Beruflich Dokumente
Kultur Dokumente
M} and
the notation ()
T
means transpose operation. The resulting
vector b
k
c
k
is fed to P parallel pre-ltering units, one for each
antenna branch. The contributions of all users are summed
chip-by-chip to form the following transmit signal vector at
the pth antenna
s
p
=
K
k=1
b
k
u
k,p
p = 1, 2, . . . , P (1)
where u
k,p
= [u
k,p
(1), u
k,p
(2), . . . , u
k,p
(M)]
T
has entries
u
k,p
(n) = c
k
(n)v
k,p
(n) n = 1, 2, . . . , M (2)
and v
k,p
= [v
k,p
(1), v
k,p
(2), . . . , v
k,p
(M)]
T
are the pre-
ltering coefcients of the kth user at the pth antenna. Finally,
each vector s
p
(p = 1, 2, . . . , P) is mapped on M subcarriers
using an OFDM modulator for each antenna branch, com-
prising an inverse fast Fourier transform (IFFT) unit and the
insertion of a cyclic prex (CP).
In the following we aim directly at the design of u
k,p
rather
than v
k,p
. In doing so we assume that the channel impulse
response of each user is practically constant over the downlink
time-slot (slow fading) and that perfect channel knowledge is
available at both the transmitter and receiver ends. The impact
of channel variations on the system performance is addressed
later by simulation.
B. MC-CDMA Receiver
The signals transmitted by the BS array propagate through
multipath channels. At the receiver side, the P incoming wave-
forms are implicitly recombined by the single receive antenna
and passed to an OFDM demodulator, which eliminates the CP
and performs a fast fourier transform (FFT) operation. Without
loss of generality, we concentrate on the jth MT. Then, the
output X
j
= [X
j
(1), X
j
(2), . . . , X
j
(M)]
T
from the FFT unit
takes the form
X
j
=
P
p=1
K
k=1
H
j,p
u
k,p
b
k
+n
j
(3)
where n
j
= [n
j
(1), n
j
(2), . . . , n
j
(M)]
T
represents thermal
noise and it is modelled as a Gaussian vector with zero-mean
and covariance matrix
2
I
M
(I
M
denotes the identity matrix
of order M) while H
j,p
is a diagonal matrix
H
j,p
= diag{H
j,p
(i
1
), H
j,p
(i
2
), ..., H
j,p
(i
M
)} (4)
collecting the channel frequency response between the pth
transmit antenna and the jth MT over the M subcarriers.
To proceed further, we dene the MP K matrix U =
[u
1
u
2
. . . u
K
] with u
k
= [u
T
k,1
u
T
k,2
. . . u
T
k,P
]
T
. Then, X
j
can be rewritten as
X
j
= H
j
Ub +n
j
(5)
where H
j
= [H
j,1
H
j,2
. . . H
j,P
] is an M MP channel
matrix and b = [b
1
, b
2
, ..., b
K
]
T
is the data vector. To keep
the complexity of the MT at a reasonable level, the decision
statistic for b
j
is obtained by feeding X
j
to a linear single
user detector. This produces
y
j
= g
H
j
X
j
(6)
where g
j
= [g
j
(1), g
j
(2), ..., g
j
(M)]
T
is a unit-norm vector
that performs both channel equalization and signal despreading
while ()
H
denotes Hermitian transposition. The scalar in (6)
can be thought of as being part of an automatic gain control
(AGC) which does not impair the signal-to-noise ratio at the
receiver. Substituting (5) into (6) yields
y
j
= q
H
j
Ub + w
j
(7)
where q
j
= H
H
j
g
j
is an MP-dimensional vector that depends
on the channel coefcients and data detection strategy while
w
j
= g
H
j
n
j
is a Gaussian random variable with zero-mean and
variance
2
. In the sequel, we consider the following detection
techniques.
1) Pure despreading (PD):
g
j
(n) = c
j
(n) n = 1, 2, ..., M (8)
2) Maximum Ratio Combining (MRC):
g
j
(n) = c
j
(n)
P
p=1
H
j,p
(i
n
)
_ 1
M
M
m=1
p=1
H
j,p
(i
m
)
2
n = 1, 2, ..., M
(9)
3) Equal Gain Combining (EGC):
g
j
(n) = c
j
(n)
P
p=1
H
j,p
(i
n
)
p=1
H
j,p
(i
n
)
n = 1, 2, ..., M. (10)
Note that the PD strategy dispenses from channel knowledge
at the receiver, thereby keeping the complexity of the MT at
a very low level.
Stacking the decision statistics of all users into a single
vector y = [y
1
, y
2
, ..., y
K
]
T
, we get
y = Q
H
Ub + w (11)
where w = [w
1
, w
2
, ..., w
K
]
T
is a Gaussian vector with zero-
mean and covariance matrix
2
I
M
whereas Q is the following
matrix with dimensions MP K
Q =
_
_
H
H
1,1
g
1
H
H
2,1
g
2
H
H
K,1
g
K
H
H
1,2
g
1
H
H
2,2
g
2
H
H
K,2
g
K
.
.
.
.
.
.
.
.
.
.
.
.
H
H
1,P
g
1
H
H
2,P
g
2
H
H
K,P
g
K
_
_
. (12)
Inspections of (11) indicates that y depends on the pre-
ltering sequences through U. In the next section we design
U according to the MMSE criterion.
III. PRE-FILTERING ALGORITHM
The pre-ltering matrix U is designed so as to minimize
the following mean square error
J = E
_
y b
2
_
(13)
where E {} represents statistical expectation which must be
computed with respect to data symbols and thermal noise. In
doing so, we impose a constraint on the overall transmit power
allocated to all active users. This means that U must fulll
the following equation
tr
_
U
H
U
_
= K (14)
where tr{} denotes the trace of the matrix. Note that this is
in contrast to the methods discussed in [6]- [7] where separate
constraints are posed for all active users, i.e., u
k
2
= 1 for
k = 1, 2, . . . , K. As is intuitively clear, the constraint (14)
ensures more degrees of freedom and allows the BS to give
more power to the weakest users so as to jointly perform signal
pre-ltering and power allocation.
We assume that the transmitted symbols belong to a PSK
constellation (i.e., |b
k
|
2
= 1). Also, symbols of different
users are modelled as independent random variables. Thus,
substituting (11) into (13) and bearing in mind that b and w
are statistically independent, produces
J = tr
_
_
Q
H
VI
K
_ _
Q
H
VI
K
_
H
_
+
2
2
K (15)
where we have dened V = U. Note that the constraint
(14) can also be rewritten as
tr
_
V
H
V
_
=
2
K (16)
so that (15) reduces to
J = tr{( Q
H
VI
K
)( Q
H
VI
K
)
H
+
2
V
H
V}. (17)
Our objective consists of determining the matrix V that
minimizes the right-hand-side (RHS) of (17). The solution to
this problem is found by setting to zero the derivative of J
with respect to V and reads [10]
V = Q(Q
H
Q+
2
I
K
)
1
. (18)
Finally, bearing in mind that U = (1/) V, we have
U =
K
tr{V
H
V}
V. (19)
In the sequel, the matrix Uis referred to as the transmit Wiener
lter (TWF). It is fair to say that TWF is reminiscent of the
method proposed by S alzer and Mottier (S&M) in [7]. The
latter employs the following pre-ltering coefcients
u
k
=
k
Q
_
Q
H
Q+
2
I
K
_
1
d
k
(20)
where
k
is chosen so as to meet the set of constraints
u
k
2
= 1 (k = 1, 2, . . . , K), while d
k
denotes the K-
dimensional vector with entries
d
k
(n) =
_
1
0
if n = k
otherwise .
(21)
Note that S&M reduces to the total interference removal (TIR)
scheme discussed in [6] as
2
becomes vanishing small.
Inspection of (18) and (19) reveals that the crux in the
calculations of U is the inversion of Q
H
Q . The latter has
dimensions K K, independently of the spreading factor
and the number of transmit antennas. Also, a single matrix
inversion is needed to compute all the pre-ltering coefcients.
This makes the proposed algorithm suitable for practical
implementations.
IV. NUMERICAL RESULTS
A. Simulations parameters
Table I summarizes the main parameters of the system under
investigation. We only consider uncoded transmissions and
assume that the system is fully-loaded (K=8). Channel A of
the European Hiperlan/2 standardization project [11] has been
extended to the case of multiple transmit antennas and taken
as the underlying channel model in the considered system.
We assume that the distance between the antenna elements
is large enough to consider P independent channels for each
active user. Unless otherwise specied, the channels are kept
xed over the downlink time-slot (static channel), but vary
from slot to slot. Also, their average energy is the same for
each user.
TABLE I
MC-CDMA SYSTEM PARAMETERS FOR SIMULATIONS
Carrier Frequency 5.2 GHz
Transmission Bandwidth 20 MHz
Number of Subcarriers N = 64
Useful MC-CDMA Symbol Duration 3.2s
Guard Length 0.45s
Spreading Length M = 8
Signal Constellation QPSK
10
-4
10
-3
10
-2
10
-1
10
0
B
E
R
18 16 14 12 10 8 6 4 2 0
E
t
/N
0
, dB
P=1
P=2
P=4
AWGN
TIR
S&M
TWF+PD
Fig. 1. BER vs. E
t
/N
0
with K=8 for P=1, 2 or 4 antennas.
B. System perfomance
The performance of the proposed pre-ltering schemes is
evaluated in terms of average BER computed over all active
users versus E
t
/N
0
, where E
t
is the transmitted energy per
bit and N
0
/2 is the two-sided noise spectral density. For a
fair comparison with a variable number of transmit antennas,
E
t
/N
0
has been scaled (in dB) by 10log(P) so as to cancel
out the corresponding array gain.
Figure 1 illustrates the BER of the TWF in conjunction
with a PD (TWF+PD). The number of transmit antennas is
P=1, 2 or 4 and the performance over the AWGN channel is
also shown as a benchmark. Comparisons are made with the
TIR and S&M schemes. As expected, the system performance
improves with P. We see that TWF+PD outperforms the other
schemes irrespective of the number of transmit antennas. Also
TIR and S&M have virtually the same performance for P=2
or 4.
Figure 2 shows the BER of the previous schemes for P = 1
and 4 in the presence of power imbalance. In particular,
we consider a system in which the averages energies of the
channel responses are 0 dB for users #1,2,3 and 4, +3 dB for
users #5, 6 and, nally, -3 dB for the last two users. Again,
TWF+PD performs remarkably better than the other schemes
and achieves a gain of approximately 2 dB at an error rate of
10
3
. The reason is that TWF re-distributes the transmission
power among available users, thereby allowing the BS to
jointly perform signal pre-ltering and power allocation. Vice
10
-4
10
-3
10
-2
10
-1
10
0
B
E
R
18 16 14 12 10 8 6 4 2 0
E
t
/N
0
, dB
P=1
P=4
AWGN
TIR
S&M
TWF+PD
Fig. 2. BER vs. Et/N
0
with K=8 for and P=1 or 4 in the presence of
power imbalance.
10
-4
10
-3
10
-2
10
-1
10
0
B
E
R
12 10 8 6 4 2 0
E
t
/N
0
, dB
MMSE
TWF+PD
TWF+MRC
TWF+EGC
AWGN
Fig. 3. BER vs. E
t
/N
0
with K=8, P=1 and different SUD schemes.
versa, both TIR and S&M allocate the same power to all users
and, thus, results a very poor performance for the weak users.
Figure 3 shows the performance of TWF in conjunction
with different SUD schemes and a single transmit antenna.
The BER of a conventional MMSE single-user receiver (i.e.,
without any pre-ltering) is also shown for comparison. We
see that TWF+EGC gives the best results. In particular, the
gain with respect to the conventional MMSE receiver is
approximately 6 dB an error rate of 10
2
. As expected,
TWF+PD performs poorly compared to the other schemes as
it does not exploit CSI at the receiver side.
The results of Figure 4 have been obtained in the same
operating conditions of Figure 3, expect that two transmit
antennas are now employed. The curve labelled STBC+MMSE
refers to the transmission technique proposed by W .Sun et
al. in [12], which employs the Alamouti space-time coding
scheme at the transmitter and an MMSE-based detector at the
receiver. Note that this technique does not require any CSI at
10
-4
10
-3
10
-2
10
-1
10
0
B
E
R
12 10 8 6 4 2 0
E
t
/N
0
, dB
STBC+MMSE
TWF+PD
TWF+MRC
TWF+EGC
AWGN
Fig. 4. BER vs. E
t
/N
0
with K = 8 for the proposed pre-ltering technique
using P=2 and different SUD schemes.
10
-4
10
-3
10
-2
10
-1
10
0
B
E
R
11 10 9 8 7 6 5 4 3 2 1 0
Mobile Speed, v (m/s)
TWF+PD
TWF+MRC
TWF+EGC
E
t
/N
0
=10dB
P=2
Fig. 5. BER vs. the mobile speed v with K = 8 for the proposed pre-ltering
technique using P=2 antennas and different SUD schemes.
the transmitter and, therefore, it cannot provide any array gain.
For this reason, the corresponding curve has not been scaled
by 10log(P). In contrast to the results of Figure 3, we see
that now TWF+MRC has the best performance and achieves
a gain of approximately 1.5 dB with respect to TWF+PD at
an error rate of 10
3
. Also, we see that TWF outperforms
STBC+MMSE for E
t
/N
0
>6 dB, irrespective to the detection
strategy employed at the receiver.
The impact of channel variations on the system performance
is addressed in Figure 5, where the BER is shown as a
function of the mobile speed v for E
t
/N
0
=10 dB and P = 2.
For all the considered schemes, the matrix U is computed
at the beginning of each downlink slot and is kept xed
over the entire slot while the channel varies continuously
due to Doppler effects. The slot has duration 1.0 ms and
consists of 300 blocks. The receiver has ideal CSI at each time
instant (i.e., we assume that the channel variations are tracked
perfectly at the receiver). As expected, the BER degrades as v
increases but the performance loss is negligible for mobile
speeds up to 5 m/s. Hence, it can be concluded that the
considered schemes are well suited for indoor applications.
V. CONCLUSION
We have discussed MMSE pre-ltering techniques in TDD
MC-CDMA downlink transmissions. Channel state informa-
tion is obtained at the BS by exploiting the channel reciprocity
which is inherent in TDD systems. By exploiting knowledge
of the SUD technique employed at the receiver, the proposed
scheme minimizes the sum of MSEs at all MTs while con-
straining the overall transmit power instead of the powers of
the single users. In this way, power allocation and signal pre-
ltering are performed jointly at the BS.
Simulations indicate that the proposed algorithm is well
suited for indoor applications and outperforms other existing
schemes. From a practical viewpoint, it can be applied in
several scenarios such as outdoor personal communication
systems (PCSs) with low mobility users or high-speed wireless
local area networks (WLANs). It can be particularly useful for
lowering the power consumption in laptops and handled units.
REFERENCES
[1] K. Fazel, Performance of CDMA/OFDM for mobile communica-
tions systems, in Proc. IEEE Int. Conf. Universal Personal Commun.
(ICUPC), vol. 2, Oct. 1993, pp. 975979.
[2] S. Hara and P. Prasad, Overview of multicarrier CDMA, IEEE
Communications Magazine, vol. 35, pp. 126133, Dec. 1997.
[3] H. Atarashi, N. Maeda, S. Abeta, and M. Sawahashi, Broadband
Wireless Access Based on VSF-OFDM and MC/DS-CDMA, in Proc.
of PIMRC 2002, vol. 3, Sept. 2002, pp. 992997.
[4] S. Moshavi, Multiuser detection for DS-CDMA communications,
IEEE Comm. Magazine, pp. 124136, Oct. 1996.
[5] K. Fazel and S. Kaiser, Multi-Carrier and Spread Spectrum Systems.
John Wiley and Sons, 2003.
[6] A. Silva and A. Gameiro, Pre-ltering antenna array for downlink TDD
MC-CDMA systems, in Proc. Vehicular Techn. Conf. VTC 2003-Spring,
vol. 1, April 2003, pp. 641645.
[7] T. S alzer and D. Mottier, Downlink strategies using antenna arrays
for interference mitigation in Multi-Carrier CDMA, in Proc. of MCSS
2003, Oberpfaffenhofen, Germany, Sept. 2003, pp. 315326.
[8] R. Irmer, W. Rave, and F. Fettweis, Minimum BER transmission for
TDD CDMA in frequency selective channels, in Personal Indoor and
Mobile Radio Communications (PIMRC), Beijing, China, Sept. 2003,
pp. 12601264.
[9] B. R. V. W.B. Jang and R. L. Pickholtz, Joint transmitter-receiver
optimization in synchronous multiuser communications over multipath
channels, IEEE Trans. Comm., vol. 46, pp. 269278, Feb. 1998.
[10] H. Lutkepohl, Handbook of Matrices. John Wiley and Sons, NY, 1996.
[11] J. Medbo, Channels Models for Hiperlan/2 in different indoor scenari-
ous. ETSI BRAN doc., 1998.
[12] W. Sun, H. Li, and M. Amin, MMSE detection for space-time coded
MC-CDMA, in IEEE International Conference on Communications,
vol. 5, May 2003, pp. 34523456.