Sunteți pe pagina 1din 5

Multi-User Spreading Codes Retaining Orthogonality through Unknown Time- and Frequency-Selective Fading

Geert Leus , Shengli Zhou , Georgios B. Giannakis


Dept. of Elec. Engr. (ESAT), Katholieke Universiteit Leuven, Kasteelpark Arenberg 10, 3001 Leuven, Belgium Dept. of Elec. and Comp. Engr., University of Minnesota, 200 Union Str. S.E., Minneapolis, MN 55455, USA time-frequency RAKE receivers. Allowing for minimal coordination among users agrees with CDMA philosophy. The rest of this paper is organized as follows. In Section II, we introduce the transceiver model that we will use. Section III then describes the proposed transceiver. In Section IV, we investigate blind channel estimation. Finally, simulations are presented in Section V. Notations: We use upper (lower) bold face letters to denote matrices (column vectors). Superscripts , , and represent conjugate, transpose, and Hermitian, respectively. We reserve for expectation, and for integer ooring. We denote E . The all-zero mathe discrete impulse function as , the identity matrix as , and trix is denoted as the unitary FFT matrix as . We dene as the th column of and as the th column of . Furthermore, we dene as the diagonal . Fimatrix with main diagonal as the th entry of matrix nally, we dene . II. T RANSCEIVER M ODEL The block diagram in Fig. 1 describes the CDMA system with block spreading studied in this paper. Before transmis) symbol stream sion, the th users ( is rst serial to parallel converted into a stream of symand then bol blocks spread by an matrix to obtain a stream of chip blocks . The chip blocks are nally parallel to serial converted into a chip stream
% % Gl " g w % Gl " g j g qxrel " g w v om % g v Re(u0 ' ( u0 i %% R d XVtsa`rl 4BVBVXql &af 9pnel kaj ' T R P " g f i x x x i % ' " g " om % " g I R ( i'  R d i x x x i  weBVB8a6 " g 4hf x % " g w om %% R d Xel &|pr~}U|{l 44BVBezl &ky " 0 T R P " g y i x x x i % 0 " g I T URP@$iP R I % q s q p i x yxBVxVi s q p i R 8rh" 7 v G t vT t wu Yc 0g(f0 7 d eI % " 7 &$ YXW T R a`PQ$ I % &$ "HGF 7 7ED 7BA 0 ( @90 % " &$ #!  0 ( 1)'     7 5 3 8642 7 VA 1 " % 7 bD T R P USQ$ I 0 ( 6C0  

Abstract Suppression of Multi-User Interference (MUI) and mitigation of time- and frequency-selective effects constitute major challenges in the design of third-generation wireless mobile systems. Relying on block spreading and judiciously chosen time-frequency guard intervals, we propose a multi-user transceiver that eliminates MUI deterministically and guarantees symbol detectability in the presence of unknown time- and frequency-selective fading. Blind channel estimation is also investigated. Simulation results demonstrate the validity of the theoretical results and show improved performance of the proposed transceiver over a multi-user time-frequency RAKE receiver.

I. I NTRODUCTION Recently, various multi-user transceivers have been proposed that eliminate MUI deterministically and guarantee symbol detectability in the presence of unknown frequencyselective fading [2], [8], [3], [9]. They serve as attractive alternatives to multi-user RAKE receivers [10], which are very complex, do not guarantee symbol detectability, and require the knowledge of the spreading codes of all active users. However, all these transceivers are based on the assumption that the channels are time-invariant over the transmitted frame. In a practical scenario, this assumption may not hold true due to high-mobility and carrier frequency and phase drifts. Relying on block spreading and judiciously chosen time-frequency guard intervals, we here propose a multi-user transceiver that eliminates MUI deterministically and guarantees symbol detectability in the presence of unknown time- and frequencyselective fading. It serves as an attractive alternative to multiuser time-frequency RAKE receivers [5], which, like the multiuser RAKE receivers, are very complex, do not guarantee symbol detectability, and require the knowledge of the spreading codes of all active users. Through judicious block spreading code design at the transmitter, the proposed orthogonal multi-user transceiver transforms a multi-user communication problem into a set of parallel single user communication problems with matched ltering, regardless of the underlying time- and frequency- selective channels. Compared with multi-user time-frequency RAKE receivers, the proposed transceiver has the following advantages: i) It is in general less complex since it does not operate on the received blocks, but on the output of a multi-user separator, which has a shorter block length. ii) It guarantees symbol detectability through controlled redundancy. iii) It does not require any knowledge of the other users. Thus the detection is uncoordinated among users, unlike multi-user
supported by the FWO-Flanders (Belgium). supported by NSF Wireless Initiative grant no. 99-79443.

where is the additive noise and is the time- and frequency-selective channel for the th user, including transmit can be and receive lters. We assume that the channel written as
i % XuzX" g 0 $ % d i " $ 4g % " $ 4g 7 Xw u t d { &% $ 4g o i " d % " &$ {

0-7803-7206-9/01/$17.00 2001 IEEE

259

hUR

After chip rate sampling (the chip rate is denoted as the received sample stream can be written as
d d q

i % " P X&$ &e#%

d " g y $ &4%

s 8q p

" $ 4g

% gd

v ag

" #!

o % &&$ "

),

(1)

Fig. 1. Discrete-time equivalent baseband system model (only th user shown).


7

III. P ROPOSED T RANSCEIVER For the proposed transceiver, we set the symbol block length and set the transmitted chip block length to . Let us dene the zeroand the inserting matrix as zero-inserting matrix as . Denoting as the Kronecker product, we then design as the matrix given by
P o 'Q( e 0 I 0 c p( g v % 5 s wA " o q vuYv 2 i  % 5 sui rti 5 s2 " r 2 A r A@ hP c T g I i q x T g q y I o g HHv o q T iP 8 e A T@ hP g I fdo e c x e 4 u T 8 ( c 'I

to

We next show that the designed transceiver pairs can indeed achieve deterministic multi-user separation and guaranteed symbol detectability, without knowing the underlying time- and frequency-selective channels. In the following derivations, we will continuously make use of the formula for the product of Kronecker products of matrices with matching dimensions, which is given by
x Ta v h I x B T I o U n T x T I x I

(4)

260

a# UV$ % d

dQ$ #`r % r " " ! o T S 7 7 br

Let us also introduce the notation Toeplitz matrix with


0 ( 0

, which represents the . First,

{Ug 8 g  g Yk i q

eel " x %

D % " j Q el kgagHel " g 7 P v %

F c E H 7

{Ug

o % {el " C

On the MUI-free output , we can then apply any single user equalizer to mitigate the Inter Symbol Interference (ISI). We can, e.g., adopt a linear single-user equalizer to obtain .
% " g el 4ba % " el &g %el "kg ` %el &g " a pm % g Frel "kaj o b

38@7 2 5

v % Gl " g 7

. To eliminate the MUI deterministically and guarantee symbol detectability for the th user, regardless of the delay and by a matrix to Doppler spread proles, we despread obtain (2)
g 6Y % Gl " C x % eel " C g % r 6WX gd 0 P $V U ` Y p % FmGl " g o $ l " wg ` "

ag Gel kgarg q  % " j

o % " &el kga w

0 e(

where
I

represents the

Toeplitz matrix with

Note that a more general transceiver design for time- and frequency-selective channels is described in [4]. and , For time-invariant channels, we can set , which reduces to the Chip Intersuch that leaved Block Spread (CIBS) CDMA proposed in [9] with the user signature codes being taken from an FFT matrix. Unlike [9], which focuses on time-invariant channels and only intro) in the time domain, we here deal duces redundancy (by with time-varying channels and introduce redundancy both in ) and in the frequency domain (by ). the time domain (by This is illustrated in Fig. 2, where we show a time-frequency plot of . is that Inter One immediate consequence of the design of Block Interference (IBI) is removed. Because the last rows of are zero, . We then only need to perform block by block processing based on the following IBI-free blocks: (3)
R q o c o A q q x %y " W q o g v

xXVR %%

A @| X c TaRSP T gP

A @S X c T gP

" Vr VVVx W i x x

x x x i A @t BVBe%( c T gP

A @| X c T gP

" Vr W "

where
g y

is the

T QP 8

v wA

A @S X c T gP

oxm

gY

and . This model, which was also used with is sufciently large, the users in [1], [6], [5], [7], is valid if is larger than are quasi-synchronous in time and frequency, or equal to the maximal delay spread plus time-offset of all users, and is larger than or equal to the maximal Doppler spread plus frequency-offset of all users. Analytical results revealing the tting accuracy of this basis expansion model can be found in [4]. At the receiver, we collect received samples into vectors. , and Dening , the block channel input/output relationship can be described by (see [8] for time-invariant channels):
pm el " D o % n8 %% R qd 0aR T  ~d %% R 0 T R U|Pl {iBVVXzl & " " x x x i % 0 " P l " i VBxVXzl " " x x i % 0 o I % xmel " C I IPI GH7 F7 E 0 0 9 BA T 60 I q A 0 9 @8

and

as the

Xel " i %

S/P

P/S

S/P

interfering users

matrix given by

matrix dened as

 

 6

 

453

 10

( '

 () $ " &%# Q VR D T % P   % el " g j g d l 4aH%Gl " g 7 " g j g v v % el " g 7   2 I P c q  {Ug %el " g 7  R # %el " g 7 o ST %  R o % el " C I

% " el &g

e%Gl " g 

6 

q VVg r y

q g y

el " wg % % Gl " q wg

HH
T% el " wg I

x I B e T 8s P I

i % Xel " D g

YX el 4aj el P % " g %

" g o % kt Gl " C g Y

A B

o % " &el &g

wg b7 I

( & $ )'  %"#!


freq.

$ %"#


99 $ " @

 



 

$ %"#

$ " 8 994
3

  
= user 2

84(
3

rag replacements

ampl.

(


= user 1

PSfrag replacements

( @

Fig. 2. Time-frequency representation of


time

we note that

ampl.

can be expressed as (see [9] for details):

From (6), (7), (9), and clear that

% d 1" ! o U r A kg8 #`rS g

g

(5)

S Hel " S g 7 g v% I

G H7 E

7 @c

g

v I

el " g T% % Gl " g

v Vr I x v Vr A x I I

Using (5) and the fact that obtain:

, we

Hence, the designed transceiver pairs eliminate the MUI deterministically, regardless of the delay and Doppler spread proles. Using (10) to (2) and (3), we arrive at the following single user output:

{Ug q

kg v   kY i Yg

x el " wg T% I

r wA T I

6
x

% ygd U #|o " !

(6)

where is given by

r tA T ' I zT (

Tel " g %

C7o
%el " g

I o pm c

I x %Gl 4t " g

Fig. 3. Resulting set of parallel single user systems.

is the

A (g

% " el &gt

2 1

o xm

v x e% T q %el " wg rwA % " vwA T g q " o x x yI I%I v 38@7k2 PT q el " g 5 % T g q o x I y I %I v q T q g "VTel " g % P r x y I I I x v Vr A % q T q g "VTel " wg % o x x y I I I " X%el " g 7 % q x T q g yI I v 2 o q %Gl " g

v 5

Next, it can be shown that

HHIG

el " %

g

. . .

, which has the following form:

..

HH

E EE DE 6 EE

(7)

..

. . .

6 F 6
% Gl " g

G x H7 E

7pom 6 2 1 v c 0
x

qg

o % " g &el kt

5
G

7 E

v c

In the Appendix, we also prove that

Therefore, a multi-user communication problem has been converted into a set of parallel single user communication problems, as depicted in Fig. 3. More important, since the mais square unitary, Maximum Liketrix lihood (ML) optimality is preserved in this multi-user separation step, as in the case of [9] for time-invariant channels. On the other hand, since the nonzero matrices from the set always have full column rank, always has full column rank. Hence, the proposed choice for guarantees symbol detectability, regardless of the delay and Doppler spread proles. , we can then apply any singleOn the MUI-free output user equalizer to mitigate the ISI. Here for brevity, we assume

Vr

Y VBVV i x x x i

Y"

o pm

Vr c

5 ' 5 '

Vr Vr c y v A % T q g "r x y I Vr Vr c x v wA T q g " G 7 E x y I

P o &%el " g 7 I

Dening plify (7) as:

and using (8), we can further sim-

% " Gl &g|

{Ug yg q

Y iyHv  g

(9)

(8)

261
x T

6
x

T % I

v wA v A x x

g T

y I q g

o y I G E " H7 F7 c

q VVixm r o

4 4
7

o c

, it is nally

matrix that (11) (10)

or the linear MMSE equalizer given by


i % eel " g

A. Spectral Efciency Within the received block , symbols are transmitted per user. Hence, the spectral efciency is
x ' % el " C I e fc I o ' @ R ( o 0

On this data model we can then apply the second deterministic blind single user channel estimation method presented in [6] (method II). Using this method, is identiable from sup if and only if sup has full row rank. Hence, idensup . Note that for sup to tiability is independent of sup . have full row rank, it is necessary that V. S IMULATION R ESULTS In this section, we illustrate the ideas presented in this paper with computer simulations. We consider BPSK modulation and additive white Gaussian noise. Test Case 1: Lets rst consider the same fading scenario as in [5] and compare the proposed transceiver with the multi-user , time-frequency RAKE receiver [5]. As in [5], we take , , and generate a set of independent channels (see (1)) with complex Gaussian distributed with variance 0.9 if and 0.05 if . We as. For sume that the receiver knows the channels corresponds to the length of the the proposed transceiver, transmitted chip block, while for the multi-user time-frequency RAKE receiver, corresponds to the CDMA spreading factor. , , and For the proposed transceiver, we consider (hence, ). Note that in chip periods, the proposed transceiver can handle ), 2 bits per user (the symbol block length is while the multi-user time-frequency RAKE receiver can only handle 1 bit per user. Keeping this important rate difference in mind, Fig. 4 depicts a comparison between the performance of the proposed transceiver with linear ZF equalization and the performance of the linear ZF multi-user time-frequency RAKE receiver (the latter performance corresponds to the one shown in [5, Fig. 3]). From Fig. 4, we observe that the performance of the proposed transceiver accommodating 7 users (that transmit 2 bits/63 chips) is comparable to the performance of the multiuser time-frequency RAKE receiver accommodating only 2 users (that transmit 1 bit/63 chips), and much better than the performance of the multi-user time-frequency RAKE receiver accommodating 7 users (that transmit 1 bit/63 chips). More . specically, we gain about 4 dB for a BER of Test Case 2: Since the previous fading scenario is rather articial, lets now consider a more realistic fading scenario. We
% el " g

IV. B LIND C HANNEL E STIMATION To perform channel equalization, channel knowledge is required at the receiver. Since the multi-user system is converted into a set of parallel single user systems, single user channel estimation methods can be applied. Here, we will focus on

262

R 9o

o f i c e

q kBR

$" @ 8 %0 T iP o ( )o

'

A@ (hP T g

$" %1

o 0 g Bo

The spreading operation requires multiplyadd operations per user and block. Hence, the spreading comper user and block. plexity is The despreading operation requires multiply-add operations per user and block. Hence, the deper user and spreading complexity is block. Let us now consider the despreading complexity at the base station (BS), where we need to extract all users information. At rst sight, it seems that the BSs despreading complexper block. However, by making ity is use of a -point FFT, the BSs despreading complexlog per ity reduces to block. Finally, the complexity of the single-user equalizer depends on the type of equalizer that is used. For the linear ZF or MMSE equalizer described earlier, we obtain a complexity of per block; or per symbol.
T dP 8 P e  T 8 I T ( @ g A e I P T  @ g A P ec e fc T e T T  T I I  A (g P 8q e P

ag G% $ " g q  d H o R

& '

B. Complexity
P Sc I

$" %#o

o q@ % uul " g

q  {Ug e% $ 4g " d R uA R o o

By tuning the parameters, we can change the spectral efciency. However, keep in mind that (1) is only valid if is larger than or equal to the maximal delay spread plus timeis larger than or equal to offset of all users and the maximal Doppler spread plus frequency-offset of all users. are determined by Hence, the minimal values for and the chip rate, the propagation environment and the time- and frequency-offsets. Maximum spectral efciency designs are discussed in [4].
8 0 A T r60 8 I A

 

 !

% el "

'

% el "

% el "

sup

sup

sup

x%%el "

g j i x x x i % aBVBeel "

% g aj " Gl "

g t o

A B

% el "

T QP 8

A g` (i c P

A (g

A @S X c T gP

 T

 f c e

 % e&$ "  &$ rU f% " f

where E straightforward.
o

. The extension to colored noise is

03

sup

sup

% el "

 

% el "

21

sup

sup

i x x x i % BVVBeel "

g "

A el " %

8g 5 e P

A(gi c h1 T P @S  I T@ g P c A ( A g (uP c I  I

g

c T P 8

mmse

%% el "

i % eel "

g

P e%l "

T% " q Gl &g

A% Gl "

g

zf

B

 wR

d E i VBV8Ug x x x i 

 

that the additive noise is white with covariance matrix . Because the matrix is unitary, the resulting noise is still white. With white noise, we can, e.g., adopt the linear ZF equalizer given by
g Y % el " D I o % &el " g a % el " D Y g o % &el " I 7

blind channel estimation. However, it is also possible to insert training symbols and use training-based or semi-blind channel estimation at the expense of a decreased spectral efciency [7]. Suppose that is a multiple of and that the users sup , with sup , are actually one and the same user, referred to as the sup th super user. For this sup th super user, we can write
q {

  

A @S ( c T gP

T  @ g A

o 

g a P  c T 8 e f c 1  I

where is the number of delays for the th user, is the number of Doppler shifts for the th delay of the th user, and , , and are the amplitude, phase and Doppler frequency for the th Doppler shift of the th delay of the th , user, respectively. We adopt Jakes model and take , and cos (this corresponds km/h for a carrier frequency of to a relative mobile speed of and 900 MHz). Further, we consider a uniformly distributed random variable in . Applying this fading scenario, the channel , dened above, can , dened in (1), not exactly be modeled by a channel , but can be approximated (in LS sense) by a channel when and (because should be larger than or equal to the maximal Doppler spread of all Hz). For a transmitted block users, which is chosen to be and a chip rate of kHz, we size of . We choose the following transceiver paramneed , , , and eters: (hence, ). Again assuming that the receiver knows the channels , Fig. 5 shows the performance of the proposed transceiver using linear ZF equalization for two scenarios: 1) we use the true channels for propagation; 2) we use the approxifor propagation. The difference mate channels in performance between these two scenarios is a measure for the validity of (1). We see that the performance of scenario 1 is comparable to the performance of scenario 2 up to a certain SNR per user. At higher SNRs per user, modeling errors cause the performance of scenario 1 to saturate. A PPENDIX : P ROOF OF (8) . The matrices in both sides of (8) are We will prove (8) column by column. The th column of is the th column of and thus can be written as:
% l GHP A P T  B@ g A P 1c I " r W P{Xa P| A(g c " Vr So i l @| % Q A T P P W I r r c q e c @T ( fQ@ g ( A l P c I g 4

BER

due to the specic structure of


g

which can easily be veried. is the th Similarly, the th column of . Because is a column shifting column of th column of will be the matrix, the th column of , which is . Comparing the latter with (12), we have veried (8). R EFERENCES
[1] G. B. Giannakis and C. Tepedelenlio lu, Basis Expansion Models and g Diversity Techniques for Blind Equalization of Time-Varying Channels, Proc. of the IEEE, pp. 1969-1986, Oct. 1998.
Ul % aP P5 w P A T l P YA I I P z P A T@ 6c A gP  I q T " Vr W r VVg y r VV r g y I g y T l P A I r VVTg y l T l

& )o

BER

(12)

" $ 4g

  u t
o g 8 g T r60 A wg I

q  ag G% $ kg " d

T  i rg 8 " g A "UR !

s 8q p

T l

o grhUR

R wBCo o )8

 s
T P  A I

$ "4g % $ &g " d

generate a set of independent channels


p g

q  {Ug G% $ 4g " d

, with

10

10

proposed: 7 users (2 bits/63 chips) multiuser TF RAKE: 2 users (1 bit/63 chips) multiuser TF RAKE: 4 users (1 bit/63 chips) multiuser TF RAKE: 7 users (1 bit/63 chips)

g T V @ w g A

o g

 %o

T dP 8



" %


e q g

{Ug G% $ " g  q  d q  {Ug e% $ " g  d

e T (g A I (o &

$% 6w A  0 #

w o

5 !

% y y

d g h g

5! 5 !

" #! P A

r r

R wB)o

(R

& # '%A

g 4

o &%

o i0 o c

R wa}o

Vr

Vr

# $8

g 8 0

" $ 4g

g

g

10

c c

10

10

10

10 SNR (dB)

15

20

Fig. 4. Average BER as a function of the SNR per user.


0

10

scenario 1 scenario 2

10

10

10

10

6 8 SNR (dB)

10

12

14

Fig. 5. Comparison between true channels and approximate channels.

[2] G. B. Giannakis, Z. Wang, A. Scaglione, and S. Barbarossa, AMOUR - Generalized Multi-Carrier Transceivers for Blind CDMA Regardless of Multipath, IEEE Trans. on Commun., pp. 20642076, Dec. 2000. [3] G. Leus and M. Moonen, MUI-Free Receiver for a Synchronous DSCDMA System Based on Block Spreading in the Presence of FrequencySelective Fading, IEEE Trans. on SP, pp. 31753188, Nov. 2000. [4] G. Leus, S. Zhou, and G. B. Giannakis, Multiple Access Regardless of Time- and Frequency-Selective Fading, IEEE Trans. on IT, submitted for publication, June 2001. [5] A. M. Sayeed, A. Sendonaris, and B. Aazhang, Multiuser Detection in Fast-Fading Multipath Environments, IEEE JSAC, pp. 16911701, Dec. 1998. [6] C. Tepedelenlio lu and G. B. Giannakis, Transmitter Redundancy for g Blind Estimation and Equalization of Time- and Frequency-Selective Channels, IEEE Trans. on SP, pp. 20292043, July 2000. [7] T. A. Thomas and F. W. Vook, Multi-User Frequency-Domain Channel Identication, Interference Suppression, and Equalization for TimeVarying Broadband Wireless Communications, Proc. of Sensor Array and Multichannel Signal Processing Workshop, 2000, pp. 444448. [8] Z. Wang and G. B. Giannakis, Wireless Multicarrier Communications: Where Fourrier Meets Shannon, IEEE SP Magazine, pp. 2948, May 2000. [9] S. Zhou, G. B. Giannakis, and C. Le Martret, Chip-Interleaved BlockSpread Code Division Multiple Access , IEEE Trans. on Commun., 2002 (to appear); see also Proc. of CISS, 2001. [10] Z. Zvonar and D. Brady, Linear Multipath-Decorrelating Receivers for CDMA Frequency-Selective Fading Channels, IEEE Trans. on Commun., pp. 650653, June 1996.

263

S-ar putea să vă placă și