slightly larger gap between single
frequency and finite bandwidth curves.
V. CONCLUSION
This work has analyzed MIMO capacity improvements pos-
sible with a simple RECAP structure for different scenarios and
system power constraints. The results indicate that very large
gains relative to fixed antenna MIMO systems are possible, es-
pecially for interference-limited and multi-user environments,
suggesting that RECAPs may enable aggressive spectral reuse.
Consideration of finite bandwidth and losses has indicated that
RECAPs can also provide most of this capacity improvement
even with these practical impairments.
REFERENCES
[1] E. Telatar, “Capacity of multi-antenna Gaussian channels,” Eur. Trans.
Telecommun., vol. 10, no. 6, pp. 585–595, 1999.
[2] R. S. Blum, “MIMO capacity with interference,” IEEE J. Sel. Areas
Commun., vol. 21, no. 5, pp. 793–801, 2003.
[3] M. A. Jensen and J. W. Wallace, “Capacity of the continuous-space
electromagnetic channel,” IEEE Trans. Antennas Propag., vol. 56, no.
2, pp. 524–531, 2008.
[4] A. Alexiou and M. Haardt, “Smart antenna technologies for future
wireless systems: Trends and challenges,” IEEE Commun. Mag., vol.
42, no. 9, pp. 90–97, 2004.
[5] L. N. Pringle, P. H. Harms, S. P. Blalock, G. N. Kiesel, E. J. Kuster, P.
G. Friederich, R. J. Prado, J. M. Morris, and G. S. Smith, “A reconfig-
urable aperture antenna based on switched links between electrically
small metallic patches,” IEEE Trans. Antennas Propag., vol. 52, no. 6,
pp. 1434–1445, 2004.
[6] J. H. Schaffner, R. Y. Loo, D. F. Sievenpiper, F. A. Dolezal, G. L.
Tangonan, J. S. Colburn, J. J. Lynch, J. J. Lee, S. W. Livingston, R. J.
Broas, and M. Wu, “Reconfigurable aperture antennas using RF MEMS
switches for multi-octave tunability and beam steering,” in Proc. IEEE
Antennas Propag. Society Int. Symp, 2000, vol. 1, pp. 321–324.
673
[7] A. F. Molisch, M. Z. Win, Y.-S. Choi, and J. H. Winters, “Capacity
of MIMO systems with antenna selection,” IEEE Trans. Wireless
Commun., vol. 4, no. 4, pp. 1759–1772, 2005.
[8] M. D. Migliore, D. Pinchera, and F. Schettino, “Improving channel
capacity using adaptive MIMO antennas,” IEEE Trans. Antennas
Propag., vol. 54, no. 11, pp. 3481–3489, 2006.
[9] C. Waldschmidt and W. Wiesbeck, “Compact wide-band multimode
antennas for MIMO and diversity,” IEEE Trans. Antennas Propag., vol.
52, no. 8, pp. 1963–1969, 2004.
[10] B. A. Cetiner, E. Akay, E. Sengul, and E. Ayanoglu, “A MIMO system
with multifunctional reconfigurable antennas,” IEEE Antennas Wire-
less Propag. Lett., vol. 5, no. 1, pp. 463–466, 2006.
[11] D. Piazza, N. J. Kirsch, A. Forenza, R. W. Heath, and K. R. Dandekar,
“Design and evaluation of a reconfigurable antenna array for MIMO
systems,” IEEE Trans. Antennas Propag., vol. 56, no. 3, pp. 869–881,
2008.
[12] A. Grau, H. Jafarkhani, and F. De Flaviis, “A reconfigurable multiple-
input multiple-output communication system,” IEEE Trans. Wireless
Commun., vol. 7, no. 5, pp. 1719–1733, 2008.
[13] F. Fazel, A. Grau, H. Jafarkhani, and F. Flaviis, “Space-time-state
block coded MIMO communication systems using reconfigurable
antennas,” IEEE Trans. Wireless Commun., vol. 8, no. 12, pp.
6019–6029, 2009.
[14] M. L. Morris and M. A. Jensen, “Network model for MIMO systems
with coupled antennas and noisy amplifiers,” IEEE Trans. Antennas
Propag., vol. 53, no. 1, pp. 545–552, 2005.
[15] R. Mehmood and J. W. Wallace, “Diminishing returns with increasing
complexity in reconfigurable aperture antennas,” IEEE Antennas Wire-
less Propag. Lett., vol. 9, pp. 299–302, 2010.
[16] G. Gonzalez, Microwave Transistor Amplifiers. Englewood Cliffs,
NJ: Prentice-Hall, 1997.
[17] MAXIM, July 2004, “Low-Noise Amplifier (LNA) Matching Tech-
niques for Optimizing Noise Figures,” MAXIM [Online]. Available:
http://pdfserv.maxim-ic.com/en/an/AN3169.pdf
[18] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst,
“Modeling the statistical time and angle of arrival characteristics of an
indoor multipath channel,” IEEE J. Sel. Areas Commun., vol. 18, no.
3, pp. 347–360, 2000.
Rashid Mehmood (S’05) received the B.Sc. degree
(cum laude) in communication systems engineering
from the Institute of Space Technology (IST),
Pakistan, in 2007 and the M.Sc. degree in electrical
engineering from Jacobs University Bremen (JUB),
Bremen, Germany, in 2010.
From 2007 to 2008, he worked as a Research As-
sociate at IST and supervised various undergraduate
laboratories. From 2008 to 2010, he worked as a Re-
search Assistant in several laboratories at JUB and
external companies. His current research interests in-
clude reconfigurable aperture antennas, antenna optimization and wireless and
optical communications.
Mr. Mehmood was a recipient of the 2009 IEEE AP-S Undergraduate Re-
search Award.
Jon W. Wallace (S’99–M’03) received the B.S.
(summa cum laude) and Ph.D. degrees in electrical
engineering from Brigham Young University (BYU),
Provo, UT, in 1997 and 2002, respectively.
He received the National Science Foundation
Graduate Fellowship in 1998 and worked as a
Graduate Research Assistant at BYU until 2002.
From 2002 to 2003, he was with the Mobile
Communications Group, Vienna University of
Technology, Vienna, Austria. From 2003 to 2006,
he was a Research Associate with the BYU Wireless
Communications Laboratory. Since 2006, he has been an Assistant Professor
of electrical engineering at Jacobs University, Bremen, Germany. His current
research interests include MIMO wireless systems, physical-layer security,
cognitive radio and UWB systems.
Dr. Wallace is serving as an Associate Editor of the IEEE TRANSACTIONS ON
ANTENNAS AND PROPAGATION and is a Co-Guest Editor of the Special Issue on
Multiple-Input Multiple-Output (MIMO) Technology.
674 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Eigen-Coherence and Link Performance of
Closed-Loop 4G Wireless in Measured
Outdoor MIMO Channels
Matthew Webb, Mythri Hunukumbure, Member, IEEE, and Mark Beach, Member, IEEE
Abstract—Employing feedback in multiple-input-mul-
tiple-output (MIMO) systems offers the potential to significantly
increase data rates over open-loop schemes. However, the impact
the propagation channel has on the behavior and performance
of the scheme must be considered and, where appropriate, the
signaling designed to suit. In this paper, measured outdoor MIMO
propagation data is used to evaluate initially the distributions and
statistics of the coherence of the eigenmodes of the channel, for
a wide range of mobility scenarios and user devices. Then, the
impact on link performance of the differing coherences in time
and frequency is evaluated in terms of the feedback delay and
spectral spacing of pilots in a practical feedback scheme used in
mobile WiMAX. It is found that not only mobility and physical
environment, but also the directivity of the antenna array affect
the eigen-coherence and also, therefore, the performance of a
feedback scheme. Bit-error rate and information capacity are
both shown to be sensitive to increased feedback delay and pilot
spacing, however, the closed-loop system always outperforms its
open-loop counterpart.
Index Terms—Coherence, eigenmode, feedback, MIMO, pilots.
I. INTRODUCTION
T HE provision of high-throughput, high-mobility wireless
communications over city-size environments offers the
possibility of radically improving the user experience and op-
erator business opportunities in such situations. The benefits of
multiple-input-multiple-output (MIMO) systems in such situa-
tions have long been appreciated in the literature [1], [2]. To
reach towards the higher end of MIMO performance, it is nec-
essary to include feedback which provides channel state infor-
mation (CSI) to the transmitting end of the link, a point well-es-
tablished in MIMO information theory [3], [4].
In any feedback scheme, there is inevitably a delay between
estimating the channel and feeding back the CSI. During this
time, the channel will naturally evolve and mobile users may
Manuscript received May 28, 2010; revised September 13, 2011; accepted
September 27, 2011. Date of publication October 25, 2011; date of current ver-
sion February 03, 2012. This work was supported in part by the UK Mobile
VCE ‘MIMO Propagation’ Elective. Parts of this work were presented at the
IET EuCAP, 2007.
M. Webb was with the Centre for Communications Research, University of
Bristol, Bristol BS8 1UB, U.K. He is now with Fujitsu Laboratories of Europe
Ltd., Hayes UB4 8FE, U.K. (e-mail: Matthew.Webb@uk.fujitsu.com).
M. Hunukumbure is with Fujitsu Laboratories of Europe Ltd., Hayes UB4
8FE, U.K. (e-mail: Mythri.Hunukumbure@uk.fujitsu.com).
M. Beach is with the Centre for Communications Research, University of
Bristol, Bristol BS8 1UB, U.K. (e-mail: M.A.Beach@bristol.ac.uk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173431
0018-926X/$26.00 © 2011 IEEE
move, and this will tend to cause performance loss. Particularly
relevant studies of this have been conducted in [5]–[7]. Fur-
thermore, in wideband multicarrier systems popular in emergent
wireless standards, such as mobile WiMAX [9] and 3GPP LTE
[10], spectral pilots are used to allow the same CSI feedback to
be used on a number of subcarriers even though they do not have
identical channel responses. Given these two limitations, perfor-
mance of a feedback-based algorithm depends on the channel’s
coherence time [11], [12] and coherence bandwidth [13]. Con-
ditions for feedback to be beneficial are derived in [14] in terms
of the coherences of the channel, and the throughput of MIMO
spatial multiplexing (SM) schemes is considered in terms of
the coherence of an eigenmode in [15]. Thus, the nature of the
propagation channel, particularly its coherence time and band-
width statistics, is critically important to the system’s correct
functioning, as are reductions in performance as these channel
parameters change.
This paper presents results from an extensive wideband out-
door MIMO propagation measurement campaign in an urban
cellular scenario to tackle this joint question. While there are
numerous studies of the coherence properties of wireless chan-
nels, for example in [16]–[18] for coherence time, in [19]–[21]
for coherence bandwidth, and in [22], [23] jointly, in this paper
the focus is the coherence of the eigenmodes—the ‘eigen-coher-
ence’ (the same terminology has a different definition in [15]).
This spatial structure of the channel holds the key to the SM ca-
pability of such systems. Having considered the distribution and
implications of the eigen-coherence, the time/frequency struc-
ture of the propagation data is used to allow the simulation of a
pilot-based CSI feedback scheme and thus the loss for incorpo-
rating feedback delay and increasing spectral pilot spacing. The
analysis in this paper offers four main contributions.
• The distribution and statistics of the eigen-coherence times
and bandwidths of a measured outdoor MIMO propagation
channel, investigated for users who are standing, walking
or driving, and carrying a personal digital assistant (PDA),
a laptop or non-device antenna array.
• The implications of different antenna arrays on user de-
vices in terms of their effect on perceived eigen-coherence
and thus feedback load.
• The impact on bit-error rate (BER) and information ca-
pacity of feedback delay and spectral pilot spacing in a
closed-loop spatial multiplexing MIMO system.
• A demonstration that, in the environment measured, feed-
back is always useful, even in the presence of considerable
delay or spectral offset.
The setting for this is codebook-based quantization of the
channel matrix—a popular means of providing CSI feedback
WEBB et al.: EIGEN-COHERENCE AND LINK PERFORMANCE OF CLOSED-LOOP 4G WIRELESS IN MEASURED OUTDOOR MIMO CHANNELS 675

with manageable computational complexity and limited feed-

back load. Such codebook schemes are prevalent across many
wireless standards: the one used here will be that specified for
IEEE 802.16e ‘mobile WiMAX’ [8], but versions are also in-
cluded in 3GPP LTE [10], IEEE 802.16m [9], and 3GPP2-UMB
[24]. The scheme used in [8] is designed to approximate singular
value decomposition (SVD) feedback [25] and the performance
Fig. 1. The three devices used in measurements: laptop, PDA, and dipoles.
is thus tightly related to the behavior and coherence properties
of the eigenmodes. Previous work on the performance of mo-
bile WiMAX includes [26] which covers some aspects of this TABLE I
codebook scheme, but in an 802.16m setting focusing on the SUMMARY OF MEASUREMENT DEVICE ANTENNA CHARACTERISTICS
throughput of various feedback methods; [27] which considers
many different physical-layer aspects, but not codebook feed-
back; and [28] which presents codebook design for system con-
figurations not supported in [8].
Section II summarizes the measurement campaign and data,
and Section III describes its use to study the eigen-coherence
properties of the channels. Section IV presents the analysis of
BER and capacity changes according to feedback delay and fre- 30 m high five-storey building overlooking the city center where
quency pilot spacing, and Section V summarizes the work. the measurements were conducted. The antennas were fitted to
metal railings, and given 3 m horizontal separation and an 8
II. MIMO PROPAGATION MEASUREMENT CAMPAIGN downtilt to improve coverage.
An extensive outdoor MIMO measurement campaign was
carried out around Bristol city center, U.K. [29], [30], using a B. Data Pre-Processing
multichannel wideband channel sounder [31]. The center fre-
In each standing and walking measurement there were 4096
quency of the measurements was 2 GHz (wavelength,
samples of each frequency finger in the 6.1 s measurement
cm), with a 20 MHz signal bandwidth. A periodic transmit
period. To improve the measured signal-to-noise ratio (SNR),
signal with a repetition period of 6.4 s was carried through
the complex gains comprising 4 consecutive samples (collected
128 discrete frequency fingers with 156.25 kHz spacing. A 4 4
within the channel’s physical coherence time) were averaged,
MIMO configuration was used throughout.
taking advantage of the zero-mean white noise statistics. There
Measurements were conducted at 56 point locations in Bristol
are thus 1024 time samples at each of the 128 frequency fingers,
city-center, U.K., and along 10 vehicular drive routes (see [29]
giving a spacing of approximately for the
for a map). At each point location, 6.1 s of measurements with
walking cases. In the vehicular measurements, there were 7944
each of the three devices described in Section II-A were col-
samples of each frequency finger in an 8.1 s measurement
lected whilst the mobile user was standing facing in each of four
period, with consecutive pairs of measurements averaged. Thus
directions separated by successive approximate 90 rotations,
there were 3972 samples over a distance of approximately 108
followed by the user walking at 1 m/s for 6 s in each of two ap-
m, giving a spacing of . The channels are then
proximately perpendicular directions. The vehicular measure-
normalized in the time-domain such that they have unit average
ments lasted 8.1 s each at an approximately constant speed of
gain over the bandwidth and all transmit-receive links.
30 mph (48 km/h).

A. Prototype Device and Antennas III. COHERENCE OF EIGENMODES

The measurement campaign deployed three different devices
shown in Fig. 1: some head-mounted ‘reference’ dipoles, a per-
sonal digital assistant (PDA), and a laptop. The purpose was
to capture propagation data relevant to possible realistic de-
ployments of MIMO technology on user devices, and to study
the fundamental properties of the channel without the variable
impacts of physical device form-factor and human body shad-
owing. The four head-worn halfwave dipoles were located in
two pairs in orthogonal planes around the cycle helmet, with
the two elements in each pair oriented at to the plane’s
vertical. The linear slots in the PDA were arranged around the
face and edges of the device, and the printed inverted-F an-
tennas (PIFAs) in the laptop on the lid’s two inside edges. Table I
summarizes the devices’ characteristics; values are the averages
across the four individual, almost-identical antennas in each
case.
The base station (BS) antennas were two dual polarized
UMTS antennas (a total of four transmit chains) mounted atop a
Eigen-coherence time, , and bandwidth, , defined pre-
cisely below, are used here to measure how quickly the spatial
modes of a MIMO channel change in their respective domains.
This is essential to a feedback-based SM system, since perfor-
mance is governed by how often feedback is updated compared
to the eigen-coherence in that domain. In that respect, eigen-co-
herence is a parameter of significant interest in such schemes
as compared to classical coherence which deals only with the
overall channel matrix. We have studied relations between clas-
sical quantities and eigen-coherences in [30].
Eigen-coherences are calculated based on the autocovariance
functions (ACFs) of the eigenvalues of the channel. Define
to be the variation of eigenvalue over time and
frequency , and its ACF in the time-domain to be
(1)
676
Fig. 2. Eigen-coherence times of the largest eigenvalue across all locations.
TABLE II
EIGEN-COHERENCE TIME STATISTICS FOR VEHICULAR ROUTES
where is the expectation over and denotes complex con-
jugation. Then is the earliest lag such that , tions, having a wider azimuth view tends to decrease the coher-
and . An equivalent procedure in the frequency
domain is used for . Figures are shown for only the largest
eigenvalue for the sake of clarity.
A. Coherence Time
Fig. 2 shows the cumulative distribution function (cdf) of
for the largest eigenvalue of the channel matrix over all point
locations. Statistics for the vehicular routes are summarized in
Table II since the range of coherence times is too small to depict
in Fig. 2. As expected, coherence times while walking are dis-
tributed at significantly lower values than when standing: the
walking median is 35 ms compared to the standing median at
110—140 ms and the vehicular medians are an order of magni-
tude lower again at around 5 ms.
In the standing measurements, the PDA’s coherence times are
smallest, and the reference dipoles the largest. Since the PDA
antennas are much more directional than the dipole array (see
Table I), this suggests that having a directional antenna in a
scattering, changing environment (but with a stationary user)
tends to estimate coherence time only in the direction(s) it can
see, whereas a wider angular ‘view’ averages out the signals
arriving from low- and high-coherence time directions. Since
most of the locations considered were in a densely scattering
city center, for antennas sensitive to a smaller angular spread
there is a higher probability that a stronger LOS component
is not detected (bringing down the coherence time) versus a
less-directional antenna which is more likely to see both LOS
and NLOS multipaths, and higher average coherence time.
The situation in the walking and vehicular measurements is
different, due to the dominant effect of the user’s motion. With
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 3. Eigen-coherence bandwidths of the largest eigenvalue across all
locations.
TABLE III
EIGEN-COHERENCE BANDWIDTH STATISTICS FOR VEHICULAR ROUTES
the multipath environment changing more quickly in all direc-
ence time and so the dipole array gives the lowest eigen-coher-
ence time distribution and the PDA the highest.
B. Coherence Bandwidth
The distribution of eigen-coherence bandwidth over all point
locations is shown in Fig. 3, where there was found to be
virtually no difference between the curves for standing and
walking measurements. The driving measurements are shown
in Table III. In each case, the median eigen-coherence band-
width is 400 kHz, although the smaller range of values in the
driving case implies a different distribution there, tending to be
to the left of the standing/walking cdfs at the higher percentiles.
Even so, only some 15% of the values for standing/walking
are above the driving laptop’s maximum of 625 kHz, showing
that the distributions are broadly similar. The reference dipoles
show the smallest eigen-coherence bandwidths, by a small
margin. Similarly to in Section III-A, the wide azimuth view
of the head-mounted dipoles means their delay response con-
tains more multipath components making the channel more
frequency selective and thus reducing coherence bandwidth.
IV. FEEDBACK DELAY AND SPECTRAL OFFSET
Two major sources of feedback imprecision are using the
same feedback over several frequency subchannels distinct
from the pilot (referred to here as spectral offset) and the
inevitable calculation and transmission delay on the feedback.
This section presents results showing how they affect the
bit-error rate (BER) and capacity of the SM feedback scheme
in mobile WiMAX. A set of standing, walking and driving
measurements are compared for the impact of feedback delay,
with their changing coherence times. Spectral offset is studied
WEBB et al.: EIGEN-COHERENCE AND LINK PERFORMANCE OF CLOSED-LOOP 4G WIRELESS IN MEASURED OUTDOOR MIMO CHANNELS 677

by comparing an NLOS to a LOS location, as these have
contrasting coherence bandwidths.
The measurements from the reference dipoles mounted on
the user’s head are used throughout this section, eliminating the
‘body shadowing’ in the laptop and PDA measurements.
A. System Model
The closed-loop SM scheme used here is a codebook quan-
tization method from mobile WiMAX [8]. The codebooks de-
fined in [8] approximate the ideal singular value decomposition
(SVD) feedback scheme described in [25]. Define a channel ma-
trix to have SVD , then the precoding needs to
approximate . A codebook of pos-
sible quantizations, or codewords of the right singular matrix of
the channel is specified, among which one, , is chosen
for optimizing a suitable metric. Call the transmitted signal ,
receive combiner , the received signal , and additive white
Gaussian noise (AWGN) , then
(2)
In the case of exact feedback, , and the
channel is diagonalized. With a codebook, , and there
is no diagonalization. To optimize the SNR, is the linear min-
imum mean squared error (MMSE) combiner for
giving a total capacity . The fact that
, along with codebook precoding, results in residual
inter-stream interference. In (6) this is modeled as additional
white i.i.d. Gaussian noise, a worst-case assumption as the inter-
stream interference is in fact correlated. computed this way
is not capacity in the strict sense, being instead a lower bound
on the actual capacity, since the self-interference is not i.i.d.
Gaussian and its specific structure could be used to obtain a
higher mutual information. In (6), a simple receiver without
knowledge of the noise structure is implied.
The SNR for all such analysis will be 20 dB. At each location,
the wideband capacity averaged over frequency at each time
sample will be reported as cdfs.
2) BER: For BER performance, QPSK with a half-rate con-
volutional forward error-correcting code (FEC) with the octal
generator (247, 371) is used. This is a constraint-length 8, max-
imal free-distance code [33]. Enough random binary bits are
generated to allow each sample, subcarrier and stream to carry a
QPSK symbol following half-rate encoding. These bits are ran-
domly interleaved before being coded and modulated. Viterbi
hard decoding is used in the receiver.
B. Feedback Delay

(3) Standing and walking measurements at the same physical lo-

cation will be used in this section, with an additional driving
route. The curves assume that calculation of and is per-
where the total transmit power is and the AWGN has covari-
formed correctly for the channel on the pilot tone and time slot
ance .
, and then actually applied to a later time slot (but still on
Codebooks with or codewords, requiring, respec-
the pilot tone) when the channel has evolved to become
tively, 3 or 6 bits of feedback, are provided in [8] for ,
. The feedback delay increments, , are therefore mul-
or antennas transmitting , or streams between
tiples of the measurement interval of 6.14 ms. For the standing
them, subject to . A full exposition of the scheme is
case, the coherence time was 144 ms and for walking it was
left to [8], with the simulations here taking:
21.6 ms (to 0.7 autocorrelation). To facilitate comparisons, the
1) and throughout, offering a mid-way
ratio in both cases, meaning that the in-
tradeoff between SM and diversity gain. Comparisons to
crement in feedback delay on adjacent BER curves represents
an open-loop system will use the same for consistency,
approximately the same multiple of coherence time. Thus, in
with (arbitrarily) the third antenna disabled, i.e., the pre-
Fig. 4, , so each curve is dis-
coder [100;010;000;001] is statically applied; disabling a
placed by 7 measurement samples relative to its neighbors, and
different antenna alters the results only slightly.
in Fig. 5, .
2) 6-bit codebooks, to outer-bound performance.
1) BER Performance: Fig. 4 shows, for the standing mea-
3) will be selected from to minimize the mean
surement, BER curves for increasing estimation delay. The long
squared error as in [32]
coherence time of 144 ms means that even large feedback de-
lays only degrade performance by around 1 dB for the codebook
(4) system over the range considered. The feedback load induced
by such a channel would therefore be low, as updates could be
This is a likely realistic metric if a system operator is tar- widely spaced in time. By comparison, when walking at about
geting quality-of-service improvement. 1 bbm/s in Fig. 5 the curves spread out significantly, with ac-
1) Capacity: Denote the original channel, on which and ceptable SNR losses of 1.5 dB after a 6 ms delay, but increasing
are calculated as , and the delayed or spectrally-offset quickly, meaning that feedback would have to be much more
channel actually used for transmission as . To simplify no- rapid if performance close to ideal was desired. Fig. 7 shows
tation, re-express (2) as the initial parts of the ACFs. Since the ACFs for walking and
driving are much steeper than for standing, moving an equal
(5) multiple of along the curve will produce greater decorrela-
with . The received SNR on tion and poorer performance. There is an error-floor appearing
stream is at the longer delays in Fig. 5; discussion of this is deferred
to Section IV-C1 where it also appears. The open-loop perfor-
(6) mance is always worse than any of the closed-loop cases and
shows slightly greater SNR loss, and more pronounced onset of
678 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 4. Impact of feedback delay on BER for standing measurements.

. Fig. 6. Impact of feedback delay on BER for driving measurements.
.

Fig. 5. Impact of feedback delay on BER for walking measurements.

. Fig. 7. Autocorrelation functions of the three measurements, with the 0.7 level
indicated.

error-floors with delayed feedback. This implies that most of the
SNR loss with precoding is actually from the spatial equaliza-
tion being unable to suppress inter-stream interference success-
fully; and also that the approximately-diagonalising effect of the
precoder can partially mitigate this problem. Clearly, more so-
phisticated channel estimation schemes would perform better
and could relax these delay constraints.
To capture data appropriately in the driving measurements,
the channel sounder’s sample spacing was reduced to 2.05 ms.
Since ms in this case, ; sig-
nificantly different to that in the earlier figures, but it was not
practically possible to move this ratio toward . Fig. 6
shows the results for the driving measurement (the closed- and
open-loop 2.05 ms delayed curves coincide). The effect of delay
is decisive: the system as modeled here clearly requires signifi-
cantly shorter delays than twice the coherence time; an observa-
tion familiar from the above results. Inferring from the standing/
walking results that equalization failure is the more significant
effect, a vehicular scenario will benefit most from using rapid
channel estimation techniques, or closer-spaced temporal pilots.
2) Capacity Performance: The capacity response (on the
basis of (6)) to feedback delay is shown in Fig. 8, for the vehic-
ular measurement. The capacity loss is about 66% after just one
delay step—even the smallest delay is in this quickly-decor-
relating scenario—but is relatively less severe than the BER re-
sult. The losses diminish after the first delay, since the channel
has already largely decorrelated. The open-loop curves are al-
ways lower than their closed-loop counterparts indicating that,
as with BER, the feedback scheme provides at least some ben-
efit. Comparing the BER and capacity effects suggests that the
principal performance losses will result from BER degradations
leading to a reduced useful throughput as a result of re-trans-
mission (e.g., via hybrid automatic repeat request (HARQ)), but
that the capacity losses are also meaningful if the BER can be
repaired.
These results could be viewed as performance degradations
due to imperfect precoding and equalization, if the rate is per-
fectly adapted with respect to the resulting channel including
imperfect precoding. An alternative view of capacity is the
outage probability. Then, since transmission is always at the
WEBB et al.: EIGEN-COHERENCE AND LINK PERFORMANCE OF CLOSED-LOOP 4G WIRELESS IN MEASURED OUTDOOR MIMO CHANNELS 679

Fig. 9. BER performance with spectral offset for NLOS location.

Fig. 8. Impact of feedback delay on capacity for driving measurements.

capacity limit, if the capacity of the precoded channel with

delay is smaller than the rate implied by the original channel
and precoder there is an outage. We have found that for the
wideband capacity, at all delays in Fig. 8 the channel is always
in capacity outage. Looking at individual subcarriers, 75–77%
of them are in outage (depending on the delay).

C. Spectral Offset
Most configurations in mobile WiMAX have pilots with a
spectral spacing of 2–4 subcarriers [8]. The subcarrier spacing
is 10.84 kHz, so feedback and equalization information is ap-
plied to subcarriers with about 20–40 kHz spectral offset from
a given pilot. This section studies the performance of the pre-
coding scheme as the feedback is applied to subcarriers with in-
creasing spectral offset from the pilot. The channel has become
, with the codeword and equalization from the pilot
channel still in use.
The frequency response of the channel was interpolated to
reduce the spacing of the measured data from 156.25 kHz by
zero-padding the end of the time-domain response. Since in this
region the response is only zero-mean noise, it negligibly al-
ters the temporal and spectral properties of the response, while
giving more frequency points following a fast Fourier trans-
form. In this way, the frequency resolution was increased by a
factor of 8 to 19.5 kHz per finger.
Results are shown for NLOS ( kHz) and LOS
( kHz) walking measurement locations.
1) BER Performance: The BER performance at the NLOS
location is shown in Fig. 9. Curves for both the closed-loop
system using the 6-bit codebook and an open-loop system with
no feedback are shown. The “pilot” curves show the BER if the
feedback information were to be used to transmit data on the
pilot subcarrier itself, i.e., with no spectral offset. For a 19.5 kHz
offset , the loss from the case of transmitting on
the pilot channel is small; much less than 1 dB at all BERs.
The losses are more pronounced for the 2-finger offset curves,
reaching nearly 3 dB at a BER of , and once the offset has
reached about 60 kHz, the losses from the ideal case are substan-
tial. The curves for open-loop transmission show that the SNR
loss caused by a given spectral offset of the receive combiner
Fig. 10. BER performance with spectral offset for LOS location.
is approximately the same as for the precoded system but that
precoding performs better at all the offsets considered here, so
that feedback is always useful.
For the LOS channel, in Fig. 10, BERs are substantially
higher due to the less-scattering conditions resulting in poorer
MIMO channels, but degradations under spectral offset are
much smaller: almost zero with a 19.5 kHz offset ,
corresponding to higher coherence bandwidth implying slower
changes in the channel over frequency. Whereas at 40 kHz
offset in NLOS the loss was 3 dB, here it is less than 1 dB. At
the largest offsets an error-floor appears. In a LOS situation
particularly, BER behavior will be dominated by the weakest
spatial stream(s), all of which have the same modulation and
coding scheme, and increasing SNR also increases the residual,
post-MMSE, inter-stream interference, resulting in an error
floor. A similar effect is analyzed in [34].
These results suggest the 20 kHz spacing used in 802.16e
is suitable in both NLOS and LOS environments in terms of
BER performance, where a pilot carrier is used over a kHz
range. However, the 2.5–3 dB loss in the NLOS environment at
a 40 kHz offset would have to be considered in the system de-
sign, and could call for other techniques to reduce the headline
680 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

that in a LOS environment, the principle system performance

degradation will be from gradual BER losses rather than ca-
pacity losses, but in NLOS, the two will interact much more
strongly, thus limiting the scope for feedback-load reduction in
such cases. Even though system performance will degrade, the
precoded capacity distributions are again always better than the
open-loop curves, so that feedback is always useful.
The outage probability behavior is as with feedback delay:
the wideband channel was found to be always in outage at the
delays considered here. At the individual subcarrier level, this is
also true in NLOS, but in LOS it reduces slightly to 90%, 97%,
and 99% of subcarriers at 1, 2, and 3 subcarrier offsets.

V. SUMMARY AND CONCLUSIONS

This paper has studied the statistics of MIMO eigenmode
coherence time and bandwidth coupled with the impact of feed-
Fig. 11. Capacity performance with spectral offset for NLOS location.
back delay and the use of spectral pilots on the performance
of a class of closed-loop spatial multiplexing scheme widely
adopted in standards. Based on measured data collected in an
urban cellular scenario, distributions of eigen-coherence were
presented encompassed standing, walking and driving users
with a variety of mobile devices. The nature of the antenna
array in use also had a distinct impact on coherence properties
and thus may be a consideration in mobile unit design in
order to avoid higher-than-necessary feedback loads. Then, the
performance of the codebook-based CSI quantization scheme
in mobile WiMAX was evaluated in terms of how quickly the
decoherence of the channel in time and frequency degrades
BER and capacity performance. Although feedback always
provided some benefit, the losses in BER were found usually to
be more significant than in capacity, and much more significant
under NLOS than LOS propagation conditions. This suggests
that future schemes which dynamically respond to the changing
propagation, e.g., by controlling the feedback rate, may fare
better than schemes that are unaware of it.
Fig. 12. Capacity performance with spectral offset for LOS location. REFERENCES
[1] V. Tarokh, N. Seshadri, and A. Calderbank, “Space-time codes for high
data rate wireless communication: Performance criterion and code con-
error-rate, such as stronger FECs. In LOS, wider spaced pilots struction,” IEEE Trans. Inf. Theory, vol. 44, no. 2, pp. 744–765, Mar.
1998.
could be tolerated in such an environment, suggesting that a dy- [2] G. J. Foschini, “Layered space-time architecture for wireless commu-
namically chosen pilot spacing, dependent on prevailing prop- nication in a fading environment when using multi-element antennas,”
agation conditions, could yield benefits. Feedback load would Bell Labs. Tech. J., pp. 41–59, Oct. 1996.
steadily reduce as coherence bandwidth increases which could [3] I. E. Telatar, “Capacity of Multi-Antenna Gaussian Channels,” Bell
Labs, Lucent Technologies, Technical Memorandum, Oct. 1995.
increase the data rate available on the forward link, providing [4] D. J. Love and R. W. Heath, “What is the value of limited feedback for
some compensation for the higher BER it will experience under MIMO channels?,” IEEE Commun. Mag., vol. 40, no. 10, pp. 54–59,
increasingly LOS conditions. Oct. 2004.
[5] A. Khrwat et al., “Feedback delay in precoded spatial multiplexing
2) Capacity Performance: The impact of spectral offset MIMO systems,” in Proc. IEEE Int. Symp. Personal Indoor and Mobile
on capacity as in (6) is shown in Figs. 11 (NLOS) and 12 Radio Commun., Cannes, France, Sep. 2008, pp. 1–4.
(LOS). Capacity losses in the NLOS case rise quickly as [6] P. Zhu, L. Tang, Y. Wang, and X. You, “Quantized beamforming with
spectral offset increases, having reduced the median level by channel prediction,” IEEE Trans. Wireless Commun., vol. 8, no. 11, pp.
5377–5382, Nov. 2009.
50% with a 60 kHz offset. However, the loss is much smaller [7] Y. Isukapalli and B. D. Rao, “Packet error probability of a transmit
at the lower end of the WiMAX spectral offsets: at 20 kHz, beamforming system with imperfect feedback,” IEEE Trans. Signal
it is 3.5 bps/Hz (15%). Comparing to Fig. 8, capacity is less Process., vol. 58, no. 4, pp. 2298–2314, Apr. 2010.
[8] IEEE Standard for Local and Metropolitan Area Networks Part
sensitive to spectral offset in NLOS than to feedback delay. The 16 Amendment 2: Physical and Medium Access Control Layers for
LOS location suffers much less, as in the BER results, because Combined Fixed and Mobile Operation in Licensed Bands and Cor-
of its higher coherence bandwidth giving greater correlation rigendum 1, IEEE Std. 802.16e-2005 and 802.16-2004/COR1-2005,
between successive spectral offsets. The loss is just 2.5 bps/Hz Feb. 2006.
[9] DRAFT Standard for Local and Metropolitan Area Networks Part 16:
(16%) at a 60 kHz spectral offset, and the decremental steps Air Interface for Broadband Wireless Access Systems, IEEE Std. 802.
in capacity are also smaller than in NLOS. These results mean 16Rev2/D6a, Jul. 2008, draft.
WEBB et al.: EIGEN-COHERENCE AND LINK PERFORMANCE OF CLOSED-LOOP 4G WIRELESS IN MEASURED OUTDOOR MIMO CHANNELS
[10] LTE-A; Evolved Universal Terrestrial Radio Access (E-UTRA); Phys-
ical Channels and Modulation, Release 10, 3GPP TS 36.211 V10.2.0,
Jun. 2011.
[11] D. Samardzija and N. Mandayam, “Pilot-assisted estimation of MIMO
fading channel response and achievable data rates,” IEEE Trans. Signal
Process., vol. 51, no. 11, pp. 2882–2890, Nov. 2003.
[12] J. Chena, R. A. Berry, and M. L. Honig, “Performance of limited feed-
back schemes for downlink OFDMA with finite coherence time,” in
Proc. IEEE Int. Symp. Inform. Theory, Nice, France, Jun. 2007, pp.
2751–2755.
[13] M. D. Larsen, A. L. Swindlehurst, and T. Svantesson, “Performance
bounds for MIMO-OFDM channel estimation,” IEEE Trans. Signal
Process., vol. 57, no. 5, pp. 1901–1916, May 2009.
[14] D. J. Love, “Duplex distortion models for limited feedback MIMO
communication,” IEEE Trans. Signal Process., vol. 52, no. 2, pp.
766–773, Feb. 2006.
[15] B. Mielczarek and W. A. Krzymień, “Influence of CSI feedback
delay on capacity of linear multi-user MIMO systems,” in Proc. IEEE
Wireless Commun. and Networking Conf., Hong Kong, Mar. 2007, pp.
1188–1192.
[16] A. M. O. Ribeiro, E. M. M. Barrientos, and E. Conforti, “Spatial corre-
lation function and coherence time characterization of 3.5-GHz micro-
cell propagation,” in Proc. Int. Microwave and Optoelectronics Conf.,
Belém, Brazil, Nov. 2009, pp. 501–505.
[17] A. Sorrentino, G. Ferrara, and M. Migliaccio, “Characterization of
NLOS wireless propagation channels with a proper coherence time
value in a continuous mode stirred reverberating chamber,” in Proc.
Eur. Wireless Technol. Conf., Rome, Italy, Sep. 2009, pp. 168–171.
[18] L. Cheng, B. Henty, F. Bai, and D. D. Stancil, “Doppler spread and
coherence time of rural and highway vehicle-to-vehicle channels at 5.9
GHz,” in Proc. IEEE Global Commun. Conf., New Orleans, LA, Nov.
2008, pp. 1–6.
[19] Q. T. Zhang and S. H. Song, “Exact expression for the coherence band-
width of rayleigh fading channels,” IEEE Trans. Commun., vol. 55, no.
7, pp. 1296–1299, Jul. 2007.
[20] J. Ø. Nielsen et al., “Measurements of indoor 16 32 wideband MIMO
channels at 5.8 GHz,” in Proc. Int. Symp. Spread Spectrum Techniques
and Applications, Sydney, Australia, Aug. 2004, pp. 864–868.
[21] A. M. O. Ribeiro, C. S. Castelli, E. M. M. Barrientos, and E. Conforti,
“Coherence bandwidth in a 1.8-GHz urban mobile radio channel,” in
Proc. Int. Microwave and Optoelectronics Conf., Salvador, Brazil, Oct.
2007, pp. 599–602.
[22] P. Kyritsi, P. C. Eggers, and A. Oprea, “Dual domain coherence mea-
sures for FWA channels in the 5–6 GHz band,” in Proc. IEEE Veh.
Technol. Conf., Fall, Vancouver, BC, Dec. 2002, vol. 1, pp. 111–115.
[23] J. R. Mendes and M. D. Yacoub, “A general bivariate Ricean model and
its statistics,” IEEE Trans. Veh. Technol., vol. 56, no. 2, pp. 404–415,
Mar. 2007.
[24] Physical Layer for Ultra Mobile Broadband (UMB) Air Interface Spec-
ification, 3GPP2 Std. C.S0084-001-0 v3.0, Aug. 2008.
[25] D.-S. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, “Fading corre-
lation and its effect on the capacity of multielement antenna systems,”
IEEE Trans. Commun., vol. 48, no. 3, pp. 502–513, Mar. 2000.
[26] K. Sivanesan, J. Xiao, R. Q. Hu, and G. Wu, “Code book based
CL-MIMO for DL Wimax Rel. 1.5: System level performance anal-
ysis,” in Proc. IEEE Int. Conf. Commun., Dresden, Germany, Jun.
2009, pp. 1–5.
[27] S. P. Alex and L. M. A. Jalloul, “Performance evaluation of MIMO
in IEEE802.16e/WiMAX,” IEEE J. Sel. Topics Signal Process., vol. 2,
no. 2, pp. 181–190, Apr. 2008.
[28] V. Raghavan, M. L. Honig, and V. V. Veeravalli, “Performance
analysis of RVQ-based limited feedback beamforming codebooks,” in
Proc. IEEE Int. Symp. Inform. Theory, Seoul, Korea, Jun. 2009, pp.
2437–2441.
[29] M. Hunukumbure and M. Beach, “MIMO channel measurements and
analysis with prototype user devices in a 2 GHz outdoor cell,” in
Proc. IEEE Int. Symp. Personal Indoor and Mobile Radio Commun.,
Helsinki, Finland, Sep. 2006, pp. 1–5.
681
[30] M. Beach, M. Hunukumbure, and M. Webb, “Dynamics of spatial
eigen modes in measured MIMO channels with different antenna
modules,” in Proc. European Conf. Antennas Propag., Edinburgh,
U.K., Sep. 2007.
[31] RUSK Channel Sounder 2010 [Online]. Available: http://www.chan-
nelsounder.de, Medav, GmbH.
[32] D. J. Love and R. W. Heath, “Limited feedback unitary precoding for
spatial multiplexing systems,” IEEE Trans. Inf. Theory, vol. 51, no. 8,
pp. 2967–2976, Aug. 2005.
[33] J. G. Proakis, Digital Communications, 3rd ed. New York: Wiley,
1974.
[34] C. Wang et al., “On the performance of the MIMO zero-forcing re-
ceiver in the presence of channel estimation error,” IEEE Trans. Wire-
less Commun., vol. 6, no. 3, pp. 805–810, Mar. 2007.
Matthew Webb received the B.A. and M.Eng. de-
grees in electrical and information sciences from the
University of Cambridge, Cambridge, U.K., in 2002,
and the Ph.D. degree in electrical engineering from
the University of Bristol, Bristol, U.K., in 2006.
From 2006 to 2011, he was a member of research
staff at the University of Bristol. He is currently
a Senior Researcher at Fujitsu Laboratories of
Europe Ltd., Hayes, U.K. His research interests
include multi-antenna wireless information theory,
radio propagation measurements and modeling,
closed-loop feedback imperfections in MIMO, and the use of information
about location and physical surroundings to augment the wireless physical and
MAC layers.
Mythri Hunukumbure (M’07) received the B.Sc.
degree in electronic and telecommunications engi-
neering from the University of Moratuwa, Sri Lanka,
in 1998, and the M.Sc. and Ph.D. degrees from the
University of Bristol, Bristol, U.K., in 2001 and
2004, respectively.
From 2005 to 2006, he was a member of research
staff at the University of Bristol. He is currently a Se-
nior Research Engineer with Fujitsu Laboratories of
Europe Ltd., Hayes, U.K. His research interests have
included code orthogonality and MIMO technology,
as well as network planning and physical layer design of 4G LTE and WiMAX
systems.
Mark Beach (M’06) received the Ph.D. degree for
research addressing the application of smart antennas
to GPS from the University of Bristol, Bristol, U.K.,
in 1989.
After graduation, he joined as a member of aca-
demic staff at the University of Bristol, Bristol. He
was promoted to Senior Lecturer in 1996, Reader in
1998, and Professor in 2003, serving as Head of the
Department of Electrical and Electronic Engineering
from 2006 to 2010. His research interests include the
application of multiple antenna technology, with par-
ticular emphasis on spatio-temporal aspects of the channel, as well as enabling
RF technologies for “green radio”.
682 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Multipath Simulator Measurements of Handset Dual

Antenna Performance With Limited Number of
Signal Paths
Paul Hallbjörner, Juan D. Sánchez-Heredia, Peter Lindberg, Member, IEEE, Antonio M. Martínez-González, and
Thomas Bolin

Abstract—Antenna pairs for diversity or MIMO functionality

are characterized under the assumption of a certain statistical dis-
tribution of the incident signals over angle and polarization, but
also assuming a signal environment with a large number of signal
paths. In many real-life environments, however, only a few signal
paths contain most of the transferred power. A multipath simu-
lator can be used to realize signal environments with a controlled
number of signal paths. This paper presents measurements of dual
antenna performance using a multipath simulator with 2–16 signal
paths. The results are analyzed in terms of statistical power distri-
butions, power imbalance, correlation coefficient, multiplexing ef-
ficiency, and diversity gain. Differences in performance depending
on the number of signal paths are noted, illustrating the value of
considering the number of signal paths in characterization.
Index Terms—Antenna measurements, multipath simulator,
signal environment, sparse environment, terminal antenna.
Fig. 1. MPS in anechoic room, with test object at the center of the MPS array.

I. INTRODUCTION

D UAL antennas in mobile terminals are used more and
more, for diversity or multiple-input multiple-output
(MIMO) functionality. There are different opinions regarding
the proper way to characterize antenna pairs for diversity or
MIMO functionality, involving different parameters [1]. One
common way to characterize dual antennas is by their mean ef-
fective gains (MEG) and correlation coefficient [2], [3]. These
parameters depend on the assumed statistical distribution of
the incident signals over angle and polarization. Dual antenna
performance is also based on the assumption of a large number
of signal paths [4], [5]. When calculating the correlation co-
efficient from the radiation patterns of the two antennas, the
integrals are in effect summations of an infinite number of
signal paths. There is, however, nothing in the integral about
the number of simultaneous signal paths at any instant.
Manuscript received April 20, 2011; revised September 20, 2011; accepted
October 10, 2011. Date of publication October 25, 2011; date of current ver-
sion February 03, 2012. The work was supported in part by MICINN (Project
TEC2008-05811) through an FPI doctoral grant (BES-2009-013764).
P. Hallbjörner is with SP Technical Research Institute of Sweden, SE-501 15
Borås, Sweden (e-mail: paul.hallbjorner@sp.se).
J. D. Sánchez-Heredia and A. M. Martínez-González are with the Univer-
sidad Politécnica de Cartagena (UPCT), E-30202 Cartagena, Spain (e-mail:
jd.sanchez@upct.es; toni.martinez@upct.es).
P. Lindberg is with TE Connectivity, SE-175 26 Järfälla, Sweden (e-mail:
peter.lindberg@te.com).
T. Bolin is with Sony Ericsson Mobile Communications, Nya Vattentornet,
SE-221 88 Lund, Sweden (e-mail: thomas.bolin@sonyericsson.com).
Digital Object Identifier 10.1109/TAP.2011.2173451
In many real-life environments, most of the transferred power
is contained in only a few signal paths [6], [7]. The performance
of an antenna pair in such environments can be studied by com-
puter simulations using the radiation patterns. Dual antenna per-
formance in multipath environments can also be measured di-
rectly, without going via the radiation patterns [8]. Reverbera-
tion chambers (RC) [9], [10] is an established technique for this.
RCs are usually designed so as to ensure that the test object is
subjected to a large number of signal paths. Multipath simu-
lators (MPS) [11]–[13] are an emerging technique that allows
better control of the signal environment than RCs. For instance,
the number of signal paths is a parameter that can be decided
arbitrarily.
This paper presents a study in which a multipath simulator is
used to characterize dual antenna performance in environments
with a limited number of signal paths. The purpose is to have
experimental indications of how performance typically differs
depending on the number of signal paths, and thereby also in-
dications regarding the need to consider the number of signal
paths in characterization.
II. SETUP
A. Multipath Simulator
An MPS comprising 16 antennas is used. The antennas are of
vertical and horizontal polarizations, and evenly distributed on
a circle around the test object, in a plane, see Fig. 1.

0018-926X/$26.00 © 2011 IEEE

HALLBJÖRNER et al.: MULTIPATH SIMULATOR MEASUREMENTS OF HANDSET DUAL ANTENNA PERFORMANCE
Fig. 2. Block diagram of the setup for S-parameter measurements on dual
antennas.
TABLE I
PHASE SWEEP RATE ON THE MPS BRANCHES, AND THE ACTIVE BRANCHES
USED FOR THE DIFFERENT NUMBER OF PATHS. FOR 2, 4, AND 8 PATHS, THREE
SETS ARE USED, EACH GIVEN BY A COLUMN
The MPS thus simulates a 2-D signal environment with up
to 16 signal paths. Each MPS antenna is fed via a sweeping
phase shifter, which simulates the Doppler shift experienced
when the terminal moves in a multipath environment. By set-
ting the sweep rate differently on the antennas, a fast fading
signal is received by the test object antennas. Each MPS an-
tenna is also fed via an attenuator, and the antenna is effectively
switched on/off by setting its attenuator to minimum/maximum
attenuation.
Fig. 2 shows a block diagram of the setup, with a network
analyzer for simultaneous measurements of the complex trans-
mission coefficients between the MPS port and each of
the two antenna ports of the test object.
Each measurement in the presented work is a 20 s sequence
of 2 6401 values between the MPS port and the two test
object antenna connectors. A maximum Doppler shift of 50 Hz
is used, to ensure that samples are taken with sufficient den-
sity with respect to the fast fading. Table I shows the phase
683
sweep rates over the MPS antenna array that realize the desired
Doppler shifts. Measurements are performed with 2, 4, 8, and 16
MPS antennas active, and the antennas that are switched on all
have equal amplitude. Which MPS antennas are active for the
respective number of signal paths is also shown in Table I, with
reference to Fig. 2 for the physical locations of the antennas. Ac-
tive antennas are marked in Table I with an “O”. With 2, 4, and
8 paths, there are three different sets of active antennas, given
by the three columns.
B. Performance Metrics
Results are analyzed in terms of the power imbalance , cor-
relation coefficient , and multiplexing efficiency [14],
according to the expressions
(1)
(2)
(3)
In (1)–(3), and are the complex transmission co-
efficients of the respective test object antenna, each being a se-
quence of 6401 samples, and denotes average over the whole
sequence. The expression in (3) differs from the one in [14] in
that normalization to the average radiation efficiency of the two
antennas is introduced. This is done because we want to study
only the effects of imbalance and correlation, but not the effect
of the absolute levels of the efficiencies.
Diversity gain [15] is also used as a figure-of-merit. Max-
imum ratio combining is assumed, and the combined signal is
thus
(4)
After this, is calculated at the 99% signal reliability levels,
%
(5)
%
where two values are calculated ( and ) with each
of the two antennas serving as the non-diversity reference case.
The reason for doing this, rather than using one specific antenna
as the reference, is that we hereby avoid assuming which an-
tenna would be used in a non-diversity case.
III. MEASUREMENTS
Two experiments are carried out, with different test objects. In
the first, a test object with near-ideal performance in an isotropic
environment is used. This makes an interesting reference case
for the change in performance when going to fewer signal paths.
In the second experiment, two terminal prototypes with less
ideal and more typical performance are used. They are proto-
types of the same terminal, but with different design solutions
for the dual antennas. This represents a realistic situation in
which different designs are characterized and compared.
684 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

A. Near-Ideal Test Object

A terminal antenna model for the 2600 MHz band is used,
with measurements performed at 2655 MHz. The model is
made of metal only (no plastic housing) and consists of a
100 mm 40 mm ground plane with two PIFA antennas fed
by coaxial cables. The cables are routed so as to minimize
influence on the antenna performance [16]. Tests with different
feed cable arrangements and with ferrite absorbers [17] are
performed to ensure that the results are not affected too much.
Electrical performance is a return loss of 12 dB and near 100%
radiation efficiency on both antennas, and a correlation coeffi-
cient of 0.25 in an isotropic environment with a large number
of signal paths.
In the MPS measurements, the test object is mounted in free-
space (no head or hand) at the center of the MPS array. It is mea-
sured in three orthogonal planes with each set of active branches
according to Table I. It is also measured in a 45 slant orienta-
tion (data mode), with 10 different orientations in azimuth. The
slant orientation measurements are done only for the leftmost set
of active branches for 2 and 4 paths, and the set of 16 branches.
The total number of measurements are thus 19/19/9/13 for the
respective 2/4/8/16 paths.
Fig. 3 shows its performance for the different number of
signal paths. Power imbalance, correlation coefficient, and
multiplexing efficiency all show a strong increase in the spread
when reducing the number of signal paths. The performance
with 16 signal paths agrees well with the performance in an
isotropic environment with a large number of signal paths, with
only small variations. With two signal paths, the correlation
coefficient is many times very high, and even with four signal
paths it varies over almost all possible values. With few signal
paths, the average correlation coefficient is also increased
significantly compared to the case of many signal paths.
Examples of time sequences and cumulative distribution
functions (CDF) of the received power are plotted for a few
measurements. The case of two signal paths are shown in
Fig. 4(a) and (b), while the case of 16 signal paths are seen in
Fig. 4(c) and (d). With 16 paths, time sequences are similar to
Rayleigh fading curves, and the CDF curves are all approxi-
mately the same, corresponding to the theoretical curves under
assumption of a large number of signal paths. With two paths,
on the other hand, strong variations are seen in time sequence
characteristics and CDF curve shapes. As a result, can be
anything from very large to almost none.

B. Two Realistic Test Objects

Two samples of the same terminal model, with dimensions
115 mm 65 mm, but with different antenna solutions, are
measured. The two test objects are referred to as Prototype A
and Prototype B. These models are more realistic than the first
one, having plastic housing and antennas with less ideal per-
formance. Their antennas are made for the 700 MHz band, and
the measurements are performed at 740 MHz. The antennas are
accessed via coaxial cables, which exit the test objects on the
middle of the long side for minimum interference. As with the
near-ideal test object, tests are conducted to ensure that the feed
cables have a negligible effect on performance. The respective
antenna performance in a 3-D environment with a large number
Fig. 3. (a) Power imbalance of the near-ideal test object, as a function of
the number of signal paths. (b) Correlation coefficient of the near-ideal test
object, as a function of the number of signal paths. (c) Multiplexing efficiency
of the near-ideal test object, as a function of the number of signal paths.
of signal paths is measured in an RC. The power imbalances
are dB for Prototype A and dB for Prototype B, and
the correlation coefficients are 0.35 for Prototype A and 0.56
for Prototype B. More interesting for comparison with the MPS
measurements is the performance in a 2-D environment. These
values are calculated from 2-D cuts of the radiation patterns in
HALLBJÖRNER et al.: MULTIPATH SIMULATOR MEASUREMENTS OF HANDSET DUAL ANTENNA PERFORMANCE 685

Fig. 4. (a) Examples of time sequences for the case of two signal paths, on the two branches (dotted and dashed). (b) Examples of cumulative distribution functions
for the case of two signal paths, with the two branches (dotted and dashed) and the combined signal using maximum ratio combining (solid). The four examples
show vastly different diversity gains. (c) Examples of time sequences for the case of 16 signal paths, on the two branches (dotted and dashed). (d) Examples
of cumulative distribution functions for the case of 16 signal paths, with the two branches (dotted and dashed) and the combined signal using maximum ratio
combining (solid). All four examples are approximately the same.

TABLE II
POWER IMBALANCE AND CORRELATION COEFFICIENT OF PROTOTYPES A AND
B, IN DIFFERENT 2-D ENVIRONMENTS
each of three orthogonal planes. The cuts are measured with a
standard far-field antenna test range. Uniform power distribu-
tion over angle and polarization is assumed when calculating
the performance parameter from these 2-D cuts. As can be
seen in Table II, the test objects are well balanced in all three
planes. Correlation coefficients are notably higher in the 2-D
environments than in the 3-D environment.
For the MPS measurements, the test objects are mounted in
free-space at the center of the MPS array. They are measured
in three orthogonal planes with each set of active branches ac-
cording to Table I. The same planes and orientations are used
for both test objects. The total number of measurements for each
test object are thus 9/9/9/3 for the respective 2/4/8/16 paths.
Results are seen in Fig. 5, with circles for Prototype A and
triangles for Prototype B. The plots show that also in this ex-
periment, reducing the number of signal paths leads to a much
greater spread in the different metrics. Power imbalance is not
as symmetrical around 0 dB as in the first experiment, and Pro-
totype A shows a larger spread than Prototype B. For both test
objects, is clearly worse with 16 paths than in a 3-D envi-
ronment with many paths, but comparable to 2-D environments
according to Table II. At 16 signal paths, is higher for Proto-
type B than Prototype A, and it is significantly higher with fewer
signal paths. A correlation coefficient of 0.7 is usually consid-
ered the limit for good diversity/MIMO performance. Table III
shows the proportion of values less than 0.7 for the two
prototypes, indicating that Prototype A is superior. The spread
686 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 5. (a) Power imbalance of Prototype A (circles) and Prototype B (trian-
gles), as a function of the number of signal paths. (b) Correlation coefficient
of Prototype A (circles) and Prototype B (triangles), as a function of the number
of signal paths. (c) Multiplexing efficiency of Prototype A (circles) and
Prototype B (triangles), as a function of the number of signal paths.
is however very large for both models, and there are many cases
in which the performance difference is reversed.
The multiplexing efficiency, being a function of both the ef-
ficiencies and the correlation, show better performance for Pro-
totype A in general, and a quicker degradation for Prototype B
TABLE III
PROPORTION OF CORRELATION COEFFICIENT VALUES LESS THAN 0.7 FOR
PROTOTYPES A AND B, FOR DIFFERENT NUMBER OF SIGNAL PATHS
TABLE IV
MEDIAN AND WORST CASE MULTIPLEXING EFFICIENCY OF PROTOTYPES A
AND B, FOR DIFFERENT NUMBER OF SIGNAL PATHS
Fig. 6. Diversity gain of Prototype A (circles) and Prototype B (triangles),
using maximum ratio combining, as a function of the number of signal paths.
TABLE V
BEST CASE, MEDIAN, AND WORST CASE DIVERSITY GAIN OF PROTOTYPES A
AND B, FOR DIFFERENT NUMBER OF SIGNAL PATHS
as the number of signal paths is reduced. Table IV summarizes
the median and worst case multiplexing efficiencies.
Diversity gain is shown in Fig. 6 and Table V. From 16 paths
down to four, the median remains stable for both prototypes,
but with an increase in the spread. With two paths, the me-
dian value for Prototype B drops drastically, while it remains
approximately the same for Prototype A. The spread increases
even more for both, and Prototype B has better best case and
worst case values. Best and worst cases are however highly
sensitive to random fluctuations. Which model has the best di-
versity gain performance depends on which number of signal
paths and which statistical parameter (median, best/worst case)
is considered.
HALLBJÖRNER et al.: MULTIPATH SIMULATOR MEASUREMENTS OF HANDSET DUAL ANTENNA PERFORMANCE
IV. CONCLUSION
Mobile terminal dual antennas are characterized using an
MPS, which provides a 2-D signal environment with up to 16
signal paths. Three test objects are characterized with 2, 4, 8,
and 16 simulated signal paths.
Several conclusions are drawn from the results. The perfor-
mance with 16 signal paths shows only small variations over
different test object orientations. For the test objects used, it is
close to the performance in a corresponding 2-D environment
with a large number of signal paths, but not necessarily close to
the performance in a 3-D environment.
As the number of signal paths is reduced, the primary effect
is a strong increase in the spread of the different metrics. Some
degradation in average performance is also seen. With only two
signal paths the correlation coefficient can have any value, with
a significant portion of the values being above 0.7, even on a test
object with good performance in a 3-D environment with many
signal paths. Two test objects of the same type of terminal, but
with different antenna solutions, are demonstrated to both have
very strong performance variations in an environment with lim-
ited number of signal paths, but also somewhat different charac-
teristics in terms of median and worst case performance. Which
antenna solution is the better depends on the exact number of
signal paths, and the choice of performance parameter.
Presented experiments show that characterization in environ-
ments with a limited number of signal paths provides a deeper
insight into dual antenna performance compared to the tradi-
tional characterization assuming a large number of signal paths,
and that an MPS is suitable for such characterization.
V. FUTURE WORK
Future work includes a study of the same type as the pre-
sented, but with the application of realistic channel models to
the MPS. Further, models and metrics intended specifically for
dual antenna performance in sparse environments should be
studied, and specific test methods should be developed that en-
able performance assessment with maximum reliability at min-
imal cost.
REFERENCES
[1] J. Villanen, P. Suvikunnas, C. Icheln, J. Ollikainen, and P. Vainikainen,
“Performance analysis and design aspects of mobile-terminal multi-
antenna configurations,” IEEE Trans. Veh. Technol., vol. 57, no. 3, pp.
1664–1674, May 2008.
[2] C. B. Dietrich, Jr., K. Dietze, J. R. Nealy, and W. L. Stutzman, “Spatial,
polarization, and pattern diversity for wireless handheld terminals,”
IEEE Trans. Antennas Propag., vol. 49, no. 9, pp. 1271–1281, Sept.
2001.
[3] T. Taga, “Analysis for mean effective gain of mobile antennas in land
mobile radio environments,” IEEE Trans. Veh. Technol., vol. 39, no. 2,
pp. 117–131, May 1990.
[4] M. J. Gans, “A power-spectral theory of propagation in the mobile-
radio environment,” IEEE Trans. Veh. Technol., vol. VT-21, no. 1, pp.
27–38, Feb. 1972.
[5] P. L. Carro, J. de Mingo, and P. G. Ducar, “Analysis of the antenna sto-
chastic effective gain in mobile environments,” presented at the IEEE
69th Veh. Technol. Conf., Barcelona, Spain, Apr. 26–29, 2009.
[6] K. Kalliola, “Experimental Analysis of Multidimensional Radio Chan-
nels,” Ph.D. dissertation, Helsinki University of Technology, Radio
Laboratory Publications, Rep. S 251, Espoo, Finland, Feb. 2002.
687
[7] A. M. Sayeed and V. Raghavan, “Maximizing MIMO capacity in
sparse multipath with reconfigurable antenna arrays,” IEEE J. Sel.
Topics Signal Process., vol. 1, no. 1, pp. 156–166, June 2007.
[8] J. S. Colburn, Y. Rahmat-Samii, M. A. Jensen, and G. J. Pottie, “Evalu-
ation of personal communications dual-antenna handset diversity per-
formance,” IEEE Trans. Veh. Technol., vol. 47, no. 3, pp. 737–746,
Aug. 1998.
[9] T. Maeda, S. Sekine, S. Obayashi, and T. Morooka, “Two methods for
estimating the diversity characteristics of built-in antennas for mobile
communication equipment,” in Proc. IEEE AP-S Int. Symp., Jun. 1995,
vol. 4, pp. 18–23, pp. 1944-1947.
[10] J. F. Valenzuela-Valdés, A. M. Martínez-González, and D. A. Sánchez-
Hernández, “Diversity gain and MIMO capacity for nonisotropic en-
vironments using a reverberation chamber,” IEEE Antennas Wireless
Propag. Lett., vol. 8, pp. 112–115, 2009.
[11] T. Sakata, A. Yamamoto, H. Iwai, K. Ogawa, J. Takada, K. Sakaguchi,
and K. Araki, “BER evaluation system for a handset antenna in a multi-
path environment using a spatial fading emulator,” in Proc. ISAP2005,
Seoul, Korea, vol. TB1-6, pp. 351–354.
[12] L. Rudant, C. Delaveaud, and M. AbouElAnouar, “Synthesizing real-
istic environments in an anechoic chamber,” in Proc. 3rd Eur. Conf. on
Antennas Propag., Berlin, Germany, Mar. 23/27, 2009, pp. 221–225.
[13] P. Hallbjörner, Z. Ying, M. Håkansson, C. Wingqvist, T. Anttila, and
J. Welinder, “Multipath simulator for mobile terminal antenna charac-
terization,” IET Microw. Antennas Propag., vol. 4, no. 6, pp. 743–750,
2010.
[14] R. Tian, B. K. Lau, and Z. Ying, “Multiplexing efficiency of MIMO
antennas,” IEEE Antennas Wireless Propag. Lett., vol. 10, pp. 183–186,
2011.
[15] D. G. Brennan, “Linear diversity combining techniques,” in Proc. IRE,
Jun. 1959, pp. 1075–1102.
[16] S. Saario, D. V. Thiel, J. W. Lu, and S. G. O’Keefe, “An assessment
of cable radiation effects on mobile communications antenna measure-
ments,” in Proc. IEEE Antennas Propag. Society Int. Symp., Montreal,
Canada, Jul. 13/18, 1997, vol. 1, pp. 550–553.
[17] C. Icheln, J. Krogerus, and P. Vainikainen, “Use of balun chokes in
small-antenna radiation measurements,” IEEE Trans. Instrum. Meas.,
vol. 53, no. 2, pp. 498–506, Apr. 2004.
Paul Hallbjörner was born in Uppsala, Sweden,
in 1966. He received the B.Sc., M.Sc., and Ph.D.
degrees, all in electrical engineering, from Chalmers
University of Technology, Göteborg, Sweden, in
1988, 1995, and 2005, respectively.
Since 1989, he has worked in the telecom industry
in the areas of pre-production engineering, product
development, and research, mainly in the field of an-
tennas and microwave technology. He has worked
with mobile terminal antennas for handsets and vehi-
cles, base station antennas, reconfigurable and steer-
able antennas, wave propagation, passive microwave circuits, electrical mate-
rial characterization, and millimeter-wave design. He has been employed by
Ericsson, Saab, Allgon, and is currently working an Antenna Researcher at SP
Technical Research Institute of Sweden, Borås, Sweden, where his main focus
is on antenna measurement techniques for mobile phones and short range com-
munication, in addition to antenna development for various applications. He is
the author of more than 70 scientific publications and the inventor of 10 patents.
Juan D. Sánchez-Heredia was born in Lorca,
Spain. He received the Telecommunication Engi-
neering degree from the Universidad Politécnica de
Cartagena, in 2009, which culminated with the Final
Degree Award, and the Master degree in information
technologies from Universidad de Murcia, in 2010.
He is currently working toward the Ph.D. degree at
the Universidad Politécnica de Cartagena, Spain.
In 2007, he worked at General Electric, Cartagena,
and was involved in several projects in relation with
the network infrastructure. In 2009 he joined the De-
partment of Information Technologies and Communications, as a Ph.D. student.
His current research areas cover MIMO communications, multimode-stirred
chambers and electromagnetic dosimetry.
688 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Peter Lindberg (M’08) was born in Uppsala,
Sweden, in 1974. He received the M.Sc. and Ph.D.
degrees in engineering physics from Uppsala Uni-
versity, Sweden, in 2000 and 2007, respectively.
Previously, he worked as an RF Engineer at
Smarteq Wireless, Enebyberg, Sweden, from 2000
and 2002, and as a Research Engineer in the Mi-
crowave Technology Group, Uppsala University,
from 2002 and 2003. He is the coauthor of more
than 30 scientific publications and co-inventor of
over 40 patents. Currently, he is with the Advanced
Technology Group, Mobile Antenna Systems Division, Laird Technologies,
Sweden, working with antenna design, measurement techniques, technology
scouting, and customer engineering support.
Dr. Lindberg received the R. W. P. King Best Paper Award from the IEEE
Antennas and Propagation Society in 2007.
Antonio M. Martínez-González received Dipl.-Ing.
degree in telecommunications engineering from the
Universidad Politécnica de Valencia, Spain, in 1998
and the Ph.D. degree from Universidad Politécnica
de Cartagena, in early 2004.
From 1998 to September 1999, he was employed
as a Technical Engineer at the Electromagnetic
Compatibility Laboratory, Universidad Politécnica
de Valencia, where he developed assessment activ-
ities and compliance certifications with European
directives related with immunity and emissions to
electromagnetic radiation from diverse electrical, electronic and telecom-
munication equipment. Since September 1999, he is an Associate Professor
at Universidad Politécnica de Cartagena. At present, his research interest is
focused on electromagnetic dosimetry, radioelectric emissions and mode stirred
chambers. In December 2006, he was one of the founders of EMITE Ing,
a technological spin-off company founded by telecommunication engineers
and doctors of the Microwave, Radiocommunications and Electromagnetism
Research Group (GIMRE), Technical University of Cartagena (Spain).
Dr. Martínez-González was awarded with the Spanish National Prize from
Foundation Airtel and Colegio Oficial de Ingenieros de Telecomunicación de
España for the best final project on Mobile Communications in 1999. In 2006
and 2008, he was a co-recipient (as a co-founder of EMITE) of the i-patentes
prize for innovation and technology transfer in the Region of Murcia (Spain).
Thomas Bolin was born in Falun, Sweden, in
1954. He received the M.Sc. degree in electrical
engineering from Linköping Technical University in
Sweden, in 1979.
He is currently working with MIMO antenna de-
sign and measurement techniques within Sony Eric-
sson Mobile Communications, Lund, Sweden. He is
one of the pioneers in the mobile handset business
having held positions in radio design and manage-
ment at Ericsson and later Sony Ericsson. Since 1996,
he founded and managed an antenna group within his
company. From 1979 to 1983, he worked for ITT Standard Radio & Telefon AB
in Vällingby, Sweden, with design of short wave radio kW transmitters. He is
the author of a number of scientific papers and holds five patents.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012 689

On Small Terminal Antenna Correlation and Impact

on MIMO Channel Capacity
Boyan Yanakiev, Jesper Ødum Nielsen, Morten Christensen, and Gert Frølund Pedersen

Abstract—Analysis of the antenna correlation at the design stage
is made, and then compared to real life performance in a typical
propagation environment and in typical use cases. A traditional
design flow is followed and conclusions are made on the perfor-
mance of several handsets. These conclusions are then contrasted
to measurements and an explanation is sought for the variations.
It is concluded that correlation estimation and optimization at the
design stage, using conventional methods, brings little benefit in
real life situations. Impact on channel capacity has been the figure
of merit.
Index Terms—Antenna, correlation, electrically small antennas,
handset antennas, MIMO systems, optical fiber, optical fiber mea-
surement applications, propagation measurements.
I. INTRODUCTION
W ITH the fast approaching worldwide launch of the long
term evolution (LTE) system, it is ever more urgent that
adequate antenna design specifications are defined. From mul-
tiple input multiple output (MIMO) theory point of view, the
received power, signal correlation and branch power ratio are
all important for the channel capacity and overall system per-
formance. The influence of each one is investigated in the liter-
ature [1] and certain requirements and rule of thumb numbers
are available [2]. It is not clear, however, how to translate these
essentially signal requirements, to antenna specifications. This
paper takes a look at the current trends in antenna design and
evaluates the relevance to real life performance. The focus is on
correlation.
A common approach is to evaluate an antenna system in terms
of diversity gain, which is known to depend on mean effective
gain [3], branch power ratio and correlation [4], [5]. While this
can give a very good estimate, of how influential each parameter
is, it is not easy to come up with design goals. A large study of
the user influence on the mean effective gain done in [6] shows,
Manuscript received June 15, 2010; revised May 18, 2011; accepted August
26, 2011. Date of publication October 25, 2011; date of current version February
03, 2012. The work of B. Yanakiev and J. Ø. Nielsen was supported in part by
the Danish Advanced Technology Foundation (Højteknologifonden) as part of
the Converged Advanced Mobile Media Platform (CAMMP) project.
B. Yanakiev is with the Molex Antenna Business Unit, DK-9220 Aalborg,
Denmark and also with the Antennas, Propagation and Radio Networking sec-
tion at the Department of Electronic Systems, Faculty of Engineering and Sci-
ence, Aalborg University, DK-9220 Aalborg, Denmark (e-mail: by@es.aau.dk).
J. Ø. Nielsen and G. F. Pedersen are with the Antennas, Propagation and
Radio Networking Section, Department of Electronic Systems, Faculty of En-
gineering and Science, Aalborg University, DK-9220 Aalborg, Denmark.
M. Christensen are with Molex Antenna Business Unit, DK-9220 Aalborg,
Denmark.
Digital Object Identifier 10.1109/TAP.2011.2173442
that up to 10 dB of variation can be expected from one user to
another. In [7] and [8] investigation is done on the measured
correlation and branch power ratio around 1.8 GHz, with the
presence of the user. It is concluded that the users generally in-
crease the correlation, but also that the branch power difference
is a more important parameter for diversity gain.
A more theoretical approach is resented in [9] and [10]where
rotation properties of the antenna radiation patterns are inves-
tigated in details and shown to contribute to the variations re-
ported earlier. An extension to this work is done in [11], for
several different environment models, highlighting the need for
realistic use cases.
A more recent work [12] further develops the concept to in-
clude a more realistic estimate of the actual diversity gain in the
presence of users and confirms the conclusions.
In the current paper the antenna performance, in terms of cor-
relation, is analyzed at two distinct stages, called here—design
and real life.
• The design stage is defined as the performance evaluation
by simulation and initial verification measurements in lab-
oratory. Antenna correlation from simulation tools is usu-
ally directly computed from the radiation patterns or S-pa-
rameters following the derivations in [4] and [13], respec-
tively. The method for measurement verification depends
on the system used—measured radiation patterns for the
anechoic chambers or sequences of transmission coeffi-
cients for the mode-stirred chambers [14]. The practical-
ities and capabilities of these two measurement methods
are different, so here the computed design stage correla-
tion is only from simulations, assuming that whatever the
verification method, it should converge to about the same
simulated value.
• The real life stage is defined as the evaluation of the
performance in a realistic propagation environment and
with realistic handling by users. To measure correlation
in such realistic conditions, large scale channel sounding
campaign was performed in a typical urban environment.
In this paper the correlation values presented as achieved
in real life scenario, are computed from the recorded
channel impulse responses, which include the transmit
and receive antennas, and the propagation channel. It
is worth noting that currently there is an effort in the
scientific community and especially industry, to emulate
such conditions in the laboratory with methods based
on both chamber types—[15]–[17] or [18]. The focus is
primarily on the end-to-end system performance in terms
of throughput with the main discussion over the realistic
modeling of the propagation channel.

0018-926X/$26.00 © 2011 IEEE

690 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Currently, rule of thumb numbers exist suggesting that for TABLE I

all other parameters being the same, the envelope correlation of HANDSET OVERVIEW
two antennas should be or even depending on
the source—[2], [4], [5] and [19],[20], respectively. Such signal
correlation value would provide good MIMO performance and
is readily adopted by antenna designers as a requirement along-
side antenna efficiency, branch power ratio etc.
The goal of this paper is to reexamine these signal correlation
requirements and evaluate the benefits of adding a particular
correlation value to the antenna design requirements. It will be
argued here that, while it is always true that low correlation
leads to better performance, high correlation is not necessarily
leading to noticeably worse performance and that designing for
low correlation can even have negative effects. Some potential
pitfalls in the current correlation measurement techniques are
also discussed.
The paper is organized as follows: Section II gives details on
the extensive work done to make realistic handset mock-ups.
Details on the measurement set-up and use cases are also given. TABLE II
LAB MEASUREMENTS FOR H1, H2 AND H4
Section III presents the theoretical background for the design
stage results. Details on processing of the measured data are
given in Section IV. Results from the design stage are presented
in Section V and channel sounding results follow in Section VI.
The impact on channel capacity is presented in Section VII with
some additional discussion. Section VIII concludes the paper.

II. TEST HANDSETS AND SCENARIOS

Since the paper aims at accurately representing both the de-
sign process as well as the real life scenarios, special atten-
tion has been paid to designing realistic handsets and evaluating
them in scenarios as close as possible to a real network.
For the design stage evaluation, standard finite difference
time domain (FDTD) based [21] software has been used for
simulation. Optimizations have been made to achieve typical
efficiency, low user influence and low coupling. All handsets
have been measured in an anechoic chamber to confirm the
simulation results. Since there is a known problem with the
coaxial cable when measuring small terminal antennas [22],
[23], [24], a special optical unit was developed [25], [26],
which allows for both accurate measurements but also natural
user handling.
A. Handsets
Seven handset mock-ups were used for the measurements.
Here the focus will be on three of them (H1, H2 and H4) for sim-
plicity, as they present the corner cases and show the main trends
well. The conclusions are true for all handsets and Table I gives
an overview. In Table I “mono” and “PIFA” stand for monopole
and planar inverted F antenna types, respectively. The hand-
sets’ form factors are given along with the electrical sizes in
columns 1 and 2 as typically defined in [27]. For example H2
is meant to represent a long, clamshell phone used when open
(Fig. 3), while H1 and H4 have sizes similar to modern touch-
screen smartphones/PDAs. All handsets have two antennas but
not all antennas are dual band. For example, antenna No.1 on H4
covers only the high band, while antenna No.2 is dual band and
is used for a SISO reference case on the low band. H4 also has
both antennas on the top, as opposed to top and bottom place-
ment in H1 and H2. Table II gives details on the measured ef-
ficiencies and coupling for the selected handsets. The coupling
is measured with coaxial cables attached. Some difference can
be expected when the cables are removed, however the optical
measurement system used does not support optical S-parameter
measurements. Simulations suggest another 2–3 dB lower cou-
pling without the coaxial cables.
All of the handsets were equipped with optical units and
plastic casings from PC-ABS material with grip position en-
gravings for accurate and natural user handling. The finger
positions were surrounded with elevated rings to avoid finger
movement during measurements. Example of H1 is shown
in Fig. 1 and details are given in [26], where the benefits of
using the optical links for radiation pattern measurements is
confirmed. The sizes listed in column 2 of Table I are the
effective electrical sizes or simply the PCB size. The PC-ABS
material has a dielectric constant of about and adds 1.5
[mm] in all directions to the overall handset size. The resolution
of the 3D rapid prototyping printer used to make the plastics,
is around 0.1 [mm], allowing for very accurate mechanical
design, antenna positioning and repeatability. Industry grade
matching components and gold plated feed springs are used to
ensure high quality and durability. During the design process
the handsets have been disassembled tens of times without any
noticeable performance change. The three selected handsets for
this paper are shown in Fig. 2. Overall the design process has
followed the traditional mobile phone antenna design path.
YANAKIEV et al.: ON SMALL TERMINAL ANTENNA CORRELATION AND IMPACT
Fig. 1. Handset one (H1). Notice the optical unit in the middle.
Fig. 2. Antennas for handsets H1, H2 and H4. Top cover of H4 shown with
screw domes and grip marks visible.
TABLE III
BASE STATIONS DETAILS
B. Real Life Measurements
Once built and verified in the lab, the handsets were mea-
sured in a realistic propagation scenario. To represent a realistic
network architecture two base stations were built—one high al-
titude, low band representing an umbrella cell and one low alti-
tude, dual band, equipped with multiple antennas, representing a
local high capacity cell—see Table III. The two test frequencies
are and . The two hands grips are freely chosen since CTIA currently
Measurements were made indoor in browsing mode in free
space and with twelve users. Aalborg University’s parallel
691
Fig. 3. Two grips used in the measurements. (a) One Hand (OH) grip for H2;
(b) Two Hands (TH) grip for H2 with the same user.
channel sounder was used to sample the complex channel
impulse response. For details on the channel sounding refer to
Section IV.
C. Use Cases
Three main cases were measured—free space (FS), one (OH)
and two (TH) hands, all in data mode operation—see Fig. 3. In
all cases the data mode is imitated by positioning the handsets
at about degrees angle from vertical position. To perform
the free space measurements, the handsets were positioned in
pockets carved into a Styrofoam block and pushed along the
measurement routes on a table with wheels, by a kneeling down
person.
The pre-defined one hand grips are an attempt to imitate the
CTIA grips described in [27] for narrow (H2, H3 and H5-7)
and wide (H1 and H4) data mode operated handsets. The case
engravings were positioned approximately where the phantom
hands would touch the handsets and the distance to the palm is
fixed with a Styrofoam piece glued to the back of the handset.
does not specify them. Fig. 3 shows the grips used with H2 and
all other narrow bar phones. H1 and H4 are thought to be touch
692 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 4. Measurement site with color coded routes.
screen devices therefore for the two hands grip users were in-
structed to touch the handset approximately in the middle imi-
tating touch interaction.
Twelve users participated in the channel sounding measure-
ments and each handset was measured with each user at each
measurement route. Further in the paper, the two distinct grips
described above, are combined in a general use case called “with
users”. In the statistical analysis later, no differentiation is made
between one and two hands grips. This is done for simplicity,
since this paper aims only to quantify the general user effect on
correlation. All conclusions made for the user influence on cor-
relation when compared to free space, are true for the two grips
individually as well as when combined.
D. Measurement Site
The indoor measurements were taken on the third floor of
Aalborg University’s building in downtown Aalborg, Denmark.
The propagation is outdoor-to-indoor in a typical urban area. On
the floor of the measurement room, a square was drawn with 4
meter long sides—A, B, C and D (Fig. 4) with the arrows indi-
cating the user orientation. The users were instructed to hold the
handsets with the predefined grips and walk forward and back-
ward along a single side, twice for each measurement, without
turning around. The total distance of 16 meters per measure-
ment, gives about 40 wavelengths of sounding distance on the
low band. The order of the user participation is random, and
measurement with the same user, were taken within several days
to perform all measurement combinations of handsets, square
sides and grips. The cable holder in the middle of the square
was used to ensure that the optical fibers do not twist or bend
too much.
Again for simplicity, the four square sides are grouped into
one general indoor case and no further differentiation is made
unless otherwise noted. In few measurements the window side
(B) leads to direct line of sight and some larger variation, most
often for H2. It is not observed in all cases most probably due
to the slight variations of the user orientation and walking path.
Statistical analysis showed that the sides are not equivalent and
that side (B) does have higher received power. This is kept in
the statistical sampling simply to represent a more general case.
III. CORRELATION AT THE DESIGN STAGE
It has previously been shown that to achieve some diversity
gain the antenna envelope correlation should be below
[2], [4], [5]. This is of course not a fixed number but rather a rule
of thumb used as a guideline. A more commonly used number
in industry is [19], [20]. These have been adopted
for MIMO as well. It is thus important to understand how to
compute the correlation and what it depends on.
A. Antenna Correlation Formulations
Antenna correlation was first formulated in [4] as
(1)
where is the cross covariance, and and are the stan-
dard deviations of the received signals. The cross covariance
can also be written in a convenient way as in [28]
(2)
where
is the cross polarization ratio of the environment as defined in
[3]. The variances can be written as
(3)
In the above equations represents both and antennas,
and if stands for both the elevation and azimuth vector
components, then and are respectively, the
electric fields and gain patterns in the far field of the i-th antenna
for both polarizations. is then the power distribution in
the environment of both polarizations in spherical coordinates.
indicates variation over both and spherical angles, and
represents the complex conjugate.
It has also been shown [29] that the envelope correlation of
Rayleigh fading signals can be written as
(4)
Equation (4) is known to have less than 10% relative error
in the approximation estimated in [19] based on the data from
[29]. In the current paper all correlation results presented are of
the envelope correlation, since it is the most commonly used.
As can be seen clearly from the above formulations, the cor-
relation depends directly on the power distribution in the
environment, which can be seen as a weighting function of the
radiation pattern differences. The total power received per an-
tenna however is normalized via making the correlation es-
timate independent of efficiency differences.
Very often, the environment is defined as isotropic for sim-
plicity. Assuming no losses and such environment, in [13] a
very fast, convenient and broadband way for correlation com-
putation from S-parameters in derived. The isotropic assump-
tion however, rarely holds and many propagation measurements
YANAKIEV et al.: ON SMALL TERMINAL ANTENNA CORRELATION AND IMPACT
have shown, that the incoming power distribution is rather di-
rective or clustered, within several main directions [30]. It will
be shown here, that this simplification can compromise the es-
timation of correlation at the design stage. Here (1)–(4) will be
used to compute the envelope correlation, which implies that a
power distribution model must be defined as well.
When measurement verification in the lab is needed, the typ-
ical procedure for anechoic chambers is to replace the simulated
radiation patterns with measured ones in (1)–(4). This requires
accurate complex pattern measurements, but allows for arbitrary
power distribution. Using mode-stirred chambers gives the ad-
vantage of directly measuring the signal correlation at the an-
tenna ports, with typically isotropic power model assumed, [14].
Solutions however exist for modeling directive environments
[31].
B. Incoming Power Distribution Models
For the purposes of this analysis three spherical incoming
power models will be used.
1) Isotropic: The power in this case is uniformly distributed
on a sphere for both polarizations and . In that case
the power is a constant independent of elevation or azimuth and
and can come out in front of the integrals in (2)
and (3) leading to the most common definition of correlation
formulation in literature and practice (e.g., [13])
(5)
2) Gaussian: The power distribution for this model is given
in [3] as Gaussian for both polarizations along elevation and
uniform along azimuth. The peak of the Gaussian distribution
indicates the mean angle of arrival and the variance is an esti-
mate of the angular spread. In this paper the mean angle of ar-
rival used is 0 degrees and the angular spread is 30 degrees for
both polarizations. Since the XPR is not defined explicitly here
it is set to 1. This is currently the most common model used in
industry.
3) AAU: A mobile specific, statistical power model for out-
door to indoor propagation is defined in [28] as a result of ex-
tensive measurements in Aalborg city, Denmark. Since the later
analysis in this paper relies on similar overall scenario (outdoor
bases and indoor mobile in a room with a single window), this
channel model should give a very good correlation estimate. It
must be noted however that the power distribution model is de-
rived about 10 years before the current measurements took place
and is a statistical description of a much richer variety of sce-
narios—multiple floors, rooms and base stations at various dis-
tances. The main feature is a single direction of arrival from the
room openings (windows) for each polarization, with different
spreads. The model is derived based on work done in [32]. Since
the model is based on measurements the cross polarization ratio
is given explicitly as and the model parame-
ters used are the ones listed in [28].
Alternative, directive power models can be found in [33].
693
IV. MEASUREMENT DATA PROCESSING
The channel sounding was done for both bands, all trans-
mitters and four handsets simultaneously. Some parameters
include: spatial sampling frequency of ,
chosen for the indoor walking speed. The impulse response
sampling is done with corresponding
to impulse response (IR) resolution. The
total IR sampling time is (or
IR samples) corresponding to frequency domain resolution
of . The total measurement time was
, which in turn produces
spatial samples for the 16 meters measurement route for each
square side. The sounding frequencies are given in Table III
and the corresponding low and high band bandwidths are
and .
The result is a complex channel impulse response with about
2 GB per measurement. In total around 300 measurements were
done, each with four handsets at a time on the four sides of the
measurement square.
A. Correlation from Measurement Data
The algorithm for computing the correlation from the mea-
surement data is described in the following steps for each an-
tenna/Tx combination.
• Perform a Fast Fourier Transform on the complex
resulting in the frequency dependent channel Transfer
Function , with complex channel gain coefficients
.
• Take the absolute value for the selected frequencies (here
only the center sounding frequency is presented) resulting
in vectors of the narrowband signal envelopes.
• Pearson’s correlation coefficient is then computed as de-
fined in [34] for a large sample from every two envelope
vectors corresponding to the two antennas receiving from
a particular transmitter. The exact treatment for a discrete
sample is given in (6),
(6)
where and are the instantaneous discrete samples
(envelope samples in this case) and and are the means.
The summation index runs through the complete sample
length .
The result is a single number for the envelope correlation for
each antenna pair and transmitter. Essentially, the procedure is
the same as the one used in a mode-stirred chamber. The ad-
vantage of the sounding data is the inherent realism and true
propagation environment. With adequate channel power mod-
eling however, the mode-stirred chamber can be a significantly
cheaper and faster solution.
B. Advanced Signal Processing
It is well known that the correlation computation defined as
above is not a robust measure, see [35] with practicalities re-
lated to channels given in [19]. Slow fading variations due to
body shadowing for example [36], can create periodic mean
694 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
power variations within the measurement, equally for both an-
tennas, essentially violating the assumptions, under which (6)
is derived. This leads to a correlation number dominated by the
slow fading pattern rather than the fast fading and should be
avoided.
The solution is to perform a demeaning procedure [19]. For
the results presented here, simple moving average filter is used.
The chosen window has width of 50 measurement samples,
which corresponds to several wavelengths at the low band for
the spatial sampling used. The number of samples has been de-
termined by investigating some of the worst case measurement
examples, where the slow fading variation is very strong. De-
meaning was performed with various window sizes and the op-
timal number was selected corresponding to a value that gives
stable estimate. Taking few window samples more or less does
not influence the correlation estimation significantly. From a
physical point of view, since the route is 16 meters, but split
in four equivalent sections, the variation within each 4 meters is
expected to be similar. Assuming constant speed, the window
size should be smaller than the 300 samples for each 4 meter
section. Details on assumptions, procedures and practicalities
of performing demeaning can be found in [19].
It must be noted that in simulations there is no slow fading
effect. This step is essential when trying to make sense of mea-
surement data compared to simulations.
C. Channel Capacity Formulation
To compute the open loop channel capacity (channel state in-
formation (CSI) known only at the receiver) the common for-
mula found in [37] is used
(7)
where is the transmit power, is the number of transmit
antennas and are the eigen values of the matrix of
rank r and
(8)
where are the instantaneous complex channel coefficients
for a 2 2 MIMO system, which is the one used in this paper.
denotes Hermitian transpose, is the antenna index and is
the transmitter index. This way, (7) is used to compute the in-
stantaneous channel capacity and then the mean is taken for the
whole measurement. The resulting ergodic capacity values are
collected for all relevant cases (for example all measurements
with users).
The upper bound capacity of such system is given in [37] by
letting , and is the signal
to noise ratio of the normalized channel.
(9)
D. Power Normalization
Normalization of the H matrix, gives the freedom to select
arbitrary SNR. There are however various ways to normalize,
leading to different parameters being highlighted. Some possi-
bilities are the following.
1) Normalization to a Reference Handset: This is usually
also done for a reference use case—most often free space. This
normalization provides an easy way to compare overall capacity
performance between handsets. The mean power from all Tx/Rx
antenna combinations for the reference scenario is computed.
The complex channel coefficients of all handsets are then nor-
malized to the square root of this mean value. This way of nor-
malizing preserves the fading, correlation and branch power as
well as handset efficiency differences and user influence infor-
mation. Similar procedure is described in [38] as first normal-
ization option. The disadvantage of this method is however,
that since all capacity influencing parameters are preserved, it
is hard to determine, which variable exactly causes the capacity
to change in the case when there is indeed some difference be-
tween measurements or handsets. The different efficiencies of
the different handsets lead to lower or higher effective SNR
when computing (7) and the capacity changes accordingly. The
problem is even further deepened when users are introduced as a
variable, although on the other hand, this can be an indicator on
the capacity degradation relative to some mean body loss factor
for example.
2) Normalization to Each Individual Measurement With Av-
eraged Antenna Powers: In this case, for each measurement,
the power collected by both antennas and all transmitters is
averaged and then the channel coefficients are normalized to
the square root of that value. The branch power ratio, corre-
lation and fading information is preserved. However handset
efficiency information and the relative body loss effect due to
the users, are lost in the normalization. This corresponds to the
second method for normalization from [38].
3) Normalization to Each Individual Receiving Antenna: In
this case the normalization is done separate for each antenna
within a single measurement. Branch power ratio information is
thus lost, leaving only correlation as a free capacity influencing
parameter. The following normalization is used
(10)
where each instantaneous channel coefficient is divided by the
square root of the mean power received by antenna , from the
corresponding transmitter , , 2 and , 2. This is
somewhat artificial way of looking at capacity and the goal is
to isolate correlation as the only capacity determining factor.
It must be noted that the correlation for both receive but also
transmit antennas is present after this normalization method.
The transmit correlation is here computed by correlating the re-
ceived power on one antenna from two different transmitters,
with the same procedure given in Section IV-A. Since in this
case, the 2 2 MIMO scenario includes only transmitters from
the near base, the propagation path to the measurement room is
nearly the same, contributing to a higher, but overall constant
transmitter correlation. Few exceptions near the room opening
are observed.
4) Normalization to Each Individual Receiving Antenna With
Slow Fading Cleared: Note that the denominator in (10) is the
mean power including the slow fading. This preserves the local
YANAKIEV et al.: ON SMALL TERMINAL ANTENNA CORRELATION AND IMPACT 695

means within the measurement leading temporarily to lower of

higher effective SNR, different from that given in (7). This can
lead to the slow fading pattern contributing to the average ca-
pacity as well as the correlation. While correlation computed
with the slow fading present, makes little sense, the presence
of slow fading when computing capacity, can be interpreted as
temporary channel variations that a user can experience. This ef-
fect can be significant since the logarithm function in (7) is not
linear. To remove the effect, the same moving average filter, is
used as for the correlation

(11)

Fig. 5. Low band envelope correlation with slow fading removed for H1.

This paper uses the third and fourth methods for the computed
capacity results.

E. Statistical Analyses Details

The means, standard deviations and percentiles of all results

presented in the later sections are based on statistical samples
from the measurement data population, for the relevant cases.
For example the measured free space correlation for H1 is based
on 16 measurements, along the four sides of the room (Fig. 4).
The antenna correlation is computed for the two low and high
band transmitters on the near base leading to a statistical sample
size of 32 values for each band. These same transmitters have
later been used for MIMO channel capacity computation. In the
case with users, since the two grips are combined, the total sta-
tistical sample size is over 100. Same is done for the other hand-
sets.
Fig. 5 shows a plot of the received envelope correlation values
for the low band of H1 with slow fading removed. In the figure
BS1:Tx1/Tx2 stand for Base station 1, transmit antenna 1/2 as
listed in Table III. The numbers under the X-axis are the users Fig. 6. Probability Density Function (PDF) of the data in Fig. 5. Percentiles
are also shown.
and the sides of the square are given above the X-axis for each
user. The multiple points for each user/side combination are due
to repeated measurements and the combination of the two grips.
This particular sample has a mean value of and stan- V. DESIGN STAGE RESULTS
dard deviation of . The probability distribution of the The antenna correlation at the design stage is computed from
sample is shown in Fig. 6 and is similar to Gaussian. For rela- simulated radiation patterns using the (1)–(4) for the three se-
tively small sample sizes, for example free space only, the cu- lected handsets and power models. For the computations given
mulative density function (CDF) has been observed and the nor- in Table IV all handsets are defined in spherical coordinates,
mality assumption has been confirmed. All data samples statisti- aligned with the standard x,y,z coordinate system as defined in
cally analyzed in this paper are considered normally distributed. [39] and then the radiation patterns are rotated around the y axis
under . Since the AAU model is directive in azimuth, in
To be able to differentiate between means, which are nearly Table IV the low/mean/high values of the correlation coefficient
the same, statistical significance testing is used. The sample are given, as the handset is rotated around azimuth. Similar
variances are tested for equality with the robust Levene’s test. is done to estimate the mean effective gain (MEG) sensitivity to
In case the null hypothesis of equal variances is confirmed, the the pattern rotation in various power distribution models in [11].
sample means are compared in an ANOVA procedure. In case As an example, only rotation around the z axis (azimuth) of H1,
the variances are not equal, a modified two sample T-test pro- results in correlation variation about for low band
cedure for not equal variances is used. in the AAU power model. This is primarily due to the ideal mod-
In all cases significance level is used unless other- eling in the simulation tool and the perfectly symmetric antenna
wise noted. All procedures described above are given in detail placement on top and bottom of the handset. For high band on
in [35]. H4 since the antennas are asymmetrical. Further
696 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
TABLE IV
DESIGN STAGE ENVELOPE CORRELATION. AAU MODEL GIVEN WITH
LOW/MEAN/HIGH VALUES
rotation in 3D including elevation can lead to even larger differ-
ences (also for the Gaussian model) but is not of interest here
since all measurements are done at about the simulated . power. This becomes clearer when the slow fading effect is re-
As can be seen from Table IV the correlation can vary sig-
nificantly with the introduction of errors by the measurements
cables and RF chokes. Although the RF chokes help prevent the
cables from radiating, they are still a significant scatterer and the
cable position cannot be controlled accurately, especially during
field measurements with moving users. Furthermore, in the sim-
ulations, the cable positions have been optimized for least influ-
ence, leading them out in the middle of the handset along the
width, where the electric field is weakest. This is also a prac-
tical way to measure in an anechoic chamber afterwards. Even
with this, the cable introduces significant errors and the benefit
of optical link measurements is again highlighted. The different
channel power models also provide different results.
The reason for selecting these particular handsets for further
analysis are the correlation values in free space—one high (H1)
and one low (H2). At this stage of the design process one might
conclude that H2 would perform significantly better as it has
much lower correlation. However, already here, one can observe
the beneficial effect of the user in the simulation with the CTIA
hand or SAM-head phantoms, leading to a much lower H1 cor-
relation. This is also seen in the results published in [12]. Note
also that the high band correlation is not significant in either
case even for very closely spaced high band antennas as in H4.
It is important however to consider that the variation in the high
band can be significant.
VI. MEASUREMENT RESULTS
Correlation computed from the channel sounding data is pre-
sented in Table V. In all cases the statistics include all four sides
TABLE V
50%/95% PERCENTILE CORRELATION VALUES IN VARIOUS CASES
of the room and in the case of users all 12 of them with both grip
types.
It can be seen that the low frequency 95% percentile is at
for H1 in the case with slow fading not removed.
Also, the high band has a 95% percentile at . It is
curious to note that, while the low band seems to be comparable
to the design stage values the high band is underestimated. One
possible reason is the presence of a slow fading pattern in the
high band. Indeed, for the given dimensions of H1 the low band
antennas utilize the whole handset body as a radiator [40], re-
sulting in a rather omni-directional radiation pattern, while the
high band antennas are more space confined and also more di-
rective. Thus the high band antennas are more selective and have
higher probability of experiencing large variations in received
moved and the two correlation numbers are adjusted.
Based on these adjusted numbers one can conclude that only
the AAU power distribution model gives somewhat realistic
estimate at the design stage. Looking at the rotation in azimuth
variation, this is generally true for the high band and cases
with users as well. This is expected result since the model has
been developed specifically for mobile reception in a room
with openings. The 50% percentile is given in Table V to
show the estimated median of the samples for comparison with
the values given in Table IV. The higher overall high band
correlation observed in measurements is most probably due to
the non ideal de-meaning procedure. Indeed some 2 dB local
mean variation can be observed after the de-meaning. Limited
port isolation at the optical unit can also be a contributing factor
for errors at the low correlation range.
The users’ orientation on the different sides of the square can
be seen as the measurement equivalence of the azimuth rota-
tion done in Table IV for the AAU model. Analysis of the per
side statistical samples shows that indeed the different sides are
statistically different. Table VI lists the results for H1 and the
same is confirmed for all other cases. The variations observed
can only be explained by a directive power model. It is however
true that changing sides, changes the location in the room as well
and it is not possible to separate the two variables with the mea-
surements performed. Therefore this investigation is taken no
further.
It is also very significant to note that in the case with users, the
correlation is always significantly different from the free space
one, thus making the free space estimate inadequate for such
different use case. For the low band, slow fading removed case,
YANAKIEV et al.: ON SMALL TERMINAL ANTENNA CORRELATION AND IMPACT
TABLE VI
50%/95% PERCENTILES FOR H1 ALONG DIFFERENT SIDES
note that the user influence can be both advantageous as in the
case of H1, also seen at the design stage but also disadvanta-
geous as in the case of H2. For H1 this can be interpreted as the
user actively changing the radiation patterns by interacting with
the near fields and thus asymmetrically changing the antennas’
reception of the incoming power leading to de-correlation. In
the case of H2, since the handset is a clamshell and features an
RF choke at the hinge (implemented with a coil and capacitor),
the user mainly interacts with the bottom antenna. Assuming the
top antenna remains undisturbed and keeping in mind that the
correlation is already rather low in free space, the user influence
can be interpreted as in [5] where the absorption from the body
in some directions is seen as limiting factor for the utilization of
all differences in the radiation patterns.
In other words, if the radiation patters are different enough in
free space, leading to low correlation when weighted with the
power distribution, the addition of the user can only block cer-
tain incoming power directions, due to absorption in the tissue,
resulting in more correlated signals. Alternatively, if the patterns
are very correlated in free space, meaning that the whole phone
radiates, the addition of the user disturbs the near fields of both
antennas in an asymmetric way, leading to lower correlation.
Both effects are observed at the low band, while at the high
band neither is obvious from the correlation numbers in Table V.
Statistical analysis on the high band showed that the increased
correlation effect is true for H1 and the mean correlation, indeed
becomes significantly higher with the addition of the user. In
the H2 and H4 cases the correlation means are confirmed equal,
most probably due to the very asymmetric antenna location on
H4 and the large distance on H2. The de-correlation effect could
not be observed at the high band, since all high band antennas
were with low correlation. Finally form Table V one can con-
clude that correlation at the high band is rather low in all situ-
ations. The correlation numbers for the center, upper and lower
regions of the bands are within . while H2 has significantly higher capacity for slow fading
Other factors such as antenna coupling have also been
shown to influence the correlation [41]. The addition of the
user can certainly change those too. [41] also suggests that for
coupling has limited effect on correlation,
which is also shown in [42]. Since dynamic tracking of the
coupling during measurements was not possible, this could not
be studied. The coupling values of the handsets however, are
close to the 10 dB rule of thumb number and it should not be
a major issue.
697
TABLE VII
MEAN CHANNEL CAPACITY FOR OF A 2 2 MIMO
CONFIGURATION
VII. DISCUSSION
Looking at Table V it is not easy to predict the impact these
correlation values would have on the end metric—channel ca-
pacity. In free space case, it looks obvious that H1
will perform much worse than H2 ( ). In the case
with users however this is not so clear— for H1 vs.
for H2. It is therefore necessary to quantify the gain
of low correlated antennas compared to high correlated ones,
especially from a design effort point of view.
A. Capacity Results
Using the normalizations described in Section IV-D-3 and
Section IV-D-4, the channel capacity is given in Table VII.
Again removing the slow fading has some effect but not as
significant as for the correlation. Overall the results relative
to each other are the same in both cases. It is important to
emphasize that these capacity numbers cannot be in any way
seen as expected for such handsets/environments as they are
produced by a very artificial normalization method with a
lot of idealizations implied. Furthermore the normalization
is not ideal and other capacity influencing effects could have
contributed to the results. The main goal is to compare them
relative to one another for the purposes of evaluating envelope
correlation influence and not to take them as absolutes.
In all cases the capacity is generally linked to
Table V—higher correlation leads to lower capacity and vice
versa. The most significant result is that with the addition of
the user on the low band, two otherwise separable in free space
performance handsets, become nearly indistinguishable—H1
and H2 with users. Statistical comparison of the two samples
shows that they are significantly different at
significance level, while being the same for with
p-value of for the case with slow fading present.
Same conclusion is valid when the slow fading is removed.
Similarly, at the high band H1 and H4 are statistically equal,
present, which disappears when the slow fading is removed
indicating that the local mean power is what causes the
statistical difference. This is most probably due to the few
extreme measurements near the windows.
When comparing the capacity in the free space case with the
capacity in the case with users for each handset, H2 low band is
the only one where the MIMO capacity is observed to be equal.
In all other cases the addition of the user leads to lower capacity
when the free space correlation is low, or higher capacity when
698
the free space correlation is high. Again some deviations due to
imperfect demeaning, normalization or not accounted for cou-
pling can be expected.
A very important note here is that whenever two values are
said to be significantly different, the statistical meaning of the
phrase is used. For example, H2 has lower correlation than H1
in all cases, thus reaching higher low band capacity with users
of 6.7 [bits/s/Hz], however H1 is very close with 6.6 [bits/s/Hz].
Statistically significant difference in the mean achieved capacity
of –0.2 [bits/s/Hz], is certainly not something signif-
icant from a design effort point of view.
Another interesting thing to note is that, the maximum ca-
pacity difference when using MIMO, observed between H1 low
band and H2 high band in free space, is only
for the two extreme correlation cases with slow
fading removed. Comparing the more realistic case of low band
only where the problem is expected in the first place, it is seen
that the maximum difference is and that
is only in free space.
The SISO capacity at the low band for H4 is also computed.
As expected, since correlation is not relevant in that case and
the power is normalized, there is no difference between the free
space and the case with users. Same is observed for all other
handsets when used in SISO mode. It also shows that even for
the high correlation low band case, some 30–40% improvement
can be achieved with the use of MIMO schemes. Of course there
is even further improvement when going to the very low corre-
lated high band.
VIII. CONCLUSIONS
The paper analyses the adequacy of correlation estimation at
the design stage, comparing it to the actual performance of the
MIMO capable mobile in the real environment with the pres-
ence of diverse users and in free space. The main conclusions
are that free space correlation computed for the isotropic envi-
ronment yields inadequate results, both when compared to free
space measurements and especially when directly used to pre-
dict the overall performance in the presence of the user. Realistic
power distribution models have to be chosen, when computing
correlation and the user influence must be included as well.
With the results presented, regarding the user influence on
the antennas and correlation, it is doubtful that it is worth in-
vesting much design effort in optimizing for low correlation an-
tennas—certainly not for free space and with isotropic environ-
ment. The richness of the user interaction brings the correlation
numbers of two handsets—one with low and one with high cor-
relations—very close to each other. Evaluating the capacity as
an end metric confirms that little can be gained by making low
correlation antennas without accounting for the user’s presence.
Large variations can be found in diverse MIMO figures of merit
when the user’s presence is accounted for. It is thus concluded
that other factors such as branch power ratio, power absorbed
and overall mean effective gain will have much more signifi-
cant effect on the perceived channel capacity and system diver-
sity than just antenna correlation.
REFERENCES
[1] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.
New York: Cambridge Univ. Press, 2005.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[2] G. Pedersen, “Antennas for Small Mobile Terminals,” Ph.D. disserta-
tion, Aalborg University, Aalborg, Denmark, 2003.
[3] T. Taga, “Analysis for mean effective gain of mobile antennas in land
mobile radio environments,” IEEE Trans. Veh. Technol., vol. 39, no. 2,
pp. 117–131, May 1990.
[4] R. Vaughan and J. Andersen, “Antenna diversity in mobile communi-
cations,” IEEE Trans. Veh. Technol., vol. 36, no. 4, pp. 149–172, Nov.
1987.
[5] G. F. Pedersen and J. B. Andersen, “Handset antennas for mobile com-
munications: Integration diversity, and performance,” Rev. Radio Sci.
1996–1999, pp. 119–137, 1999, Oxford Univ. Press.
[6] G. Pedersen, J. Nielsen, K. Olesen, and I. Kovacs, “Measured varia-
tion in performance of handheld antennas for a large number of test
persons,” in Proc. 48th IEEE Veh. Technol. Conf., May 1998, vol. 1,
pp. 505–509.
[7] G. Pedersen and S. Skjaerris, “Influence on antenna diversity for a
handheld phone by the presence of a person,” in Proc. 47th IEEE Veh.
Technol. Conf., May 1997, vol. 3, pp. 1768–1772.
[8] G. Pedersen, J. Nielsen, K. Olesen, and I. Kovacs, “Antenna diver-
sity on a UMTS handheld phone,” presented at the 10th IEEE Int.
Symp. Personal, Indoor and Mobile Radio Commun., Osaka, Japan,
Sep. 1999.
[9] K. Ogawa and T. Matsuyoshi, “An analysis of the performance of a
handset diversity antenna influenced by head, hand, and shoulder ef-
fects at 900 MHz .I. Effective gain characteristics,” IEEE Trans. Veh.
Technol., vol. 50, no. 3, pp. 830–844, May 2001.
[10] K. Ogawa, T. Matsuyoshi, and K. Monma, “An analysis of the perfor-
mance of a handset diversity antenna influenced by head, hand, and
shoulder effects at 900 MHz .II. Correlation characteristics,” IEEE
Trans. Veh. Technol., vol. 50, no. 3, pp. 845–853, May 2001.
[11] J. O. Nielsen and G. Pedersen, “Mobile handset performance evalu-
ation using radiation pattern measurements,” IEEE Trans. Antennas
Propag., vol. 54, no. 7, pp. 2154–2165, Jul. 2006.
[12] V. Plicanic, B. K. Lau, A. Derneryd, and Z. Ying, “Actual diversity
performance of a multiband diversity antenna with hand and head ef-
fects,” IEEE Trans. Antennas Propag., vol. 57, no. 5, pp. 1547–1556,
May 2009.
[13] S. Blanch, J. Romeu, and I. Corbella, “Exact representation of antenna
system diversity performance from input parameter description,” Elec-
tron. Lett., vol. 39, no. 9, pp. 705–707, May 2003.
[14] P. Hallbjorner, “Accuracy in reverberation chamber antenna correla-
tion measurements,” in Proc. IWAT, Mar. 2007, pp. 170–173.
[15] J. Welinder, L. Fast, T. Bolin, and L. Manholm, “The multi path simu-
lator for over the air testing,” in Proc. EuCAP, Apr. 2010, pp. 1–4.
[16] A. Yamamoto, T. Sakata, T. Hayashi, K. Ogawa, J. O. Nielsen, G. F.
Pedersen, J. Takada, and K. Sakaguchi, “Effectiveness of a fading em-
ulator in evaluating the performance of MIMO systems by comparison
with a propagation test,” in Proc. EuCAP, Apr. 2010, pp. 1–5.
[17] T. Sakata, A. Yamamoto, K. Ogawa, and J.-I. Takada, “MIMO channel
capacity measurement in the presence of spatial clusters using a fading
emulator,” in Proc. IEEE 20th Int. Symp. Personal, Indoor Mobile
Radio Commun., Sep. 2009, pp. 97–101.
[18] J. Valenzuela-Valdes, A. Martinez-Gonzalez, and D. Sanchez-Her-
nandez, “Emulation of MIMO nonisotropic fading environments with
reverberation chambers,” IEEE Antennas Wireless Propag. Lett., vol.
7, pp. 325–328, 2008.
[19] R. Vaughan and J. B. Andersen, “Channels, propagation and antennas
for mobile communications,” IEE Electromagnetic Waves Series, vol.
50, 2003.
[20] B. K. Lau, J. Andersen, G. Kristensson, and A. Molisch, “Impact of
matching network on bandwidth of compact antenna arrays,” IEEE
Trans. Antennas Propag., vol. 54, no. 11, pp. 3225–3238, Nov. 2006.
[21] A. Taflove and S. C. Hagness, Computational Electrodynamics: The
Finite-Difference Time-Domain Method, 3rd ed. Norwood, MA:
Artech House, 2005.
[22] W. A. T. Kotterman, G. F. Pedersen, and P. Eggers, “Cable-less mea-
surement set-up for wireless handheld terminals,” in Proc. PIMRC,
Sep. 2001, pp. B112–B116.
[23] C. Icheln, J. Krogerus, and P. Vainikainen, “Use of balun chokes in
small-antenna radiation measurements,” IEEE Trans. Instrum. Meas.,
vol. 53, no. 2, pp. 498–506, Apr. 2004.
[24] C. Icheln, J. Ollikainen, and P. Vainikainen, “Reducing the influence of
feed cables on small antenna measurements,” Electron. Lett., vol. 35,
no. 15, pp. 1212–1214, Jul. 1999.
[25] B. Yanakiev, P. Eggers, G. Pedersen, and T. Larsen, “Assessment of
the physical interface of UHF passive tags for localization,” in Proc.
1st Int. EURASIP Workshop on RFID Technol., Vienna, Austria, Sep.
2007, pp. 25–28.
YANAKIEV et al.: ON SMALL TERMINAL ANTENNA CORRELATION AND IMPACT
[26] B. Yanakiev, J. O. Nielsen, and G. F. Pedersen, “On small antenna mea-
surements in a realistic MIMO scenario,” in Proc. EuCAP, Apr. 2010,
pp. 1–5.
[27] “Method of Measurement for Radiated rf Power and Receiver Perfor-
mance,” CTIA, Oct/ 2009 [Online]. Available: http://www.ctia.org/,
CTIA, Tech. Rep., CTIA Certification Test Plan for Mobile Station
Over The Air Performance. rev. 3.0
[28] M. Knudsen and G. Pedersen, “Spherical outdoor to indoor power spec-
trum model at the mobile terminal,” IEEE J. Sel. Topics Quantum Elec-
tron., vol. 20, no. 6, pp. 1156–1169, Aug. 2002.
[29] J. Pierce and S. Stein, “Multiple diversity with nonindependent fading,”
Proc. IRE, vol. 48, no. 1, pp. 89–104, Jan. 1960.
[30] M. Narandzic, C. Schneider, R. Thoma, T. Jamsa, P. Kyosti, and X.
Zhao, “Comparison of SCM, SCME, and WINNER channel models,”
in Proc. VTC, Apr. 2007, pp. 413–417.
[31] J. Sánchez-Heredia, J. Valenzuela-Valdés, A. Martínez-González, and
D. Sánchez-Hernández, “Emulation of MIMO Rician-fading environ-
ments with mode-stirred reverberation chambers,” IEEE Trans. An-
tennas Propag., vol. 59, no. 2, pp. 654–660, Feb. 2011.
[32] J. Andersen and K. Pedersen, “Angle-of-arrival statistics for low res-
olution antennas,” IEEE Trans. Antennas Propag., vol. 50, no. 3, pp.
391–395, Mar. 2002.
[33] K. Kalliola, K. Sulonen, H. Laitinen, O. Kivekas, J. Krogerus, and P.
Vainikainen, “Angular power distribution and mean effective gain of
mobile antenna in different propagation environments,” IEEE Trans.
Veh. Technol., vol. 51, no. 5, pp. 823–838, Sep. 2002.
[34] Mathworks May 2010, Matlab [Online]. Available: http://www.math-
works.com/help/toolbox/stats/corr.html
[35] D. Montgomery, Design and Analysis of Experiments: 7th Interna-
tional Student Edition. New York: Wiley, 2009.
[36] I. Kashiwagi, T. Taga, and T. Imai, “Time-varying path-shadowing
model for indoor populated environments,” IEEE Trans. Veh. Technol.,
vol. 59, no. 1, pp. 16–28, Jan. 2010.
[37] E. Telatar, “Capacity of multi-antenna Gaussian channels,” Eur. Trans.
Telecommun. vol. 10, no. 6, pp. 585–595, 1999 [Online]. Available:
http://dx.doi.org/10.1002/ett.4460100604
[38] H. Ozcelik, M. Herdin, R. Prestros, and E. Bonek, “How MIMO ca-
pacity is linked with single element fading statistics,” in Proc. Int. Conf.
Electromagn. Adv. Appl., Sep. 2003, pp. 775–778.
[39] C. A. Balanis, Antenna Theory: Analysis and Design, 2nd ed. New
York: Wiley, 1997.
[40] O. Kivekäs, J. Ollikainen, T. Lehtiniemi, and P. Vainikainen, “Effect
of the chassis length on the bandwidth, SAR, and efficiency of internal
mobile phone antennas,” Microw. Opt. Technol. Lett. vol. 36, no. 6,
pp. 457–462, 2003 [Online]. Available: http://dx.doi.org/10.1002/mop.
10789
[41] J. Thaysen and K. B. Jakobsen, “Design considerations for low antenna
correlation and mutual coupling reduction in multi antenna terminals,”
Eur. Trans. Telecommun. vol. 18, no. 3, pp. 319–326, 2007 [Online].
Available: http://dx.doi.org/10.1002/ett.1111
[42] I. Salonen, C. Icheln, and P. Vainikainen, “Pattern correlation and mis-
match in two-element antenna arrays,” Microw. Opt. Technol. Lett. vol.
48, no. 1, pp. 41–43, 2006 [Online]. Available: http://dx.doi.org/10.
1002/mop.21254
Boyan Yanakiev received the B.S. degree in physics
from Sofia University, Bulgaria, in 2006, and the
M.S. degree in wireless communication from Aal-
borg University, Denmark, in 2008, where he is
currently pursuing the Ph.D. degree.
His current position is as an industrial Ph.D.
student in cooperation with Molex Antenna Business
Unit. His primary interests are in the area of small
integrated mobile antennas, optical antenna measure-
ment techniques and radio channel measurements.
He has been involved in the design and development
of multiple RF-to-optical convertors, for onboard handset measurements.
699
Jesper Ødum Nielsen received the M.S. degree in
electronics engineering in 1994 and the Ph.D. degree
in 1997, both from Aalborg University, Denmark.
He is currently employed at Department of Elec-
tronic Systems, Aalborg University, where his main
areas of interests are experimental investigation
of the mobile radio channel and the influence on
the channel by mobile device users. He has been
involved in channel sounding and modeling, as
well as measurements using the live GSM network.
In addition he has been working with handset
performance evaluation based on spherical measurements of handset radiation
patterns and power distribution in the mobile environment.
Morten Christensen was born in 1973. He received
the M.Sc. degree in electrical engineering from Aal-
borg University, Denmark in 1998.
In 1998, he joined Bosch Telecom A/S, Pandrup,
Denmark (acquired by Siemens Mobile Phones in
2000), where he designed integrated antennas for
mobile terminals. In 2006, he joined Motorola A/S,
Mobile Devices Aalborg where he was heading
the EMC and Antenna department. He is now with
Molex Antenna Business Unit responsible for the RF
Research activities. His areas of interests includes
handset antenna design, performance evaluation methods and radio propagation
models.
Gert Frølund Pedersen was born in 1965. He re-
ceived the B.Sc.E.E. degree (hons), in electrical engi-
neering from College of Technology, Dublin, Ireland,
and the M.Sc.E.E. and Ph.D. degrees from Aalborg
University, Aalborg, Denmark, in 1993 and 2003.
He has been employed by Aalborg University
since 1993 where he is now a Full Professor heading
the Antenna, Propagation and Networking Group
and is also the Head of the Doctoral School on
Wireless which has enrolled approximately 100
Ph.D. students. His research has focused on radio
communication for mobile terminals especially small antennas, diversity
systems, propagation and biological effects and he has published more than 75
peer reviewed papers and holds 20 patents. He has also worked as consultant
for developments of more than 100 antennas for mobile terminals including the
first internal antenna for mobile phones in 1994 with lowest SAR, first internal
triple-band antenna in 1998 with low SAR and high TRP and TIS, and lately
various multi antenna systems rated as the most efficient on the market. He
has been one of the pioneers in establishing over-the-air measurement systems.
The measurement technique is now well established for mobile terminals with
single antennas and he was chairing the COST2100 SWG2.2 group with liaison
to 3GPP for over-the-air test of MIMO terminals.
700 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Compensating for Non-Linear Amplifiers in MIMO
Communications Systems
Seyed Alireza Banani, Student Member, IEEE, and Rodney G. Vaughan, Fellow, IEEE
Abstract—MIMO systems, including those for mobile channels,
are well studied as combinations of antennas, other microwave
components, and sophisticated digital signal processing. The
communications performance is based on many assumptions,
including that of linear channels. But in practice, key compo-
nents such as microwave amplifiers are non-linear. This paper
addresses MIMO compensation of non-linear amplifiers at both
the transmitter and the receiver. The in-band distortion is mod-
eled using Busgang’s theorem. Based on this model, a decision
algorithm obtains an optimized initial estimate of the transmitted
symbols which are used for subsequent channel estimation. The
final channel matrix estimate is obtained through two alternative
methods: statistical linearization; or analytical linearization.
The approach is blind in the sense that no pilots are used for
sounding the channel. Performance is evaluated by simulation,
allowing comparison with the benchmark of coherent detection
with perfect channel knowledge. We establish the maximum SNR
value for which the nonlinearities are essentially compensated.
The performances of the two methods converge as the MIMO
system becomes linear. The linear case allows fair comparison
with known results of the conventional decision-directed Kalman
filtering, and two pilot-aided systems. Finally we compare the
performance with a look-up table technique for compensating the
non-linearity.
Index Terms—Blind channel estimation, MIMO communica-
tions, mobile channels, non-linear amplifier, non-linear modeling
for MIMO, unscented transformation.
I. INTRODUCTION
T HE use of multiple antennas at both the transmitter and
receiver in wireless communications provides increased
spectral efficiency compared to single antenna systems [1]. Nev-
ertheless, there is not a significant commercial presence for full-
MIMO systems that use large numbers of elements, which is
where the potential increase of spectral efficiency becomes dra-
matic. One reason for this lack of commercial uptake is that the
extra hardware for many-element systems is too expensive for
commercial viability. Being able to use low cost amplifiers is
particularly important.
The performance of MIMO systems has been studied exten-
sively, and simulation is the basic tool for estimating the com-
munications performance. Most studies assume that both the
Manuscript received June 13, 2010; revised June 21, 2011; accepted July 02,
2011. Date of publication October 25, 2011; date of current version February
03, 2012. This work was supported by the Natural Sciences and Engineering
Research Council (NSERC) of Canada.
The authors are with the School of Engineering Science, Simon Fraser Uni-
versity Canada (e-mail: sab8@sfu.ca; rvaughan@sfu.ca).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TAP.2011.2173441
0018-926X/$26.00 © 2011 IEEE
transmitter and receiver amplifiers are operated in the linear
region and so that the MIMO channel matrix is linear. A mi-
crowave amplifier can indeed operate as a quasi-linear device
under small signal conditions, and low signal distortion is pos-
sible with low power efficiency and higher cost (because they
are higher power) amplifiers. It follows that there is a tradeoff
between power efficiency and the resulting signal distortion.
In communications systems, this tradeoff is governed by the
need to limit the out-of-band interference in the radio spectrum.
Nonlinear distortion at the transmitter causes interference both
inside and outside the signal bandwidth. The in-band compo-
nent determines a degradation of the communications perfor-
mance, often expressed as bit-error rate (BER) [2], [3], whereas
the out-of-band component affects users in adjacent frequency
bands.
For more powerful MIMO systems (i.e., those with more an-
tenna elements) to emerge as economically viable, there is a
need for highly efficient power amplifiers, especially for bat-
tery-powered terminals. In other communications systems, such
as satellite, the payload weight, including the on-board high
power amplifiers (HPAs) is critical, and the amplifiers must run
at high efficiency. In these applications, distortion compensa-
tion is also possible using signal processing.
At the receiver, the low noise amplifiers (LNAs) are key com-
ponents because their gain and noise tend to dominate the sen-
sitivity. In fact, the LNA design involves many tradeoffs. These
are between noise figure, gain, linearity, impedance matching,
power dissipation and cost. With large-dimensioned MIMO sys-
tems where the capacity efficiency, with sufficient assumptions,
becomes proportional to the number of antennas, a large number
of high performance LNAs are required. This can make the
cost prohibitive, and lower cost (with greater distortion) LNAs
are always a pragmatic solution. However, the impact of their
non-linearities will need to be compensated. The cost for the
compensation using signal-processing is relatively low in the
sense that the digital processors are in place anyway and the
extra processing is relatively modest.
Nonlinear LNAs and their problems appear in many applica-
tions. For example, designers strive to minimize the intermod-
ulation distortion at the receiver by a minimum of nonlinear
LNA cascades [4], [5]. Also, to enable consumer products (e.g.,
GPS, etc.), an integrated receiver should minimize the number
of off-chip components, particularly the number of passive fil-
ters which are relatively expensive. These considerations mo-
tivate research into highly integrated CMOS solutions which
typically feature nonlinearity in the LNAs [6], [7]. Another ex-
ample is satellite diversity (a form of MIMO, using many PAs),
used for fading channels in low earth orbit systems [8]. These
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS
issues motivate research into the impact of nonlinear amplifiers
in MIMO systems.
A related problem in communications is estimating the
channel in order to deploy signal processing for reliable
communications. The time-varying MIMO channel has to
be estimated and this is undertaken either by using pilot
symbols or in a blind manner. Many channel estimation
techniques for linear MIMO systems have been reported.
Examples of pilot-aided systems are [9]–[11]. The cost of
using training-based approaches includes: a reduced payload
bandwidth because the pilots bite into the communications
bandwidth; and the added complexity of the implementation.
Also, the training symbols can produce inaccurate channel
estimates owing to the limited duration and number of training
intervals available for time-varying channels [12].
Blind techniques have become topical in research, because
of the unique advantage they offer, viz., the channel is effec-
tively sounded without biting into the communications band-
width. In blind and semi-blind techniques, the channel estima-
tion is developed based on second or higher-order statistics of
the fading process, see, for example, [13]–[32], and other refer-
ences too numerous to list. However, most blind methods that
employ higher order statistics typically require a large number
of data symbols, and a shortfall of symbols causes poor con-
vergence which means poor channel estimates and decreased
channel efficiency. The convergence problems associated with
blind techniques can be avoided by using a semiblind technique
[20]–[25], i.e., employing a reduced number of training sym-
bols together with blind statistical information.
One semi-blind approach for identification of the main
eigenmode, without estimating the channel matrix itself, is
presented in [23]. Two whitening-rotation-based algorithms for
semi-blind estimation of the flat MIMO channel are presented
in [24] and [25]. Such estimation procedures arise naturally in
the ICA-based source separation [26]. The use of higher-order
statistics based ICA is widespread in multiuser detection, e.g.,
[27], [28]. The main advantage of ICA techniques is that, under
mild mathematical conditions (independence of the sources),
signal recovery is guaranteed regardless of the source con-
stellation and spectral characteristics [29]. But this guarantee
is for static or slow-changing channels only. Channel estima-
tion/tracking based on Kalman/particle filtering is also well
established, e.g., [30], [31]. These are based on the tenet that
the recursive least squares (RLS) algorithm and the Kalman
filtering algorithm are both better than the LMS algorithm, in
convergence rate and tracking capability. Kalman filtering has
long been used to extend forms of the recursive least-squares
(RLS) algorithm which tracks better than the standard RLS and
LMS forms [32].
In this paper, a novel approach for joint blind channel es-
timation and data recovery for MIMO systems with nonlin-
earity in both transmitter and receiver amplifiers, is presented.
The mobile MIMO channel considered is time-selective and has
Rayleigh flat fading, i.e., it is narrowband—which is where the
power of MIMO dominates other communications techniques
for gaining spectral efficiency. Bussgang’s theorem [33], [34] is
used to model the amplifier-induced nonlinearity in the received
signal. The model has time-varying coefficients which depend
701
on both the data and channel. Using this model, a primary data
vector estimate of the transmitted signal is obtained based on the
constrained linear minimum mean square error (LMMSE) cri-
terion. Then, the channel matrix is estimated/tracked using two
alternative, recursive methods: statistical linearization via un-
scented transformation; or analytical linearization which results
in a nested iterative scheme for updating the channel matrix es-
timate. Finally, the transmitted data vector is recovered using
the channel matrix estimates.
This extends the work in [35] where the nonlinearity is in-
troduced only at the receiver amplifier. The main thrust of this
paper is to introduce and describe two new blind channel es-
timation techniques and evaluate the performance of the sys-
tems with nonlinear amplifiers at both the transmitter and the
receiver. For a nonlinear system, we must use the simulations
to compare the results with the benchmark performance of co-
herent detection with perfect channel state information (CSI) at
the receiver. We also compare the results of the two lineariza-
tion techniques with a look-up table (LUT) approach to the
nonlinear problem. The statistical/analytical linearization ap-
proaches are marginally better for low SNR (the usual region for
wireless) but the LUT is better for high SNR. In addition, for the
linear system, the performance comparison is made with the re-
sult of the conventional decision-directed Kalman filtering and
two pilot-aided systems already known in the signal-processing
community. Here, improved performance is observed over the
known techniques. The robustness of the proposed blind tech-
nique to time variations of the channel is also quantified and
compared with that of conventional channel estimation tech-
niques. Finally, the impact of assumptions in the channel mod-
eling is quantified using simulation, offering a feel for the per-
formance with time-correlation coefficient mismatch between
the channel model and the assumed model at the receiver.
The rest of the paper is organized as follows. Section II de-
scribes the system model, and this, and the other sections, are
couched in terms of communications signal processing nota-
tion for MIMO. The new blind estimation technique is formu-
lated in Section III and the simulation results are presented in
Section IV. Section V concludes the paper.
The notation is conventional, as follows. Symbols for
matrices (in capital letters) and vectors are in boldface. The
notations and stand for conjugate transpose,
transpose, and complex conjugate, respectively. is the iden-
tity matrix and denotes expectation. Also, in order to
distinguish the amplifiers at the transmitter and the receiver, the
sub-(super-) script “T” refers to the amplifiers at the transmitter
side and the sub-(super-) script “R” refers to the amplifiers at
the receiver.
II. NONLINEAR MIMO SYSTEM MODEL
The baseband equivalent representation of a non-linear
MIMO system with transmit and receive antennas em-
ploying a spatial multiplexing scheme is displayed in Fig. 1.
At the transmitter, data is picked up from a constellation set
with set size and the uncoded data stream is demultiplexed to
branches at each symbol time with symbol time duration .
Then data symbols are passed through transmit amplifiers (each
with nonlinear function ) before launching from transmit
702 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Fig. 1. Baseband representative of nonlinear MIMO system.

antennas. Assuming a Rayleigh flat fading MIMO channel, the Gaussianity of the inputs of the receive amplifiers is guaranteed.
received signal from the th antenna is modeled as (We note, therefore, that the formulation presented here is not
applicable to 16-QAM, etc.) Under such a condition,
and the receive amplifier output can be represented by [33]

(3)
(1)
where is an arbitrary deterministic complex coefficient (ex-
where and pressed as a function, below), and is suitably introduced
is the nonlinear function introduced by the receiver additive noise term. It is desirable to have zero-mean noise, un-
amplifier. The vector is the transmit signal vector correlated to the input process . This can be achieved by
with i.i.d. data symbol entries each with variance , and is setting as [3]
the zero mean additive white Gaussian noise with variance .
(In practice, for a terrestrial system, this noise is dominated by (4)
interference from other users, and other systems, i.e., the inter-
ference is assumed here to be Gaussian.) is the th row of
channel matrix , comprising Denoting , the derivatives appearing
i.i.d. complex Gaussian entries, , each in (4) are formulated as
with variance 1; index is the time index showing that at each
iteration (symbol time) the channel matrix is changing.
For both amplifiers’ nonlinearity ( and ), we con-
sider the general memoryless AM/AM and AM/PM characteris-
tics [36], [37]. In particular, denoting a transmitter/receiver am-
plifier complex input as , the
signal at the output of a nonlinear block becomes
(5)

(2) and as a result, we obtain

where and are the real functions for (6)

AM-to-AM and AM-to-PM conversions respectively. The
super-script T/R in and means that (2) is Also, the variance of the noise term in (3) can be obtained as
written in the general format and holds for any amplifier at the
transmitter or receiver with AM/AM and AM/PM characteris-
tics. (7)
Based on Bussgang’s theorem [33], [34], the output of a non-
linear memoryless amplifier excited by a Gaussian distributed Applying Bussgang’s theorem to (1), we get
signal can be represented by the scaled version of the original
signal plus an additive noise term. Here, we make use of Buss-
gang’s theorem to approximate the nonlinearity introduced at
the receiver side, i.e., , with a linearized model having
time-varying coefficients. By using the constant amplitude sig-
naling with , e.g., -PSK, however, the (8)
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS 703

where we have substituted with its instantaneous value
statistics, i.e., the correlation function values, relative to the a
priori information assumed by the receiver.
On reception of the th data symbol, the receiver first em-
ploys a new decision algorithm to obtain a primary data vector
estimate of the transmitted signal. The detail is described below.

A. Decision Algorithm
Having the unbiased MMSE estimate of the channel matrix
(9)
coefficients, , and the corresponding error variances
associated with the estimation process, , obtained
since calculating the expectation in (6) may be difficult. from the previous symbol interval (this is a standard type of as-
In order to recover the transmitted data vector from sumption in deriving iterative algorithms), the channel matrix
, the channel matrix estimation process at the th symbol time can be expressed
has to be estimated. by
The idea is to estimate the channel matrix in a blind manner (11)
at each symbol time without making use of pilot symbols or
training sequences. where and is a matrix with
zero-mean independent white Gaussian noise entries having the
III. BLIND CHANNEL ESTIMATION corresponding variances , and
Many authors have tried to approximate the time variation of also (12), shown at the bottom of the page.
the fading channel by different dynamic models depending on Moreover, the optimal channel coefficients’ linear predic-
the application. However, the results in [38] have shown that tions, given that the channel follows the AR(1) model of
the first-order autoregressive model provides a sufficiently ac- (10), but ignoring the addition of received data measurements
curate model for time-selective fading channels. Therefore, the , are
relation between and can be approximated by
(13)
(10)
with the corresponding prediction error variances
where is the value of the channel coefficients’ auto-
correlation function, , evaluated at and (14)
is a matrix with zero-mean, independent, white Gaussian noise
entries each with variance 1. When the fading statistics are un- From (10)(13) and the set of (8) in matrix form, the received
known, can be usually estimated from the data in a vector is approximated as
training-assisted mode or decision-directed mode [39]. How-
ever, we are not trying to estimate this statistic here, and for
simulation, it is taken as where is the zero-
order Bessel function of the first kind and is the maximum (15)
Doppler frequency. This form encompasses a commonly used,
where
but major, assumption—a 2D-omnidirectional incident uncorre-
lated power distribution and 2D-omnidirectional antennas. The
exact form of the correlation function is not important here in
the sense that the algorithm below uses only one sample value
of it. In practice, if this correlation sample value changes with
time, it will do so slowly, and this allows time to track its esti-
mate for its application in this algorithm. A sensitivity analysis
(shown below) indicates that the proposed blind technique is
robust for a moderate mismatch of the channel’s second-order (16)

(12)
.. .. .. ..
. . . .
704 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

in which

(17) (23)

In calculating , the noise variance is

(18) substituted by its instantaneous value since calculating
the expectation in (7) is difficult. It is worth noting
In the above expressions, the notation that is a function of which, in turn, depends on
is used to emphasize that . Consideration of
the noise vector and matrix depend on these in the above equations makes the new decision algorithm
, and . The equation of (15) can be yield a more accurate primary symbol data vector estimate .
re-written as
B. Channel Estimation
Having the primary data vector estimate at hand, the
channel matrix estimate can now be refined in the form of two,
alternative methods: using statistical linearization via unscented
(19) transformation; or using analytical linearization which results in
a locally (nested) iterative scheme for updating the channel ma-
trix estimate. Both of these methods are presented below.
where . In the arrangement of (19), the key idea 1) Statistical Linearization: Here, the knowledge gained
of introducing is to treat it as an unknown deterministic vari- from observing the measurements
able and we try to estimate it. The receiver searches over the
are used to refine the predicted channel vector
-dimensional transmit data vector constellation set
.
to see which candidate of yields
At first, the new set of observations is generated as
the MMSE estimate of according to the observations con-
strained by .
The primary data vector estimate is

(20)
(24)

where is the unbiased LMMSE of obtained by The Unscented Transformation (UT) [40], [41] is to handle
the nonlinearities in (24). However, the standard UT is charac-
(21) terized with real-valued random variables. As a result, to fit the
UT principles to our problem involved with complex random
variables, a summary of UT is described below.
with
Consider propagating a -dimensional real-valued random
vector with mean and covariance , through an arbitrary
nonlinear function , to produce a random
variable . A set of points, called sigma points,
(22) are generated by the following algorithm whose sample mean
and sample covariance are and , respectively:
and is the noise covariance matrix
calculated at . According to (12) and since the elements
in are independent, matrix is a diagonal matrix
with th diagonal element as

(25)

where is a scaling parameter such that

, and is the th column of the matrix square
root of , and is the weight that is associated with
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS 705

the th point. The weights are normalized; that is, they satisfy sigma points are passed through the nonlinear function
. The set of samples chosen by (25) are guaran- to yield the prediction measure-
teed to have the same sample mean, covariance, and all higher, ment
odd-ordered central moments, as the distribution of the random
vector . The matrix square root and affect the fourth and (34)
higher order sample moments of the sigma points [41]. Now
each sigma point is propagated through the nonlinear function
, with variance

(26)

and the first two moments of are computed as follows

(27) (35)

The problem of refining the channel vector estimate

(28)
, is finalized with a new estimation problem as fol-
lows. We wish to linearly estimate the random vector
To fit the UT principle to our problem, we stack all the real as in (29) with mean and covariance matrix
and imaginary parts of into
, from
vector as (29), shown at the bottom of the page.
Considering the above arrangement along with (13)–(14), measurement in the minimum mean
would be a random vector with mean and square sense ( is the innovation term). The LMMSE esti-
diagonal covariance matrix as (30) and (31), shown at mator is [42]
the bottom of the page.
With the assumption of wide sense stationary uncorrelated
scattering (WSSUS), and are uncorre-
lated random variables with nearly equal variances. (Asymp- (36)
totically they are equal). As a result, for a complex channel co-
efficient, , we get where

(32)

Thus, we approximate in (31) by (33), shown at the

bottom of the page. The associated estimation error covariance matrix is
Now, in (24) is taken as the nonlinear function
with the corresponding output, and sigma
points, are generated according to (25). These (37)

(29)

(30)

(31)

(33)
706 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012

Finally, which is the th element of vector is with and

obtained as
(43)

and the associated measurement noise variance.

The strategy of finding a minimum is to use Newton-Raphson
iteration starting from . At the beginning of the th it-
(38) eration step, we have already an estimate obtained from
with approximated estimation variance previous step. Now the set of equations (8) is better approxi-
mated by

Note that each of the channel vectors is

obtained individually from the algorithm described above.
2) Analytical Linearization: Here, the set of equations (8) (44)
are rewritten as
with noise variance

(39)

where is the value of evaluated at

. (45)
For brevity, let and
At this point, we expand around to a second order
denote the set of all sampled re- Taylor series approximation:
ceived signal from th antenna up to th symbol time. Using
Bayes’ theorem on the conditional density for memoryless
sensor systems yields
(46)

where and denote

the Gradient and Hessian matrix of with respect to com-
plex vector , respectively.
The estimate is the minimum of the approximation of (46).
(40)
It is found by equating the gradient of the approximation to zero.
Differentiation of (46) with respect to complex vector gives
with a normalization constant
(47)
(41) which results in
(48)
The approximation is made such that both the predicted
channel matrix elements in (13) and the noise term in (39) are The Gradient and Hessian of , in explicit form, are ob-
considered normally distributed. Thus, the posterior probability tained from (42) as
density , which is the product of two Gaussians, is
also a Gaussian. Therefore, the MMSE estimate coincides with
the MAP estimate and the task is now to find the maximum of
. Equivalently, we can maximize its logarithm. After
the elimination of the irrelevant constants and factors, it boils
down to minimization the following function

(49)
(42)
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS
(50)
Substitution of (49)–(50) into (48) yields the following itera-
tion scheme:
(51)
We note that, converges very fast and it is possible to fix
the number of iterations to some small number , e.g.,
. (The effect of on system error performance is analyzed
through simulation below). The final result is set to the last iter-
ation, i.e., . at least the th iteration. The initial channel estimates obtained
The factor in (51)
can be regarded as the error covariance matrix associated with
, i.e.,
(52)
This insight gives another connection to the last term in (51)
because, in comparison with the standard Kalman filter recur-
sion formulas, the term can be regarded
as the Kalman gain matrix during the th iteration.
Note that each of the channel vectors has
to be obtained individually from the locally iterative algorithm
described above.
C. Data Vector Recovery
Here, the receiver recovers the transmitted data vector using
the channel matrix estimate
and the primary data vector estimate . The minimum dis-
tance receiver [44] chooses the vector that solves
(53)
where denotes the Frobenius norm and the equation
shown at the bottom of the page.
In (53) the search is performed over all the candidate vector
symbols, , and the decoding complexity of the receiver is ex-
ponential in . Note that in fact, (53) is the optimum ML de-
coder when the MIMO system is linear.
Finally, the estimated symbols are fed to the hard de-
cision block to yield the detected transmitted symbols
by setting the optimal thresholds in the constellation re-
gions of the transmitted signal. The channel matrix es-
707
timate, and the corre-
sponding variances associated with the estimation processes,
, also help the decision
algorithm in the subsequent th symbol interval to obtain
the subsequent primary data vector estimate . We note
that for initialization, it is sufficient for the proposed algorithm
to use any of the training approaches developed in [43] only
once at the beginning of whole data sequence transmission. To
obtain a meaningful estimate of at the beginning, we need
at least as many training measurements as unknowns, which
implies that at the first th iteration, any kind of training-based
scheme, e.g., LS or MMSE pilot-aided channel estimation, is
used for an initial estimate of the channel, . From
the th iteration, the algorithm switches to the presented
iterative algorithm and uses the channel estimate that is
at hand. This means that the algorithm switches to the blind
approach (in the sense that no further pilots are required) after
this way are accurate enough for the algorithm to have fast
convergence for the typical application of systems with fade
rates of 0.005.
IV. SIMULATION RESULTS
The optimal receiver performance sets a lower bound on
the error rate probability of sub-optimal receivers. However,
no exact optimal analytical solution is available for error
probability, even for linear MIMO systems. (For linear systems
several upper bounds on error probability for the ML as well
as other sub-optimal receivers have been derived.) Thus, we
simulate the SER performance results of coherent detection
with perfect CSI as benchmark, and evaluate and compare the
results of the presented blind system with the associated perfect
CSI reference ones.
We take i.i.d. QPSK data symbols with zero mean and vari-
ance 1, and for convenience of interpretation, the data symbol
duration is ms (i.e., 10 k symbols/sec) and the channel
fade rate is . Also, throughout the simulations,
the signal-to-noise ratio is
dB (54)
It is worth noting that the proposed algorithm is formu-
lated in a general form and it can be applied to any linear
and nonlinear MIMO systems with nonlinearities introduced
by and in (1). The last sub-section below, IV.B, is
dedicated to linear MIMO systems since with this choice, more
direct assessment of the proposed blind channel estimation
technique on system error performance is possible. Also, this
allows fair performance comparison with known results of
the conventional decision-directed Kalman filtering [30] and
two pilot-aided systems (least-squares (LS), and an MMSE
pilot-aided system [11]).
708
Fig. 2. SER performance of a 2 2 nonlinear MIMO system with different
nonlinearity configurations for transmit and receive amplifiers, when the statis-
tical linearization technique is used, and with fade rate 0.001.
Fig. 3. The effect of nonlinear parameter (representing the nonlinearity at
the transmitter) on the system SER performance with fixed . increases, less degradation in performance is observed since
Here, the value of is assumed to be
known, i.e., we are not including estimation of these statistics
as part of the algorithm. Including this estimate is a relatively
straightforward extension, but the goal here is to quantify the be-
havior of the blind estimation with known channel statistics. The
impact of any mismatch in the time-correlation coefficient mis-
match (perhaps caused by an incorrect estimate of , or simply
because the model is incorrect) on the proposed system’s error
performance is analyzed by simulation below. We show that the
proposed blind technique is robust for a mild mismatch of the
channel’s second-order statistics, i.e., the correlation function
values which relate to , relative to the a priori information
assumed by the receiver. However, a large degradation in perfor-
mance is demonstrated as the mismatch increases, as expected,
and this is demonstrated below.
A. Nonlinear MIMO Systems
Despite nonlinearities being usually small, they are known to
be very difficult to deal with [44]. Here, for both transmit and
receive amplifiers, we provide some examples from the family
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
of solid state power amplifier (SSPA) nonlinearity which is de-
fined as [36], [37]
(55)
where defines the smoothness of the transition from linear op-
eration to saturation, and is the saturation output amplitude
where for simulation it is set as . is a standard pa-
rameter which the designer must assign to the SSPA amplifier.
This may require a calibration measurement. Designers would
be able to decrease the cost of the amplifiers if a larger non-lin-
earity could be tolerated. For large values of , the SSPA model
approaches the soft limiter model that is commonly used to rep-
resent the clipping operation [45] to reduce the dynamic range
of the OFDM signal. The soft limiter model is specified as
(56)
We note that the model used in (55) is just an example of
amplifier nonlinearity provided here for simulation purposes.
Any kind of memoryless AM/AM and AM/PM characteristics
can be used instead. Furthermore, throughout the simulations,
and are used as the corresponding transmit and receive
amplifier smoothness parameter, respectively.
Fig. 2 illustrates the SER performance of a 2 2 nonlinear
MIMO system with different nonlinearity configurations for
transmit and receive amplifiers when the statistical linearization
technique is used. In general, as either of the parameters or
the nonlinearity becomes smoother.
A useful result is that, for a given receiver nonlinearity repre-
sented by parameter, the ensuing system performance curves
are essentially shifted versions of the curves obtained with the
ideal linear transmit amplifier . This is seen in Fig. 3
where the effect of nonlinear parameter (representing the
nonlinearity at the transmitter) on the system SER performance
with fixed is illustrated. As an example, there is a shift
of 1.1 dB in SNR, except near the error floor region. As a re-
sult, in order to assess the impact of the receiver nonlinearity
(represented by ) on system error performance more directly,
in the rest of the simulations, we evaluate the SER performance
of nonlinear MIMO systems with ideal linear amplifiers at the
transmitter by setting . Here we can also benchmark the
two statistical/analytical linearization techniques with a look-up
table (LUT) approach to the nonlinear problem.
1) Statistical/Analytical Linearization Approaches:
: The impact of nonlinearity parameter on the SER per-
formance of a 2 2 nonlinear MIMO system using the statis-
tical linearization technique is shown in Fig. 4. Comparison
is also made with the linear system error curve which corre-
sponds to large values of (soft limiter model). The nonlin-
earity is nearly compensated for values of dB for
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS 709

Fig. 5. SER performance of a nonlinear SISO and a 2 2 nonlinear MIMO

Fig. 4. Impact of nonlinearity parameter on SER performance of a 2 2 system with the nonlinearity parameter and . Dashed and
nonlinear MIMO system employing the statistical linearization technique while solid lines correspond to coherent detection with perfect CSI, and the proposed
. blind systems, respectively. The analytical linearization parameter is .

. The performance curves with analytical lineariza-

tion also follow the same trend as the curves with statistical lin-
earization. The SNR value of 13 dB would change with different
fading rates.
At large values of SNR, the difference between the curves of
the statistical/analytical linearization approaches and that for a
linear system increases with SNR. The performance degrada-
tion associated with the statistical/analytical approaches can be
explained via (8): it is found via simulations (not shown here)
that the noise variance (7) increases with SNR. As a result,
the Bussgangs’ approximation in (8) becomes less accurate for
larger values of SNR. In turn, this causes inaccuracies in the de-
cision algorithm which is built up from (8).
Fig. 5 shows the SER performance of a nonlinear SISO and a
2 2 nonlinear MIMO system with the nonlinearity parameter Fig. 6. Impact of the analytical linearization parameter on SER performance
and perfect CSI at the receiver. The corresponding of a 2 2 nonlinear MIMO system, for .
curves of the presented blind system employing statistical and
analytical linearization techniques are also illustrated. The
analytical linearization technique finishes the locally iterative
(nested) scheme in steps. The effect of on system
performance is studied below. The degradation in performance
of a 2 2 nonlinear system is more than that in the nonlinear
SISO system. This trend is expected since more nonlinear am-
plifiers contribute more nonlinearity as the number of receive
antennas increases. Fig. 5 reveals the details.
Finally, the impact of the analytical linearization parameter
on SER performance is shown in Fig. 6. Typically, the per-
formance gets better as the total number of iterations in the pro-
posed local iterative scheme increases from 1 to 3. For ,
the local iterative scheme nearly converges to a steady state
which is the locally optimum value, and as a result no further
improvement in performance is observed.
Fig. 7. Effect of mismatch parameter (57) on the relative SER performance of
So far, we assumed that the normalized time-correlation system employing the presented blind channel estimation with statistical lin-
function of the channel coefficients is known beforehand. In earization at dB.
the channel estimation process we make use of one value
of the channel normalized correlation coefficient function,
i.e., . However, a real-world, local even in the main lobe region, and almost never (or another
autocorrelation is seldom exactly (or some other model), model) away from the main lobe. Furthermore, estimating a
710
Fig. 8. Receiver nonlinearity compensation by means of LUT, here for the th antenna branch and
to the multiple antenna decoder block for channel estimation and data recovery, designed for linear MIMO systems.
correlation coefficient from finite samples introduces uncer-
tainty depending on the number of samples used. In order to get
a reliable estimate where the function becomes small, a very
large number of samples is required [46].
In order to test the performance sensitivity to the second-
order statistics, we define a simple percentage correlation func-
tion sample mismatch as
where the normalized time-correlation function used at the re-
ceiver takes on different values from .
Fig. 7 illustrates the effect of the mismatch parameter
on SER performance, given as the relative
SER change, , at
dB using the statistical linearization. The system
remains moderately insensitive to the mismatch of (57)—up to
20% for fate rate 0.005, but for higher mismatch, there will be a
larger degradation in the SER performance. Also, there is more
sensitivity to the mismatch for higher fade rates. The reason is
that, for a given mismatch, the higher fade rate results in the
addition of an error term to (19) which in turn causes more
signal estimation inaccuracies in the decision algorithm. Thus,
for the typical applications with fade rates 0.005, the
proposed blind technique is robust to mild mismatches of the
channel’s second-order statistics. The figure helps to quantify
the sensitivity of the proposed receiver to the modeling of the
temporal second order statistics of the channel.
2) Look-Up Table Approach Followed by Linear Blind
Channel Estimation: : We may also attempt to directly
compensate the nonlinearity at the receiver before proceeding to
channel estimation and data recovery by means of an LUT ap-
proach. First, by processing as shown
in Fig. 8, the estimation values of are
obtained. The estimation process at this step, can be modeled as
where is a vector with i.i.d. complex Gaussian entries each
with variance 1 and is the associated estimation error vari-
ance reflecting the inaccuracies in estimating .
Next, these estimates are fed to the multiple antenna decoder
block, designed for linear MIMO systems where blind channel
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
. The outputs of all branches, , are fed
(57)
Fig. 9. SER performance curves of a 2 2 system with nonlinearity compen-
sation using LUT, for and . The effect of gridding size on
the error performance is also illustrated and comparison is made with the pre-
sented statistical/analytical linearization approaches of Section III.
estimation and data recovery is carried out. From the SNR point
of view, we have
(59)
Equation (59) states that the proposed blind channel estima-
tion/data recovery designed for the linear system is going to op-
erate on the signal (i.e., ) that has the reduced SNR value of
in comparison to . Simulations show that de-
pends mainly on the gridding size (quantization for LUT) and
does not change with SNR. As a result, for a fixed gridding size,
the curves of LUT-based system are essentially shifted versions
of the curve for a linear system. We note that the inversion op-
eration shown in Fig. 8 is performed by means of LUT.
The SER performance curves of system based on LUT com-
pensation are shown in Fig. 9 for the nonlinearity parameters
. The effect of different values of gridding sizes (quan-
tization for the LUT) is also illustrated. Here, the input signal
full scale amplitude is taken as 0 to 4, so a grid size of 0.005
means there are 800 LUT entries. The curves are essentially
(58) shifted versions of the curve for a linear system which confirm
(59). However, as the gridding size increases (less LUT entries),
a larger shift is observed.
The comparison can also be made with the SER results of
the techniques proposed in Section III using statistical/ana-
lytical linearization and Bussgang’s theorem. The proposed
statistical/analytical linearization approaches outperform the
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS
Fig. 10. Comparison of SER performance of the presented algorithm with that
of LUT-based approach in a 2 2 system with nonlinear transmit and receive
amplifiers ( and various values of ).
system based on LUT for SNR values smaller than about 13.5
dB. Since the error performance of statistical/analytical lin-
earization approaches coincide with the curves of linear system
for low values of SNR (where the nonlinearity is completely
compensated), it is concluded that the main reason that the
performance of LUT-based approach falls behind the statis-
tical/analytical approaches, is the gridding size. In particular,
for very small values of gridding sizes, , the resultant
LUT-based curve coincides with the corresponding curve for
the linear system (refer to (59)) and the LUT-based approach
and the statistical/analytical approaches yield the same per-
formance for small values of SNR ( dB for the
example provided). On the other hand, at large values of SNR,
the difference between the curves of the statistical/analytical
approaches, and that for a linear system, increases with SNR
while the LUT-based curves experience a constant degradation
in all SNRs. As a result, for large values of SNR the LUT-based
approach outperforms the presented statistical/analytical lin-
earization approaches.
Comparison is also made between the performance of the pre-
sented algorithm and that of LUT-based approach in a 2 2
system with nonlinear amplifiers at both transmitter and receiver
(both and are small). The results are illustrated in Fig. 10
for and various values of the transmit nonlinearity pa-
rameter, . The LUT has a grid size of 0.005. For small values
of , the presented algorithm outperforms the LUT-based ap-
proach in SER performance for all values of SNR. In particular,
the difference between SER performances becomes larger as
decreases. This is simply because the presented LUT-based ap-
proach is not designed to compensate the transmit nonlinearity.
The result in Fig. 10 lays out that the presented algorithm is
more effective than the LUT-based approach in MIMO systems
with nonlinearities at both transmit and receiver amplifiers.
We note that all presented techniques require knowledge of
the nonlinearity characteristics, and . How-
ever, the systems based on LUT require the extra memory of
the table. The size of the extra memory increases, as the grid-
ding size decreases.
711
Fig. 11. SER performance of a linear 2 2 MIMO system employing ML,
MMSE, ZF, and MMSE-OSUC receivers. Dashed and solid lines correspond to
coherent detection with perfect CSI and proposed blind system, respectively;
.
In order to assess the impact of the proposed blind channel es-
timation technique on system error performance more directly,
the next section considers linear MIMO systems. This also al-
lows fair performance comparison with a known blind system
employing Kalman filtering to track the channel [30] and two
pilot-aided systems (least-squares (LS), and an MMSE pilot-
aided system [11]).
B. Linear MIMO System: ,
For the linear MIMO systems ( and ), the
signal model (1) simplifies to
(60)
The proposed blind channel estimation can be joined with any
decoding structure (e.g., ML, MMSE, ZF and MMSE-OSUC) at
the receiver. Each decoding structure provides a specific order
of diversity. The zero forcing (ZF) receiver provides
order diversity [47] (the same as MMSE and successive can-
cellation (SUC) receivers but with different SNR loss). As an
alternative, one may take advantage of an ordered successive
cancellation (OSUC) receiver [48] which may have more than
order diversity because of the ordering (selection)
process [47], or use the optimal ML receiver which extracts
order diversity with the expense of high decoding complexity
(exponential in ).
For different decoding structures, we compare the SER per-
formance of the system employing the proposed blind channel
estimation with the associated coherent detection curve with
perfect CSI. The results are illustrated in Fig. 11 for a 2 2
system. Each pair of curves bearing the same color (also labeled)
corresponds to a specific decoding structure (the “dotted line”
is associated with the “coherent detection with perfect CSI” and
the “solid line with marker” is the performance when the effect
of blind channel estimation is added to that system).
The statistical and analytical linearization techniques yield
the same performance in this case since no nonlinearity ex-
ists. With each receiver type (decoding structure), the presented
712 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
Fig. 12. Comparison of SER performances of linear 2 2 MMSE-OSUC
receivers employing different channel estimation techniques using SM-HE,
QPSK signaling and fade rate .
system exploits the same diversity order as the coherent detec-
tion with perfect CSI but experiences an SNR loss. However,
the SNR loss depends on the system parameters and the type of
decoding structure. In particular, it is within 3 dB (specifically,
2.5 dB) of the perfect CSI curves in an ML (in MMSE-OSUC)
receiver at fade rate and SER of . tion). The pilot-aided systems are more sensitive to time-vari-
We may also compare the results of a system employing the
presented blind channel estimation with that of a system using
other channel estimation techniques for a given decoding struc-
ture. Here, a fair comparison is possible with a known blind
channel estimation based on Kalman filtering (conventional de-
cision-directed Kalman filtering [30]), and two known pilot-
aided channel estimations: Least Squares (LS) and an MMSE
pilot-aided system [11]. The results are shown in Fig. 12 using
the MMSE-OSUC receiver. The presented system outperforms
both pilot-aided channel estimation techniques but with just a
slight improvement in performance over the decision-directed
Kalman-based estimation at moderate-high values of SNR.
Recall that the technique of [30] is a decision-directed
algorithm which uses channel predictions to obtain the coarse
symbol estimate for formulating the measurement equations
in the Kalman filtering. However, obtaining the coarse symbol
estimate is undertaken without considering the associated
channel estimation error variances, , whereas these
have been taken into consideration in the proposed decision
algorithm ((15)–(22)). Compared to pilot-aided systems, the
proposed blind system has superior performance for all SNRs,
at least for the fade rate used here. This is expected, because
in pilot-aided systems, the channel is estimated only after at
least consecutive symbols, whereas in the presented blind
technique, the channel is estimated at each symbol time which
results in more accurate channel estimates in the time-varying
environment.
The robustness of the proposed blind system to the time-vari-
ations of the channel can also be quantified by simulation and
compared with that of other conventional channel estimation
techniques. The MMSE-OSUC decoding structure is used for
all the systems. The effect of fade rate (maximum Doppler shift)
on error performance is shown in Fig. 13 for
Fig. 13. Effect of maximum Doppler shift on blind/pilot-aided system error
performance employing MMSE-OSUC receiver at dB.
Hz at dB. For the proposed blind system,
there is a maximum performance degradation of 15% in the
SER (compared to 60% in pilot-aided systems) for a fading rate
0.01 or 100 Hz, relative to the case of block
fading in a quasi-static channel (the Kalman-based estimation
has the same trend as with the proposed blind channel estima-
ations of the channel. However, in general, the system perfor-
mances depend on many factors such as channel fade rate, SNR
and pilot scheme used. For example, the LS pilot-aided system
is always worse than the MMSE pilot-aided system on this mea-
sure, and at very low fade rates (static fading), all the algorithms
(two-pilot aided systems, the Kalman-based estimation, and the
proposed blind channel estimation technique) yield nearly the
same performance at all values of SNR. However, as the fade
rate increases, the deference between the SER curves increases.
It is worth noting that techniques that are based on the three
conventional channel estimators under comparison (the two
LS/MMSE pilot-aided channel estimator and decision-directed
Kalman filtering) and/or detectors such as ZF, MMSE and
OSUC, become ineffective in nonlinear MIMO systems. This
is because these channel estimators and detectors are designed
based on the linearity in a system. As an example, it is well
known that the Kalman filter is the optimum linear state es-
timator when the set of state and observation equations are
linear [42], [30]. The performance of a Kalman filter degrades
significantly even when a mild nonlinearity exists in the system.
More severely, in most nonlinear cases the Kalman filter does
not converge at all if an inappropriate initial state value is set
within the algorithm [42]. As a result, a blind Kalman-based
channel estimator is not capable of tracking the time-variation
in the channel when the system in nonlinear, and the system
remains running without appropriate channel estimates, and the
receiver fails to detect the transmitted symbols correctly.
This is not the case with the presented algorithm. As shown
earlier, the presented algorithm is also effective for nonlinear
MIMO systems. As a result, this can be considered as one of
the advantages of the presented algorithm over the other con-
ventional techniques (including detection and channel estima-
tion) designed exclusively for linear systems.
ALIREZA BANANI AND VAUGHAN: COMPENSATING FOR NON-LINEAR AMPLIFIERS IN MIMO COMMUNICATIONS SYSTEMS
V. SUMMARY AND CONCLUSIONS
The amplifiers in a practical MIMO system can be perfor-
mance-limiting. Amplifier non-linearity is modeled here with
a memoryless, AM/PM amplifier characterization. The theory
and signal-processing is presented for compensating the link
degradation caused by the non-linearity. This includes a new
blind approach in the form of two alternative channel estima-
tion/tracking methods (statistical and analytical linearization)
for nonlinear MIMO systems. The MIMO performance is
estimated by simulation. For a fixed nonlinearity in each
receiver amplifier, the resultant system error curves are essen-
tially shifted versions of the curves obtained with ideal linear
transmit amplifiers. Furthermore, for the examples provided,
the proposed approach is capable of nearly compensating the
nonlinearity induced at the receiver side for values of SNR
less than about 13 dB. That is, in this range of SNR, the error
curves coincide with those of the linear MIMO system when
the amplifiers are linear. The two channel estimation methods,
statistical linearization and analytical linearization, have similar
performance, being within a fraction of a dB in SNR. They
both follow the same trend as coherent systems with perfect
CSI for small- to midsized SNR, and develop an error floor for
large SNR. The look-up table (LUT) approach also offers good
performance, being more effective than the presented algorithm
in compensating the receiver nonlinearity at high SNR when
the transmit amplifier is linear. This is at the expense of extra
memory. For smaller values of the SNR which are typical of
wireless at extended ranges, the performance is similar, with
the linearization outperforming the LUT approach by about
a half dB. However, with nonlinearities at both transmit and
receiver amplifiers, the presented algorithm is more effective
than the LUT-based approach in MIMO systems at all values
of SNR.
As the MIMO system becomes linear, typically possible
by using expensive amplifiers, the performances of the sta-
tistical and analytical linearization approaches converge. The
blind schemes presented here compare favorably against
known blind schemes such as conventional decision-directed
Kalman filtering, and two established, pilot-aided systems.
The proposed blind channel estimation is also more robust
to the time-variations of the channel compared to LS/MMSE
pilot-aided systems.
The comparison of simulation results with benchmark and
other known results puts a focus on the importance of checking
the channel modeling. We show that the proposed blind tech-
nique is robust for a moderate mismatch of the channel’s
second-order statistics, i.e., the correlation function values,
relative to the a priori information assumed by the receiver.
However, a large degradation in performance is demonstrated
as the mismatch increases, as expected.
REFERENCES
[1] G. J. Foschini and M. J. Gans, “On limits of wireless communications
in a fading environment when using multiple antennas,” Wireless Per-
sonal Commun., vol. 6, no. 3, pp. 311–335, Mar. 1998.
[2] R. O’Neil and L. B. Lopes, “Performance of amplitude limited mul-
titone signals,” in Proc. IEEE Veh. Technol. Conf., Jun. 1994, pp.
1675–1679.
713
[3] D. Dardari, V. Tralli, and A. Vaccari, “A theoretical characterization of
nonlinear distortion effects in OFDM systems,” IEEE Trans. Commun.,
vol. 48, no. 10, pp. 1755–1764, Oct. 2000.
[4] W. A. Morgan, “Minimize IM distortion in GaAs FET amplifiers,” Mi-
crowaves RF, vol. 25, no. 10, pp. 107–110, 1986.
[5] V. M. Vladimirov, S. N. Kulinich, and Y. Y. Shikhov, “LNA—Ac-
tive bandpass filter for receiver-indicator of Glonass GPS,” pre-
sented at the Int. Conf. Information, Commun. and Energy Systems
and Technol., ICEST 2002, Oct. 2002.
[6] T. H. Lee, The Design of CMOS Radio Frequency Integrated Cir-
cuits. Cambridge, U.K.: Cambridge Univ. Press, 1998.
[7] T. H. Lee, “5-GHz CMOS wireless LANs,” IEEE Trans. Microw.
Theory Tech., vol. 50, no. 1, pp. 268–280, Jan. 2002.
[8] M. Ibnkahla et al., “High-speed satellite mobile communications:
Technologies and challenges,” IEEE Proc., vol. 92, no. 2, Feb. 2004.
[9] Q. Sun, D. C. Cox, H. C. Huang, and A. Lozano, “Estimation of con-
tinuous flat fading MIMO channels,” IEEE Trans. Wireless Commun.,
vol. 1, no. 2, pp. 549–553, Oct. 2002.
[10] X. Ma, G. B. Giannakis, and S. Ohno, “Optimal training for block trans-
mission over doubly selective wireless fading channels,” IEEE Trans.
Signal Process., vol. 51, no. 5, pp. 1351–1366, May 2003.
[11] M. Biguesh and A. B. Gershman, “Training-Based MIMO Channel Es-
timation: A study of estimator tradeoffs and optimal training signals,”
IEEE Trans. Signal Process., vol. 54, no. 3, pp. 884–893, Mar. 2006.
[12] B. Hassibi and B. M. Hochwald, “How much training is needed in mul-
tiple-antenna wireless links?,” IEEE Trans. Inf. Theory, vol. 49, no. 4,
pp. 2515–2528, Apr. 2003.
[13] C.-Y. Chi and C.-H. Chen, “Cumulant based inverse filter criteria for
MIMO blind deconvolution: Properties, algorithms, and application to
DS/CDMA systems in multipath,” IEEE Trans. Signal Processing, vol.
49, no. 7, pp. 1282–1299, Jul. 2001.
[14] Z. Ding and T. Nguyen, “Stationary points of a kurtosis maximization
algorithm for blind signal separation and antenna beamforming,” IEEE
Trans. Signal Process., vol. 48, no. 6, pp. 1587–1596, Jun. 2000.
[15] C. Y. Chi, C. Y. Chen, C. H. Chen, and C. C. Feng, “Batch processing
algorithms for blind equalization using higher-order statistics,” IEEE
Signal Process. Mag., vol. 2, no. 1, pp. 25–49, Jan. 2003.
[16] J. Liang and Z. Ding, “Blind MIMO system identification based on
cumulant subspace decomposition,” IEEE Trans. Signal Process., vol.
51, no. 6, pp. 1457–1468, Jun. 2003.
[17] C. Shin, R. W. Heath, and E. J. Powers, “Blind channel estimation for
MIMO-OFDM systems,” IEEE Trans. Veh. Technol., vol. 56, no. 2, pp.
670–685, Mar. 2007.
[18] B. Chen and A. P. Petropulu, “Frequency domain blind MIMO system
identification based on second and higher order statistics,” IEEE Trans.
Signal Process., vol. 49, no. 8, pp. 1677–1688, Aug. 2001.
[19] T. Acar, Y. Yu, and A. P. Petropulu, “Blind MIMO system estima-
tion based on PARAFAC decomposition of higher order output ten-
sors,” IEEE Trans. Signal Process., vol. 54, no. 11, pp. 4156–4168,
Nov. 2006.
[20] M. A. Khalighi and S. Bourennane, “Semi-blind channel estimation
based on superimposed pilots for single-carrier MIMO systems,” in
Proc. IEEE Veh. Technol. Conf., Apr. 2007, pp. 1480–1484.
[21] J. Gao and H. Liu, “Low-complexity MAP channel estimation for mo-
bile MIMO-OFDM systems,” IEEE Trans. Wireless Commun., vol. 7,
no. 3, pp. 774–780, Mar. 2008.
[22] Z. Ding, T. Ratnarajah, and C. F. N. Cowan, “HOS-based semi-blind
spatial equalization for MIMO Rayleigh fading channels,” IEEE Trans.
Signal Process., vol. 56, no. 1, pp. 248–255, Jan. 2008.
[23] T. Dahl, N. Christophersen, and D. Gesbert, “Blind MIMO eigenmode
transmission based on the algebraic power method,” IEEE Trans.
Signal Process., vol. 52, no. 9, pp. 2424–2431, Sep. 2004.
[24] A. K. Jagannatham and B. D. Rao, “Whitening-rotation-based semi-
blind MIMO channel estimation,” IEEE Trans. Signal Process., vol.
54, no. 3, pp. 861–869, Mar. 2006.
[25] A. Medles and D. T. M. Slock, “Semiblind channel estimation for
MIMO spatial multiplexing systems,” in Proc. Vehicular Technol.
Conf., Oct. 2001, vol. 2, pp. 1240–1244.
[26] R. Everson and S. Roberts, Independent Component Analysis, Princi-
ples and Practice. Cambridge, U.K.: Cambridge Univ. Press, 2001.
[27] V. Zarzoso and A. K. Nandi, “Blind MIMO equalization with optimum
delay using independent component analysis,” Int. J. Adapt. Control
Signal Process., vol. 18, no. 3, pp. 245–263, Mar. 2004.
[28] A. Mansour, “A mutually referenced blind multiuser separation of
convolutive mixture algorithm,” Signal Process., vol. 81, no. 11, pp.
2253–2266, 2001.
714 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 60, NO. 2, FEBRUARY 2012
[29] V. Zarzoso and A. K. Nandi, “Exploiting non-Gaussianity in blind
identification and equalization of MIMO FIR channels,” IEE Proc. Vi-
sion, Image Signal Process., vol. 151, no. 1, pp. 69–75, Feb. 2004.
[30] C. Komninakis, C. Fragouli, A. H. Sayed, and R. D. Wesel,
“Multi-input multi-output fading channel tracking and equaliza-
tion using Kalman estimation,” IEEE Trans. Signal Process., vol. 50,
no. 5, pp. 1065–1076, May 2002.
[31] W. H. Chin, D. B. Ward, and A. G. Constantinides, “Semi-blind
MIMO channel tracking using auxiliary particle filtering,” in Proc.
GLOBECOM, Nov. 2002, vol. 1, pp. 322–325.
[32] S. Haykin, A. H. Sayed, J. R. Zeidler, P. Yee, and P. C. Wei, “Adaptive
tracking of linear time-variant systems by extended RLS algorithms,”
IEEE Trans. Signal Process., vol. 45, no. 5, p. 11 18-1 128, May 1997.
[33] J. J. Bussgang, “Cross Correlation Function of Amplitude-Distorted
Gaussian Input Signals,” M.I.T., Cambridge, MA, Tech. Rep. 216,
Mar. 1952, vol. 3, Res. Lab Electron..
[34] J. Minkoff, “The role of AM-to-PM conversion in memoryless non-
linear systems,” IEEE Trans. Commun., vol. 33, no. 2, pp. 139–144,
Feb. 1985.
[35] S. A. Banani and R. G. Vaughan, “Blind channel estimation for
MIMO systems with nonlinearities at the receiver,” in Proc. IEEE Veh.
Technol. Conf., May 2010, pp. 1–5.
[36] S. Benedetto, E. Biglieri, and V. Castellani, Digital Transmission
Theory. Englewood Cliffs, NJ: Prentice-Hall, 1987.
[37] A. A. M. Saleh, “Frequency independent and frequency dependent
nonlinear model of TWT amplifiers,” IEEE Trans. Commun., vol.
COM-29, pp. 1715–1720, Nov. 1981.
[38] H. Wang and P. Chang, “On verifying the first-order Markovian
assumption for a Rayleigh fading channel model,” IEEE Trans. Veh.
Technol., vol. 45, no. 2, pp. 353–357, May 1996.
[39] L. M. Davis, I. B. Collings, and R. J. Evans, “Coupled estimators for
equalization of fast-fading mobile channels,” IEEE Trans. Commun.,
vol. 46, no. 10, pp. 1262–1265, Oct. 1998.
[40] S. Julier, J. Uhlmann, and H. F. Durrant-White, “A new method for
nonlinear transformation of means and covariances in filters and es-
timators,” IEEE Trans. Automatic Contr., vol. 45, no. 3, pp. 77–482,
Mar. 2000.
[41] E. A. Wan and R. van der Merwe, “The unscented Kalman filter for
nonlinear estimation,” in Proc. IEEE Symp. Adaptive Systems Signal
Proc., Comm. Control (AS-SPCC), 2000, pp. 153–158.
[42] A. H. Sayed, Fundamentals of Adaptive Filtering. New York: Wiley,
2003.
[43] M. K. Tsatsanis, G. B. Giannakis, and G. Zhou, “Estimation and equal-
ization of fading channels with random coefficients,” Signal Process.,
vol. 53, no. 2–3, pp. 211–229, Sep. 1996.
[44] J. G. Proakis, Digital Communications, 5th ed. New York: McGraw-
Hill, 2008.
[45] R. Prasad, OFDM for Wireless Communications Systems. Norwood,
MA: Artech House, 2004.
[46] R. Vaughan and J. B. Anderson, Channels, Propagation and Antennas
for Mobile and Personal Communications. New York: Peregrinus/
IEE, 2003.
[47] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless
Communications. Cambridge, U.K.: Cambridge Univ. Press, 2003.
[48] G. J. Foschini, G. D. Golden, R. A. Valenzuela, and P. W. Wolniansky,
“Simplified processing for high spectral efficiency wireless communi-
cation employing multi-element arrays,” IEEE J. Sel. Areas Commun.,
vol. 17, no. 11, pp. 1841–1852, Nov. 1999.
Seyed Alireza Banani (S’07) received the B.Sc.
and M.Sc. degrees in electrical communication
engineering from Shiraz University, Shiraz, Iran,
in 2004 and 2007, respectively. He is currently
working toward the Ph.D. degree with the School
of Engineering Science, Simon Fraser University,
Burnaby, BC, Canada.
His research interests include target tracking,
independent components analysis, nonlinear
state estimation, blind channel estimation, mul-
tiple-input-multiple-output (MIMO) capacity, and
signal processing for MIMO wireless communications systems.
Mr. Alireza Banani has been a recipient of a Canada graduate scholarship
from the Natural Science and Engineering Research Council of Canada since
2009.
Rodney G. Vaughan (M’84–SM’89–F’07) received
the B.E. and M.E. degrees in electrical engineering
from the University of Canterbury, New Zealand, and
the Ph.D. degree from Aalborg University, Denmark.
He was with the New Zealand Post Office (now
Telecom NZ), the New Zealand Department of Scien-
tific and Industrial Research, and Industrial Research
Limited (IRL). During this time, he undertook me-
chanical and electrical projects from heating and ven-
tilation designs to network analysis and traffic fore-
casting, as well as microprocessor applications from
abattoir automation to communications networks. In 1992, when he was with
IRL, he developed research programs and personnel in communications tech-
nology. Here, his industrial projects included antennas for personal, cellular, and
satellite communications; sonar array processing; large-N multiple-input mul-
tiple-output (MIMO) communications systems design and capacity theory; and
statistical field theory. Since 2003, he has been a Professor of electrical engi-
neering and the Sierra Wireless Chair in Communications with the School of
Engineering Science, Simon Fraser University, Burnaby, BC, Canada. His re-
cent projects include mobile communications; bio-implantable antennas; wide-
band elements, compact multielement antenna design and evaluation; optimal
circular polarization purity antennas, multifaceted arrays; MIMO capacity real-
ization, blind techniques and interference mitigation for orthogonal frequency
division multiplexing; channel modeling and estimation; and on body propaga-
tion analysis.
Prof. Vaughan was an International Union of Radio Science (URSI) Young
Scientist in Fields and Waves and in Electromagnetic Theory and a 2003 Fellow
of the BC Advanced System Institute. He is an URSI Correspondent and con-
tinues as New Zealand’s URSI Commission B Representative.
Digital Object Identifier 10.1109/TAP.2012.2186031
Digital Object Identifier 10.1109/TAP.2012.2186033
 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION
Editorial Board
Michael A. Jensen, Editor-in-Chief Brenn Ellsworth, Editorial Assistant
Department of Electrical and Computer Engineering e-mail: tap.editor@ieee.org
459 Clyde Building (801) 422-3903 (voice)
Brigham Young University
Provo, UT 84602
e-mail: tap.editor@ieee.org
(801) 422-5736 (voice) Senior Associate Editor
Karl F. Warnick

Associate Editors
Dimitris Anagnostou Stuart G. Hay Derek McNamara Satish Sharma
Hiroyuki Arai Sean V. Hum Andrea Neto Jamesina Simpson
Ozlem Aydin Civi Ramakrishna Janaswamy Claude Oestges Mei Song Tong
Zhi Ning Chen Buon Kiong Lau George W. Pan Jon W. Wallace
Jorge R. Costa Jin-Fa Lee Athanasios Panagopoulos Fan Yang
George Eleftheriades Kwok Wa Leung Patrik Persson Ali Yilmaz
Magda El-Shenawee Duixian Liu K.V. S. Rao Zhengqing Yun
Lal C. Godara Andrea Massa Shanker Balasubramaniam Zhijun Zhang
Yang Hao Yue Ping Zhang

INFORMATION FOR AUTHORS

MANUSCRIPTS should be specially prepared for publication in this TRANSACTIONS and submitted electronically via Manuscript Central, the manuscript administration
software. The preferred format of the manuscript is in the IEEE two-column template with figures and captions included in the text. This template can be downloaded
from http://www.ieee.org/web/publications/authors/transjnl/index.html. Please follow instructions given in the template. Color figures should be avoided unless the paper
is intended to be published with color figures at extra cost. For details refer to “Author Digital Tool Box” on the IEEE website at http://wwww.ieee.org/web/publications/
authors/index.html.

ELECTRONIC SUBMISSION of manuscripts is required for both initial review and for publication. To submit your manuscript electronically for review go to the Manuscript
Central website at http://mcmanuscriptcentral.com/tap-ieee. For additional information about the electronic submission process and creating files for electronic submission,
visit the AP TRANSACTIONS website at: http://ieeeaps.org/aps_trans/

COPYRIGHT: It is the policy of the IEEE to own the copyright to the technical contributions it publishes on behalf of the interests of the IEEE, its authors, and their employers,
and to facilitate the appropriate reuse of this material by others. To comply with the U.S. Copyright Law, authors are required to sign and submit an IEEE Copyright Form
before publication. This form returns to authors and their employers the full rights to reuse their material for their own purposes. Authors must submit a signed copy of this
form before their manuscript can be published. The copyright form can now be completed electronically in Manuscript Central, and downloaded from the AP TRANSACTIONS
website as a PDF file.

PAGE CHARGES: All Authors are requested to honor page charges of $110 per printed page for the first eight pages of a Paper and for the first four pages of a Communication.
Payment of page charges for this IEEE TRANSACTIONS, like journals of other professional societies, is not a necessary prerequisite for publication. The author will receive
100 free reprints (without covers) only if the page charge is honored. Detailed instructions will accompany the proofs. A Page Charge Form must be completed and returned
to the Editor before a manuscript can be published. The page charge form, in PDF format, may be downloaded from the AP TRANSACTIONS website.

CHARGES FOR EXCESSIVE LENGTH: A mandatory EXCESSIVE PAGE LENGTH charge of $200 per page is required for Papers longer than eight (8) printed pages
and Communications longer than four (4) printed pages. If the author will not agree to pay mandatory charges, the author will be required to reduce the manuscript to eight
printed pages or less (four printed pages for Communications) prior to final acceptance and scheduling for publication. This reduction will occur after technical review of the
manuscript. The above policy has no relationship to requests by the Editor to reduce very long manuscripts prior to technical review.

PAPERS may be any length from one to about 25 typewritten double-spaced pages and are subjected to review by the Editor, Associate Editors, and their staff of reviewers.
Each paper must have an abstract of about 150 words. Authors should allow approximately 12 weeks for the initial review cycle. Authors are encouraged to make their initial
submission as short as possible. Because of limited space available in the TRANSACTIONS and the difficulty in finding reviewers for very long papers, manuscripts (either
single or multiple-part papers) that exceed 40 pages of double-spaced text and figures will not be considered. Upon being informed that your paper has been accepted for
publication, be sure that the final version is accompanied by a disk or CD. The electronic form of the paper should match the hard copy exactly and should include the
references, biographies, footnotes, and figure captions. Most popular IBM-PC-compatible, Macintosh, and UNIX formats are acceptable. A more complete IEEE authors’
electronic guide or an IEEE\LaTeX style file can be obtained by email at “help@ep.ieee.org” or by fax at +1 732 562 0545.

COMMUNICATIONS are narrower in scope and shorter than PAPERS. They will typically cover an extension of previous work or provide new insight on a specialized
topic. As a result, the review of past work is expected to be limited to the key material. The manuscript length is normally up to four journal pages (two-column template)
and should not exceed six journal pages. Page charges apply after four pages. COMMUNICATIONS are subjected to the same rigorous review as used for PAPERS.

COMMENTS are brief notes containing remarks (and Authors’ replies) as well as corrections on previously published manuscripts. Usually, they are reviewed only by the
Editor, and are published at his discretion.

REVIEWS AND TUTORIALS are full length papers which need not contain original work. They primarily review the state of the art and developments leading thereto.
Please contact the Editor prior to undertaking preparation of a review or tutorial.

ILLUSTRATIONS should be kept to a minimum. Figures and photos should be original proofs and not photocopies. Photos must be glossy prints with no screens. Laser
prints are not acceptable in place of photos or where gray scale graphics are involved. All line drawings and photos should be in black and white unless specifically requested.
Letters and symbols should be large enough to be readily legible when the drawing is reduced to a one-column width as much as 4:1 reduction from the original. Materials
should be no larger than 22 28 cm . For final versions, electronic graphics must be provided as separate files.

PAGE PROOFS are sent to each principal author just prior to publication of the paper. Minor and typographical errors should be noted and corrected according to the
instructions enclosed with the proofs. Major changes required to correct errors in the author’s manuscript will not, in general, be permitted. In order to avoid the need for a
major change, the author should proofread his manuscript very carefully.

PUBLICATION POLICY: The author must ascertain that any clearance, company or foundation release, necessary for publication of the paper in this TRANSACTIONS has
been obtained. If the paper or any modification of it has been published in the public literature, the author must indicate this in the letter of transmittal accompanying the
submitted paper.

NEWS ITEMS concerning the Society members and activities should be sent to the AP Magazine Editor-in-Chief, Dr. W. R. Stone, Stoneware Limited, 1446 Vista Claridad,
La Jolla, CA 92037. Telephone: +1 619 459 8305.

Digital Object Identifier 10.1109/TAP.2012.2186029

 INSTITUTIONAL LISTINGS
The IEEE Antennas and Propagation Society is grateful for the support given by the organizations listed below and invites applications for Institutional
Listings from other firms interested in the field of Antennas and Propagation.

The charge for an Institutional Listing is $75 per issue, $375 for six consecutive issues, and $725 for 12 issues. Agency fee is not granted on Institutional
Listings. Applications for Institutional Listings, should be sent to the Editor-in-Chief or Susan Schneiderman, IEEE Media, 445 Hoes Lane, Piscataway,
NJ 08854, ss.ieeemedia@ieee.org. Please include name of company, company contact information, billing address, name of publication, dates of issue,
artwork in the form of a pdf should be sent to Felicia Spagnoli, f.spagnoli@ieee.org, +1 732-562-6334.

Digital Object Identifier 10.1109/TAP.2012.2186411