Coding and Signal Processing for MIMO Communications - A Primer

Inaki Berenguer

and Xiaodong Wang

Department of Electrical Engineering
Columbia University
New York, NY 10027
{ib, wangx}@ee.columbia.edu
Abstract
Rapid growth in mobile computing and other wireless multimedia services is inspiring many re-
search and development activities on high-speed wireless communication systems. Main challenges
in this area include the development of ecient coding and modulation signal processing tech-
niques to improve the quality and spectral eciency of wireless systems. The recently emerged
space-time coding and signal processing techniques for wireless communication systems employing
multiple transmit and receive antennas oer a powerful paradigm for meeting these challenges.
This paper provides an overview on the recent development in space-time coding and signal pro-
cessing techniques for multiple-input multiple-output (MIMO) communication systems. We rst
review the information theoretic results on the capacities of wireless systems employing multiple
transmit and receive antennas. We then describe two representative categories of space-time sys-
tems, namely, the BLAST systems and the space-time block coding systems. Signal processing
techniques for channel estimation and decoding in space-time systems are also discussed. Fi-
nally, some other coding and signal processing techniques for wireless systems employing multiple
transmit and receive antennas are also briey touched upon.
Keywords: Multiple antennas, wireless communications, channel capacity, space-time coding.

I. Berenguer is supported by a Fulbright fellowship.

1
1 Introduction
Multiple-input multiple-output (MIMO) communication technology has received signicant re-
cent attention due to the rapid development of high-speed broadband wireless communication
systems employing multiple transmit and receive antennas. Information theoretic results show
that MIMO systems can oer signicant capacity gains over traditional single-input single-output
channels [17, 45]. This increase in capacity is enabled by the fact that in rich scattering wireless
environments, the signals from each individual transmitter appear highly uncorrelated at each of
the receive antennas. When conveyed through uncorrelated channels between the transmitter and
the receiver, the signals corresponding to each of the individual transmit antennas have attained
dierent spatial signatures. The receiver can exploit these dierences in spatial signatures to
separate the signals originated from dierent transmit antennas.
Many MIMO techniques have been proposed targeting at dierent scenarios in wireless com-
munications. The Bell-Labs Layered Space Time (BLAST) system [16, 47] is a layered space-time
architecture originally proposed by Bell-Labs to achieve high data rate wireless transmissions. In
this scheme, dierent symbol streams are simultaneously transmitted from all transmit antennas
(i.e., they overlap in frequency and in time). The receive antennas receive the superposition of all
symbol streams and recover them via proper signal processing. On the other hand, in Space-Time
Coding (STC) systems [2, 40, 41, 43], the same information symbol stream is transmitted from
dierent transmit antennas in appropriate manner to obtain transmit diversity. Hence, in STC
systems the MIMO channel is exploited to provide more reliable communications, whereas in
the BLAST system the MIMO channel is used to provide faster communications. By employing
higher signal constellations the STC systems can achieve high throughput just like the BLAST
system. In this paper, we give a general overview of the capacity results for MIMO systems as
well as the BLAST and STC techniques.
The remainder of this paper is organized as follows. In Section 2 we summarize the capacity
results for MIMO systems and discuss the impact of antenna correlation on capacity. In Section
3, we describe the BLAST system and related decoding and channel estimation techniques. In
Section 4, we discuss space-time coding techniques and in particular the space-time block codes.
Performance comparisons between the BLAST system and the space-time block coding system
are also made. Finally, in Section 5, we briey touch upon some other space-time coding and
signal processing techniques.
2 Capacity of MIMO Systems
In this section, we summarize the information theoretic results on the capacities of MIMO chan-
nels, developed in the late 1990s [45, 17]. These results show the signicant potential gains in
2
channel capacity by employing multiple antennas at both the transmitter and receiver ends; and
inspired an enormous surge of world-wide research activities to develop space-time coding and
signal processing techniques that can approach the MIMO channel capacity.
2.1 Capacity Results
Tx

x
nT
x
1
x
2
Rx
y
nR
y
1
y
2






h
2,1
Figure 1: Schematic representation of a MIMO system.
Consider a MIMO system with n
T
transmit antennas and n
R
receive antennas signaling
through at fading channels, as shown in Figure 1. The input-output relationship of this system
is given by
y = Hx +v, (1)
where x =
_
x
1
, x
2
, ..., x
n
T

T
is the (n
T
1) transmitted signal vector, y =
_
y
1
, y
2
, ..., y
n
R

T
is
the (n
R
1) received signal vector, v =
_
v
1
, v
2
, ..., v
n
R

T
is the received noise vector and
H =
_

_
h
11
h
21
h
1,n
T
h
21
h
22
h
2,n
T
.
.
.
.
.
.
.
.
.
h
n
R
,1
h
n
R
,1
h
n
R
,n
T
_

_
(2)
is the (n
R
n
T
) MIMO channel matrix with h
ij
representing the complex gain of the channel
between the j th transmit antenna and the i th receive antenna.
It is assumed that the noise sample v
i
, i = 1, 2, ..., n
R
, is a circularly symmetric complex
Gaussian random variable with zero mean and variance
2
, denoted as v
i
N
c
(0,
2
). That is,
{v
i
} N(0,

2
2
), {v
i
} N(0,

2
2
), and they are independent. It is assumed that the complex
channel gains h
ij
N
c
(0, 1). Note that in general, the channel gains may be correlated.
Assuming that the channel matrix H is known at the receiver, but not at the transmitter,
the ergodic (mean) capacity of the MIMO channel with an average total transmit power P (i.e.,
tr
_
E
_
xx
H
_
_
P) is given by [45, 17]
C = E
_
log det
_
I
n
R
+
1
n
T
P

2
HH
H
__
3
= E
_
log det
_
I
n
T
+
1
n
T
P

2
H
H
H
__
bits/s/Hz, (3)
where the expectation is taken with respect to the distribution of the random channel matrix H.
To gain some insight on the capacity expression in (3), denote = P/
2
. Then the capacity
can be expressed as
C =
p

k=1
E{log(1+

n
T

k
)}, (4)
where p = min{n
T
, n
R
} and
1
, ...,
p
are the eigenvalues of the matrix HH
H
or H
H
H. Note that
the matrices HH
H
and H
H
H have the same eigenvalues which are all real and non-negative. If
we compare (4) with the capacity of a single-input single-output (SISO) channel [11], we observe
that the capacity of a MIMO system is equivalent to the sum of p parallel SISO channels, each
one with an equivalent signal-to-noise ratio equal to
i
.
Furthermore, it can be shown that when both n
T
and n
R
increase, the capacity increases
linearly with respect to min{n
T
, n
R
}. On the other hand, if n
R
is xed and n
T
increases, then
the capacity saturates at some xed value; whereas if n
T
is xed and n
R
increases, the capacity
increases logarithmically with n
R
. These asymptotic behaviors of the ergodic capacity are shown
in Figure 2.
0 2 4 6 8 10 12 14 16
0
10
20
30
40
50
60
70
number of antennas
C

i
n

b
i
t
/
c
h
a
n
n
e
l

u
s
e
Capacity with respect the number of antennas for = 15dB
: Increase both n
T
and n
R
: n
T
= 1 and increase n
R
: n
R
= 1 and increase n
T
Figure 2: Ergodic capacities of uncorrelated MIMO channels. The channel is known at the receiver
but not at the transmitter.
Another notion that is frequently used in practice is the outage capacity. Dene the instanta-
4
neous capacity as
(H, ) = log det
_
I
n
R
+

n
T
HH
H
_
. (5)
Obviously (H, ) is a random variable since H is random. Given a certain outage probability
P
out
, the corresponding outage capacity C
out
is dened through the following equation,
P {(H, ) C
out
} = P
out
. (6)
So far we have assumed that the channel matrix H is known at the receiver but not at the
transmitter. Another scenario is that the channel is known at both the transmitter and receiver.
This is the case, for example, when the system employs time-division duplex (TDD) so that the
uplink and downlink channels are reciprocal to each other. In this case, the instantaneous capacity
is given by the following water-lling equation [38]
(H, ) =
n
T

i=1
[log (
i
)]
+
bits/s/Hz, (7)
where
1
, ...,
n
T
are the eigenvalues of the matrix H
H
H, is chosen such that =
n
T

i=1
_

1

i
_
+
and the operator ()
+
is specied as
(x)
+
=
_
x if x > 0,
0 if x 0.
(8)
The ergodic capacity is then given by C = E{(H, )}. Moreover, the outage capacity in this
case is specied by
P {(H, ) C
out
} = P
out
. (9)
Figure 3 shows the 10% outage capacity of uncorrelated MIMO channels with and without
water-lling. It is seen that by knowing the channel at the transmitter, some capacity gain can
be obtained at low signal-to-noise ratio.
2.2 Eects of Antenna Correlations
It has been observed that antennas placed with large enough separations will receive essentially
uncorrelated signals [30]. However, in handsets or terminals, large separations among the antennas
may not be feasible. On the other hand, when the transmitter or receiver is not surrounded
by scatterers, no local scattering or diversity occurs, and the spatial fading at the antennas is
correlated. Hence, insucient antenna spacing and lack of scattering cause the individual antennas
to be correlated.
5
0 5 10 15 20 25 30
0
10
20
30
40
50
60
70
(dB)
C
1
0

i
n

b
i
t
/
c
h
a
n
n
e
l

u
s
e
10% outage capacities (without antenna correlation) n
T
= n
R
: n
T
= n
R
= 1 without waterfilling
: n
T
= n
R
= 1 waterfilling
: n
T
= n
R
= 2 without waterfilling
: n
T
= n
R
= 2 waterfilling
: n
T
= n
R
= 4 without waterfilling
: n
T
= n
R
= 4 waterfilling
: n
T
= n
R
= 8 without waterfilling
: n
T
= n
R
= 8 waterfilling
Figure 3: 10% outage capacities of uncorrelated MIMO channels with and without employing
water-lling.
d
scatterers N
d
d
source
y
x
Rx
y
x
Rx
Rx
y
x
r
Cluster
mean angle
of arrival
mean angle
of arrival
angular
spread
Figure 4: Model with local scatterers. Incident wave is approximately plane at the receiving array.
6
We next discuss the correlation model and the eect of antenna correlation on capacity. Fol-
lowing [8], assuming correlations at both the transmitter and receiver, the (n
R
n
T
) channel
response matrix can be modeled as
H = R
1/2
r
H
w
R
1/2
t
(10)
with H
w
being an uncorrelated (n
R
n
T
) matrix with i.i.d. N
c
(0,1) entries and R
t
and R
r
, of
size (n
T
n
T
) and (n
R
n
R
), representing the covariance matrices inducing transmit and receive
correlations respectively. Note that for the case of uncorrelated transmitter (receiver), we have
R
t
= I (R
r
= I).
The form of cross-correlation between the waves impinging on antenna elements (i.e., R
r
or
R
t
) has been studied and modeled in several references [3, 8, 10, 14, 38]. These models use
similar parameters to characterize the correlation. Specically, assuming that no line of sight
exists between the transmit and the receive antennas, the signal reaching the receive antennas
can be modeled as arriving from a number of equivalent point sources or scatterers in the vicinity
of the receiver as shown in Figure 4. Assuming that the antennas are omnidirectional (i.e. they
radiate and receive from all directions in space), there are three main parameters that characterize
the correlation between antennas (see Figure 4):
Distance d between antennas in terms of wavelengths,
Angular spread of the arrival incident waves
R
o
,
Mean angle of arrival of incident waves
R
o
.
Large values of the angular spread
R
o
result in uncorrelated signals at each of the antennas. The
angular spread is a function of the distance of the cluster to the antenna array and radius of the
cluster. For example, in an outdoor environment, a cluster could be a building located far away
from the antenna array deriving in a small angular spread
R
o
. In an indoor environment, the
cluster of scatterers will be the walls surrounding the array. In this case, there will be signals
impinging the antenna array from all directions resulting in a large angular spread value; therefore,
uncorrelated fading among the antennas can be expected. Figure 5 depicts dierent scattering
scenarios similar to those dened for COST-259 models [38]. In this representation, the circle
represents a cluster of scatterers. The ve dierent scenarios correspond to:
Uplink: This scenario corresponds to a base station operating as a receiver from some
high point without any scatterer nearby. The receiver, usually a handset or terminal, will
be surrounded by scatterers. The angular spread corresponding to the receiver (i.e., base
station) is very low resulting in correlation among the receive antennas.
7
Downlink: This scenario is similar to the uplink but with the base station acting as a
transmitter.
Urban area: Medium size angular spread for both the transmitter and the receiver. Scatterer
clusters represent buildings.
Rural area: Low angular spread for both the transmitter and the receiver. Scatterer clusters
represent mountains and hills.
Indoor: Large angular spread for both the transmitter and the receiver. Impinging waves
arrive from all directions in the space.
d
Cluster
Rx
d
Tx
d
Cluster
Rx
d
Tx
o
60
d
Cluster
Rx
d
Tx
o
6
o
60
d
Cluster
Rx
d
Tx
o
6
d
Cluster
Rx
d
Tx
Scenario A
Typical Downlink
Scenario E
Typical Picocell/indoor
Scenario D
Typical Rural area
Scenario C
Typical Urban area
Scenario B
Typical Uplink
angular spread
mean angle
of arrival
Figure 5: MIMO scattering scenarios.
Figure 6 shows the 10% outage capacities for the dierent scenarios dened in Figure 4 with
n
T
= n
R
= 4 and an antenna spacing of d = 0.5. We assume that the channel is known at the
8
receiver but not at the transmitter. We have used the correlation model described in [3]. We
also show the SISO capacity for comparison. It is seen that urban and indoor scenarios with rich
scattering oer much higher MIMO capacities than rural environments.
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
(dB)
C
1
0

i
n

b
i
t
/
c
h
a
n
n
e
l

u
s
e
10% outage capacities n
T
= n
R
= 4, antenna spacing 0.5
: Scenario A
: Scenario B
: Scenario C
: Scenario D
: Scenario E/Indoor
: Uncorrelated
: SISO
Uncorrelated
MIMO
Scenario E
Indoor
SISO
Scenario D
(rural)
Scenario C
(urban)
Scenario B
(uplink)
Scenario A
(downlink)
Figure 6: MIMO outage capacities for dierent channel scenarios described in Figure 5.
Figure 7 shows the 10% outage capacities of a correlated MIMO channel with and without
water-lling. The correlation scenario corresponds to an urban area depicted in Figure 5 with
an antenna spacing of d = 0.5. Comparing with Figure 3, it is seen that signicant capacity
gain can be achieved when there are antenna correlations and the channel is known at both the
transmitter and the receiver.
3 The BLAST System
The information theoretical results from the preceding section indicates the enormous capacity
gain by employing multiple antennas at both ends of the communication systems. Realizing such
a potential gain, researchers at Bell-Labs developed the rst MIMO architecture for high-speed
wireless communications the BLAST systems.
BLAST (Bell-Labs Layered Space Time) [16, 47] is a high speed wireless communication
scheme employing multiple antennas at both the transmitter and the receiver. In a BLAST
system, the transmitted data is split equally into n
T
transmit antennas and then simultaneously
sent to the channel overlapping in time and frequency. The signals are received by n
R
receive
antennas as shown in Figure 8 and signal processing at the receiver attempts to separate the
received signals and recover the transmitted data. The input-output relationship of a BLAST
9
0 5 10 15 20 25 30
0
5
10
15
20
25
30
35
40
(dB)
C
1
0

i
n

b
i
t
/
c
h
a
n
n
e
l

u
s
e
10% outage capacities with antenna correlations (urban environment) n
T
= n
R
: n
T
= n
R
= 1 without waterfilling
: n
T
= n
R
= 1 waterfilling
: n
T
= n
R
= 2 without waterfilling
: n
T
= n
R
= 2 waterfilling
: n
T
= n
R
= 4 without waterfilling
: n
T
= n
R
= 4 waterfilling
: n
T
= n
R
= 8 without waterfilling
: n
T
= n
R
= 8 waterfilling
Figure 7: 10% outage capacities of a correlated MIMO channel corresponding to an urban scenario,
with and without employing water-lling.
Tx

s
nT
s
1
s
2
y
nR
y
1
y
2






h
2,1
Tx
Tx
Rx
Rx
Rx
BLAST
signal
processing:
estimate
and decode
Vector
encoder
Tx
Data
Rx
Data
Figure 8: Schematic representation of a BLAST system.
10
system can be expressed as
y =
_

n
T
Hs +v (11)
where s =
_
s
1
, s
2
, ..., s
n
T

T
is the (n
T
1) transmit signal vector with s
i
belonging to a nite
constellation A, v =
_
v
1
, v
2
, ..., v
n
R

T
is the (n
R
1) receive noise vector with v
i
N
c
(0, 1),
H is dened in (2) and is the total signal-to-noise ratio independent of the number of transmit
antennas. It is assumed unitary power for the transmitted symbols, E
_
|s
i
|
2
_
= 1.
3.1 BLAST Detection Algorithms
It is seen from (11) that the receive antennas see the superposition of all the transmitted signals.
The task of a BLAST detector is to recover the transmitted data s from the received signal y.
In what follows, we describe several BLAST detection algorithms [18, 20]. Here we assume the
channel matrix H is known at the receiver. We will discuss channel estimation algorithms in
Section 3.2.
Maximum Likelihood (ML) Receiver
The ML detector is the optimal receiver in terms of bit error rate. Let A be the symbol con-
stellation set (e.g., QPSK or M-QAM) whose size is M. Then, the ML detection rule is given
by
s = arg min
sA
n
T
_
_
_
_
y
_

n
T
Hs
_
_
_
_
2
. (12)
Note that the minimization problem is performed over all possible transmitted signal vectors s
in the set A
n
T
. The computational complexity of an exhaustive search is then O(M
n
T
). Hence,
although the ML receiver is optimal, its complexity grows exponentially with the number of trans-
mit antennas. A low complexity local search method called sphere decoding whose complexity
is O(M
3
) is developed in [12, 15].
Zero Forcing and Cancellation Receiver
A simpler receiver is the zero forcing (ZF) receiver. The ZF receiver considers the signal from
each transmit antenna as the desired signal and the remainder as interferers. Nulling is performed
by linearly weighting the received signals to satisfy the ZF criterion, i.e., inverting the channel
response. Furthermore, a superior performance can be obtained by using nonlinear techniques by
means of symbol cancellation. Using symbol cancellation, the already detected and sliced symbol
from each transmit antenna is subtracted out from the received signal vector, similarly to decision
feedback equalization or multiuser detection with successive interference cancellation. Therefore,
the next signal to be decoded will see one interferer less.
11
For simplicity, assume n = n
T
= n
R
. Denote the QR factorization of H as H = QR where Q
is unitary, i.e., QQ
H
= I and R is upper triangular. The nulling operation of the received vector
y is performed by
z = Q
H
y =
_

n
T
RS +Q
H
v; (13)
that is
_

_
z
1
z
2
.
.
.
z
n
_

_
=
_

n
_

_
r
1,1
r
1,2
. . . r
1,n
0 r
2,2
. . . r
2,n
.
.
.
.
.
.
.
.
.
.
.
.
0 . . . 0 r
n,n
_

_
_

_
s
1
s
2
.
.
.
s
n
_

_
+
_

_
w
1
w
2
.
.
.
w
n
_

_
. (14)
Note that since Q is unitary, there is no noise amplication, i.e., w = Q
H
v is also N
c
(0, I). In
(14), the decision statistic z
n
is just a noisy scaled version of s
n
which can be directly estimated
and then subtracted from z
n1
. Repeating the estimating and subtracting operations until all
transmitted signals are decoded, the algorithm can be summarized as follows
s
n
= Quantize
_
1
rn,n
_
n

z
n
_
s
n1
= Quantize
_
1
r
n1,n1
_
_
n

z
n1
r
n1,n
s
n
__
.
.
.
s
i
= Quantize
_
1
r
i,i
_
_
n

z
i

k=i+1
r
i,k
s
k
__
.
.
.
s
1
= Quantize
_
1
r
1,1
_
_
n

z
n1

k=2
r
1,k
s
k
__
where the quantizer takes values from the constellation A.
Nulling and Cancellation Receivers with Ordering
In the decoding algorithm discussed above, a wrong decision in the detection of a symbol adds
interference to the next symbols to be detected. It is shown in [18, 20] that it is advantageous
to rst nd and detect the symbol s
k
with the highest signal to-noise ratio, i.e., with the highest
reliability. The detected symbol is then subtracted from the rest of the received signals. Therefore,
after cancelling s
k
, we have a system with n
T
1 transmit antennas and n
R
receive antennas, i.e.,
the corresponding channel matrix is obtained by removing column k from H. The same process is
then applied on this (n
T
1, n
R
) system and the algorithm continues until all transmitted symbols
have been decoded. That is, the nulling and cancellation operation is performed from the more
reliable symbols to the less reliable ones.
The nulling operation can be performed by means of ZF or minimum mean-square error
(MMSE). Similarly to the ZF equalization in single antenna systems, the ZF criterion yields
the following two problems: (1) The algorithm can encounter singular matrices that are not
12
invertible; and (2) ZF focuses on cancelling the interference (i.e., overlapping signals) completely
at the expense of enhancing the noise, possibly signicantly. On the other hand, the MMSE
criterion minimizes the error due to the noise and the interference combined. In the ordering
operation, the MMSE method nulls the component with the smallest MSE. Following [20], the
BLAST decoding algorithm based on the MMSE nulling and cancellation with ordering is as
follows:
G = H
r = y
FOR i = 1 : n
T
DO
P = (

n
G
H
G+I)
1
k
i
= argmin{P
j,j
}, j / {k
1
, k
2
, ..., k
i1
} (ordering: nd min MSE)
w = (GP)(:, k
i
) (nulling vector)
z = w
H
r
s
k
i
= Quantize(z)
r = r
_

n
H(:, k
i
) s
k
i
(cancellation)
G = G\ H(:, k
i
) (remove column of that transmit antenna)
END
Figure 9 compares the BER performance of the four detection methods discussed above in a
BLAST system with n
T
= n
R
= 4 antennas and QPSK modulation. It is seen that the ML decoder
has the best BER performance although for every transmitted code vector, the receiver needs to
evaluate (12) over 4
4
= 256 possibilities. On the other hand, the MMSE nulling and cancellation
algorithm with ordering exhibits the best performance among the suboptimal algorithms.
3.2 MIMO Channel Estimation Algorithms
So far, we have assumed that the MIMO channel matrix H is known at the receiver. In practice,
the receiver needs to estimate this matrix prior to the start of the decoding process. We next
discuss the channel estimation methods based on a training preamble [34].
Suppose T n
T
MIMO training symbols s(1), s(2), ..., s(T) are used to probe the channel.
The received signals corresponding to these training symbols are
y(i) =
_

n
T
Hs(i) +v(i), i = 1, 2, ..., T. (15)
Denote Y =
_
y(1), y(2), ..., y(T)

, S =
_
s(1), s(2), ..., s(T)

and V =
_
v(1), v(2), ..., v(T)

.
Then (15) can be written as
Y =
_

n
T
HS +V. (16)
13
0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
10
0
(dB)
B
i
t

E
r
r
o
r

R
a
t
e
BER for different BLAST decoding algorithms ( n
T
= n
R
= 4 and QPSK modulation)
: ML decoder
: ZF Nulling and Cancellation
: ZF Null & Canc + Ordering
: MMSE Null & Canc + Ordering
Figure 9: BER performance of dierent BLAST decoding algorithms with n
T
= n
R
= 4 and
QPSK. Uncorrelated MIMO channels and perfect channel knowledge at the receiver are assumed.
The maximum likelihood estimate of the channel matrix H is given by

H
ML
= arg min
H
_
_
_
_
Y
_

n
T
HS
_
_
_
_
2
=
_
n
T

YS
H
(SS
H
)
1
. (17)
According to [34], the optimal training symbol sequence S that minimizes the channel estimation
error should satisfy
SS
H
= T I
n
T
. (18)
One way to generate such optimal training sequences is to use the Hadamard matrices [26] (when
they exist for n
T
). As an example, consider a system with n
T
= 4 and a training sequence length
T = 16 symbol intervals. We rst generate a (4 4) Hadamard matrix as
A =
1 + i

2
_

_
1 1 1 1
1 1 1 1
1 1 1 1
1 1 1 1
_

_
. (19)
Then the optimal training sequence can be constructed by concatenating four A matrices as
S =
_
A A A A

. (20)
As an alternative to the ML channel estimator, the linear MMSE channel estimator is obtained
as a linear transformation of the received signals Y that minimizes the estimation error and it is
14
given by

H
MMSE
=
_

n
T
YS
H
_

n
T
SS
H
+I
_
1
. (21)
4 6 8 10 12 14 16 18 20
10
2
10
1
T
B
i
t

E
r
r
o
r

R
a
t
e
BER of different lengths of the training sequence using the MMSE IC with ordering decoder .
Assume = 10dB and n
T
= n
R
= 4 and QPSK modulation
: Perfect Channel Estimate
: ML Channel Estimate
: MMSE Channel Estimate
Figure 10: Eect of the training length T on the BER performance.
We next give a simulation example. Consider a BLAST system with n
T
= n
R
= 4 antennas
and QPSK modulation. We assume uncorrelated fading and a signal-to-noise ratio = 10dB.
Figure 10 shows the BER of dierent channel estimation algorithms for dierent lengths of the
optimal training sequence. As a decoder we use the MMSE nulling and cancellation with ordering
algorithm. It is seen that the MMSE and ML channel estimators have similar performance. Figure
11 compares the BER performance of the MMSE nulling and cancellation with ordering decoder
using the ML channel estimator with dierent lengths of the optimal training sequence.
4 Space-Time Coding
In the previous section, we discussed the BLAST system which increases the data rate by simul-
taneously transmitting symbols from multiple transmit antennas. However, the BLAST approach
suers from two major drawbacks: (1) it requires n
R
n
T
that is not always feasible when the
receiver is a small or battery operated device; and (2) the performance of the suboptimal BLAST
decoding algorithms is limited by error propagation. In this section, we discuss the space-time
coding approach that exploits the concept of diversity.
15
0 2 4 6 8 10 12 14 16
10
3
10
2
10
1
10
0
(dB)
B
i
t

E
r
r
o
r

R
a
t
e
BER with ML channel estimator and different length of the training sequence.
Assume n
R
= n
T
= 4 and MMSE IC + ordering decoder
: Perfect channel estimate
: ML channel estimator, T = n
T
: ML channel estimator, T = 2n
T
: ML channel estimator, T = 4n
T
Figure 11: BER performance of the ML channel estimator with dierent lengths of the optimal
training sequence.
4.1 The Concept of Diversity
With space-time codes (STC) [2, 40, 41, 43], instead of transmitting independent data streams
as in BLAST, the same information is transmitted in appropriate manner simultaneously from
dierent transmit antennas to obtain transmit diversity. The main idea of transmit diversity
is that if a message is lost in a channel with probability p and we can transmit replicas of the
message over n independent such channels, the loss probability becomes p
n
. Using diversity, more
reliability is given to the symbols which allows employing higher order constellation resulting in
higher throughput just like the BLAST system. The main dierence between BLAST and STC
can be summarized as: (1) BLAST transmits more symbols, i.e., n
T
symbols/channel use; and
(2) STC transmits only (at most) 1 reliable symbol/channel use by means of diversity.
As an example, consider a systems willing to transmit 4 bit/s/Hz with 2 transmit anten-
nas. BLAST would use QPSK symbols per antenna, i.e., 4 bit/s/Hz. STC can only send 1
symbol/channel use, therefore 16-QAM symbols would need to be employed. In the end, the
same data is transmitted through higher order constellations. There are two main types of STCs,
namely space time trellis codes (STTC) [43] and space time block codes (STBC) [41].
The STTC is an extension of trellis coded modulation [6] to the case of multiple transmit
and receive antennas. It provides both full diversity and coding gain. However, it has the
disadvantage of high decoding complexity which grows exponentially with the number of antennas.
Specic space-time trellis codes designed for two or four antennas perform very well in slow fading
16
environments and come within 2-3 dB of the outage capacity. STTCs are designed to achieve full
diversity and then, among the codes that achieve full diversity, maximize the coding gain. For
further references on STTC refer to [4, 43].
In hope of reducing the exponential decoding complexity of STTC, Alamouti proposed a
simple space-time coding scheme using two transmit antennas [2]. Later, the STBC introduced in
[40], generalized the Alamouti transmission scheme to an arbitrary number of transmit antennas.
STBC achieve full diversity as the STTC although they do not provide any coding gain. This is
not a problem since they can be concatenated with an outer channel code [5]. Besides achieving
full diversity, the main property of STBC is that there is a very simple ML decoding algorithm
based only on linear processing. These codes are based on some specic linear matrices and the
reduced complexity receiver is due to the orthogonal properties of these matrices.
4.2 Space-Time Block Codes
We assume a wireless communication system where the transmitter is equipped with n
T
and the
receiver with n
R
antennas. A space time block code matrix is represented as
C
p,n
T
=
space
_

_
c
1,1
c
1,2
. . . c
1,n
T
c
2,1
c
2,2
. . . c
2,n
T
.
.
.
.
.
.
.
.
.
c
p,1
c
p,2
. . . c
p,n
T
_

time

(22)
At each time slot t, signals c
t,i
, i=1,2,. . . ,n
T
, are transmitted simultaneously from the n
T
transmit
antennas as shown in Figure 12. Therefore, at time t, transmitter antenna i will transmit c
t
,
i
in
the matrix (1 t p and 1 i n
T
, with p being the length of the block code). Next, we
describe the encoding and decoding operations of the STBC for two transmit antennas, namely
the Alamouti code.
Tx

c
t,nT
c
t,1
c
t,2
y
t,nR
y
t,1
y
t,2






h
2,1
Tx
Tx
Rx
Rx
Rx
ML
decoder
(linear
processing)
STBC
C
p,nT
=
G( s
1
,...,s
k
)
Tx
Data
[ s
1
...,s
k
]
Rx
Data
Figure 12: Schematic representation of an STBC system.
17
STBC with n
T
= 2: Alamouti Code
The Alamouti code is an STBC using n
T
= 2 transmit antennas and any number of receive
antennas. The Alamouti code matrix O
c,2
is dened as [2]
O
c,2
=
_
x
1
x
2
x

2
x

1
_
. (23)
Consider transmitting symbols of a signal constellation A of size 2
b
. Every two time slots, 2b bits
arrive at the encoder and select constellation signals s
1
and s
2
. Setting x
1
= s
1
and x
2
= s
2
in
O
c,2
, we arrive at the following transmission matrix
C
2,2
=
_
s
1
s
2
s

2
s

1
_
. (24)
Then, in the rst time slot, antenna 1 transmits s
1
and antenna 2 transmits s
2
. In the next time
slot, antenna 1 transmits s

2
and antenna 2 transmits s

1
. Since two time slots are needed to
transmit two symbols (s
1
, s
2
), the rate of the code is R = 1 symbol/channel use.
At the receiver, the received signal by antenna i during two consecutive time slots (t=1,2) is
_
y
1,i
y
2,i
_
=
_

2
C
2,2
h
i
+v
i
=
_

2
_
s
1
s
2
s

2
s

1
_ _
h
i,1
h
i,2
_
+
_
v
1,i
v
2,i
_
, i= 1, 2, (25)
which can be rewritten as
_
y
1,i
y

2,i
_
. .
y
i
=
_

2
_
h
i,1
h
i,2
h

i,2
h

i,1
_
. .
H
i
_
s
1
s
2
_
. .
s
+
_
v
1,i
v
2,i
_
. .
v
i
, i= 1, 2. (26)
We note that the orthogonality of the code O
c,2
implies the orthogonality of H
i
, i.e., H
H
i
H
i
=
_
|h
i,1
|
2
+|h
i,2
|
2
_
I
2
. Assuming that the receiver has knowledge of the channel coecients h
i,j
, we
form a decision statistic at each receive antenna by left multiplying the received vector in (26) by
H
H
i
which results in
z
i
=
_
z
1,i
z
2,i
_
= H
H
i
y
i
=
_

2
H
H
i
H
i
s +H
H
i
v
i
. (27)
Hence, using the orthogonality property of H
i
it yields
z
i
=
_
z
1,i
z
2,i
_
=
_

2
_
|h
i,1
|
2
+|h
i,2
|
2
_
_
s
1
s
2
_
+
_
w
1,i
w
2,i
_
. (28)
Adding all the statistics from all n
R
receive antennas we obtain
z =
_
z
1
z
2
_
=
n
R

i=1
_
z
1,i
z
2,i
_
=
_

2
n
R

i=1
_
|h
i,1
|
2
+|h
i,2
|
2
_
_
s
1
s
2
_
+
n
R

i=1
_
w
1,i
w
2,i
_
. (29)
18
In (29), in the absence of noise, z
1
will be just an scaled version of s
1
and z
2
will be an scale
version of s
2
without any cross dependency. To estimate the symbols that were sent, we just scale
and quantize the decisions statistics in (29) as
s
1
= Q(z
1
),
and s
2
= Q(z
2
). (30)
We recall that the decoupling has been possible because of the orthogonality of the Alamouti code
matrix.
4 2 0 2 4 6 8 10 12 14
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
0
E
b
/N
o
(dB)
B
i
t

E
r
r
o
r

R
a
t
e
BER performance comparison: Alamouti vs BLAST
( n
T
= n
R
= 2, R = 2 bit/s/Hz and without antenna correlations)
: BLAST with ML decoder
: BLAST with MMSE Null & Canc + Ord decoder
: Alamouti
Figure 13: BER performance comparison between BLAST (BPSK modulation) and Alamouti
(QPSK modulation) with n
T
= n
R
= 2 (transmission rate R = 2 bit/s/Hz). Uncorrelated MIMO
channel and perfect channel knowledge at the receiver are assumed.
We now compare the performance of the Alamouti scheme with that of the BLAST system
discussed in the previous section. For both systems, we consider n
T
= n
R
= 2. We assume
that both schemes have a transmission rate R = 2 bit/s/Hz. This rate can be achieved using
BLAST with BPSK or using the Alamouti code with QPSK modulation. For a fair comparison,
we compare the two systems in terms of signal-to-noise ratio per bit, i.e., E
b
/N
o
. Assuming perfect
channel estimation at the receiver and no antenna correlations, Figure 13 shows that Alamouti
performs better than BLAST and this improvement is greater at higher signal-to-noise ratio. We
next compare their performance in correlated MIMO channels. We consider a medium level of
correlation typical of urban environments as described in Figure 5. It is seen from Figure 14 that
Alamouti performs much better than BLAST in such a scenario.
19
5 0 5 10 15 20 25
10
5
10
4
10
3
10
2
10
1
10
0
E
b
/N
o
(dB)
B
i
t

E
r
r
o
r

R
a
t
e
BER performance comparison: Alamouti vs BLAST
( n
T
= n
R
= 2, R = 2 bit/s/Hz and with antenna correlations)
: BLAST with ML decoder
: BLAST with MMSE Null & Canc + Ord decoder
: Alamouti
Figure 14: BER Performance comparison between BLAST (BPSK modulation) and Alamouti
(QPSK modulation) with n
T
= n
R
= 2 (transmission rate R = 2 bit/s/Hz). Correlated MIMO
channel (urban environment in Figure 5) and perfect channel knowledge at the receiver are as-
sumed.
General STBC Based on Orthogonal Designs (n
T
2)
The Alamouti scheme presented above works only with two transmit antennas. This scheme
was later generalized in [40, 41] to an arbitrary number of transmit antennas. Similarly to the
Alamouti code in (23), the general STBC is dened by a code matrix with orthogonal columns.
Just like in the Alamouti scheme, a simple linear receiver is also obtained due to the orthogonality
of the columns of the code matrix. In general, an STBC is dened by a (p n
T
) matrix G. The
entries of the matrix G are linear (possibly complex) combinations of the variables x
1
, x
2
, ..., x
k
(representing symbols). The columns of the matrix represent antennas and the rows time slots.
Therefore, p time slots are needed to transmit k symbols, resulting in a code rate R = k/p
symbols/channel use. It is of special interest code matrices achieving the maximum transmission
rate permitted by the STC theory, i.e, R = 1 symbol/channel use. For a xed n
T
, among the
code matrices that achieve the maximum rate, we will be interested in those with minimum
values of p or equivalently, minimum number of time slots needed to transmit a block. These
code matrices are referred as delay optimal and they are interesting because they minimize the
memory requirements at the transmitter and at the receiver (i.e., encoding and decoding delay).
We recall that p n
T
.
STBC for real constellations
20
For real signal constellations such as PAM, the entries of the code matrices are only real linear
combinations of x
1
, x
2
, ..., x
k
. General STBC based on real orthogonal designs achieving full
diversity and full rate, can be found for any number of transmit antennas n
T
[43]. Using n
T
= 2,
4 and 8 antennas, STBC code matrices can be found with p = n
T
(i.e., minimum possible delay
in STBC). As an example, an STBC suitable for real constellations with n
T
= 4 is
G
4
=
_

_
x
1
x
2
x
3
x
4
x
2
x
1
x
4
x
3
x
3
x
4
x
1
x
2
x
4
x
3
x
2
x
1
_

_
(31)
for which it can be veried that G
T
4
G
4
=
_
4

i=1
x
2
i
_
I
4
. The encoding process at the transmitter is
similar to that for the Alamouti code, as follows. Consider a real constellation of size 2
b
. At time
1, 4b bits arrive at the encoder and select symbols s
1
, s
2
, s
3
, s
4
. Let x
i
= s
i
in matrix G
4
in (31)
to obtain the code matrix C
4
. At time t = 1,2,3 and 4, the t-th row of C
4
is transmitted from the
four transmit antennas simultaneously. Therefore, with n
T
= 4 transmit antennas and employing
the code matrix C
4
, four symbols are transmitted during four symbol intervals achieving R =
1 symbol/channel use, i.e., the maximum rate allowed by the STC theory. At the receiver, the
orthogonality of the matrix C
4
simplies the ML decoder decoupling the detection of each of the
transmitted symbols.
STBC for complex constellations
Complex STBC are analogous to the real ones except that the code matrices contain entries
x
1
, x
2
, ..., x
k
, their conjugates, and multiples of them by

1, making them useful for com-
plex constellations such as M-PSK or M-QAM. As an example, an STBC with n
T
= 4 for complex
constellations can be constructed using the real orthogonal design in (31) as
G
c,4
=
_
G
4
G

4
_
=
_

_
x
1
x
2
x
3
x
4
x
2
x
1
x
4
x
3
x
3
x
4
x
1
x
2
x
4
x
3
x
2
x
1
x

1
x

2
x

3
x

4
x

2
x

1
x

4
x

3
x

3
x

4
x

1
x

2
x

4
x

3
x

2
x

1
_

_
. (32)
As before, the code C
c,4
can be obtained substituting x
i
by the data symbols s
i
in G
c,4
. In
this code, transmitting each row at a time, 8 symbols intervals are needed to transmit 4 symbols,
therefore having a rate R = 1/2 symbol/channel use, i.e., half of the maximum rate permitted by
the STC theory. Complex STBC of R = 1/2 achieving full diversity can be built for any number
of transmit antennas n
T
from real STBC using G
c,n
T
=
_
G
n
T
G

n
T
_
.
21
It has been shown that complex STBC with full rate (i.e., R = 1) exists only for n
T
= 2, i.e.,
the Alamouti code. In this sense, the Alamouti codes is quite unique. Codes that achieve a rate
R = 3/4 with complex constellations have been found with n
T
= 3 and n
T
= 4 [40].
5 Further Topics and Conclusions
In this paper, we have discussed the huge increase in capacity that can be obtained in rich
scattering environments by using multiple antennas at the transmitter and the receiver; and we
have given an overview of the main classes of space-time techniques recently developed in the
literature. In conclusion, the area of space-time coding and signal processing is new, active and
full of challenges. The following is a list of some other important topics related to MIMO systems
and space-time coding and signal processing:
Space-time trellis codes (STTC): An STTC is basically a trellis-coded modulation (TCM)
code, which can be dened in terms of a trellis tree. Rather than transmitting the output
code symbols serially from a single transmitter antenna as in the traditional TCM scheme,
in STTC all the output code symbols at each time are transmitted simultaneously from
multiple transmitter antennas. The rst STTC communication system was proposed in
[43]. Some design criteria and performance analysis for STTC in the presence of channel
estimation error are given in [44]. Some improved STTC codes found by exhaustive computer
search are given in [4].
Dierential space-time codes: Previous sections assumed that the receiver had knowledge
of the channel matrix before starting the detection algorithms. In some situations, this is
not possible since no training symbols are available. In some other situations, the channel
changes so rapidly that channel estimation is dicult or requires to send training symbols
very often. That is the reason why it is interesting to consider dierential techniques that do
not require estimation of the channel response neither at the receiver nor at the transmitter.
Dierential STBC based on orthogonal designs are proposed in [29, 39] and that based on
unitary group codes were proposed in [28]. Similarly to the SISO case, dierential decoding
incurs a performance penalty of about 3dB compared with coherent detection.
Space-time precoding: The space-time coding schemes presented in this paper only require
channel knowledge at the receiver. In some cases, channel status can be fedback to the
transmitter or directly estimated by the transmitter such as in a TDD system. In such
scenarios, the performance can be improved if the transmitter uses this channel information.
Dierent precoding schemes have been proposed in [36].
22
MIMO antenna selection: Usually, the RF chain (amplier, digital-to-analog converters,
etc.) in wireless devices is one of the most expensive blocks. A promising approach for
reducing the cost and complexity while retaining a reasonably large fraction of the high
potential data rate of a MIMO system is to employ a reduced number of RF chains at
the receiver (or transmitter) and attempt to optimally allocate each chain to one of a
larger number of receive (transmit) antennas. In this way, only the best set of antennas is
used, while the remaining antennas are not employed, thus reducing the number of required
RF chains. Dierent approaches to selecting those antennas are recently proposed in the
literature [21, 22, 25].
MIMO applications in OFDM and CDMA systems: Recently, the use of MIMO systems in
frequency-selective fading channels in combination with orthogonal frequency division mul-
tiplexing (OFDM) and coherent detection has been considered [1, 7]. Code design criteria
for the MIMO OFDM systems are given in [32, 33], and specic code designs are given in [9].
Moreover, MIMO coding and signal processing techniques for code-division multiple-access
(CDMA) systems are developed in [27, 35].
Turbo processing for MIMO systems: Iterative or turbo demodulation and decoding for
coded BLAST or coded STC systems have been investigated in [13, 23, 31, 33, 37, 46].
Other space-time coding schemes: Other classes of codes are being developed for MIMO
systems. As an example, linear dispersion (LD) codes [24] can be used with any conguration
of transmit and receive antennas and they are designed to optimize the mutual information
between the transmitted and received signals. The LD codes can be decoded using any
BLAST detection algorithm. Moreover, layered space-time coding schemes are developed
in [19, 42].
References
[1] D. Agrawal, V. Tarokh, and N. Seshadri. Space-time coded OFDM for high data-rate wireless com-
munication over wideband channels. In Proc. IEEE Vehicular Technol. Conf. (VTC), Ottawa, ON,
Canada, May 1998.
[2] S. M. Alamouti. A simple transmit diversity technique for wireless communications. IEEE J. Select.
Areas Commun., 16(8):14511458, Oct. 1998.
[3] D. Asztely. On antenna arrays in mobile communications systems: Fast fading and GSM base station
receiver algorithms. Technical Report S3-SB-9611, KTH, Mar. 1996.
[4] S. Baro, G. Bauch, and A. Hansmann. Improved codes for space-time trellis-coded modulation. IEEE
Commun. Lett., 4(1):2022, Jan. 2000.
[5] G. Bauch. Concatenation of space-time block codes and Turbo-TCM. In Proc. IEEE Int. Conf.
Commun. (ICC), Vancouver, BC, Canada, June 1999.
23
[6] E. Biglieri, D. Divsalar, P. J. McLane, and M. K. Simon. Introduction to Trellis-Coded Modulation
with Applications. New York: Macmillan, 3rd edition, 1991.
[7] H. Bolcskei, D. Gesbert, and A. J. Paulraj. On the capacity of OFDM-based spatial multiplexing
systems. IEEE Trans. Commun., 50(2):225234, Feb. 2002.
[8] H. Bolcskei, D. Gesbert, and A. J. Paulraj. On the capacity of OFDM-based multi-antenna systems.
In Proc. IEEE Int. Conf. Acoust., Speech, Sig. Proc. (ICASSP), Istanbul, Turkey, June 2000.
[9] H. Bolcskei and A. J. Paulraj. Space-frequency codes for broadband fading channels. In Proc. IEEE
Int. Symp. on Inform. Theory (ISIT), Washington, D.C., June 2001.
[10] C. N. Chuah, D. N. C. Tse, J. M. Kahn, and R. A. Valenzuela. Capacity scaling in MIMO wireless
systems under correlated fading. IEEE Trans. Inform. Theory, 48(3):637650, Mar. 2002.
[11] T. M. Cover and J. A. Thomas. Elements of Information Theory. Wiley, 1991.
[12] M. O. Damen, A. Chkeif, and J. C. Belore. Lattice code decoder for space-time codes. IEEE
Commun. Let., 4(5):161163, May 2000.
[13] B. Dong, X. Wang, and A. Doucet. A new class of soft MIMO demodulation algorithms. IEEE Trans.
Sig. Proc., to appear.
[14] R. B. Ertel, P. Cardieri, K. W. Sowerby, T. S. Rappaport, and J. H. Reed. Overview of spatial channel
models for antenna array communication systems. IEEE Personal Commun., 5(1):1022, Feb. 1998.
[15] U. Fincke and M. Pohst. Improved methods for calculating vectors of short length in a lattice, including
a computational complexity. Mathematics for Computation, 44:463471, Apr. 1995.
[16] G. J. Foschini. Layered space-time architecture for wireless communication in a fading environment
when using multi-element antennas. Bell Labs. Tech. J., 1(2):4159, 1996.
[17] G. J. Foschini and M. J. Gans. On the limits of wireless communications in a fading environment
when using multiple antennas. Wireless Personal Commun., 6(3):311335, 1998.
[18] G. J. Foschini, G. D. Golden, R. A. Valenzuela, and P. W. Wolniansky. Simplied processing for high
spectral eciency wireless communication employing multi-antenna arrays. IEEE J. Select. Areas
Commun., 17(11):18411852, Nov. 1999.
[19] H. El Gamal and B. M. Hochwald. A new approach to layered space-time coding and signal processing.
IEEE Trans. Inform. Theory, 47(6):23212334, Sep. 2001.
[20] G. D. Golden, G. J. Foschini, R. A. Valenzuela, and P. W. Wolniansky. Detection algorithm and
initial laboratory results using the V-BLAST space-time communication architecture. IEEE Electron.
Lett., 35(1):1415, 1999.
[21] D. A. Gore and A. J. Paulraj. MIMO antenna subset selection with space-time coding. IEEE Trans.
Sig. Proc., 50(10):25802588, Oct. 2002.
[22] A. Gorokhov. Antenna selection algorithms for MEA transmission systems. In Proc. IEEE Int. Conf.
Acoust. Speech, Sig. Proc. (ICASSP), Orlando, Florida, May 2002.
[23] D. Guo and X. Wang. Blind detection in MIMO systems via sequential Monte Carlo. IEEE J. Select.
Areas Commun., 21(3):453-464, Apr. 2003.
[24] B. Hassibi and B. M. Hochwald. High-rate codes that are linear in space and time. IEEE Trans.
Inform. Theory, 48(7):18041824, July 2002.
[25] R. Heath and A. Paulraj. Antenna selection for spatial multiplexing systems based on minimum error
rate. In Proc. IEEE Int. Conf. Commun. (ICC), Helsinki, Finland, June 2001.
[26] A. S. Hedayat, N. J. A. Sloane, and J. Stufken. Orthogonal Arrays: Theory and Applications. Springer-
Verlag, New York, 1991.
24
[27] B. Hochwald, T. L. Marzetta, and C. B. Papadias. A transmitter diverisity scheme for wideband
CDMA systems based on space-time spreading. IEEE J. Select. Areas Commun., 19(1):4860, Jan.
2001.
[28] B. L. Hughes. Dierential space-time modulation. IEEE Trans. Inform. Theory, 46(7):25672578,
Nov. 2000.
[29] H. Jafarkhami and V. Tarokh. Multiple transmit antenna dierential detection from generalized
orthogonal designs. IEEE Trans. Inform. Theory, 47(6):26262631, Sep. 2001.
[30] W. C. Y. Lee. Mobile Communications Design Fundamentals. Wiley, 2nd edition, 1992.
[31] B. Lu and X. Wang. Iterative receivers for multiuser space-time coding systems. IEEE J. Select. Areas
Commun., 18(11):23222335, Nov. 2000.
[32] B. Lu and X. Wang. A space-time trellis code design method for OFDM systems. Wireless Personal
Commun., 24(3):403418, Feb. 2003.
[33] B. Lu, G. Yue, and X. Wang. Performance analysis and design optimization of LDPC-coded MIMO
OFDM systems. IEEE Trans. Sig. Proc., to appear.
[34] T. L. Marzetta. BLAST training: estimation channel characteristics for high-capacity space-time
wireless. In Proc. 37th Annual Allerton Conf. Commun., Comput. & Control, Monticello, IL, Sep.
1999.
[35] D. Reynolds, X. Wang, and H. V. Poor. Blind adaptive space-time multiuser detection with multiple
transmitter and receiver antennas. IEEE Trans. Sig. Proc., 50(6):1261-1276, June 2002.
[36] H. Sampath, P. Stoica, and A. J. Paulraj. Generalized linear precoder and decoder design for MIMO
channels using the weighted MMSE criterion. IEEE Trans. Commun., 49(12):21982206, Dec. 2001.
[37] M. Sellathurai and S. Haykin. Turbo-BLAST for wireless communications: theory and experiments.
IEEE Trans. Sig. Proc., 50(10):25382546, Oct. 2002.
[38] M. Stege, M. Bronzel, and F. Fettweis. MIMO-Capacities for COST 259 scenarios. In Proc. Int.
Zurich Seminar on Broadband Commun., ETH Zurich, Switzerland, Feb. 2002.
[39] V. Tarokh and H. Jafarkhami. A dierential detection scheme for transmit diversity. IEEE J. Select.
Areas Commun., 18(7):11691174, July 2000.
[40] V. Tarokh, H. Jafarkhami, and A. R. Calderbank. Space-time block codes from orthogonal designs.
IEEE Trans. Inform. Theory, 45(5):14561467, July 1999.
[41] V. Tarokh, H. Jafarkhami, and A. R. Calderbank. Space-time block coding for wireless communica-
tions: Performance results. IEEE J. Select. Areas Commun., 17(3):451460, Mar. 1999.
[42] V. Tarokh, A. Naguib, N. Seshadri, and A. R. Calderbank. Combined array processing and space-time
coding. IEEE Trans. Inform. Theory, 45(4):11211128, May 1999.
[43] V. Tarokh, N. Seshadri, and A. R. Calderbank. Space-time codes for high data rate wireless commu-
nications: Performance criterion and code construction. IEEE Trans. Inform. Theory, 44(2):744765,
Mar. 1998.
[44] V. Tarokh, A. Naguib, N. Seshadri, and A. R. Calderbank. Space-time codes for high data rate
wireless communication: Performance criteria in the presence of channel estimation errors, mobility
and multiple paths. IEEE Trans. Commun., 47(2):199207, Feb. 1999.
[45] I. E. Telatar. Capacity of multi-antenna Gaussian channels. Euro. Trans. Telecommun., 10(6):585595,
Nov. 1999.
[46] S. ten Brink, G. Kramer, and A. Ashikhmin. Design of low-density parity-check codes for multi-
antenna modulation and detection. IEEE Trans. Commun., submitted.
[47] P. W. Wolniansky, G. J. Roschini, G. D. Golden, and R. A. Valenzuela. V-BLAST: an architecture
for realizing very high data rates over the rich-scattering wireless channel. In Proc. 1998 Int. Symp.
Sig. Sys. Elect. (ISSSE98), Pisa, Italy, Sep. 1998.
25