University of Calgary PRISM: University of Calgary's Digital Repository Graduate Studies Legacy Theses 2001 Group-blind multiuser detection for CDMA systems Yu, Jae-Chon Yu, J. (2001). Group-blind multiuser detection for CDMA systems (Unpublished master's thesis). University of Calgary, Calgary, AB. doi:10.11575/PRISM/14237 http://hdl.handle.net/1880/40698 master thesis University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. Downloaded from PRISM: https://prism.ucalgary.ca
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
University of Calgary
PRISM: University of Calgary's Digital Repository
Graduate Studies Legacy Theses
2001
Group-blind multiuser detection for CDMA systems
Yu, Jae-Chon
Yu, J. (2001). Group-blind multiuser detection for CDMA systems (Unpublished master's thesis).
University of Calgary, Calgary, AB. doi:10.11575/PRISM/14237
http://hdl.handle.net/1880/40698
master thesis
University of Calgary graduate students retain copyright ownership and moral rights for their
thesis. You may use this material in any way that is permitted by the Copyright Act or through
licensing that has been assigned to the document. For uses that are not allowable under
copyright legislation or licensing, you are required to seek permission.
Downloaded from PRISM: https://prism.ucalgary.ca
THE UMNERSITY OF CALGARY
Group-Blind Multiuser Detection
for CDMA Systems
by
Jae-Chon Yu
A THESIS
SUBMITTED TO THE FACULTY OF GRADUATE STUDIES
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF MASTER OF SCIENCE
DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING
CALGARY, ALBERTA
JANUARY, 2001
@ Jae-Chon Yu 2001
National Library 1+1 dcalada BiblioWque nationale du Canada
uisitions and 9- Acquisitions el Bib lographic Services services bibliographiques
The author has granted a non- exclusive licence allowing the National Library of Canada to reproduce, loan, distriiute or sell copies of this thesis in microform, paper or electronic formats.
The author retains ownership of the copyright in this thesis. N e i k the thesis nor substantial extracts fiom it may be printed or otherwise reproduced without the author's permission.
L'auteur a accord6 me licence non exclusive pennettant A la Bibliothwe nationale du Canada de reproduke, pr&ery distxibuer ou vendre &s copies de cette these sous la forme de microfiche/film, de reproduction sm papier ou sur format electronique.
L'auteur conserve la propriete du droit d'auteur qui proege cette Wse. Ni la these ni des extraits substantiels de celle-ci ne doivent &re imprimes ou autrement reproduits sans son autorisation.
Abstract
We will discuss group-blind multiuser detectors which can reduce intra-cell and
inter-cell interference efficiently. These detectors have better performance com-
pared to conventional multiuser detectors and combat the near-far problem.
Estimated groupblind multiuser detectors which use blind channel estimation
and subspace tracking algorithm are proposed for real time implementation in
DS-CDMA systems. These estimated group-blind multiuser detectors have rea-
sonable performance with low calculation complexity.
Due to the asymmetric allocation of uplink and downlink time slots in Univer-
sal Mobile Telecommunications System Terrestrial Radio Access Time Division
Duplex (UTRA-TDD), the performance of mobile stations can be degraded se-
riously by inter-cell interference. Groupblind multiuser detectors are therefore
proposed to mitigate this problem.
Acknowledgments
The author would like to thank Dr. Andeis Hast-Madsen for his supenision and
guidance of the work presented in this thesis. The author would also like to thank
TRLabs and the University of Calgary for their funding support. Finally, this work
would never have been completed without the encouragement, understanding, and
is found. With the eigen component and the received signal, we can estimate
the information data.
@
4.2 Synchronous Estimated Detectors
cakum E i m
-POnm wfth S W sum=. Tracking
The estimated multiuser detectors for the synchronous system are listed as fol-
Iows:
E8timate C o t T e I ~
matrlx R
Blind Linear MMSE detector: Direct method
Group=blind MUD
& [n] = sgn (s~R- ' [n] r [n] )
e
Blind MMSE detector: subspace method
% w = sgn (s~u, [n] A; [n] c[n]r[n])
4
4.2 Synchronous Estimated Detectors 84
Blind Linear Zero-forcing detector: Subspace method
Groupblind Linear Zero-forcing Detector: Form I
A
T -Ts -1-T 6[n] = sgn (1, (S S
x [I - ~ [ n ] 6, [la] (is [n] - b21) -'u: [n]]r [n]
Groupblind Linear Zer-forcing Detector: Form I1
Group-blind Linear Hybrid Detector: Form I
4.3 Asynchronous Estimated Detectors 85
Group-blind Linear Hybrid Detector: Form I1
Group-blind Linear MMSE Detector: Form I
Group-blind Linear MMSE Detector: Form I1
4.3 Asynchronous Estimated Detectors
The estimated multiuser detectors for the asynchronous system are as follows.
a Blind Linear MMSE detector: Direct method
& [n] = sgn [n] r [ I )
4.3 Asvnchronous Estimated Detectors 86
Blind MMSE detector: subspace method
$[n] = sgn (hf~,[n]A;'[n]~f[n]r[n])
Blind Linear Zero-forcing detector: Subspace method
ik [n] = sgn (~,Hu, [n] (A, [n] - BZI) -'u,H [n] r [n]) (4.17)
a Group-blind Linear Zero-forcing Detector: Form I
gc[n] = s g n ( ~ $ ~ + ~ ( R ~ R ) - l ~ ~
x [I - ~ [ n ] U, [n] (A, [n] - d21) -6: [n] ] r [n]
Group-blind Linear Zero-forcing Detector: Form I1
Groupblind Linear Hybrid Detector: Form I
4.4 Simulation Results 87
Group-blind Linear Hybrid Detector: Form I1
Bk [n] = s p n ( I$~+, [HHU. [n]~; [n]~ ," [ n ] ~ ]
Group-blind Linear MMSE Detector: Form I
C
( T ( i jq + &)-I HH 6 k b I = sgn I,,,
' * ) ) (4.22) x (I - b [ n ] ~ . [ n ] ~ , [n]Us [n] r[n]
Group-blind Linear MMSE Detector: Form I1
- bk[n] = sgn (~TK,,, ( H ~ H + C ~ I ) - ' H ~ [I - ( ~ . [ n ] i ~ ; ~ [ n ] )
x (fi [n] A, [n] fir [n]) (4. [n]~;~[n]) fi[n]] r [n]) (4.23)
4.4 Simulation Results
In this simulation, several multiuser detectors are compared with the conven-
tional detector, the partial-MMSE detector, and the full MMSE detector. While
the partial MMSE detector can reduce only intra-cell interference, the full MMSE
detector can reduce both intra-cell interference and inter-cell interference with
the assumption that all spreading codes of both known and unknown users are
4.4 Simulation Results 88
known.
In Figure 4.2 and 4.4, the performance of the blind MMSE multiuser detec-
tors were evaluated. For both synchronous systems and asynchronous systems,
the blind MMSE multiuser detector using the subspace method outperforms the
blind MMSE multiuser detector using the direct method because it gains high
resolution from the subspace decomposition. After some symbols, the SINR of
the blind MMSE detector using the subspace method crosses over the partial
MMSE detector and converges to the full MMSE detector. In the case of random
code, it crosses over the partial MMSE detector and the conventional detector
faster than in the case of gold code because the partial MMSE detector and the
conventional detector have worse performance in the case of random code.
Figure 4.3 and 4.5 show performances of groupblind multiuser detectors. In
most cases, the groupblind multiuser detectors outperform the blind MMSE
detector using the subspace method.
4.4 Simulation Results 89
(a) gold code
(b) random code
Figure 4.2: Estimated blind MMSE multiuser detectors in synchronous DS-CDMA systems ( N=31, 6 known users, 4 unknown users, SIR=3dB, SNR=20dB)
4.4 Simulation Results 90
(a) gold code
(b) random code
Figure 4.3: Estimated hybrid groupblind multiuser detectors in synchronous DS-CDMA systems ( N=31, 6 known users, 4 unknown users, SIR=3dB, SNR=BOdB)
4.4 Simulation Results 91
(a) gold code
(b) random code
Figure 4.4: Estimated blind MMSE multiuser detectors in asynchronous DS-CDMA systems ( N=31, 6 known users, 4 unknown users, SIR=3dB, SNR=SOdB)
4.4 Simulation Results 92
(a) gold code
(b) random code
Figure 4.5: Estimated hybrid groupblind multiuser detectors in asynchronous DS-CDMA systems ( N=31, 6 known users, 4 unknown users, SIR=3dB, SNR=20dB)
4.5 Blind Channel Estimation 93
4.5 Blind Channel Estimation
In this section, we will discuss the estimation problem of the channel of a desired
user in asynchronous DS-CDMA systems. The channel of a desired user can be
estimated blindly with the knowledge of its own spreading code and the received
signal. The performance of the blind multiuser detector and the group-blind
multiuser detector will be evaluated by a simulation. To solve this problem, we
introduce the discrete time channel model in 4.5.1.
4.5.1 Discrete-time Channel Model
F'rom (3.5) and (3.9) , the n-th composite channel response during j-th symbol
is given by
where j = O , - - = ,6k; n = O,--- . P - 1. Decimation of hkb7n] into p sub-
sequences is written as
4.5 Blind Channel Estimation 94
From the fact that T = NTc and Tc = PA, hk,& i] can be given by
From the composite channel response ~ ( t ) given in (3.6), The sequence gk[2] is
obtained by sampling at rate & = e:
The length @pk) of the sequence {ijk[i]) is determined by the length of support
of &(t). Since gk(t) is non-zero only on the interval [dk + T ~ I , dk + r k ~ + Tc], we
have
The sequences gk,&] in (4.27) are obtained by down-sampling the sequence
{&[i]) by a factor of p, i.e., gki*,[i] = gk[ip+q], i = 0, -# , w - 1; q = 0, * - . ,p-1.
4.5 Blind Channel Estimation 95
From (4.27), ha, can be expressed as the convolution of q and a,,:
Denote
Then (4.29) can be written in matrix form as
4.5 Blind Channel Estimation 96
Finally, denote
Then, the composite channel matrix can be written as
where ek is an ( L ~ + 1) P x ppk matrix formed from the spreading code of k-th
user. For instance, when the over-sampling factor p = 2, we have
For other values of p, the matrix Cr is similarly constructed. Suppose that the
4.5 Blind Channel Estimation 97
user k is the user of interest and his spreading sequence (ck[O], - , c k [ N - 11)
is known to the receiver (and therefore ck is known). We next consider the
problem of estimating the channel vector & in (4.31) based on the received
signal r[i] in (3.12).
4.5.2 Blind Channel Estimation in White Noise
The correlation matrix of the received signal r[t] can be written as
where U, is the signal subspace orthonormal eigenvectors, U, is the noise sub-
space orthonormal eigenvectors, and A, is the signal subspace diagonal eigen-
value matrix. Fkom this the channel response gk can be estimated from the
orthogonality relationship [20] :
since U, is orthogonal to the column space of Hz and hk is in the column
space of H. Thus an estimation of the channel response gk can be obtained by
computing the minimum eigenvector of the matrix ( c~u~u ,c~) . We are here
using Kalman tracking [28] for the channel estimation. Kalman tracking has
4.5 Blind Channel Estimation 98
8 ( N x pk) complexity where pk is the length of the impulse response gk.
The estimation of the signal subspace Us will be outlined below. The pro-
jected received signal z( i) onto the noise subspace is obtained from
from the fact that U,U: = I - u , U ~ . Using (4.33), we have
We consider the following constrained adaptive filtering technique to estimate
the channel state:
min E { ~ ~ ~ c ~ z ( z ) I*) gk
subject to 1 lgkll = 1.
Among a number of algorithms that can be employed to solve the above con-
strained optimization problem. Here we use the following Kalman-type of algorithm[28]
for channel estimation.
4.6 Subspace Xkacking 99
with the initial condition K(0) = I. Once an estimate of the channel state gk
is obtained, the composite signature waveform of the desired user is given by
4.6 Subspace Tracking
Due to change of multipath and moving mobile stations, the channel for a user is
non-stationary in a real communications environment. SVD or EVD is need for
the high resolution of estimation. However, the calculation complexity of SVD
or EVD is very high ( 0(N3), where N is the dimension of the correlation matrix
of the received signal). Therefore, in a real-time implementation of the group
bIind multiuser detector, a reduced complexity updating algorithm for finding
the eigenvalues and eigenvectors is needed. There exist many subspace tracking
algorithms in the literature with various complexities, i. e., O ( N K ) , O(N K2),
0(N2K), or 0(N2). A survey of subspace tracking algorithms can be founded
in [29]. in the next two sections, we will briefly introduce two subspace tracking
4.6 Subspace Xkacking 100
algorithms which are well suited for the group-blind multiuser detectors.
4.6.1 FASIR Algorithm
FASIR stands for FAst Subspace Iteration with Ritz acceleration which has
6(NK2) complexity where N is the processing gain and K is the number of
known users. Consider the class of matrices of rank K:
fP ( t ) = us ( t ) C , ( t ) 2 ~ s ( t ) (4.38)
where Us(t) is an N by K matrix with orthonormal columns and C,(t) is diag-
onal real. If R(t) is replaced by its low-rank approximation, then we have
R(t) c &(t) is "the FASIR approximation." This shows why the approxima-
tions of FASIR and R(t - l)Us(t - 1 ) = U.(t - l)A,(t - 1 ) are equivalent if
attention is restricted to the subspace iteration. A simple algorithm [29] can be
4.6 Subspace Tracking 101
given by
U.(O) arbitrary m x K ; U.(O)~U.(O) = I;
C(0) = I, the identity matrix
For t = 0,1,2, - - -
w = [BU.(t - l ) E , ( t - I), r( t ) ] ;
Compute the N x K and K x K matrixUs ( t ) , C , ( t ) ,
In the SVD u.(~)c, ( t ) y H = W
goto t = t + 1 , (4.40)
where p = 0.99. The FASIR algorithm should satisfy following conditions:
Compute only a restricted subset of K eigenpairs, in order to decrease the
complexity.
Use the estimate of the previous time step (t - 1) as initial guess at step
a If possible, compute the K left singular pairs of a matrix D instead, such
that D D ~ = R, in order to reduce rounding errors.
4.6 Subspace Tkackhg 102 - - -
4.6.2 Noise Average Cross-terms Singular Value Decom-
position (NA-CSVD)
Although the PASTd (Projection Approximation Subspace Tkacking with defla-
tion technique) algorithm [32] has a O ( N K ) complexity, the deflation technique
causes stronger loss of orthonormality between eigen vectors and a slightly in-
creased computational complexity if N >> K. On the other hand, the NA-
CSVD algorithm has the advantage of maintaining the orthonormality of eigen
vectors and the descending order of eigen values by careful choice of the type of
Givens rotation. In addition, it has a O ( N K ) complexity. This led us to choose
the NA-CSVD algorithm. The NA-CSVD algorithm can be described as follows
[311.
InitiaIizat ion
- Initialize withus = U N ~ ~ , C = C K + I ~ K + I
F o r n = l , ... ,oo
xs = u,nx
z=x-Usxs
V~ = ~ / l l ~ l l
4.6 Subspace ?backing 103
QR step
compute the angle 4i to zero y ( 2 ) as in
end
Refinement step
for I = K downto 1
Choose the type of rotation
end
End
For the hybrid groupblind multiuser detectors, (4.11) and (4.20), the input
to the N.4-CSVD algorithm is the projection of the received signal onto the
4.7 Simulation Results for Synchronous DS-CDMA systems 104
subspace orthogonal to subspace spanned by known users:
In the hybrid group-blind multiuser detectors, (4.11) and (4.20), the matrix
Us(n) is the direct output of U(n) of the NA-CSVD algorithm, while As(n) =
C(n)2.
In the NA-CSVD algorithm, we assume that the number of users, i.e., the
dimension of the signal subspace, is fixed and known. However, in real CDMA
systems, it is possible for some users to appear and disappear. Therefore, another
algorithm to find the number of users is needed. Rank estimation of the signal
subspace with the NA-CSVD algorithm was developed by P. A. Pango [30] and
the hybrid groupblind detector using NA-CSVD with the rank estimation could
be further studied.
4.7 Simulation Results for Synchronous DS-CDMA
systems
We consider CDMA systems with a variable number of both known and unknown
users to compare the performance between them. The users are assigned purely
random codes of length N = 31.
An ensemble of 50 different random code assignments for each user is gener-
4.7 Simulation Results for Synchronous DS-CDMA systems 105
ated. To investigate the subspace tracking ability, bit-by-bit detect ion is imple-
mented. The mean signal to interference and noise ratio (SINR) is calculated
over all known users with a moving window which has the length of 20.
Figure 4.6, Figure 4.7, and Figure 4.8 show the performance comparison
among Werent multiuser detectors with a various number of both known and
unknown users. It has previously been shown that the hybrid groupblind de-
tector using the SVD algorithm has better performance than other detectors. In
this section, the performance of the hybrid groupblind multiuser detector using
the NA-CSVD is evaluated. Since the NA-CSVD algorithm has a low complex-
ity, i.e., it is less accurate, the performance of the hybrid groupblind detector
using the NA-CSM algorithm is of course worse than when the SVD or FASIR
algorithms are used. However, in all cases, it still has a better performance
than the partial MMSE (non-blind MMSE) detector and has the advantage of
low complexity. Also, it is better than the blind MMSE detector for the case
of K = 7,k = 4. &om the three figures, we can easily recognize that the
performance of the NA-CSVD algorithm critically depends on the subspace di-
mension. As the subspace dimension increases, the performance decreases. As
can be seen from the figures, it is obvious that the hybrid groupblind multiuser
detector using the NA-CSVD has a much better performance than the blind
MMSE detector using the NA-CSVD. The reason is that the hybrid groupblind
detector need only track K eigenvalues and eigenvectors, while the blind MMSE
4.8 Simulation Results for Asynchronous DS-CDMA systems 106
detector must track K + K eigendues and eigenvectors.
Figure 4.9 shows the BER performance of multiuser detectors. 100 different
ensembles of 10,000 bits for each user were generated. For each ensemble of
10,000 bits, the detectors were estimated over the first 300 bits. Figure 4.9
shows that the hybrid groupblind detector has better performance than the
blind MMSE detector. In most cases, the hybrid groupblind detector and the
blind MMSE detector using the NA-CSVD are worse than when using SVD
or FASIR. However, the hybrid groupblind detector using the NA-CSVD has
the advantage of low complexity. While this conclusion applies only to the
NA-CSVD, it can be expected to hold true for other low complexity subspace
tracking algorithms, as these seem to work best for low subspace dimensions.
Thus, because of the lower subspace dimension in group-blind type algorithms,
these can be expected to work considerably better than the blind algorithms.
4.8 Simulation Results for Asynchronous DS-
CDMA systems
We consider an asynchronous CDMA system with 7 known users and 3 unknown
users. The users are assigned purely random codes of length N = 31. The chip
pulse is a raised cosine pulse which has roll-off factor 0.5. Each user's initial
delay dk is uniform on [0,4Tc]. The channel of each user has L = 3 paths. The
4.8 Simulation Results for Asynchronous DS-CDMA systems 107
delay of each path T-J is uniform on [0, 4Tc]. Hence the maximum delay spread
is 8Tc. The fading gain of each path in each user's channel is generated from a
complex Gaussian distribution and fixed over the duration of one signal frame.
The path gains in each user's channel are normalized so that each user's signal
arrives at the receiver with the same power. An ensemble of 50 different random
code assignments for each user is generated. To investigate the subspace tracking
ability, bit-by-bit detection is implemented. The mean signal to interference and
noise ratio (SINR) is calculated as a moving average over all known users with
a window length of 20.
Figure 4.10 shows a performance comparison for different multiuser detec-
tors. It has previously been shown that the groupblind multiuser detector using
the SVD algorithm has better performance than other detectors. Although the
FASIR algorithm has low complexity, it has very good performance as can be
seen in previous work [22]. But, because of the inaccuracy of the channel esti-
mation, the performance of the groupblind multiuser detector using the FASIR
algorithm is worse than when using SVD. However, in all cases, it still has a - better performance than the non-blind MMSE detector and has the advantage
of low complexity. Also, it is better than the blind MMSE detector in the case
of K = 7,W = 3 .
Figure 4.11 shows the BER performance of multiuser detectors. 100 differ-
ent ensembles of 10,000 bits for each user were generated. For each ensemble of
4.8 Simulation Results for Asynchronous DS-CDMA systems 108
10,000 bits, the detectors were estimated over the first 1000 bits. As can be seen
for Fig. 4-11, the groupblind multiuser detector has better performance than
the blind MMSE detector. In most cases, the groupbli~1.d multiuser detector
and the blind MMSE detector using the FASIR algorithm have worse perfor-
mance than using SVD. However, the groupblind detector using the FASIR
algorithm has the advantage of low complexiiy. Unlike the synchronous case
[22], blind channel estimation is the main performance degradation factor in the
asynchronous CDMA systems model.
4.8 Simulation ftesults for Asynchronous DS-CDMA systems 109
Figure 4.6: Performance comparison of multiuser detectors with respect to bits: K = 7, K = 4, and SNR=20dB
Figure 4.7: Performance comparison of multiuser detectors with respect to bits: K = 7, K = 10, and SNR=2OdB
4.8 Simulation Results for Asynchronous DS-CDMA systems 110
.-. . SNGlE USER
Figure 4.8: Performance comparison of multiuser detectors with respect to bits: K = 2, K = 10, and SNR=20dB
Figure 4.9: BER of multiuser detectors with respect to SNR: K = 7, K = 4
4.8 Simulation Results for Asmchronous DS-CDMA systems 111
Figure 4.10: Performaace of the Group-blind linear hybrid detector imple- mented by the FASIR algorithm and Kalman tracking: SNR=POdB, 7 known users and 3 unknown users
Figure 4.11: BER performance of multiuser detectors with respect to SNR: K = 7 , k = 3
Chapter 5
Group-blind Multiuser Detection
for UTRA-TDD
5.1 UMTS Terrestrial Radio Access
In the European third generation mobile radio system, Universal Mobile Telecom-
munications System (UMTS), there is a complex UMTS air interface called
UMTS Terrestrial Radio Access (UTRA) for the requirements of different ser-
vices. The UTRA consists of two modes, the UTRA-FDD (Frequency Division
Duplex) [25] which uses the different frequencies for the uplink and downlink
transmissions and UTRA-TDD (Time Division Duplex) [26] which uses the same
frequency for the uplink and downlink transmissions. The UMTS spectrum was
depicted in Figure 5.1. The basic technologies for the UTRA are wideband code
5.1 UMTS Terrestrial Radio Access 113
1- 1- 1- m o larr nto 1170 m
Figure 5.1: UMTS spectrum allocation
division multiple access (WCDMA) for the FDD mode and time-division CDMA
(TD-CDMA) for the TDD mode as described in Figure 5.2.
Duplex Scheme UTRA-FDD UTRA-TDD Multiple Access Scheme WCDMA TD-CDMA hlodulation QPSK QPSK Frame Length 10 ms 10 ms Pulse Shaping Root Raise Cosine, r=0.22 Root Raise Cosine, r=0.22 Number of time slots per frame 15 slots 15 slots Chip rate 3.84 Mchips/s 3.84 Mchips/s Bandwidth 5MEz 5 MHz Multirate concept multicode multicode , multislot
and orthogonal variable and OVSF spreading factor (OVSF)
Channel allocation no dynamic channel slow and fast allocation (DCA) required DCA supported
Capacity allocation 5 MHz for uplink 5 MHz carrier divided between uplink and 5 MHz for downlink between uplink and downlink downlink (2-14 out of 15 slots)
Table 5.1: Basic system parameters of UTRA-TDD and FDD
The basic system parameters of UTRA-TDD and FDD are described in Table
5.1 There are some characteristics of UTRA-TDD systems listed below.
a Reciprocal channel: In UTRA-FDD, the fast fading of uplink is different
from down link because the fast fading is up to the frequency. However, the
same frequency is employed in both uplink and downlink in UTRA-TDD,
the fast fading is the same in both uplink and downlink. This enables
the transceiver to estimate the fast fading for its transmission from the
5.1 UMTS Terrestrial Radio Access 114
received signal.
a Unpaired band: While UTRA-FDD requires a pair of bands, UTRA-
TDD can be implemented on an unpaired band.
a Flexible capacity allocation: In UTRA-TDD, there exists flexible ca-
pacity allocation between the uplink and the downlink. If the capacity
requirement is asymmetric between the uplink and the downlink, the ca-
pacity can be adjusted by duplex switching point.
a Interference between uplink and d o d n k : Since both uplink and
downlink use the same frequency, the transmitted signal of uplink can
interfere with the received signal of downlink, and vice versa.
In UTRA-TDD mode , the duration of a frame is lOms and it is subdivided
into 15 time slots of 625ps duration. Within each time slot, orthogonal variable
spreading factor (OVSF) codes of length 16 are used for user signal separation.
The TDD frame is divided into downlink and uplink parts as depicted in Figure
5.3. To change the transmit direction, the switching points (SWPs) are used. By
varying the position of the SWP, asymmetrical data rates can easily be realized.
There are two benefits in UTRA-TDD mode. First, the TDD mode is well
suited for microcell/picocell environment for high bit rates and low mobility
applications through the use of variable asymmetric t r d c . Second, the TDD
mode benefits from the reciprocal nature of the channel, i.e., we can use the
5.2 Interference between Uplink and Downlink in UTRA-TDD 115
impulse response of the uplink channel for the downlink channel of a user.
5.2 Interference between Uplink and Downlink
in UTRA-TDD
The primary limiting factors of the TDD mode are synchronization difficulties
and the associated interference problems. The asymmetric allocation of traffic
leads to an interference scenario that will impact the overall spectral efficiency
of a TDD mode. Figure 5.4 depicts this scenario. two neighboring cells use
the same frequency and have different uplink/downlink asymmetric traffic and
the MS2 is near border of cell and transmitting signal with full power. MS1
has more downlink traffic than MS2. In this case, the uplink transmission from
MS2 to BS2 can block the downlink transmission from BS1 to MS1 causing the
inter-cell interference. the inter-cell interference can be avoided using a dynamic
resource allocation (DRA) algorithm.
5.3 Group-Blind Multiuser Detection for UTRA-
TDD
In this section, we will discuss the groupblind linear hybrid multiuser detector
in UTRA-TDD mode. Consider an asynchronous time duplex code division mul-
5.3 GroupBlind Multiuser Detection for UTRA-TDD 116
tiple access (TD-CDMA) system with K known users in a cell and K interfering
users from adjacent cells. To use QPSK modulation, complex values are used
for data symbols The received signal can be expressed as
where H and fi are the channel matrices for the in-cell users and other-cell
users, b and 6 are the data symbols, v is the additive Gaussian noise with unit
power, and a2 is the variance of noise. The data symbols consist of asymmetric
uplink/downlink time slots as depicted in Figure 5.4. Let pL be the orthogond
projection onto the space orthogonal to the in-cell users channel matrix H given
by
where I is the identity matrix. The orthogonal projection of the correlation
matrix R = E[rrH] can then be decomposed as
- - u,"
u,H
u; - -
PLRfiL = [uuu]
r - A, 0 0
0 $1 ,, 0 0 0 - -
5.3 G~OUD-Blind Multiuser Detection for UTRA-TDD 117
where us, u ~ , A, are the signal subspace eigenvector matrix, the noise subspace
eigenvector matrix, and the signal subspace eigenvalue matrix, respectively. The
exact groupblind hybrid multiuser detector [20] is then given by
The groupblind linear hybrid multiuser detector can reduce intra-cell interfer-
ence from the c e l and the inter-cell interference from adjacent cells efficiently.
Figure 5 -5 compares the performance of the exact groupblind multiuser de-
tector with the performance of the traditional partial MMSE detector, which
ignores interference from adjacent cells. It is seen that the group-blind linear
hybrid multiuser detector has a better performance than the partial MMSE de-
tector for the time slots 4-10 where the information is seriously corrupted by
interfering users from adjacent cells. While the partial MMSE detector can only
reduce intra-cell interference, the groupblind linear hybrid multiuser detector
can reduce both intra-cell and inter-cell interference. The correlation matrix
was estimated in each time slot for the estimated groupblind linear hybrid mul-
tiuser detector. The estimated groupblind linear hybrid multiuser detector can
be expressed as
5.3 GrouwBlind Multiuser Detection for UTRA-TDD 118
Figure 5.6 shows a performance of the estimated hybrid groupblind multiuser
detector with respect to time slots. After time slot 5, the SINR of the estimated
group-blind linear hybrid group-blind multiuser detector cross over the SINR of
the partial MMSE detector. In the time slot 11-15, sincc! there are no interfering
users from adjacent cells, the groupblind linear hybrid multiuser detector is the
same as the zero-forcing detector, i.e., &[n] = sgn (1; ( g H $ % ) - l ~ H r [n]) . Thus,
the problem of UTRA-TDD can be solved with the group-blind linear hybrid
multiuser detector.
5.3 GrouwBlind Multiuser Detection for UTRA-TDD 119
m
channel bandwidth
UTRA-TDD
- channel bandwidth
WCDMA
UTRA-FDD
Figure 5.2: UMTS Terrestrial Radio Access (UTRA)
Figure 5.3: Frame structure of UTRA-TDD
5.3 Group-Blind Multiuser Detection for UTRA-TDD 120
Figure 5.4: Interference scenario and UTRA-TDD frame structure
Time Slots
1 -3 I I 4-1 0 11-15 I I I
500 low lSOO 2006 2500 symbo's
Figure 5.5: Performance of the exact group-blind linear hybrid detector in the UTRA-TDD mode with SIR=-20dB, SNR=20dB, 6 in-cell users, and 4 interfer- ing users from adjacent cells.
5.3 GrouwBlind Multiuser Detection for UTRA-TDD 121
/ time slots
30
25
20
Symbols
r I I I
-
- Group-Blind
Figure 5.6: Performance of the estimated group-blind linear hybrid detector in the UTRA-TDD mode with SIR=-20dB, SNR=20dB, 6 in-cell users, and 4 interfering users from adjacent cells.
Chapter 6
Conclusion and Future Work
6.1 Conclusion
In this thesis, we have demonstrated that group-blind multiuser detectors reduce
both intra-cell and inter-cell interference efficiently. The groupblind multiuser
detectors were introduced and evaluated for both synchronous and asynchronous
systems. In most cases, the groupblind multiuser detectors have better perfor-
mance compared to traditional multiuser detectors such as the conventional
detector, the blind MMSE detector, and the partial MMSE detector.
Exact groupblind multiuser detectors which use the exact correlation matrix
have the same SINR as the ideal full MMSE detector when inter-cell interference
occurs. However, estimated groupblind multiuser detectors need to be trained
to get the exact correlation matrix with the time average method. But they
6.2 Future Work 123
converge in performance to the ideal full MMSE detector. Blind channel esti-
mation and subspace tracking algorithms to update the eigen components with
low calculation complexities have been studied.
Because the hybrid groupblind multiuser detector has an excellent perfor-
mance when the time slots are corrupted by inter-cell interference, the hybrid
group- blind multiuser detector is effective for UTRA-TDD systems which suffer
from serious inter-cell interference.
In this thesis, my contributions axe an adaptation of two subspace tracking
algorithms to groupblind multiuser detection to reduce calculation complexity,
and the application of group-blind multiuser detection in UTR-4-TDD.
6.2 Future Work
In this thesis, we assumed that the number of users in the received signal is
known to the receiver. However, a user can appear or disappear in cellular
systems. This information is very important for proper separation of the noise
subspace and signal subspace in SVD or subspace tracking. An estimation of the
number of users required to implement a group-blind multiuser detector should
be examined.
So far, the groupblind multiuser detector considers one antenna. A perfor-
mance improvement is expected when space-time signal processing with lower
complexity is used. Although there axe fast DSP processors available, the devel-
6.2 Future Work
opment of a complexity reduced subspace tracking algorithm and an iterative
implementation of the groupblind multiuser detector are suggested.
Bibliography
[l] C. Shannon, "A Mathematical Theory of Communication," Bell System
Technical Journal, Vol. 27, 1948, pp. 379-423 and 623-656.
[2] 'LA conversation with Claude Shannon," IEEE Commun. Mag., Vol. 22, No.
5, May 1984, pp.123-126.
[3] Scholtz, "The Evolution of Spread-Spectrum Multiple- Access Communica-
tions," in Code Division Multiple Access Communications (S. G. Glisic and
P. A. Leppanen, Eds., Kluwer Academic Publishers, 1995.
[4] Cooper and R. W. Nettleton, "A spread-spectrum techniques for high-
capacity mobile communications," IEEE Trans. Veh. Tech., Vol. 27, No.
4, Nov. 1978, pp. 264-275.
[5] S. Verdu, "Minimum Probability of Error for Asynchronous Gaussian Mul-
tiple Access," IEEE Trans. on IT., Vol. IT-32, No. l, Jan. 1986, pp. 85-96.
[6] H. V. Poor and S. Verdu, "Single-user detectors for multiuser channels,"