Documente Academic
Documente Profesional
Documente Cultură
SUMMARY
We address the downlink of a cellular multi-carrier CDMA (MC-CDMA) system taking into account
channel estimation. The system performance in presence of a synchronization mismatch between two
interfering base stations (BS) is analyzed in the way that a mobile terminal receives the perfectly
synchronized signal from the desired BS as well as the signal from one interfering BS with a
synchronization offset. It is demonstrated through simulations that spreading and cell specific random
subcarrier interleaving effectively decorrelates the interfering signal, independent of the synchronization
offset. Furthermore, the robustness of the channel estimator to cellular interference is examined.
Copyright # 2004 AEI.
1. INTRODUCTION
Multicarrier (MC) modulation, in particular orthogonal
frequency division multiplexing (OFDM) [1], has been
successfully applied to various digital communications
systems. OFDM can be efficiently implemented by using
the discrete Fourier transform (DFT). Furthermore, for
the transmission of high data rates its robustness in transmission through dispersive channels is a major advantage.
For MC-CDMA, spreading in frequency and/or time direction is introduced in addition to the OFDM modulation [2
4]. MC-CDMA has been deemed a promising candidate
for the downlink of future mobile communications systems
[5, 6], and has recently been implemented by NTT DoCoMo in an experimental system [7].
Recently, there has been growing interest in applying
an OFDM-based air interface to cellular systems. We
focus on a system which should be robust against interference, rather than trying to avoid interference, as this
ultimately would require inter-cell synchronization, which
comes along with a significant signaling overhead. This
means that inter-cell interference can be significant, especially if the system is to operate with high frequency reuse
factor.
For a cellular multicarrier system where adjacent base
stations (BS) are not synchronized, the cellular interference is generally dependent on the synchronization offset
between interfering base stations. For an unsynchronized
system, the interference observed at a certain subcarrier
stems from all interfering subcarriers. So, the interference
can be modeled as white Gaussian noise. For a perfectly
synchronized system, on the other hand, orthogonality
between subcarriers is preserved and the interference can
be described on the subcarrier level, which may not be
Gaussian. In such a case, spreading, cell specific random
interleaving, and scrambling of subcarriers can be used
to decorrelate the interfering and desired signal [8].
Cellular interference not only corrupts the transmitted
data but also the pilot symbols used for channel estimation.
We focus on pilot-symbol aided channel estimation
(PACE), where pilot symbols are periodically inserted in
the time-frequency grid of the multicarrier signal. Channel
* Correspondence to: Gunther Auer, DoCoMo Euro-Labs, Landsberger Strasse 312, 80687 Munchen, Germany. E-mail: auer@docomolab-euro.com
y
This paper has been presented in part at the 4th International Workshop on Multi-Carrier Spread Spectrum (MC-SS 2003), Oberpfaffenhofen, Germany.
174
G. AUER ET AL.
T
dk d 1 k; . . . ; dNu k , consisting of the kth symbol
of all Nu users, the spreading operation results inz
sk CL dk s1 k; . . . ; sL kT
where CL represents the L Nu spreading matrix. The system load of the MC-CDMA system is Nu =L and can be
adjusted between 1 and 1=L. The spreading operation is
used to achieve a multiple access scheme for Nu users.
The spreading factor L can be significantly smaller than
the number of available subcarriers Nc . In this case each
user may transmit Nd Nc =L data streams in parallel.
The output of the spreader (1) is grouped into Nd blocks,
s sT Nd ; . . . ; sT 1Nd 1T , to yield Nc spread
chips per OFDM symbol. Given a MC-CDMA system
having Nc subcarriers and a frame length of Nframe OFDM
symbols, the block length of the code word of one particular user is MNframe Nc =L bits, which corresponds to
Nframe Nc =L spread blocks sk per OFDM frame, with
0 4 k < Nd Nframe .
Generally, the multiple access for an OFDM-based air
interface is very flexible. CDMA can be combined with
FDMA or TDMA, termed M&Q modification in Reference [10]. To this end, the spread blocks sk may also
be assigned to different users, which allows to support a
maximum of Nc users per OFDM symbol. Moreover, one
may choose to assign various spreading codes to one user,
in which case one user can use all Nc subcarriers. For the
sake of simplicity, we will restrict to a MC-CDMA system
described above, having Nu 4 L users, each user transmitting Nd Nc =L symbols per OFDM symbol.
Subsequently, the spread chips of the th OFDM symbol, s , is frequency interleaved by the inner interleaver,
m
Pin , over one OFDM symbol to maximize the diversity
gain. More specifically, we choose a cell specific random
m
interleaver for Pin , where m identifies the BS. The purpose
m
of Pin is twofold: first, by increasing the distance between
adjacent spread symbols a diversity gain is achieved; second, the inter-cell interference between adjacent BSs is
randomized. The interleaver produces the output
h
iT
m
m
m
m
X Pin s X;1 ; . . . ; X;Nc
2
where the interleaved symbol of the th OFDM symbol,
m
at subcarrier i, is denoted by X;i . In order to distinguish
Figure 1. Block diagram of the MC-CDMA system; (a) transmitter, (b) receiver.
Copyright # 2004 AEI
z
Given a matrix X, the operators X , XT , XH and X1 denote the conjugate complex, transpose, Hermitian transpose, and inverse of X respectively.
ANALYSIS OF CELLULAR INTERFERENCE FOR MC-CDMA AND ITS IMPACT ON CHANNEL ESTIMATION
y;n y n Nsym
zm; n Nsym n n Nsym
Y;i
m
X ;i ,
m;
H ;i
m;
X ;i H ;i
N;i
N;i
where
and
denote the transmitted symbol
from BS m having an energy per symbol of Es , the channel
transfer function (CTF) from BS m to receive antenna
and AWGN with zero mean and variance N0 respectively.
Copyright # 2004 AEI
175
NR
X
m; m;
Y;i
H ;i
1
m
X ;i
NR
NR
X
m; 2 X m;
H ;i N;i
H ;i
1
|{z}
1
m2
H
;i
m2
where H
;i 2 R accounts for the coherently combined
m;
power of the CTFs. In a practical system, the CSI H ;i
m;
^
is replaced by its estimate H
;i . By using MRC receive
antenna diversity provides an NR -fold improvement of
the averge signal to noise ratio (SNR), as well as an additional NR -fold diversity gain, due to the mutually uncorrelated fading assumption.
A block diagram of a MC-CDMA receiver is depicted in
Figure 1(b). Due to frequency selective fading of the multipath fading channel and the random interleaving of the
spread chips, the orthogonality of the spreading sequences
cannot be maintained and multiple access interference
(MAI) occurs [12]. Various detection schemes for MCCDMA have been proposed in the literature, both linear
and non-linear [10, 12, 13]. An efficient compromise
between reducing MAI and utilizing the diversity of the
frequency selective channel is the linear minimum mean
squared error (MMSE) detector [14]. Applying the MMSE
m m
criterion to the MRC output Y;i =H
;i with the constraint
of a one tap equalizer, the linear MMSE detector becomes
^ m
X
;i
m
Y;i
m2 1
H
;i
g
m1
is subsequently deinterleaved by Pin . Next, the subblock rk r1 k; . . . ; rL kT containing symbol k of
the Nu users is despread
h
iT
^k CH rk d^1 k; . . . ; d^Nu k
d
L
176
G. AUER ET AL.
u
of one frame, fd^ kg, are combined to a serial data
stream. The symbol demapper maps the data symbols into
bits, by also calculating the log-likelihood ratio (LLR) for
each bit, which serves as reliability information for the
decoder [15, 16]. According to References [15, 16] LLRs
are the optimum values which can be exploited by a
Viterbi decoder. The symbol demapper assumes the MAI
u
of fd^k g to be white Gaussian noise with zero mean and
appropriately scaled variance [10, 13]. The codebits are
deinterleaved and finally decoded using a soft-in soft-out
channel decoder. We use the Max-Log MAP algorithm
for the channel decoder [17, 18], which is an approximation of the optimum maximum a posteriori (MAP) symbol-by-symbol detector [19].
hm; t;
Q0
X
m;
hm;
t
q
q
q1
m;
H;i
m;
0m;
00m;
Q0
X
m;
m;
h;q ej2p q
i=T
10
q1
where Tsym NDFT NGI Tspl and T NDFT Tspl represent the OFDM symbol duration with and without the
guard interval.
The discrete two dimensional (2D) frequency correlam; m;
m;
tion function, EH;i H;ii RHH i; , specifies the correlation between subcarriers and OFDM
Copyright # 2004 AEI
11
RHH i
Q0
X
m;
2 j2p q
m;
e
q
i=T
12
q1
m; 2
m;
m;
symbols spaced i=T Hz and Tsym sec apart. It is generally assumed that the fading in time and frequency direcm;
tion is independent. Thus, RHH i; can be expressed
in the product form
13
zm; t T m n t
14
177
ANALYSIS OF CELLULAR INTERFERENCE FOR MC-CDMA AND ITS IMPACT ON CHANNEL ESTIMATION
15
m;
N
c 1
X
m;
16
k0
k6i
|{z}
Es =E, having a timing offset T I and a carrier frequency offset f I . So, E accounts for the difference
in received signal power between the two interfering
BSs. It is well known that the effects of a carrier frequency
offset for OFDM on the link level are very severe [23, 25,
26]. On the other hand, considering an interfering signal
the situation is somewhat different. Basically, the power
of the interference will not change due to a synchronization offset. However, the characteristics of the interference
might depend on the synchronization offset. After sampling at the mobile terminal the received signal at receive
antenna is in the form
ICI
where
m
m
Jk;i
ejpEk;i
=NDFT
NDFT ejpEm
k;i
m
sin pEk;i
m
sin pEk;i =NDFT
accounts for the ICI from subcarriers k to i. The cross-subcarrier local frequency offsets are
m
Ek;i f m T k i
18
178
G. AUER ET AL.
1
E
NR
NE0s NLu
21
2.4.1. Gaussian approximation. A very simple approximation is to model the entire interference as Gaussian noise,
by appropriately scaling the variance of the AWGN term.
By applying the Gaussian approximation (GA) the
received signal after OFDM demodulation in Equation
(20) is approximated by
0;
Y;i H ;i X ;i ;i
22
and frequency. For 2D-PACE the pilot symbols are scattered throughout the time-frequency grid, yielding a 2D
pilot grid. A scattered pilot grid is used, for example in
the terrestrial digital TV standard DVB-T. 2D filtering
algorithms have been proposed for PACE, based on 2D
Wiener filter interpolation [28, 29].
To describe PACE, it is useful to define a subset of the
received signal sequence containing only the pilots,{
m
~ m
fX
g fX;i g, with ~Dt and i ~iDf . The quantities
~;~i
Df and Dt denote the pilot spacing in frequency and time,
respectively. If a scattered pilot grid is used, the received
OFDM frame is sampled in two dimensions, with rate
Df =T and Dt Tsym in frequency and time, respectively. In
order to reconstruct the signal, there exists a maximum
Df and Dt , dependent on the maximum delay of the channel, max , and the maximum Doppler frequency fD;max. By
applying the sampling theorem, the following relation
must be satisfied [29]:
Df max =T 4 bf
1
b
2 t
23
3. CHANNEL ESTIMATION
Pilot-symbol aided channel estimation (PACE) is based on
periodically inserting known symbols, termed pilot symbols, in the transmitted data sequence. PACE was first
introduced for single carrier systems and required a flatfading channel [27]. If the pilot spacing is sufficiently
close, the channel response of data symbols at an arbitrary
position can be reconstructed by exploiting the correlation
of the recived signal. When extending the idea of PACE to
multi-carrier systems, it must be taken into account that
the received signal is correlated in two dimensions, in time
Copyright # 2004 AEI
ANALYSIS OF CELLULAR INTERFERENCE FOR MC-CDMA AND ITS IMPACT ON CHANNEL ESTIMATION
Mt
X
Wn00
n1
Mf
X
~ ~
Wm0 i H
n;im
24
25
26
It is important to note that the parameters of the robust estimator should always be equal or larger than the worst case
channel conditions, i.e. largest propagation delays and
maximum expected velocity of the mobile user, so
filter 5 max and fD;filter 5 fD;max . Furthermore, the average
SNR at the filter input, gfilter , which is used to generate the
filter coefficients, should be equal or larger than actual
average SNR, so gfilter 5gc . In order to determine the channel estimator only filter, fD;filter and gfilter are required.
m1
179
4. SYSTEM SCENARIOS
In case the mobile is near the cell/sector boundary, E in
Equation (19) will be close to one, so the carrier to interference ratio, gc , for a single antenna receiver approaches
0 dB. In order to maintain a reliable connection the system
may not be operated fully loaded, so Nu < L. Furthermore,
we employ NR receive antennas in order to exploit spatial
diversity and to benefit from an NR fold SNR gain. However, to achieve these gains reliable channel estimation is
essential. Moreover, the spectral efficiency is compromised if Nu < L for the entire system, even for users which
observe little interference. This problem can be mitigated
if users are grouped together which are close to the cell
boundaries with 50% load (Nu L=2) and QPSK modulation; other users which experience little interference may
be allocated 100% load and higher order modulation.
Alternatively, a frequency reuse factor of 3 may be used
for outer parts of the cell/sector as suggested in Reference
[32]. Even for such a scheme it is of interest to closely analyze cellular interference, since the resulting sector
throughput increases if the portion of the sector having a
frequency reuse above one can be kept as small as
possible. In any case, optimization of the throughput per
cell, sector or beam is beyond the scope of this paper.
Euro. Trans. Telecomms. 2004; 15:173184
180
G. AUER ET AL.
5. SIMULATION RESULTS
The bit error rate (BER) performance of the cellular MCCDMA system is evaluated by computer simulations. The
system parameters of the MC-CDMA system and of the
channel model were taken from Reference [7] and are
shown in Table 1. All BSs are using exactly the same system parameters, i.e. the same spreading length L, number
of active users Nu etc. This implies that if Nu decreases, the
cellular interference also decreases. However, the difference in received signal power between interfering BSs,
E, does remain constant. The channel is modeled by a
tap delay line model with Q0 12 taps, a tap spacing of
16 Tspl , with an exponential decaying power delay
ANALYSIS OF CELLULAR INTERFERENCE FOR MC-CDMA AND ITS IMPACT ON CHANNEL ESTIMATION
the fully synchronized system. The reason is, for the linear
MMSE equalizer of Equation (7) the interference term is
assumed to be WGN, which is exactly the case for the GA.
However, it is seen in Figure 5 that the GA curves closely
match the performance of the fully synchronized cellular
system, especially for high system loads. Moreover, it
was reported in Reference [8] that the GA is accurate even
for lower spreading factors. Obviously, if results for the
GA and a perfectly synchronized system are equivalent,
cellular interference with arbitrary synchronization offsets
are accurately modeled by the GA. This is a somewhat
unexpected result since on a link level the MAI of MCCDMA cannot be modeled by the GA. Only for the single
user case the difference between GA and the fully synchronized system is more than 0.5 dB. In this case, the interfering signal of the fully synchronized system is caused by
one users signal only. Therefore, the degradation with
respect to the GA increases.
Since the GA accurately describes the system performance even for a fully synchronized system, the signal
from the interfering BSs is observed as WGN at the
I
mobile. This means that the interfering signal, Z;i
I I
X;i H;i from Equation (20), is sufficiently decorrelated
by cell specific interleaving, scrambling and spreading.
Figure 6(a) shows the BER versus E for the considered MC-CDMA system with half load, Nu 8, having
NR 1 and two receive antennas. Clearly, there is a significant diversity gain if two receive antennas are used. This
is true for any system scenario. It is seen that the difference
between System A and the receiver with perfect CSI for
one receive antenna is about 1 dB. It appears that the
degradation of the channel estimator due to interference
is proportional to the overall performance. If the interference is high the performance is so poor that the additional
degradation due to channel estimation is not a major problem. On the other hand, if the interference is low, the pilot
symbols are not corrupted by interference as well. The performance for System B, where the pilots are received without interference, is roughly in-between the curves for
System A and perfect CSI. At high interference levels
the performance of System B is close to the case of
perfect CSI. Clearly, as the interference decreases (E
becomes larger) System B approaches the performance
of System A.
Figure 6(b) shows the BER versus E for a MC-CDMA
system with a synchronization offset between the interfering BSs and the mobile of f I 0:5=T and nI 300.
With a synchronization offset the performance of System B
remains superior, especially for high interference levels
(E is low). Since, the receiver which utilizes antenna
Copyright # 2004 AEI
181
diversity performs better at low E, this effect is more pronounced with NR 2. It appears that the interference
levels at pilot positions of System B are still lower than
for System A. Thus, the synchronization requirements
between two interfering BSs are not as strict as between
the transmitting BS and the mobile receiver. For instance,
on the link level the frequency offset between transmitter
and receiver should not exeed 25% of the subcarrier spacing [25]. On the other hand, the frequency offset between
two interfering BSs can be significantly higher than that,
without a degradation in performance.
Note, System B transmits zero subcarriers at a ratio of
2/25, due to the higher pilot reuse. This will result in
decreasing ICI and ISI in case of a synchronization offset,
compared to System A. This should be taken into account
when comparing System A and B.
Figure 7 compares the BER performance between
System A and B for a two antenna receiver, dependent
on the interferer frequency offset f I T. The time synchronization offset was set to zero. All other parameters
Euro. Trans. Telecomms. 2004; 15:173184
182
G. AUER ET AL.
6. CONCLUSIONS
The performance of the downlink of a celluar MC-CDMA
system with 2 1-D PACE has been analyzed.
Copyright # 2004 AEI
ACKNOWLEDGEMENT
REFERENCES
1. Weinstein S, Ebert P. Data transmission by frequency division multiplexing using the discrete fourier transform. IEEE Transactions on
Communications Technology 1971; 19:628634.
2. Fazel K, Papke L. On the Performance of Convolutionally-Coded
CDMA/OFDM for Mobile Communication Systems. In Proceedings
of IEEE International Symposium on Personal, Indoor and Mobile
Radio Communication (PIMRC93), Yokohama, Japan, September
1993, pp. 468472.
3. Yee N, Linnartz J, Fettweis G. Multi-Carrier CDMA in Indoor Wireless Radio Networks. In Proceedings of IEEE International Symposium on Personal, Indoor and Mobile Radio Communication
(PIMRC93), Yokohama, Japan, September 1993, pp. 109113.
4. Chouly A, Brajal A, Jourdan S. Orthogonal Multicarrier Techniques
Applied to Direct Sequence Spread Spectrum CDMA Systems. In
Proceedings of IEEE Global Telecommunication Conference (Globecom93). Houston, TX, USA, November 1993, pp. 17231728.
5. Abeta S, Atarashi H, Sawahashi M. Performance of Coherent MultiCarrier/DS-CDMA and MC-CDMA for Broadband Packet Wireless
Access. IEICE Transactions on Communications 2001; E84-B:406
414.
6. Atarashi H, Abeta S, Sawahashi M. Variable Spreading FactorOrthogonal Frequency and Code Division Multiplexing (VSFOFCDM) for Broadband Packet Wireless Access. IEICE Transactions on Communications 2003; E86-B:291299.
7. Kishiyama Y, Maeda N, Higuchi K, Atarashi H, Sawahashi M.
Experiments on Throughput Performance Above 100-Mbps in Forward Link for VSF-OFCDM Broadband Packet Wireless Access. In
ANALYSIS OF CELLULAR INTERFERENCE FOR MC-CDMA AND ITS IMPACT ON CHANNEL ESTIMATION
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
183
AUTHORS BIOGRAPHIES
Gunther Auer received his Dipl.-Ing. Degree in Electrical Engineering from Universitat Ulm, Germany, in 1996, and his Ph.D. from
the University of Edinburgh, U.K., in 2000. From 2000 to 2001 he was a research and teaching assistant with Universitat Karlsruhe
(TH), Germany. Since 2001, he is a senior research engineer at NTT DoCoMo Euro-Labs, Munich, Germany. His research interests
include multi-carrier based communication systems, multiple access schemes and statistical signal processing, with an emphasis on
channel estimation and synchronization techniques.
Stephan Sand received his M.Sc. degree in electrical engineering from the University of Massachusetts Dartmouth, U.S.A. and the
Dipl.-Ing. degree in communications technology from the University of Ulm, Germany, in 2001 and 2002 respectively. He is currently
working toward his Ph.D. at the Institute of Communications and Navigation, German Aerospace Center (DLR), Oberpfaffenhofen,
Germany. His main research interests include various aspects of mobile communications and signal processing, such as time-frequency methods for signal processing, space-time signal processing, MC-CDMA, channel estimation and multiuser detection.
Armin Dammann received his Dipl.-Ing. degree from the University of Ulm, Germany, in 1997. He is currently working toward his
Ph.D. at the Institute of Communications and Navigation, German Aerospace Center (DLR), Oberpfaffenhofen, Germany.
Stefan Kaiser received his Dipl.-Ing. degree and Ph.D. in electrical engineering from the University of Kaiserslautern, Germany, in
1993 and 1998 respectively. Since 1993, he is with the Institute of Communications and Navigation of the German Aerospace Center
(DLR) in Oberpfaffenhofen, Germany, where he is currently head of the Mobile Radio Transmission Group. In 1998, he was a visiting
researcher at the Telecommunications Research Laboratories (TRLabs) in Edmonton, Canada, working in the area of wireless
Copyright # 2004 AEI
184
G. AUER ET AL.
communications. His current research interests include multi-carrier communications, multiple access schemes and space time processing for mobile radio applications. Dr. Kaiser is co-organizer of the international workshop series on multi-carrier spread spectrum
(MC-SS), and he is co-author of the book Multi-Carrier and Spread Spectrum Systems (John Wiley & Sons, 2003) and co-editor of the
book series Multi-Carrier Spread Spectrum & Related Topics (Kluwer Academic Publishers, 20002004). He is also guest editor of
several special issues on multi-carrier spread spectrum of the European Transactions on Telecommunications (ETT). He is co-chair of
the IEEE ICC 2004 Communication Theory Symposium. Moreover, Dr. Kaiser is organizer and lecturer of the seminar series on Wireless LANs at the Carl-Cranz-Gesellschaft (CCG) in Oberpfaffenhofen, Germany. He is a senior member of the IEEE and member of
the VDE/ITG.