This action might not be possible to undo. Are you sure you want to continue?
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 97
MIMO Wireless
Communication
Daniel W. Bliss, Keith W. Forsythe, and Amanda M. Chan
■ Wireless communication using multipleinput multipleoutput (MIMO)
systems enables increased spectral efficiency for a given total transmit power.
Increased capacity is achieved by introducing additional spatial channels that are
exploited by using spacetime coding. In this article, we survey the environmental
factors that affect MIMO capacity. These factors include channel complexity,
external interference, and channel estimation error. We discuss examples of
spacetime codes, including spacetime lowdensity paritycheck codes and space
time turbo codes, and we investigate receiver approaches, including multichannel
multiuser detection (MCMUD). The ‘multichannel’ term indicates that the
receiver incorporates multiple antennas by using spacetimefrequency adaptive
processing. The article reports the experimental performance of these codes and
receivers.
M
uiriiiiixiur multipleoutput (MIMO) sys
tems are a natural extension of developments
in antenna array communication. While the
advantages of multiple receive antennas, such as gain
and spatial diversity, have been known and exploited for
some time [1, 2, 3], the use of transmit diversity has
only been investigated recently [4, 5]. The advantages
of MIMO communication, which exploits the physi
cal channel between many transmit and receive anten
nas, are currently receiving significant attention [6–9].
While the channel can be so nonstationary that it can
not be estimated in any useful sense [10], in this article
we assume the channel is quasistatic.
MIMO systems provide a number of advantages
over singleantennatosingleantenna communication.
Sensitivity to fading is reduced by the spatial diversity
provided by multiple spatial paths. Under certain envi
ronmental conditions, the power requirements associ
ated with high spectralefficiency communication can
be significantly reduced by avoiding the compressive re
gion of the informationtheoretic capacity bound. Here,
spectral efficiency is defined as the total number of in
formation bits per second per Hertz transmitted from
one array to the other.
After an introductory section, we describe the con
cept of MIMO informationtheoretic capacity bounds.
Because the phenomenology of the channel is impor
tant for capacity, we discuss this phenomenology and
associated parameterization techniques, followed by ex
amples of spacetime codes and their respective receivers
and decoders. We performed experiments to investigate
channel phenomenology and to test coding and receiver
techniques.
Capacity
We discuss MIMO informationtheoretic performance
bounds in more detail in the next section. Capacity in
creases linearly with signaltonoise ratio (SNR) at low
SNR, but increases logarithmically with SNR at high
SNR. In a MIMO system, a given total transmit power
can be divided among multiple spatial paths (or modes),
driving the capacity closer to the linear regime for each
mode, thus increasing the aggregate spectral efficiency.
As seen in Figure 1, which assumes an optimal high
spectralefficiency MIMO channel (a channel matrix
with a flat singularvalue distribution), MIMO systems
enable high spectral efficiency at much lower required
energy per information bit.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
98 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
The informationtheoretic bound on the spectral ef
ficiency is a function of the total transmit power and
the channel phenomenology. In implementing MIMO
systems, we must decide whether channel estimation
information will be fed back to the transmitter so that
the transmitter can adapt. Most MIMO communica
tion research has focused on systems without feedback.
A MIMO system with an uninformed transmitter
(without feedback) is simpler to implement, and at high
SNR its spectralefficiency bound approaches that of an
informed transmitter (with feedback).
One of the environmental issues with which com
munication systems must contend is interference, ei
ther unintentional or intentional. Because MIMO sys
tems use antenna arrays, localized interference can be
mitigated naturally. The benefits extend beyond those
achieved by singleinput multipleoutput systems, that
is, a single transmitter and a multipleantenna receiver,
because the transmit diversity nearly guarantees that
nulling an interferer cannot unintentionally null a large
fraction of the transmit signal energy.
Phenomenology
We discuss channel phenomenology and channel pa
rameterization techniques in more detail in a later sec
tion. Aspects of the channel that affect MIMO system
capacity, namely, channel complexity and channel sta
tionarity, are addressed in this paper. The first aspect,
channel complexity, is a function of the richness of scat
terers. In general, capacity at high spectral efficiency
increases as the singular values of the channel matrix
increase. The distribution of singular values is a mea
sure of the relative usefulness of various spatial paths
through the channel.
SpaceTime Coding and Receivers
In order to implement a MIMO communication sys
tem, we must first select a particular coding scheme.
Most spacetime coding schemes have a strong connec
tion to wellknown singleinput singleoutput (SISO)
coding approaches and assume an uninformed trans
mitter (UT). Later in the article we discuss spacetime
lowdensity paritycheck codes, spacetime turbo codes,
and their respective receivers. Spacetime coding can
exploit the MIMO degrees of freedom to increase re
dundancy, spectral efficiency, or some combination
of these characteristics [11]. Preliminary ideas are dis
cussed elsewhere [6].
A simple and elegant solution that maximizes diver
sity and enables simple decoupled detection is proposed
in Reference 12. More generally, orthogonal spacetime
block codes are discussed in References 13 and 14. A
general discussion of distributing data across transmit
ters (linear dispersive codes) is given in Reference 15.
High SNR design criteria and specific examples are giv
en for spacetime trellis codes in Reference 16. Unitary
codes optimized for operation in Rayleigh fading are
presented in Reference 17. Spacetime coding without
the requirement of channel estimation is also a com
mon topic in the literature. Many differential coding
schemes have been proposed [18]. Under various con
straints at the transmitter and receiver, information
theoretic capacity can be evaluated without condition
ing on knowledge of the propagation channel [19, 20].
More recently, MIMO extensions of turbo coding have
been suggested [21, 22]. Finally, coding techniques for
informed transmitter systems have received some inter
est [23, 24].
Experimental Results
Because informationtheoretic capacity and practical
performance are dependent upon the channel phenom
enology, a variety of experiments were performed. Both
channel phenomenology and experimental procedures
are discussed in later sections. Experiments were per
0 5 –5 10
20
15
10
5
0
E
b
/N
0
(dB)
S
p
e
c
t
r
a
l
e
f
f
i
c
i
e
n
c
y
(
b
i
t
s
/
s
e
c
/
H
z
)
M = 16
M = 8
M = 4
M = 1
FIGURE 1. Spectralefficiency bound as a function of noise
spectraldensitynormalized energy per information bit
(E
b
/N
0
). The graph compares four different M × M multiple
input multipleoutput (MIMO) systems, assuming channel
matrices with flat singularvalue distribution.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 99
formed in an outdoor nonstationary environment in a
mixed residential, industrial, and light urban settings.
Intentional highpower interference was included.
InformationTheoretic Capacity
The informationtheoretic capacity of MIMO systems
has been widely discussed [7, 25]. The development of
the informed transmitter (“water filling”) and unin
formed transmitter approaches is repeated in this sec
tion, along with a discussion of the relative performance
of these approaches. (The concept of “water filling” is
explained in the sidebar entitled “Water Filling.”) In
addition, we introduce the topic of spectralefficiency
bounds in the presence of interference, and we discuss
spectralefficiency bounds in frequencyselective envi
ronments. Finally, we summarize alternative channel
performance metrics.
Informed Transmitter
For narrowband MIMO systems, the coupling between
the transmitter and receiver for each sample in time can
be modeled by using
z Hx n + ,
(1)
where z is the complex receivearray output,
H∈
×
n n
R T
is the n
R
× n
T
(number of receive by transmit antenna)
WATER FI LLI NG
W
arii iiiiixc is a metaphor for the solution
of several optimization problems related to
channel capacity. The simplest physical example is
perhaps the case of spectral allocation for maximal
total capacity under a total power constraint. Let x
k
denote the power received in the kth frequency cell,
which has interference (including thermal noise) de
noted n
k
. If the total received power is constrained
to be x, then the total capacity is maximized by
solving
max log( )
max
{ }
{ }
x x x
k
k k
x x x
k k
k
k k
k
x n
:
:
∑
∑
∑
+ /
1
kk
k k
k
k
n x n
∑ ∑
+ − . log( ) log( )
Use Lagrange multipliers and evaluate
∂
∂
+ − −

(
'
'
`
J
J
J
]
]
]
]
∑ ∑
x
n x x x
k
j
j j
j
j
log( ) µ
to find a solution. The solution satisfies x
k
+ n
k
=
µ
–1
for all nonzero x
k
. Figure A illustrates the solu
tion graphically as an example of water filling. The
difference between the water level (blue) and the
noise level (red) is the power allocated to the signal
Noise
Frequency
P
o
w
e
r
FIGURE A. Notional waterfilling example.
in each frequency cell. The volume of the water is
the total received power of the signal. Note that cells
with high levels of interference are not used at all.
A similar solution results when the capacity is ex
pressed by
log( ) 1+
∑
g x
k k
k
for gains g
k
. One can write the gains as g n
k k
−1
and use the waterfilling argument above. In this
context, cells with low gains may not be used at all.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
100 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
channel matrix, x is the transmitarray vector, and n is
zeromeancomplex Gaussian noise.
The capacity is defined as the maximum of the mu
tual information [26]
I ( ,  ) log
(  , )
(  )
, z x H
z x H
z H
]
]
] 2
p
p
(2)
over the source conditional probability density p(  ) x H
subject to various transmit constraints, where the ex
pectation value is indicated by the notation . Not
ing that the mutual information can be expressed as the
difference between two conditional entropies
I ( ,  ) (  ) (  , ), z x H z H z x H − h h
(3)
that
h h n e
R n
(  , ) ( ) log ( ), z x H n
2
2
π σ
and that h(  ) z H is maximized for a zeromean Gauss
ian source x, the capacity is given by
C
n n
n n
R
R
+
, sup log
†
† †
x x
I H xx H
I
2
2
2
σ
σ
(4)
where the notation indicates determinant, † indi
cates Hermitian conjugate, and I
n
R
indicates an identity
matrix of size n
R
. A variety of possible constraints ex
ist for xx
†
, depending on the assumed transmitter
limitations. Here we assume that the fundamental limi
tation is the total power transmitted. Optimization over
the n
T
× n
T
noisenormalized transmit covariance ma
trix, P xx /
†
, σ
n
2
is constrained by the total noise
normalized transmit power P
o
. By allowing different
transmit powers at each antenna, we can enforce this
constraint by using the form tr{ } P ≤ P
o
. The informed
transmitter (IT) channel capacity is achieved if the
channel is known by both the transmitter and receiver,
giving
C
P
n
o
R
IT
tr
+ .
;
sup log
( )
†
P P
I HPH
2
(5)
To avoid radiating negative power, we impose the addi
tional constraint P > 0 by using only a subset of channel
modes.
The resulting capacity is given by
C
P
n
o
IT
tr
+
.
−
+
log
{ }
2
1
D
D
(6)
A waterfilling argument establishes that the entries d
m
in the diagonal matrix
D∈
+ +
×
n n
contain the n
+
topordered eigenvalues of HH
†
. The
values d
m
must satisfy
d
n
P
m
o
>
+
.
+
−
tr{ } D
1
(7)
If Equation 7 is not satisfied for some d
m
, it will not be
satisfied for any smaller d
m
.
In this discussion we assume that the environment is
stationary over a period long enough for the error asso
ciated with channel estimation to vanish asymptotically.
In order to study typical performance of quasistationary
channels sampled from a given probability distribution,
capacity is averaged over an ensemble of quasistationary
environments. Under the ergodic assumption (that is,
the ensemble average is equal to the time average), the
mean capacity C
IT
is the channel capacity.
Uninformed Transmitter
If the channel is not known at the transmitter, then
an optimal transmission strategy is to transmit equal
power from each antenna P I / P n
o T n
T
[7]. Assum
ing that the receiver can accurately estimate the chan
nel, but the transmitter does not attempt to optimize its
output to compensate for the channel, the uninformed
transmitter (UT) maximum spectral efficiency bound
is given by
C
P
n
n
o
T
R
UT
+ . log
†
2
I HH
(8)
This is a common transmit constraint, as it may be dif
ficult to provide the transmitter channel estimates. The
sidebar entitled “Toy 2 × 2 Channel Model” discusses
an example of IT and UT capacities for a simple line
ofsight environment.
Capacity Ratio
At high SNR, C
IT
and C
UT
converge. This can be ob
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 101
served in the large P
o
limit of the ratio of Equations 6
and 8,
C
C
P
n
o
min
o
T
min
n
P
n
IT
UT
tr
→
+
+
→
−
log
{ }
log
l
2
2
1
D
D
I D
oog log
log log
log
2 2
2 2
2
P n
P n
o min n
o
min
( ) ( )
( )
− +
−
D
TT n
min
( ) +
→
log
,
2
1
D
(9)
where the n
min
diagonal entries in D contain all non
zero eigenvalues of H
†
H. If n
T
> n
R
, then the conver
gence to one is logarithmically slow.
At low SNR the ratio C
IT
/C
UT
is given by
C
C
P d d
o max max
n
P
n
R
o
T
IT
UT
→
+ /
+
log [( ) ]
log
†
2
2
1
I HH
+
+

(
'
`
J
log( { })
log
†
†
1 P
o
n
P
n
R
o
T
maxeig
tr
HH
I HH
JJ

¦




≈ ,
maxeig
tr
{ }
{ }
†
†
HH
HH
1
n
T
(10)
using Equation 6 with n
+
= 1 and Equation 8. Given
this asymptotic result, we can make a few observations.
The spectralefficiency ratio is given by the maximum
to the average eigenvalue ratio of H
†
H. If the channel
is rank one, such as in the case of a multipleinput sin
gleoutput (MISO) system, the ratio is approximately
equal to n
T
. Finally, in the special case in which H
†
H
has a flat eigenvalue distribution, the optimal transmit
covariance matrix is not unique. Nonetheless, the ratio
C
IT
/C
UT
approaches one.
Interference
By extending the discussion in the previous section [8,
27], we can calculate capacity in the presence of unco
operative (worst case) external interference η, in addi
tion to the spatiallywhite complex Gaussian noise n
considered previously. The mutual information is again
given by Equations 2 and 3, where entropy h(  , ) z x H
in the presence of the external interference becomes
h( ) n +η ,
h e
n n
(  ) log z x H I R , ≤ +
¦ ¦
,
2
2 2
π σ σ
and σ
n
2
R is the spatialinterference covariance matrix.
Equality is achieved if and only if the interference am
plitudes have a Gaussian distribution. Thus the worst
case informed capacity, the maximumminimum mu
tual information,
C I
p
p
int
,  ,

sup inf ( )
( )
( )
z H
z x H
η
(12)
becomes
C
P
o
IT int
tr
,
;
 +  , sup log
( )
†
P P
I HPH
2
(13)
using
H I R H ≡ + .
− /
( )
1 2
(14)
Gaussian interference corresponds to a saddle point of
the mutual information at which the maximummini
mum capacity is achieved. The capacity in the pres
ence of Gaussian interference has a form identical to
Equation 6 under the transformation D D →
, where
D contains the eigenvalues of
HH
†
. The transmitted
noisenormalized power covariance matrix
P is calcu
lated by using
H. Similarly, the uninformed transmit
ter spectralefficiency bound in the presence of noise is
given by the same transformation of H H →
.
In the limit of high spectral efficiency for n
J
infinite
J/S jammers, the loss in capacity approaches
C
n n n
n n
C
T R J
T R
int
→
, −
,
.
min( )
min( )
(15)
In general, the theoretical capacity is not significantly
affected as long as the number of antennas is much
larger than the number of jammers. This resistance to
the effects of jammers is demonstrated experimentally
later in the article.
FrequencySelective Channels
In environments in which there is frequencyselec
tive fading, the channel matrix H(f ) is a function of
frequency. Exploiting the orthogonality of frequency
channels, the capacity in frequencyselective fading can
be calculated by using an extension of Equations 6 and
8. For the uninformed transmitter, this leads to the fre
quencyselective spectralefficiency bound
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
102 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
TOY 2 × 2 CHA NNEL MODEL
beamwidths closely approximates many ad hoc
definitions for physical arrays. Figure A displays the
eigenvalues µ
1
and µ
2
as a function of generalized
beamwidth separation. When the transmit and re
ceive arrays are small, indicated by a small separa
tion in beamwidths, one eigenvalue is dominant.
As the array apertures become larger, indicated by
a larger separation, one array’s individual elements
can be resolved by the other array. Consequently,
the smaller eigenvalue increases. Conversely, the
larger eigenvalue decreases slightly.
Equations 6 and 7 in the main article are em
ployed to determine the capacity for the 2 × 2 sys
tem. The “waterfilling” technique (explained in a
previous sidebar) must first determine if both modes
in the channel are employed. Both modes are used
if the following condition is satisfied,
µ
µ µ
µ µ
2
1 1
2 1
1 2
2
1 2
2
2
1 1
1
1 2
>
+ +
,
> − >
−

(
P
P
a
o
o
v v
v v
†
†
''
`
J
J
,
assuming µ
1
> µ
2
.
If the condition is not satisfied, then only the
0 0.2 0.4 0.6 0.8 1.0
–30
–20
–10
0
Generalized beamwidth separation
E
i
g
e
n
v
a
l
u
e
/
a
2
(
d
B
)
FIGURE A. Eigenvalues of HH
†
for a 2 × 2 lineof
sight channel as a function of antenna separation.
B
icausi rui oisriinuriox of channel matrix
eigenvalues is essential to the effectiveness of
multipleinput, multipleoutput (MIMO) commu
nication, we employ a toy example for the purposes
of introduction, and we discuss the eigenvalue dis
tribution of a 2 × 2 narrowband MIMO system in
the absence of environmental scatterers. To visualize
the example, we can imagine two receive and two
transmit antennas located at the corners of a rect
angle. The ratio of channel matrix eigenvalues can
be changed by varying the shape of the rectangle.
The columns of the channel matrix H (in Equation
1 in the main article) can be viewed as the receiver
array response vectors, one vector for each transmit
antenna,
H v v ( ), 2
1 1 2 2
a a
where a
1
and a
2
are constants of proportionality
(equal to the rootmeansquared transmittoreceive
attenuation for transmit antennas 1 and 2 respec
tively) that take into account geometric attenuation
and antenna gain effects, and v
1
and v
2
are unit
norm array response vectors. For the purpose of this
discussion, we assume a = a
1
= a
2
, which is valid if
the rectangle deformation does not significantly af
fect overall transmittertoreceiver distances.
The capacity of the 2 × 2 MIMO system is a
function of the channel singular values and the total
transmit power. Eigenvalues of HH
†
are given by
µ
1 2
2
1 2
2 1
,
±
( )
, a v v
†
where the absolute value is denoted by . The
separation between receive array responses can be
described in a convenient form in terms of general
ized beamwidths [40],
b
12 1 2
2
¦ ¦
.
π
arccos
†
v v
For small angular separations, this definition of
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 103
stronger channel mode is employed and the capac
ity, from Equation 6, is given by
C P
a P
o
o
IT
+ ,
+ +
( )
]
]
;
log ( )
log
†
2 1
2
2
1 2
1
1 2 1
µ
v v
otherwise, both modes are used and the capacity is
given by
C
P
P
o
IT
+ +

(
'
`
J
J
,
log
log
2
1 1
1
2
2
2
1 2
1 2
2
0
0
µ µ
µ
µ
µ µ
oo
a
+ +

(
'
`
J
J

¦








,
−
µ µ
µ µ
1 2
1 2
2
2
1
2
1
2 1 log v
††
†
log
v
v v
2
2
2 1 2
2
1
1

(
'
`
J
J

¦








+
¦ ¦
− − .
P
o
Figure B displays the resulting capacity as a func
tion of a
2
P
o
(mean singleinput singleoutput SNR)
for two beamwidth separations, 0.1 and 0.9. At low
values of a
2
P
o
the capacity associated with small
beamwidth separation performs best. In this regime,
capacity is linear with receive power, and small
beamwidth separation increases the coherent gain.
At high values of a
2
P
o
large beamwidth separation
produces a higher capacity as the optimal MIMO
system distributes the energy between modes.
The total received power is given by
2
1 2
2
v v
†
a P
o
when using one mode, and
FIGURE B. The informed transmitter capacity of a
2 × 2 lineofsight channel, assuming antenna beam
width separations of 0.1 (solid line) and 0.9 (dashed
line).
–10 –5 0 5 10 15 20
0.5
1
2
5
10
a
2
P
0
(dB)
S
p
e
c
t
r
a
l
e
f
f
i
c
i
e
n
c
y
(
b
/
s
e
c
/
H
z
)
2
2
1
2
1 2
2
1 2
2
a P
o
+
−

(
`
J
v v
v v
†
†
,
when using two modes, where P
o
is the total noise
normalized power. In both cases, the total received
power is much larger than a
2
P
o
.
In complicated multipath environments, small
arrays employ scatterers to create virtual arrays of
a much larger effective aperture. The effect of the
scatterers upon capacity depends on their number
and distribution in the environment. The individual
antenna elements can be resolved by the larger ef
fective aperture produced by the scatterers. As dem
onstrated in Figure A, the ability to resolve antenna
elements is related to the number of large singular
values of the channel matrix and thus the capacity.
C
df C P f
df
f
o
n
n
P f
o
UT, FS
UT
;
≈
∆ +
∫
∫
∑
( ( ))
log
H
I
1
2 nn n n
n
n
T
f
f f
f
H H ( ) ( )
†
∑
∆
1
≈ + ,
1
2
n
P
n
f
o
T
log
†
I HH
(16)
where the distance between frequency samples is given
by ∆f and the n
f
bin frequencypartitioned channel
matrix is given by
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
104 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
H
H
H
H
≡

(
'
'
'
'
`
J
J
J
J
( )
( )
( )
f
f
f
n
f
1
2
0 0 0
0 0 0
0 0
JJ
.
(17)
For the informed transmitter channel capacity, pow
er is optimally distributed amongst both spatial modes
and frequency channels. The capacity can be expressed
C
n
f
IT FS ,
≈ + , max log
†
P
I HPH
1
2
(18)
which is maximized by Equation 6 with the appropriate
substitutions for the frequencyselective channel, and
diagonal entries in D in Equation 7 are selected from
the eigenvalues of
HH
†
. Because of the block diagonal
structure of H, the ( ) ( ) n n n n
T f T f
⋅ × ⋅ spacefrequency
noisenormalized transmit covariance matrix
H is a
block diagonal matrix, normalized so that
tr{ } .
P ≤ n P
f o
Other Performance Metrics
The informationtheoretic capacity is not the only pos
sible metric of performance. As an example, another
useful performance metric is the outage capacity [16],
or the achievable spectralefficiency bound, assuming a
given probability of errorfree decoding of a frame. In
many practical situations this metric may be the best
measure of performance, for example, in the case in
which the system can resend frames of data.
Channel Phenomenology
In this section we describe tools for modeling, estimat
ing, and characterizing MIMO channels. These topics
are discussed in greater detail elsewhere [25, 28]. First
we introduce the standard model and simple modifica
tions to it. Then we discuss the simplest channel char
acterization, which is mean receive power, followed by a
description of channel estimation techniques, methods
for determining how much channels have changed, and
channel parameterization and estimation techniques.
Standard Model
A variety of techniques are used to simulate the channel
matrix [29]. The simplest approach is to assume that
all the entries in the channel matrix are sampled from
identical independent complex Gaussians H G . This
assumption corresponds to an environment with com
plicated multipath scattering. While this approach is
convenient from the perspective of performing analytic
calculations, it may provide a channel eigenvalue distri
bution that is too flat. At the other extreme, channels
can be characterized by a diversity order [30], which is
used to indicate an effective cutoff in the eigenvalue
distribution induced by spatial correlation. A number of
approaches that introduce spatial correlations have been
suggested. One approach uses the form
H M GM .
L R
(19)
The above model results in a ( ) ( ) n n n n
T R T R
⋅ × ⋅ link
bylink covariance matrix of the Kronecker product
form ( ) ( )
† †
M M M M
L L R R
⊗
∗
for the entries in the chan
nel matrix H. This product structure can arise from a
spherical Green’s function model of propagation, pro
vided several additional conditions are met. First, scat
terers are concentrated around (but not too close to) the
transmitter and receiver. Second, multiple scattering of
a particular kind (from transmitter element to trans
mitter scatterer to receiver scatterer to receiver element)
dominates propagation. Third, scatterers are sufficient
ly separated in angle when viewed by their associated
array.
Received Power
It is often convenient to parameterize the incoming
signal power in terms of a
2
P
o
, where a
2
is the mean
squared link attenuation. It can be employed to eas
ily compare performance by using different constraints
and environments. This choice corresponds to the typi
cal noisenormalized received power for a single receive
and single transmit antenna radiating power σ
n o
P
2
.
However, this choice can be mildly misleading because
the total received power will, in general, be much larger
than a
2
P
o
. In general, a
2
is defined by the Frobenius
norm squared of the channel matrix normalized by the
number of transmitters and receivers,
a
n n
T R
2
.
tr{ }
†
HH
(20)
The total received noisenormalized power produced
by a set of orthogonal receive beamformers is given by
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 105
tr{ }
†
HPH . The uninformed transmitter rate is maxi
mized by sending equal power to all transmit antennas
so that tr{ }
†
HPH becomes P n n a P
o T R o
/ tr{ }
†
HH
2
.
It is worth noting that P is not in general optimized by
the informed transmitter to maximize received power
but to maximize capacity.
The total received power for the capacityoptimized
informed transmitter, given an arbitrary channel ma
trix, is
tr tr
tr
{ }
( )
†
HP H
D
D I
IT
o
n
P
n
+

(
'
`
J
J
−
−
+
+

¦





1



+
−
.
+
−
+
+
P
n
n
n
o
tr tr tr { } { } { } D D D
1 2
(21)
The first term in Equation 21 is bounded from below
by
P
n
P
n n
n n a P
o o
T R
T R
tr tr
min
max
{ } { }
{ }
{ }
†
D HH
+
≥
,
≥ ,
2
oo
.
(22)
The second term in Equation 22 is bounded from be
low by zero. Consequently, the total received power is
greater than or equal to max{ } n n a P
T R o
,
2
.
For very small a
2
P
o
, far from the nonlinear regime of
the Shannon limit, the optimal solution is to maximize
received power. This is done by transmitting the best
mode only, setting n
+
= 1. In this regime the total re
ceived power is given by
tr maxeig { } { }
† †
HP H HH
IT o
P → .
(23)
This result is bounded from above by n n a P
T R o
2
, which
is achieved if there is only a single nontrivial mode in
the channel.
Channel Estimation
The Gaussian probability density function for a multi
variate, signalinthemean, statistical model of the re
ceived signal Z, assuming T ∈
×
n n
T s
is the transmit
sequence, is given by
p
e
n n n
s R s
( )
[( ) ( )]
†
Z R H T
R
Z HT R Z HT
 , ,
 
− − −
−
tr
1
π
,,
(24)
where R is the noiseplusinterference covariance ma
trix. The maximumlikelihood estimate of H is given
by
ˆ
( )
† †
H ZT TT ,
−1
(25)
assuming that the reference signals in T are known and
TT
†
is nonsingular.
The previous channelestimation discussion explicit
ly assumed flat fading. However, the frequencyselective
channels can be estimated by first estimating a finite
impulseresponse MIMO channel, which can be trans
formed to the frequency domain.
A finite impulseresponse extension of Equation 1
is given by introducing delayed copies of T at delays
δ δ δ
1 2
, , ,
n
taps
,
T
T
T
T
≡

(
'
'
'
'
`
J
J
J
J
J
,
( )
( )
( )
δ
δ
δ
1
2
n
taps
(26)
so that the transmit matrix has dimension ( ) n n n
T taps s
⋅ × .
The resulting wideband channel matrix has the dimen
sion ( ) n n n
T taps s
⋅ × ,
[
ˆ
( )
ˆ
( )
ˆ
( )]
( )
† †
H H H
ZT TT
δ δ δ
1 2
1
n
taps
.
−
(27)
Using this form, an effective channel filter is associated
with each transmittoreceive antenna link. By assum
ing regular delay sampling, we can use a discrete Fou
rier transform to construct the explicit frequencyselec
tive form,
[
ˆ
( )
ˆ
( )
ˆ
( )]
[
ˆ
( )
ˆ
( )
ˆ
(
H H H
H H H
f f f
n
taps
1 2
1 2
δ δ δδ
n n n
taps taps T
)]( ) ⊗ , I
(28)
where the npoint discrete Fourier transform is repre
sented by
n
and the Kronecker product is represented
by ⊗.
ChannelDifference Metrics
A variety of metrics are possible. In investigating chan
nel variations, no one metric will be useful for all situa
tions. As an example, two completely different channels
can have the same capacity. Depending upon the issue
being investigated, we may wish to think of these matri
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
106 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
ces as being similar or very different. Here two metrics
are discussed. Both metrics are ad hoc, but motivations
are provided. The first metric measures differences in
channel singularvalue distributions. The second metric
is sensitive to differences in both the singularvalue dis
tribution and the channel eigenvector structure.
EigenvalueBased Metric
As was mentioned earlier, MIMO capacity is only a
function of the channel singular values. Equivalently,
capacity is invariant under channelmatrix transforma
tions of the form
H WHW → ,
1 2
†
(29)
where W
1
and W
2
are arbitrary unitary matrices. Con
sequently, for some applications it is useful to employ a
metric that is also invariant under this transformation.
Because capacity is a function of the structure of the
channel singularvalue distribution, the metric should
be sensitive to this structure.
The channel capacity is a function of HH
†
. A natu
ral metric would employ the distance between the ca
pacity for two channel matrices at the same average to
tal received power, that is, the same a
2
P
o
,
∆ +
− + .
C
a P
n
a P
n
o
T
a a
o
T
b b
UT
log
log
†
†
2
2
2
2
I
H H
I
H H
(30)
However, there are two problems with this definition.
First, the difference is a function of P
o
. Second, there is
degeneracy in H singular values that gives a particular
capacity. To address the first issue, the difference can be
investigated in a high SNR limit, giving
∆ ≈ −
,
∑
C
a a b b
m
n n
T R
UT
log log
log
† †
min( )
2 2
1
2
H H H H
λλ λ
m a a m b b
( ) log ( )
† †
H H H H
− ,
2
where λ
m
( ) X indicates the mth largest eigenvalue of X.
To increase the sensitivity to the shape of the eigenvalue
distribution, the metric is defined to be the Euclidean
difference, assuming that each eigenvalue is associated
with an orthogonal dimension, giving
δ
λ
2
1
2 2
( , )
[log ( ) log
min( )
†
H H
H H
a b
m
n n
m a a
T R
≡ −
,
∑
λλ
m b b
( )]
†
H H .
Fractional Receiver Loss Metric
In this section we introduce a powerweighted mean
cos
2
θ metric. The metric takes into account both the
eigenvalue and eigenvector structure of the channels. It
is motivated by the effect of receivebeamformer mis
match on capacity. Starting with Equation 8, the low
SNR uninformed transmitter capacity approximation is
given by
C
P
n
e
P
n
o
T
o
T
+
≈

¦








log
log ( )
†
†
2
2
I HH
HH tr
llog ( )
log ( )
†
†
2
2
2
e
P
n
e
P
n
o
T
m
m m
o
T
m
m m
m
m
∑
∑
≡
h h
w h
w
hh
h
m
,
(33)
where h
m
is the column of the channel matrix associat
ed with transmitter m, and indicates the l
2
norm.
In the low SNR limit, the optimal receive beamformer
is given by the matched response given in w
m
. If some
other beamformer is employed, labeled ′ w
m
, then signal
energy is lost, adversely affecting the capacity,
′ ≈ ′ .
∑
C e
P
n
o
T
m
m m
log ( )
†
2
2
w h
(34)
One possible reason that a beamformer might use the
wrong matched spatial filter is channel nonstationarity.
The fractional capacity loss is given by
′
≈
′
∑
∑
∑
∑
′
′ C
C
m
m m
m
m
m
m
m
m
m
m
m
m
2
2
2
2
2
†
†
†
w h
h
h
h
h
h
h
h
mm
m m
m
m
∑
∑
,
2
2
2
h
h
cos θ
(35)
(31)
(32)
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 107
which is the powerweighted mean cos
2
θ
m
estimate,
where cos θ
m
is defined to be the inner product between
the “good” and “bad” unitnorm array responses for the
mth transmitter. It is generally desirable for metrics to
be symmetric with respect to H and ′ H, thus avoiding
moral attributions with regard to channel matrices. Us
ing the previous discussion as motivation, a symmetric
form is given by
γ
θ
( )
cos
H H
h h
h h
, ′ ≡
′
′
,
∑
∑
m
m m m
m
m m
2
(36)
where the “powerweighted” expectation is evaluated
over transmitters.
Singular Values
The singularvalue distribution of H, or the related ei
genvalue distribution of HH
†
, is a useful tool for un
derstanding the expected performance of MIMO com
munication systems. From the discussion earlier, we can
see that the channel capacity is a function of channel
singular values, but not the singularvector structure of
the channel. Thus channel phenomenology can be in
vestigated by studying the statistics of channel singular
value distributions.
Channel Parameterization
A commonly employed model assumes the channel is
proportional to a matrix G, where the entries are inde
pendently drawn from a unitnorm complex circular
Gaussian distribution. While the distribution is conve
nient, it does suffer from a singularvalue distribution
that is overly optimistic for many environments. As was
previously discussed, one solution is to introduce spatial
correlations using the transformation F M GM b
L R
†
[29]. While this approach is limited, it produces simply
more realistic channels than the uncorrelated Gaussian
model. The spatial correlation matrices can be factored
so that M UA U
L
L
α
†
and M VA V
R
R
α
†
, where U
and V are unitary matrices, and A
α
L
and A
α
R
are posi
tivesemidefinite diagonal matrices.
Assuming that the number of transmit and receive
antennas are equal and have similar spatial correlation
characteristics, the diagonal matrices can be set equal,
A A A
α α α
L R
, producing the new random channel
matrix F, where
F UA U G VA V
UA GA V
′
b
b
α α
α α
† †
†
(37)
and
A
α
α α α
α α α
, , ,
, , ,
−
−
n
n
n
diag
tr(diag
{ }
{ }
0 1 1
0 1 1
22
)
,
(38)
where b is used to set overall scale, n is given by the size
of A
α
, and U and V indicate random unitary matrices.
Used here is the fact that arbitrary unitary transforma
tions do not affect the statistics of the Gaussian matrix.
The form of A
α
given here is somewhat arbitrary, but
has the satisfying characteristics that as α → 0, a rank
one channel matrix is produced, and as α →1, a spa
tially uncorrelated Gaussian matrix is produced. Fur
thermore, empirically this model provides good fits to
experimental distributions. The normalization for A
α
is
chosen so that the expected value of F
F
2
is b n n
T R
2
,
where
F
2
indicates the Frobenius norm.
Channel Parameter Estimation
An estimate for
ˆ
α associated with particular transmit
and receive locations is given by minimizing the mean
square metric given in Equation 32,
ˆ
arg min [ ( )]
^ ^
α δ α , ,
2
H F
(39)
where
ˆ
X indicates the estimated value of X. Here the
expectation, denoted by , indicates averaging is over
an ensemble of F for a given α and an ensemble of H for
given transmit and receiver sites.
It is worth noting that this approach does not neces
sarily provide an unbiased estimate of α. Estimates of α,
using the metric introduced here, are dependent upon
the received SNR. Data presented later in the article
have sufficiently high SNR such that α can be estimat
ed within ±0.02.
SpaceTime LowDensity ParityCheck Codes
This section of the article introduces lowdensity parity
check (LDPC) codes, which were studied extensively
by R.G. Gallager [31]. The significance of modern im
plementations of LDPC codes rests on iterative decod
ing algorithms that, for LDPC codes, are applications
of techniques formulated for Bayesian belief networks,
which are introduced below. This section also discusses
a simple application of LDPC to spacetime codes.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
108 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
BAY ESI A N BELI EF NETWOR KS
G
iaiuicai oicooixc is often
based on the Bayesian belief
networks popularized by Pearl, in
the context of machine learning, in
a wellknown monograph [1]. An
interpretation of various decoding
algorithms in terms of Bayesian be
lief networks is presented elsewhere
[2].
To appreciate the use of belief
networks for decoding, consider
the probability density function de
noted p x x x x ( )
1 2 3 4
, , , in Figure A.
This function factors in the man
ner shown in the figure, expressing
simpler variable dependencies than
those allowed by the multivariate
notation p x … x
l
( )
1
, , . The factor
ization can be represented by a di
rected acyclic graph as shown, with
directed arrows expressing condi
tional probabilities of the more gen
eral form p x u … u
l
( )  , ,
1
.
For decoding purposes, each
node in the graph maintains an al
phabet (for example, the symbol al
phabet for coding applications) and
several (probability) distributions
over this alphabet. One probability
distribution, denoted π(x), can be
interpreted as a prior density on the
alphabet while another (nonnega
tive, but not a normalized density)
distribution, denoted λ(x), can be
interpreted as a likelihood function
on the alphabet.
In addition, each node keeps
track of a belief function that is the
product of priors and likelihoods:
π(x) λ(x). The maximum of the be
lief function can be used as a deci
sion on the value of the node’s al
phabet.
To evaluate a consistent set of
distribution functions, messages
are received and transmitted from
each node. Messages that flow from
parent to child are denoted π
k
P
k
u ( )
and are treated as if they were priors,
while messages that flow from child
to parent are denoted λ
k
C
x ( ) and are
treated as if they were likelihoods.
At each node, messages received
from parents and children are used
to update the internal (for that
node) prior and likelihood func
tions π(x) and λ(x) for the node’s
alphabet.
Nodes are activated in any order,
subject only to the requirement that
all incoming messages are available.
When a node is activated, it calcu
lates its internal prior and likeli
hood functions and then makes its
messages available to its parent and
child nodes. Initial settings of the
internal functions are provided (but
not shown in the figure) to enable
the process to start.
LowDensity ParityCheck Codes
LDPC codes were developed by Gallager, who studied
their distance properties and decoding in a wellknown
monograph [31]. With the advent of graphical decod
ing techniques, softdecision decoding of LDPC codes
became practical, resulting in renewed interest in these
codes. Subsequent developments in code design and
decoding have led to codes that achieve levels of per
formance astonishingly close to the Shannon capacity
[32], albeit at the cost of extremely long codewords.
However, decoding complexity of LDPC codes scales
linearly (with a fixed number of iterations) with the
code length, making relatively long codes practical.
LDPC codes are linear block codes defined by a
paritycheck matrix. Each symbol in the codeword is
involved in only a few paritycheck equations. Con
sequently, most entries in the paritycheck matrix are
zero. Regular LDPC codes have n
C
paritycheck equa
tions for each symbol, and each paritycheck equation
involves n
R
symbols. Thus, if the dimensionality of the
paritycheck matrix is r × c, we have rn
R
= cn
C
for regu
lar LDPC codes. As an example, the LDPC code used
for some of the experiments described later satisfies (r, c)
= (512, 1024) and (n
R
, n
C
) = (8, 4). More powerful
codes that are not regular are also known [33].
LDPC Decoding
Recently, graphical decoding techniques have motivat
ed practical code design. Bayesian belief networks [34]
can be used to formulate decoders for LDPC codes and
turbo codes (the sidebar entitled “Bayesian Belief Net
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 109
For loopless graphs, the order of
activation does not matter and the
process converges. Unfortunately,
for decoding applications, interest
ing graphs have loops, so order of
activation matters and convergence
is not guaranteed. Typically, nodes
are activated in a repetitive pattern
for a certain number of iterations
until a stopping criterion is met.
Symbol decisions are based on the
belief function.
References
1. J. Pearl, Probabilistic Reasoning in In
telligent Systems: Networks of Plausible
Inference (Morgan Kaufmann, San
Mateo, Calif., 1988).
2. R.J. McEliece, D.J.C. MacKay, and
J.F. Cheng, “Turbo Decoding as an
Instance of Pearl’s ‘Belief Propagation’
Algorithm,” IEEE J. Sel. Areas Com
mun. 16 (2), 1998, pp. 140–152.
Variable dependencies
Loopless directed acyclic graph (DAG)
Directed Markov field
Bayesian belief network
Messages passed
likelihoods
priors
Received Sent
Node calculations and messages
Belief
p(x
1
,
x
2
,
x
3
,
x
4
) = p(x
4
x
3
)p(x
3
x
1
)p(x
2
x
1
)p(x
1
)
x
1
x
2
x
3
x
4
Child nodes
(x) =
Σ
p(xu
1
,…,u
s
)
Π
P
(u
k
)
k=1 u
s
k
π π
P
π
C
π
λ(x) (x) π
Parent nodes (alphabets u
k
B
k
) ∈
Update node
(alphabet x A) ∈
C
λ
P
λ
(x) =
Π
C
(x)
k
k
λ λ
C
(x) = (x)
Π
C
(x)
j k
k≠j
π π λ
P
(u
j
) =
Σ
(x)p(xu
1
, …, u
s
)
Π
P
(u
k
)
j k
k≠j x,u
k
: k≠j
π λ λ
FIGURE A. Bayesian belief networks provide a framework for representing conditional probabilities in a graphi
cal manner. Each node has a symbol alphabet on which it maintains a belief function that factors as a product of a
priorlike function and a likelihoodlike function. Beliefs are updated by passing messages among nodes in a man
ner suggested by the terminology. Initial states and a node update order must be chosen. Only in special cases do
the iterations converge to a Bayesian decision, but for many interesting applications, the iterative technique is both
practical and effective. Turbo codes and low density paritycheck codes have decoders based on this paradigm.
works” provides more information). However, beyond
connecting the decoding algorithm of LDPC codes to
Bayesian belief networks, a thorough explanation of the
steps in this algorithm is outside the scope of this ar
ticle; we present only a concise summary.
For LDPC codes, Figure 2 shows a graph illustrat
ing data and paritycheck dependencies for the code
words. In general, each nonzero entry in the parity
check matrix indicates the edge of a graph connecting a
paritycheck node (row index) and a codeword symbol
(column index). The example in Figure 2 is a single par
itycheck code on four symbols. The graph shows the
symbol nodes c
1
,…, c
4
, the data nodes z
1
,…, z
4
, and
the paritycheck nodes, labeled by zeroes. Each edge
between a paritycheck node and a symbol node corre
sponds to a nonzero entry in the paritycheck matrix.
Decoding occurs by treating the graph as a Bayes
ian belief network using the conditional probabilites
p z c
k k
( )  , which express the likelihood ratios, and
p c … c c
i i i
k
l k
( ) , 0
1
 , ,
( ) ∑
δ
which expresses the paritycheck relation. The resulting
algorithm can be viewed as sweeping through the rows
and columns of the paritycheck matrix, updating like
lihood ratios l
k
for each nonzero entry in the matrix.
The notation below denotes l
i
j
as the likelihood ratio
stored with the i
j
th (nonzero) entry in a fixed (for the
given step) row or column of the paritycheck matrix. In
this form, the iterative steps of the algorithm are sum
marized for the simple case of a binary symbol alphabet
by the equations:
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
110 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
1. Row sweeps
tanh tanh
( )
( )
l
l
i
j k
i
k
j
new
old
2 2
]
]
]
]
]
]
≠
∏
2. Column sweeps [column c with loglikelihood ra
tio l
c
( ) LLR
]
l l l
i
j k
i c
k j
( ) ( ) ( ) new new LLR
+
≠
∑
3. Bit decisions (column c)
sign
LLR
k
i c
l l
k
∑
+
]
]
]
]
( )
.
For the code used in the experiments, each row sweep
involves eight l
i
j
per row and each column sweep four l
i
j
per column.
Each of the row (column) operations is independent
of any other row (column operation) and hence can be
implemented in any order or in parallel. This allows
considerable acceleration of hardware decoders. Decod
ing can be halted after a fixed number of iterations or
after the paritycheck equations are satisfied.
Some simplifications that are not possible for nonbi
nary symbol alphabets are involved in the binary case.
In this more general context, the row/column sweeps
are expressed by:
1. Row sweeps
π β π β ( ) ( )
( ) ( )
i i
n
n j k n i i
k k j j
new old
p U U p
]
≠
2
]]
]
2. Column sweeps
i j k i c
k j
( ) ( ) ( ) new old LF
p p p ∝
]
]
]
≠
3. Symbol decisions
m i c
m
p p
( )
( )
.
LF
Paritycheck matrix
1 1 0 0
0 1 1 0
0 0 1 1
Node firing order: z 0 c z 0 c ···
Stopping rule: parity check satisfied
Some initialization
Flat priors for codeword nodes: (c
i
)
Fixed likelihoods for
evidentiary nodes:
k
(x) = (x − z
k
)
π
λ δ
δ p(0c
i
1
, . . . ,c
i
s
) = (
Σ
c
i
k
)
Evidentiary nodes
(observations)
Codeword
component
Parity checks
Bayesian belief network
z
1
c
1
z
2
c
2
z
3
c
3
z
4
c
4
0 0 0
p(zc)
k
FIGURE 2. Application of Bayesian belief networks to lowdensity paritycheck codes. Softdecision
decoding of low density paritycheck codes can be based on Bayesian belief networks. Both the re
dundancies in codewords c
k
and the relationship between the codewords and the data z
k
can be rep
resented graphically. The datacodeword relationship is expressed through the probability densities
p(zc), which are assumed to be independent sampletosample. Redundancies in the codewords are
expressesed in a similar notation as p(0c
i
1
,…, c
i
s
) where the symbols c
i
k
, 1 ≤ k ≤ s, are involved in a par
ity check. In this manner, all depedancies are expressed graphically through conditional probability
densities as required for the formalism of Bayesian belief networks.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 111
The components of the vector p
k
express probabilities
for the values of the kth symbol, the permutation π
k
in
dicates the effect of a particular nonbinary coefficient
in the paritycheck equation, U
n
is a WalshHadamard
matrix, and the notation denotes the Hadamard
(component by component) product.
SpaceTime Extension LDPC
There are a variety of extensions of LDPC codes to
spacetime codes, which are introduced and explained
in the sidebar entitled “SpaceTime Codes.” For the ex
periments described below, only one type of extension
was considered.
Each spacetime channel transmits one of several
possible quadrature phaseshift keying (QPSK) wave
forms with slightly offset carrier frequencies. The dif
ferential frequencies are sufficiently large to effectively
decorrelate the transmitted waveforms over the length
of a codeword (1024 bits) even if the data sequences in
each channel are identical. These differential frequen
cies are also large compared to the expected Doppler
spreads and small compared to the signal bandwidth.
In the simplest example of such a code, the I and
Q components of a transmitter represent, respectively,
two different LDPC codewords. Each transmitter sends
the same complex baseband sequence (QPSK) shifted
in frequency. The transmitter outputs, viewed collec
tively as a vector at any instant, vary in time and thus
effectively probe the environment characterized by the
channel matrix. Since the transmitted vector varies sig
nificantly over the duration of a codeword, the coding
provides spatial diversity. Decoding occurs by forming
likelihood ratios based on channelmatrix estimates and
then using the iterative decoder described above. Note
that the channel matrix can change during the code
word, in which case channelmatrix estimates can vary
sample to sample.
The LDPC spacetime code just described exhibits
full spatial redundancy among all transmitters. Less
redundancy, and therefore higher data rates, can be
achieved by dividing the transmitters into subsets, each
of which is fully redundant yet different from any other
subset. For example, the spacetime code discussed lat
er, in the section on experiments, uses four transmitters.
The first two transmitters send two bits (redundant
in I and Q) of a symbol of an LDPC codeword over
GF(16). The remaining two transmitters send the oth
er two bits of the same symbol. Decoding is based on
likelihood functions built over GF(16) using estimates
of the channel matrices. Again, differential frequencies
among transmitters enable spatial diversity.
SpaceTime Turbo Code and Multichannel
Multiuser Detectors
While the theoretical performance is determined by
the channel phenomenology, practical MIMO perfor
mance requires the selection of a spacetime code and
an appropriate matched receiver. In this section we dis
cuss the spacetime turbo code used in this example.
We develop a maximumlikelihood formulation of a
multipleantenna multiuser receiver, and we discuss
suboptimal implementations of the receiver. We also
introduce minimummeansquarederror extensions of
the receiver, and we discuss the value and use of train
ing data.
SpaceTime Turbo Code
Turbo codes, introduced elsewhere [35], illustrate that
codes constructed with simple components, such as
with interleavers and convolutional encoders, combined
with an iterative decoding process can achieve near
Shannon capacity performance. The iterative decod
ing process, taking advantage of information exchange
among component decoders, provides a feasible way to
approach optimal performance. For each component
decoder, the best decoding algorithm is the maximum
a posteriori (MAP) algorithm or the BCJR algorithm
[36], which is derived from the MAP principle. Modi
fications of the MAP algorithm include logMAP and
maxlogMAP [37]. Recently, implementation of turbo
decoders has been carried out and high datarate decod
ing is possible [38].
A number of spacetime extensions of turbo coding
have been suggested [21, 22]. The approach used here,
which was introduced elsewhere [39], provides a 2bit/
sec/Hz link for a 4 × 4 MIMO system with independent
QPSK waveforms from each transmitter. A single data
stream is turbo encoded and the encoded data stream is
distributed redundantly amongst the transmitters. The
turbo encoder employs a rate1/3, 16state convolution
al encoder twice with two different 4096bit random
interleavers. The distribution of systematic bits is such
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
112 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
SPACETI ME CODES
S
iacirixi coois are used with
multiple transmitters to provide
spatial as well as temporal redun
dancy in the data received by an ar
ray of antennas. There are two basic
approaches to spacetime coding.
In the first approach, the transmit
ter can be informed of the propa
gation channel by the receiver and
thus adjust its coding accordingly.
This approach offers the largest in
formationtheoretic capacity but
can be difficult to accomplish in a
dynamic environment. The second
approach, which is taken here, uses
fixed codes of various rates that of
fer good performance on average
(over all channels). These codes
share transmitted power equally
among all spatial channels.
The number of different types of
spacetime codes is too large to pro
vide a useful overview here. Instead
we briefly describe two important
categories of spacetime codes that
are not treated in the text.
Block Orthogonal Codes
For data Z and channel matrix H,
consider a set of matrix symbols S
contained in S. The information
bits are encoded in matrices that
are constrained to lie in the class S.
This class is defined by the property
that SS
†
is proportional to the iden
tity matrix with a fixed (indepen
dent of S) proportionality constant.
The maximumlikelihood decision
for S is based on finding
S
S
Z HS
ZS H
∈
∈
−
,
S
S
argmin
tr
2
arg max Re ( )
† †
which involves a linear function in
the entries of S. For some simple
classes S, linearity of the likelihood
function decouples decisions on the
data symbols. For example, consid
er the Alamouti code [1].
S
s s
s s
:
−

(
'
`
J
J

¦








.
∗
∗
S S
1 2
2 1
The information symbols s
1
and
s
2
are sent redundantly over both
channels. The likelihood function
is linear in each s
k
, decoupling de
modulation decisions.
Another example of an orthogo
nal matrix code is
S S S
S
− − −
−
−
∗
[ ]
0 0
0
1 2 3 4
2 1 4 3
3 4 1
with
s s s s
s s s s
s s s ss
s s s s
2
4 3 2 1
−
.

(
'
'
'
'
'
'
'
`
J
J
J
J
J
J
J
J
SpaceTime Trellis Codes
Figure A provides an example of a
spacetime trellis code. A pair of bits
( ) I I
t t
1 2
, at time t enters a convolu
tional encoder with integer coeffi
cients a
k
p
and b
k
p
at the pth lag in
the kth channel. The input bits are
interpreted as the integers 0 or 1.
Computations occur modulo 4 and
result in an integer value between
0 and 3 for each channel. A fixed
mapping between these four inte
gers and the quadrature phaseshift
keying (QPSK) alphabet completes
the coding and modulation.
The trellis code is defined by
the coefficients { } a b
k
p
k
p
, . These are
often chosen under one of several
design criteria, also shown in the
figure. Each codeword is a matrix
symbol C. The probability of an er
ror in deciding between two such
that each systematic bit is sent twice on two different
transmitters. The parity bits are sent once, distributed
randomly amongst the transmitters. The difference in
weighting between the systematic and parity bits pro
vides an effective puncturing of the code. Because more
energy is dedicated to systematic bits, remodulation er
rors have a reduced effect on subtraction performance,
in principle improving the performance of the iterative
multiuser detection for a given bit error rate.
Multichannel Multiuser Detector
The multichannel multiuser detector (MCMUD) algo
rithm, discussed elsewhere [3, 39, 40], is a minimum
meansquarederror (MMSE) extension to an iterative
implementation of a maximumlikelihood multiplean
tenna receiver. The MCMUD algorithm employed for
this analysis iteratively combines a blind spacetimefre
quency adaptive beamformer with a multiuser detector.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 113
Example of spacetime trellis code (a
k
,
b
k
{0, 1, 2, 3})
Design criteria for spacetime trellis codes (4 ≤ rn
R
or rn
R
≥ 4)
transmitters
receivers
Notation: rank r matrix codeword C
n
T
n
R
kth transmitter
data: bit pair
(I
t
1
, I
t
2
) x
t
k
trellis coding codeword
symbol
QPSK
modulation
i
x
t
k
∈
Σ I
t
1
–p
a
p
k
+ Σ I
t
2
–q
b
q
k
mod4
p=0 q=0
v
1
v
2
p
e
≤
Π
[(C
1
– C
2
)(C
1
– C
2
]
†
–n
R
E
–rn
R
k=1
r
k
λ

4N
0
p
e
≤
1
4
–e
–n
R 4N
0
tr[(C
1
–C
2
)(C
1
–C
2
)
†
]

E
FIGURE A. Spacetime trellis codes introduce spatial as well as temporal
redundancy in the transmitted data. Code design often involves a pruned
search over a class of codes based on a simple figure of merit. For exam
ple, the minimum leastsquares distance between codewords (represent
ed by spacetime matrices C
k
) can be maximized. In the example shown,
an alphabet consisting of the integers modulo 4 is used for convolutional
encoding at each transmitter. The resulting output symbols are mapped to
a QPSK alphabet. The coefficients a
k
, b
k
determine the code. Note that the
spectral efficiency is 2 bits/sec/Hz.
symbols can be bounded by (Bhat
tacharyya bound)
p e
e
E
N
≤ .
− − −
4 1 2 1 2
0
( ) ( )
† †
C C H H C C
The approximation H H
†
≈ n I
R n
T
motivates one of the design crite
ria shown in the figure. Integrating
over H motivates the other.
In both cases r denotes the rank
of the matrix difference C
1
– C
2
.
Constrained searches over the code
coefficients are commonly used to
find codes with the smallest pos
sible error between closest code
words under either criterion. When
4 ≥ rn
R
, it is important to ensure
that the rank of the matrix differ
ence is not too small. When 4 < rn
R
,
maximizing the Euclidean distance
between the two codewords C
k
be
comes important.
Reference
1. S.M. Alamouti, “A Simple Transmit Di
versity Technique for Wireless Com
munications,” IEEE J. Sel. Areas Com
mun., 16 (8), 1998, pp. 1451–1458.
We present here the results of the maximum likeli
hood (ML) formulation of MCMUD, employing a
quasistatic narrowband MIMOchannel model. The
number of receive antennas n
R
by number of samples, n
s
data matrix, Z ∈
×
n n
R s
, is given by
Z HT N + , (44)
where the channel matrix H∈
×
n n
R T
contains the
complex attenuation between each transmit antenna
and receive antenna; T ∈
×
n n
T s
is the transmitted se
quence; and N∈
×
n n
R s
is additive Gaussian interfer
ence plus noise. The probability density for a multivari
ate signalinthemean model is given by
p
e
n n n
s R s
( )
( ) ( )
†
Z R H T
R
Z HT R Z HT
 , ,
 
− − −
¦ ¦
−
tr
1
π
,,
(41)
where R indicates the spatial covariance matrix of the
interference plus noise,   indicates the determinant
of a matrix, † indicates the Hermitian conjugate, and tr
indicates the trace of a matrix. Maximizing the prob
ability density with respect to H is equivalent to mini
mizing the tr{ } in Equation 41,
tr ( )( )
†
Z HT Z HT R − − ,
−
¦ ¦
1
(42)
which is satisfied by
ˆ
( )
† †
H ZT TT ,
−1
(43)
assuming TT
†
is not rank deficient. Substituting
ˆ
H,
p
e
n n n
s R s
→
 
,
−
⊥ −
¦ ¦
tr ZP Z R
T
R
† 1
π (44)
where the matrix P T TT T
T
≡
− † †
( )
1
projects onto the
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
114 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
row space spanned by T, and P I P
T T
⊥
−
n
s
projects
onto the orthogonal complement of the row space of T.
Maximizing with respect to an internal parameter of R
gives
tr tr { } { }
†
R ZP Z R R R R
− ⊥ − −
− ,
1 1 1
0
T s
n
(45)
where
R indicates the derivative of R with respect to
some internal parameter. This relationship is satisfied
when
ˆ
†
R
ZP Z
T
,
⊥
n
s
(46)
assuming that R is not rank deficient. Using these re
sults, the ML statistic for estimating T is given by
max ( )
†
R H
T
Z R H T ZP Z
,
−
⊥ −
 , ,

(
'
`
J
J
  p
e
n
n n
s
n
s R
s
π
. .
(47)
The determinant of ZP Z
T
⊥ †
is minimized to demodu
late the signals for all transmitters jointly.
Although it is theoretically possible to use the statis
tic ZP Z
T
⊥ †
directly for demodulation, an iterative ap
proach is much more practical. We define T T T
A B
≡ ( )
† † †
to be a partitioned form of T, where the n
A
× n
s
ma
trix T
A
contains the signals associated with a particular
subset of n
A
transmit antennas and the (n
T
– n
A
) × n
s
matrix T
B
contains the signals associated with all other
transmit antennas. By factoring P X X
T
B
⊥
†
, the rows
of X form an orthonormal basis for the complement
of the row space of T
B
such that XX I
†
, where the
symmetric identity matrix has a dimension of n
s
minus
the number of rows in T
B
. By defining Z ZX
X
≡
†
and
T T X
X A
≡
†
, we can show that
ZP Z Z P Z
T X T X
X
⊥ ⊥
.
† †
(48)
The determinant can be factored into terms with and
without reference to T
A
,
Z P Z Z Z I P P
X T X X X T Z
X X X
⊥
− .
† †
n
s
(49)
Because the first term is free of T
A
, demodulation is per
formed by minimizing the second term. This form sug
gests an iterative approach, where the signal associated
with each transmitter, in turn, is considered to be user
A and is demodulated by minimizing I P P
T Z
X X
n
s
− .
If T
A
is a row vector, such that n
A
= 1, then the sec
ond term can be simplified and interpreted in terms of
a beamformer
I P P T T T P T
w Z T
T Z X X X Z X
A X X
X X X
n
s
n
− − ,
−
−
1
1
1
( )
† †
† †
ss
s
n
,
− ,
⊥
1
w ZP T
A T A
B
† †
(50)
where
w R H
R Z Z ZP Z
H Z T
A X A
X X X T
A X
B
,
≡ ,
≡
−
⊥
1
1 1
ˆ ˆ
ˆ
ˆ
† †
n n
s s
XX X X
T A A T A
T T
ZP T T P T
B B
† †
† †
( )
( )
−
⊥ ⊥ −
.
1
1
(51)
The n
R
× 1 vector w
A
contains the receive beamforming
weights,
ˆ
R
X
is the interferencemitigated signalplus
noise covariance matrix estimate, and
A H
ˆ
is the chan
nel estimate associated with T
A
. It is worth noting that
the form for
A H
ˆ
is simply the column of
ˆ
H, given in
Equation 43, associated with T
A
.
ˆ
( )
† †
† †
†
†
H ZT TT
Z T T
T T T T
T T
A B
A A A B
B A
−

(
'
`
J
J
−
1
1
††
†
† †
T T
Z T T
M M
M
B B
A B

(
'
'
'
`
J
J
J
J

(
'
`
J
J
−
, ,
≡
1
11 1 2
1,, ,

(
'
'
'
`
J
J
J
J

(
'
`
J
J
, ,
⋅
−
2 2 2
11 1 2
†
† †
(
M
Z T T
M M
A B
MM M
M M M M M
2 2
1
1 2
1
2 2
1
1 2 11 1 2 2 2
,
−
,
−
,
−
, , , ,
−
− −
†
†
)
(
11
1 2
1
M
,
−

(
'
`
J
J
.
†
)
(52)
By focusing on the first column and substituting in for
M
1,1
, M
1,2
, and M
2,2
, we can find A H
ˆ
.
A A A A A B B B B A
B
H ZT T T T T T T T T
ZT
ˆ
( [ ] )
[
† † † † †
†
−
−
− − 1 1
TT T T T
T T T T T T T T
B B B A
A A A B B B B A
† †
† † † †
]
( [ ] )
−
− −
× −
1
1 11
1
,
⊥ ⊥ −
ZP T T P T
T A A T A
B B
† †
( )
(53)
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 115
which is the same form found in Equation 51.
Demodulation is performed by maximizing the
magnitude of the inner product of the beamformer out
put w Z
A
†
and the interferencemitigated reference sig
nal T P
A T
B
⊥
.
Suboptimal Implementation
A variety of suboptimal but computationally more ef
ficient variants are possible. In general, these approxi
mations become increasingly valid as the number of
samples in the block increases.
The first computational simplification is found by
noting that the normalization term of the channel esti
mate in Equation 51 can be approximated by
A T A A T A
T A A A
H ZP T T P T
ZP T T T
B B
B
ˆ
( )
( )
† †
† †
⊥ ⊥ −
⊥ −
1
1//
− / − / −
× −
×
2
1 2 1 2 1
( [ ] [ ] )
(
† † †
I T T T P T T T
T
A A A T A A A
B
AA A
T A A A
T
ZP T T T
B
†
† †
)
( )
− /
⊥ −
≈ .
1 2
1
(54)
(We did not assume that T
A
is a row vector in the previ
ous discussion.)
The second approximation reduces the computation
cost of the projection operator. The operator that proj
ects on the orthogonal complement of the row space of
M is given by
P I M MM M
M
⊥ −
− .
† †
( )
1
(55)
This operator can be approximated by
P I M M M M
M
⊥ −
≈ −
]
]
,
∏
m
m m m m
† †
( )
1
(56)
where m indicates the mth row in the matrix. By re
peating the application of this approximate projection
operator, we can reduce the approximation error at the
expense of additional computational complexity.
MMSE Extension
Because of the effects of delay and Dopplerfrequency
spread, the model given in Equation 40 for the received
signal is incomplete for many environments, adversely
affecting the performance of the spatialbeamformer
interpretation of the ML demodulator. Because turbo
codes require relatively long block lengths to be effec
tive, they are particularly sensitive to Doppler offsets.
Extending the beamformer to include delay and Dop
pler corrects this deficiency. With this approach, the
spatialbeamformer interpretation presented in Equa
tion 50 is formally the same, but all projectors are ex
tended to include delay and Doppler spread. The data
matrix is replaced with
Z Z Z
Z
X X
X
STF
n
t f t f
t f
≡ , ,
,
[ ( ) ( )
(
† †
†
δ δ δ δ
δ δ
δ
1 1 1 2
1
ff t f
t f
n n
) ( )]
† †
Z
X
δ δ
δ δ
, ,
(57)
which is a ( ) n n n n
R f t s
⋅ ⋅ ×
δ δ
matrix that includes pos
sible signal distortions. The new channel estimate has
dimension ( ) n n n n
R f t T
⋅ ⋅ ×
δ δ
, but T remains the same.
The MMSE beamformer is given by
w w Z T
Z Z Z T
X STF STF STF
STF STF STF
argmin −
−
†
†
( )
2
1
XX
†
.
(58)
Figure 3 shows a diagram for this demodulator (MC
MUD).
S
p
a
c
e

t
i
m
e
d
e
m
u
l
t
i
p
l
e
x
e
r
S
p
a
c
e

t
i
m
e

f
r
e
q
u
e
n
c
y
a
d
a
p
t
i
v
e
b
e
a
m
f
o
r
m
e
r
T
e
m
p
o
r
a
l
s
u
b
t
r
a
c
t
i
o
n
T
u
r
b
o
d
e
c
o
d
e
r
T
u
r
b
o
e
n
c
o
d
e
r
S
p
a
c
e

t
i
m
e
m
u
l
t
i
p
l
e
x
e
r
Block for each transmitter
Info
bits
Channel
estimation
FIGURE 3. Diagram of a multichannel multiuser detector
(MCMUD) spacetime turbocode receiver. The receiver
iteratively estimates the channel and demodulates the sig
nal. The spacetime frequencyadaptive beamformer com
ponent compensates for spatial, delay, and frequencyoff
set correlations. By iteratively decoding the signal, previous
signal estimates can be used to temporally remove contri
butions from other transmitters, which is a form of multius
er detection.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
116 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
Training Data
In principle, there is no need for training data, because
the channel and information can be estimated jointly.
Furthermore, the use of training data competes directly
with information bits. For reasonably stationary chan
nels, the estimate for the previous frame can be em
ployed as an initial estimate for the demodulator. How
ever, for more quickly moving channels some training
data is useful. Here, a small amount of training data
is introduced within a frame (20%). This provides an
initial channel estimate for the spacetimefrequency
adaptive beamformer.
In the experiment, knowledge of the encoded sig
nal is used to provide that training data. Because the
number of training samples is relatively small, it is use
ful to use a small number of temporal and frequency
taps during the first iteration. Larger dimension space
timefrequency processing is possible by using estimates
of the data.
Phenomenological Experiment
This section presents channelcomplexity and channel
stationarity experimental results for MIMO systems.
We introduce the experiments and then discuss chan
nel mean attenuation and channel complexity. We then
discuss the variation of MIMO channels as a function
of time and as a function of frequency.
Experimental System
The employed experimental system is a slightly modi
fied version of the system used previously at Lincoln
Laboratory [3, 41]. The transmit array consists of up to
eight arbitrary waveform transmitters. The transmitters
can support up to a 2MHz bandwidth. These trans
mitters can be used independently, as two groups of
four coherent transmitters, or as a single coherent group
of eight transmitters. The transmit systems can be de
ployed in the laboratory or in vehicles. When operat
ing coherently as a multiantenna transmit system, the
individual transmitters can send independent sequences
by using a common local oscillator. Synchronization
between transmitters and receiver and transmitter geo
location is provided by GPS receivers in the transmitters
and receivers.
The Lincoln Laboratory array receiver system is a
highperformance sixteenchannel receiver system that
can operate over a range of 20 MHz to 2 GHz, sup
porting a bandwidth up to 8 MHz. The receiver can
be deployed in the laboratory or in a stationary “bread
truck.”
MIT Campus Experiment
The experiments were performed during July and Au
gust 2002 on and near the MIT campus in Cambridge,
Massachusetts. These outdoor experiments were per
formed in a frequency allocation near the PCS band
(1.79 GHz). The transmitters periodically emitted 1.7
sec bursts containing a combination of channelprobing
and spacetimecoding waveforms. A variety of coding
and interference regimes were explored for both mov
ing and stationary transmitters. The spacetimecoding
results are discussed later in the article [39, 40]. Chan
nelprobing sequences using both four and eight trans
mitters were employed.
The receive antenna array was placed on top of a
tall onestory building (at Brookline Street and Henry
Street), surrounded by two and threestory buildings.
The transmit array was located on the top of a vehicle
within two kilometers of the receive array. Different
four or eightantenna subsets of the sixteenchannel re
ceiver were used to improve statistical significance. The
receive array had a total aperture of less than 8 m, ar
ranged as three subapertures of less than 1.5 m each.
The channelprobing sequence supported a band
width of 1.3 MHz with a length of 1.7 msec repeated
0 500 1000 1500 2000
Link range (m)
a
2
(
d
B
)
–30
–20
–10
0
10
20
FIGURE 4. Scatter plot of the peaknormalized mean
squared singleinput singleoutput (SISO) link attenuation
a
2
versus link range for the outdoor environment near the
PCS frequency allocation. The error bars indicate a range
of plus or minus one standard deviation of the estimates at
a given site.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 117
ten times. All four or eight transmitters emitted nearly
orthogonal signals simultaneously.
Attenuation
Figure 4 displays the peaknormalized meansquared
SISO attenuation averaged over transmit and receive
antenna pairs for a given transmit site for the outdoor
environment. The uncertainty in the estimate is evalu
ated by using a bootstrap technique.
Channel Complexity
We present channel complexity by using three differ
ent approaches: variation in a
2
estimates, eigenvalue cu
mulative distribution functions (CDF), and α estimate
CDFs. Table 1 is a list of transmit sites used for these
results. The table includes the distance (range) between
transmitter and receiver, the velocity of the transmitter,
the number of transmit antennas, and the estimated α
for the transmit site. Uncertainty in α is determined by
using the bootstrap technique [42]. The CDF values re
ported here are evaluated over appropriate entries from
Table 1. The systematic uncertainty in the estimation
of α caused by estimation bias, given the model, is less
than 0.02.
Figure 5 displays CDFs of a n n
T R
2
/ tr{ } ( )
†
HH
estimates normalized by mean a
2
for each transmit site.
CDFs are displayed for narrowband SISO, 4 × 4, and
8 × 8 MIMO systems. Because of the spatial diversi
ty, the variation in mean antennapair received power
decreases dramatically as the number of antenna pairs
increases, as we would expect. This reduction in varia
tion demonstrates one of the most important statistical
effects that MIMO links exploit to improve commu
nication link robustness. For example, if we wanted to
Table 1. List of Transmit Sites
Site Location Range Velocity Number of α
(m) (m/sec) antennas
1 Henry and Hasting 150 0.0 8 0.79 ± 0.01
2 Brookline and Erie 520 0.0 8 0.80 ± 0.01
3 Boston University (BU) 430 0.0 8 0.78 ± 0.01
4 BU at Storrow Drive 420 0.0 4 0.72 ± 0.01
5 Glenwood and Pearl 250 10.0 4 0.85 ± 0.01
6 Parking lot 20 0.1 4 0.78 ± 0.02
7 Waverly and Chestnut 270 0.2 4 0.67 ± 0.02
8 Vassar and Amherst 470 0.7 4 0.68 ± 0.02
9 Chestnut and Brookline 140 0.1 4 0.70 ± 0.02
10 Harvard Bridge 1560 11.6 4 0.69 ± 0.02
11 BU Bridge 270 2.7 4 0.83 ± 0.04
12 Vassar and Mass Ave 1070 7.6 4 0.59 ± 0.01
13 Peters and Putnam 240 9.1 4 0.87 ± 0.05
14 Glenwood and Pearl 250 5.2 4 0.76 ± 0.02
15 Brookline and Pacific 780 7.2 4 0.86 ± 0.03
16 Pearl and Erie 550 0.1 4 0.71 ± 0.04
17 Storrow Drive and BU Bridge 410 9.2 4 0.85 ± 0.03
18 Glenwood and Magazine 370 0.0 4 0.78 ± 0.02
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
118 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
operate with a probability of 0.9 to close the link, we
would have to operate the SISO link with an excess
SISO SNR (a
2
P
o
) margin of over 15 dB. The MIMO
systems received the added benefit of array gain, which
is not accounted for in the figure.
Figures 6 and 7 present CDFs of eigenvalues for 4 × 4
and 8 × 8 meansquaredchannelmatrixelementnor
malized narrowband channel matrices, eig{ }
†
HH . The
CDFs are evaluated over all site lists. Some care must be
taken in interpreting these figures because eigenvalues
are not independent. Nonetheless, the steepness of the
CDFs is remarkable. We might interpret this to indicate
that optimized spacetime codes should operate with a
relatively high probability of success.
Figure 8 shows the CDFs for α estimates. The mean
values of α for each environment are 0.76 for 4 × 4 sites
and 0.79 for 8 × 8 sites, where the form x ± y indicates
the estimated value x with statistical uncertainty y es
timated by using a bootstrap uncertainty estimation
technique. While we might expect smaller variation in
the 8 × 8 systems because of the much larger number of
paths, this effect may have been exaggerated in Figure 8
because of the limited number of 8 × 8 sites available in
the experiment.
Channel Stationarity
Figure 9 displays the temporal variation of eigenvalues
of HH
†
for stationary and moving transmitters. In this
figure the normalization is fixed, allowing for overall
shifts in attenuation. As we would expect, the eigenval
ues of the moving transmitter vary significantly more
than those of the stationary environment. However, the
eigenvalues of the stationary transmitter do vary some
what. While the transmitters and receivers are physi
cally stationary, the environment does move. This effect
is particularly noticeable near busy roads. Furthermore,
while the multiple antennas are driven with the same lo
cal oscillator, given the commercial grade transmitters,
there are always some small relativefrequency offsets.
The example variation is given for transmit sites 7 and
14 from Table 1.
0 –5 –15 –10 5
1.0
0.8
0.6
0.2
0.4
0
a
2
(dB)
C
D
F
o
f
a
2
4 × 4
8 × 8
SISO
FIGURE 5. Cumulative distribution function (CDF) of chan
nel a
2
estimates, normalized by the mean a
2
for each site, for
SISO, 4 × 4, and 8 × 8 MIMO systems.
10 0 –40 –30 –20 –10 20
1.0
0.8
0.6
0.2
0.4
0
Eigenvalue
(dB)
C
D
F
o
f
c
h
a
n
n
e
l
e
i
g
e
n
v
a
l
u
e
s
10 0 –40 –30 –20 –10 20
1.0
0.8
0.6
0.2
0.4
0
Eigenvalue
(dB)
C
D
F
o
f
c
h
a
n
n
e
l
e
i
g
e
n
v
a
l
u
e
s
FIGURE 6. CDF of narrowband channel eigenvalue distribu
tions for 4 × 4 MIMO systems.
FIGURE 7. CDF of narrowband channel eigenvalue distribu
tions for 8 × 8 MIMO systems.
0.9 0.8 0.5 0.6 0.7 1.0
1.0
0.8
0.6
0.2
0.4
0
α
α
C
D
F
o
f
4 × 4
8 × 8
FIGURE 8. CDF of α estimates for 4 × 4 and 8 × 8 MIMO sys
tems.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 119
While the movingtransmitter eigenvalues fluctuate
more than those of the stationary transmitter, the values
are remarkably stable in time. Conversely, an example
of the time variation of the powerweighted mean cos
2
θ
metric (from Equation 36), displayed in Figure 10,
varies significantly for the moving transmitter within
10 msec. This variation indicates that the eigenvector
structure varies significantly, while the distribution of
eigenvalues tends to be more stable. In the example, the
stationary transmitter is located at site 7, and the mov
ing transmitter is located at site 14. Over the same pe
riod the stationary transmitter is relatively stable. Fig
ures 11 and 12 display CDFs for stationary and moving
transmitters. The significant variation of the moving
transmitter is an indication that implementing an in
formed transmitter MIMO system would be very chal
lenging for the moving transmitter, but might be viable
for some stationary MIMO systems.
FrequencySelective Fading
Figure 13 gives an example of the frequency variation of
the powerweighted mean cos
2
θ. The variation is indi
cated by using the metric presented in Equation 36. In
the example, the stationary transmitter is located at site
0.05 0.10 0 0.15
20
10
0
–10
–20
Time (sec)
(
d
B
)
λ
0.05 0.10 0 0.15
20
10
0
–10
–20
Time (sec)
(
d
B
)
λ
(a)
(b)
FIGURE 9. Eigenvalues (λ) of HH
†
as a function of time for
(a) stationary and (b) moving transmitters. The same over
all attenuation, estimated at t = 0, is used for all time sam
ples.
0 0.02 0.04 0.06
0.2
0.4
0.6
0.8
1.0
Time (sec)
Stationary
Moving
γ
{
H
(
t
0
)
,
H
(
t
)
}
FIGURE 10. Example time variation of powerweighted mean
cos
2
θ, γ { ( ), ( )}
0
H H t t , for stationary and moving 4 × 4 MIMO
systems.
0 0.02 0.04 0.06
0.2
0.4
0.6
0.8
1.0
Time (sec)
0.2
0.1
γ
{
H
(
t
0
)
,
H
(
t
)
}
0 0.02 0.04 0.06
0.2
0.4
0.6
0.8
1.0
Time (sec)
0.9
0.7
0.5
0.3
0.1
γ
{
H
(
t
0
)
,
H
(
t
)
}
FIGURE 11. CDF of time variation of powerweighted mean
cos
2
θ, γ { ( ), ( )}
0
H H t t , for a stationary 4 × 4 MIMO system. The
graph displays contours of CDF probabilities of 0.1, 0.2, 0.3,
0.4, 0.5, 0.6, 0.7, 0.8, and 0.9. Because there is little variation,
all curves are compressed near a γ value of 1.0
FIGURE 12. CDF of time variation of powerweighted mean
cos
2
θ, γ { ( ), ( )}
0
H H t t , for a moving 4 × 4 MIMO system. Con
tours of CDF probabilities of 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8,
and 0.9 are displayed.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
120 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
7. Relatively small frequency offsets induce significant
changes in γ { ( ) ( )} H H f f
0
, . Figure 14 shows the CDF
of the frequencyselective channel variation. This sen
sitivity indicates that there is significant resolved delay
spread and that, to safely operate with the narrowband
assumption, bandwidths less than 100 kHz should be
employed. We note that delay spread, and the result
ing frequencyselective fading, are both a function of
environment and link length. Consequently, some care
must be taken in interpreting this result.
SpaceTime LowDensity
ParityCheckCode Experiments
A lowdensity paritycheck code over GF(16) provides
the basis of the example of experimental and simulated
results shown in Figure 15. The code used is half rate
with length 1024. The MIMO wireless link is realized
with four cohered transmitters located on a stationary
van several hundred meters away from an array of re
ceivers situated on a onestory building. The environ
ment consists predominantly of two and threestory
residential buildings and some commercial buildings of
similar heights in an urban setting. Propagation delay
spreads are typically several microseconds and Dop
pler spreads are at most a few hundred hertz. There is
typically no identifiable lineofsight component in
the propagation. The signal has a pulseshaped QPSK
modulation and bandwidth of about 100 kHz. Coding
provides a spectral efficiency of 2 bits/sec/Hz.
The receiver consists of sixteen channels fed by low
gain elements with wide azimuth beamshapes. The
elements are oriented in various directions, not neces
sarily pointing at the sources. For the example below,
four element subarrays are chosen at random to provide
multichannel receivers. In other words, C(16, 4) 4 × 4
0.2
0.4
0.6
0.8
1.0
Frequency (kHz)
γ
{
H
(
f
0
)
,
H
(
f
)
}
–600 –400 –200 0 200 400 600
0.2
0.4
0.6
0.8
1.0
Frequency (kHz)
γ
{
H
(
f
0
)
,
H
(
f
)
}
–600 –400 –200 0 200 400 600
0.9
0.7
0.5
0.3
0.1
FIGURE 13. Example of frequencyselective variation of the
powerweighted mean cos
2
θ, γ { ( ), ( )}
0
H H f f .
FIGURE 14. CDF of frequencyselective variation of the
powerweighted mean cos
2
θ, γ { ( ), ( )}
0
H H f f . The graph dis
plays contours of CDF probabilities of 0.1, 0.2, 0.3, 0.4, 0.5,
0.6, 0.7, 0.8, and 0.9.
0 2 4 6 8 10 12
0
0.05
0.10
0.15
0.20
0.25
Excess E
b
/N
0
(dB)
B
i
t

e
r
r
o
r
p
r
o
b
a
b
i
l
i
t
y
Measured
Simulated
• Spectral efficiency
2 bits/sec/Hz
FIGURE 15. Measured and simulated results in bit error
rate probability for a spacetime lowdensity paritycheck
(LDPC) code at a spectral efficiency of 2 bits/sec/Hz. Bit
error rates are evaluated for an ensemble of 4 × 4 MIMO
systems. The estimated channel matrices are used in the
simulation to model propagation. Each estimated channel
matrix suppports a theoretical capacity that can be in ex
cess of 2 bits/sec/Hz. The matrix is scaled until it supports
a capacity of exactly 2 bits/sec/Hz. The resulting scale fac
tor is used to evaluate the excess (beyond Shannon) E
b
/N
0
associated with the (unscaled) channel matrix. Agreement
between measured and simulated results are good to about
1 dB. About 4 to 5 dB excess E
b
/N
0
is required to reliably
complete the link.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 121
MIMO links can be evaluated. The channel transfer
matrices have a random structure that varies from sub
array to subarray.
Figure 15 shows symbolerror probability as a func
tion of excess E
b
/N
0
, which is related to the excess spec
tral efficiency (beyond 2 bits/sec/Hz) predicted by a
capacity bound, given the measured channel transfer
matrix. For this example, in a comparatively station
ary environment, the channel transfer matrices are used
both for the simulated results and for the computation
of excess E
b
/N
0
. As the figure shows, about 4.5 dB ex
cess E
b
/N
0
is required to complete the link at 2 bits/sec/
Hz. Simulations agree to within about 1 dB.
SpaceTime TurboCode Experiments
In this section we present the experimental performance
of a spacetime turbo code. We begin by discussing the
experimental parameters, and then we summarize the
performance of the MIMO system with stationary
transmitter and receiver in a dynamic environment. Ad
ditionally, for an even more complicated environment,
we describe performance results with a mobile transmit
ter and multiple strong interferers.
Experimental Parameters
Outdoor experiments were performed in a frequency
allocation near the PCS band, using a sixteenchannel
receiver. A variety of coding and interference regimes
were explored for both moving and stationary transmit
ters. Channelprobing sequences and four and eight
transmitter spacetime codes were transmitted. This
section reports on the outdoor performance results of
spacetime turbo codes for 4 × 4 MIMO systems. The
outdoor experiments were performed during July and
August 2002 on and near the MIT campus. The receive
antenna array was placed on top of a onestory building
(at Brookline Street and Henry Street) surrounded by
two and threestory buildings.
For the examples discussed in this article, quadra
turephaseshiftkey (QPSK) signals were transmitted
on four antennas at 123 × 10
3
chips per second for a
total data rate of 246 kb/sec, using the spacetime code
discussed earlier. A 160kHz spectral limit was enforced
by using a rootraisedcosine pulse shaping. Total trans
mit power was approximately 100 mW, radiated from
0dBi antennas. We discuss two examples with differ
ent transmit locations. In both examples the link does
not have line of sight. Different fourantenna subsets of
the sixteenchannel receiver were used to improve statis
tical significance.
Example 1. In this example, the transmitter was lo
cated in the parking lot at Boston University (Universi
ty Road and Storrow Drive) with about a halfkilometer
separation between the transmitter and receiver. Figure
16 shows the geometry of the experiment. Traffic on
Storrow Drive is typically heavy and the posted speed
limit of 45 mph is generally misinterpreted as the mini
mum allowed speed. While the transmitter is stationary,
the environment is nonstationary because of the traffic.
Example 2. The transmitter was moving at 10 m/sec
at a range of 500 m. Figure 17 shows the geometry of
the experiment. To simulate the effects of local oscilla
tor errors, we introduced artificial frequency offsets at
the transmitters. These errors were within ±80 Hz.
Two wideband jammers were transmitting at a range
of 100 m. Each jammer was received at a jammerto
noise ratio (JNR) of approximately 25 dB. Figure 18
shows the eigenvalues of the noisenormalized interfer
enceplusnoise spatial covariance matrix. The “noise”
eigenvalues of the jammer spatial eigenvalue distribution
Boston
University
Receive
array
Cambridge
MIT
Transmit
array
FIGURE 16. Example 1: map of MIMO communication exper
iment near the MIT campus, including the locations of the
transmitter and receiver.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
122 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
are slightly higher than we would naively expect, given
the 0dB noise normalization. This behavior is probably
an indication of either delay spread or nonstationarity in
the received jammer signal. Either of these explanations
presents additional challenges to the receiver.
Both the delay and the Doppler spread affect the
design and performance of the receiver. Here a space
timefrequency adaptive processor is employed. The
number of delay and frequency taps in the adaptive pro
cessor depends upon the phenomenology. Delay spread
was found to be less than ±4 µsec. For the stationary
environment, in quiet regions (no nearby traffic), no
Doppler spread was detected. For the stationary trans
mitter near heavy traffic in experimental example 1, the
Doppler spread was found to be within ±150 Hz. For
the moving transmitter in experimental example 2, the
Doppler spread was found to be within ±180 Hz.
Experimental Example 1
Bit error rates for various detector alternatives are re
ported as a function of mean singleinput singleoutput
(SISO) SNR, a
2
P
o
. Here, a
2
is the meansquared link
attenuation. Figure 19 displays the results for four de
tection variations:
1. Trainingdatabased adaptive spatial beamform
ing (three turbo iterations).
2. Coarse trainingdatabased spacefrequency beam
forming (one turbo iteration; Doppler taps: {–1, 0, 1}).
3. Spacetimefrequency beamforming employing
decisiondirected channel estimation without multiuser
detection (three turbo iterations; Doppler taps: {–4/3,
–2/3, 0, 2/3, 4/3}; temporal taps: {–1/2, 0, 1/2}).
4. MCMUD with spacetimefrequency beamform
ing (three turbo iterations; Doppler taps: {–4/3, –2/3,
0, 2/3, 4/3}; temporal taps: {–1/2, 0, 1/2}), and where
Doppler taps are represented in resolution cells (60
Hz).
Performance improves with receiver complexity; the
algorithm, however, must bootstrap up in complexity
iteratively. Starting with the highest complexity on the
first iteration increases the probability of converging to
the wrong solution. Because the channel contains sig
nificant Doppler spread, the spatial beamformer per
forms poorly. With the relatively long block lengths of
the turbo code, Doppler beamforming is required in
this environment. We note that experimental perfor
mance is essentially the same as was found in simula
tions. Furthermore, the experimental performance is
FIGURE 17. Example 2: map of MIMO communication ex
periment near the MIT campus, including the locations of
the transmitter, receiver, and jammers at a fixed jammerto
noise ratio (JNR).
Receive
array
Transmitter
25 mph
Jammers
25 dB JNR
100 m
1 2 3 4
0
10
20
30
40
Eigenvalue number
R
e
l
a
t
i
v
e
p
o
w
e
r
(
d
B
)
FIGURE 18. Eigenvalue distribution of the noisenormalized
interferenceplusnoise spatial covariance matrix.
10
–1
10
–2
10
–3
10
–4
a
2
P
o
(dB)
B
i
t
e
r
r
o
r
r
a
t
e
3.0 3.5 4.0 4.5 5.0 5.5 6.0
Spatial beamforming
Trainingdatabased
SFAP
STFAP
MCMUD
FIGURE 19. Bit error rate of 4 × 4, 2bit/sec/Hz spacetime
turbo code as a function of mean SISO signaltonoise ra
tio (SNR) (a
2
P
o
) for a Boston University transmit location,
using adaptive spatial beamforming, coarse trainingdata
based spacefrequency adaptive beamforming (SFAP),
spacetimefrequency adaptive beamforming (STFAP) em
ploying decisiondirected channel estimation, and MCMUD
with spacetimefrequency adaptive beamforming.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 123
similar to the simulated performance of the best space
time codes.
Experimental Example 2
The experimental data includes the effects of two high
power wideband jammers, a moving transmitter, and
local oscillator errors. Experimental performance of this
spacetime turbo code for a stationary transmitter in the
absence of interference is discussed elsewhere [39].
Figure 20 shows the bit error rate of the spacetime
turbo code using the MCMUD receiver. The bit error
rate is displayed in terms of the mean SISO signalto
interferenceplusnoise ratio (SINR). This is the aver
age SINR at a given receive antenna, assuming that
all power of the transmit array is transmitted from a
single transmit antenna. We note that this experimen
tal system in this difficult environment operates at an
SINR that is 25 dB better than the informationtheo
retic SISO bound, and operates probably at least 35 dB
better than a practical SISO system. Furthermore, there
is only approximately a 3dB loss in a
2
P
o
performance
compared to the performance in an environment with
out jammers. The effectiveness of the receiver is due in
part to its ability to compensate for delay and frequency
spread. The MCMUD employs a spacetimefrequen
cy adaptive processor that uses a fourantenna receiver
with temporal and frequency taps that cover a range of
±4 microseconds and ±200 Hz.
Summary
In this article we addressed informationtheoretic, phe
nomenological, coding, and receiver issues for MIMO
communication. Performance bounds assuming either
an informed transmitter or an uninformed transmitter
were presented for flatfading, frequencyselective, and
jammed environments. A channel phenomenology pa
rameterization was introduced. Experimental phenom
enological results were reported, the results indicating
that the observed channels can be typically character
ized by high degrees of complexity. Furthermore, for
environments with transmitters on moving vehicles,
the channel varies significantly on a time scale less than
10 msec. Two spacetime coding techniques were in
troduced, one based on LDPC and the other on turbo
codes. Experimental demodulation performance results
were presented for a variety of environments, including
those with wideband jammers. In the presence of the
jammer, the MIMO system (using the MCMUD re
ceiver) operated dramatically better than SISO systems.
Acknowledgments
The authors would like to thank Peter Wu of Lincoln
Laboratory for his help developing the spacetime turbo
code, and Naveen Sunkavally of MIT and Nick Chang
of the University of Michigan for their help with the ex
periment. The authors would also like to thank the ex
cellent Lincoln Laboratory staff involved in the MIMO
experiment, in particular Sean Tobin, Jeff Nowak, Lee
Duter, John Mann, Bob Downing, Peter Priestner, Bob
Devine, Tony Tavilla, and Andy McKellips. We also
thank Ali Yegulalp of Lincoln Laboratory and Vahid
Tarokh of Harvard University for their thoughtful com
ments, and Dorothy Ryan of Lincoln Laboratory for
her helpful comments. Finally, the authors would like
to thank the MIT New Technology Initiative Commit
tee for their support.
10
–1
10
–2
10
–3
10
–4
SISO SINR (dB)
B
i
t
e
r
r
o
r
r
a
t
e
–21 –20 –19 –18 –17
FIGURE 20. Bit error rate of 4 × 4, 2bit/sec/Hz spacetime
turbo code using the MCMUD receivver with spacetime
frequency beamforming as a function of mean SISO signal
tointerferenceplusnoise ratio (SINR).
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
124 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
R EFER ENCES
1. W.C. Jakes, Microwave Mobile Communications (Wiley, New
York, 1974).
2. R.A. Monzingo and T.W. Miller, Introduction to Adaptive
Arrays (Wiley, New York, 1980).
3. K.W. Forsythe, D.W. Bliss, and C.M. Keller, “Multichan
nel Adaptive Beamforming and Interference Mitigation in
Multiuser CDMA Systems,” ThirtyThird Asilomar Conf. on
Signals, Systems & Computers 1, Pacific Grove, Calif., 24–27
Oct. 1999, pp. 506–510.
4. A. Wittneben, “Basestation Modulation Diversity for Digital
SIMULCAST,” Proc. IEEE Vehicular Technology Conf., St.
Louis, Mo., 19–22 May 1991, pp. 848–853.
5. V. Weerackody, “Diversity for DirectSequence Spread Spec
trum Using Multiple Transmit Antennas,” Proc. IEEE Int.
Communications Conf. 3, Geneva, 23–26 May, 1993, pp.
1775–1779.
6. G.J. Foschini, “Layered SpaceTime Architecture for Wire
less Communication in a Fading Environment When Using
MultiElement Antennas,” Bell Labs Tech. J. 1 (2), 1996, pp.
41–59.
7. I.E. Telatar, “Capacity of MultiAntenna Gaussian Chan
nels,” Eur. Trans. Telecommun. 10 (6), 1999, pp. 585–595.
8. D.W. Bliss, K.W. Forsythe, A.O. Hero, and A.L. Swindle
hurst, “MIMO Environmental Capacity Sensitivity,” Thir
tyFourth Asilomar Conf. on Signals, Systems & Computers 1,
Pacific Grove, Calif., 29 Oct.–1 Nov. 2000, pp. 764–768.
9. D.W. Bliss, K.W. Forsythe, and A.F. Yegulalp, “MIMO
Communication Capacity Using Infinite Dimension Ran
dom Matrix Eigenvalue Distributions,” ThirtyFifth Asilo
mar Conf. on Signals, Systems & Computers 2, Pacific Grove,
Calif., 4–7 Nov. 2001, pp. 969–974.
10. T.L. Marzetta and B.M. Hochwald, “Capacity of a Mobile
MultipleAntenna Communication Link in Rayleigh Flat
Fading,” IEEE Trans. Inf. Theory 45 (1),1999, pp. 139–157.
11. L. Zheng and D.N.C. Tse, “Diversity and Freedom: A Fun
damental Tradeoff in MultipleAntenna Channels,” IEEE
Trans. Inf. Theory 49 (9), 2003, pp. 1076–1093.
12. S.M. Alamouti, “A Simple Transmit Diversity Technique for
Wireless Communications,” IEEE J. Sel. Areas Commun. 16
(8), 1998, pp. 1451–1458.
13. V. Tarokh, H. Jafarkhani, and A.R. Calderbank, “Space–
Time Block Codes from Orthogonal Designs,” IEEE Trans.
Inf. Theory 45 (5), 1999, pp. 1456–1467.
14. G. Ganesan and P. Stoica, “Space–Time Block Codes: A
Maximum SNR Approach,” IEEE Trans. Inf. Theory 47 (4),
2001, pp. 1650–1656.
15. B. Hassibi and B. Hochwald, “HighRate Linear SpaceTime
Codes,” Proc. IEEE Int. Conf. on Acoustics, Speech, and Sig
nal Processing 4, Salt Lake City, Utah, 7–11 May 2001, pp.
2461–2464.
16. V. Tarokh, N. Seshadri, and A.R. Calderbank, “Space–Time
Codes for High Data Rate Wireless Communication: Perfor
mance Criterion and Code Construction,” IEEE Trans. Inf.
Theory 44 (2), 1998, pp. 744–765.
17. B.M. Hochwald and T.L. Marzetta, “Unitary Space–Time
Modulation for MultipleAntenna Communications in Ray
leigh Flat Fading,” IEEE Trans. Inf. Theory 46 (2), 2000, pp.
543–564.
18. B.M. Hochwald and W. Sweldens, “Differential Unitary
SpaceTime Modulation,” IEEE Trans. Com. 48 (12) 2000,
pp. 2041–2052.
19. K.W. Forsythe, “Capacity of FlatFading Channels Associat
ed with a SubspaceInvariant Detector,” 34th Asilomar Conf.
on Signals, Systems and Computers 1, Pacific Grove, Calif., 29
Oct.–1 Nov. 2000, pp. 411–416.
20. K.W. Forsythe, “Performance of SpaceTime Codes over a
FlatFading Channel Using a SubspaceInvariant Detector,”
36th Asilomar Conf. on Signals, Systems and Computers 1, Pa
cific Grove, Calif., 3–6 Nov. 2002, pp. 750–755.
21. A. Stefanov and T.M. Duman, “Turbo Coded Modulation
for Wireless Communications with Antenna Diversity,” Proc.
IEEE Vehicular Technology Conf. 3, Amsterdam, 19–22 Sept.
1999, pp. 1565–1569.
22. Y. Liu, M.P. Fitz, and O.Y. Takeshita, “Full Rate Space–Time
Turbo Codes,” IEEE J. Sel. Areas Commun. 19 (5), 2001, pp.
969–980.
23. H. Sampath and A.J. Paulraj, “Joint Transmit and Receive
Optimization for High Data Rate Wireless Communication
Using Multiple Antennas,” Conf. Record ThirtyThird Asilo
mar Conf. on Signals, Systems & Computers 1, Pacific Grove,
Calif., 24–27 Oct. 1999, pp. 215–219.
24. N. Sharma and E. Geraniotis, “Analyzing the Performance
of the SpaceTime Block Codes with Partial Channel State
Feedback,” Proc. Wireless Communications and Networking
Conf. 3, Chicago, 23–28 Sept., 2000, pp. 1362–1366.
25. D.W. Bliss, K.W. Forsythe, A.O. Hero, and A.F. Yegulalp,
“Environmental Issues for MIMO Capacity,” IEEE Trans.
Signal Process. 50 (9), 2002, pp. 2128–2142.
26. T.M. Cover and J.A. Thomas, Elements of Information Theory
(Wiley, New York, 1991).
27. F. R. Farrokhi, G. J. Foschini, A. Lozano, and R.A. Valen
zuela, “LinkOptimal Space–Time Processing with Multiple
Transmit and Receive Antennas,” IEEE Comm. Lett. 5 (3),
2001, pp. 85–87.
28. D.W. Bliss, A.M. Chan, and N.B. Chang, “MIMO Wireless
Communication Channel Phenomenology,” IEEE Trans. An
tennas Propag., 52 (8), 2004, pp. 2073–2082.
29. D. Gesbert, H. Bölcskei, D.A. Gore, and A.J. Paulraj, “Perfor
mance Evaluation for Scattering MIMO Channel Models,”
ThirtyFourth Asilomar Conf. on Signals, Systems & Computers
1, Pacific Grove, Calif., 29 Oct.–1 Nov. 2001, pp. 748–752.
30. H. Bölcskei and A.J. Paulraj, “Performance of SpaceTime
Codes in the Presence of Spatial Fading Correlation,” Thirty
Fourth Asilomar Conference on Signals, Systems & Computers
1, Pacific Grove, Calif., 29 Oct.–1 Nov. 2000, pp. 687–693.
31. R.G. Gallager, LowDensity ParityCheck Codes (MIT Press,
Cambridge, Mass., 1963).
32. S.Y. Chung, G.D. Forney, Jr., T.J. Richardson, and R. Ur
banke, “On the Design of LowDensity ParityCheck Codes
within 0.0045 dB of the Shannon Limit,” IEEE Commun.
Lett. 5 (2), 2001, pp. 58–60.
33. T.J. Richardson, M.A. Shokrollahi, and R.L. Urbanke, “De
sign of CapacityApproaching Irregular LowDensity Pari
tyCheck Codes,” IEEE Trans. Inf. Theory 47 (2), 2001, pp.
619–637.
34. R.J. McEliece, D.J.C. MacKay, and J.F. Cheng, “Turbo De
coding as an Instance of Pearl’s ‘Belief Propagation’ Algo
rithm,” IEEE J. Sel. Areas Commun. 16 (2), 1998, pp. 140–
152.
35. C. Berrou, A. Glavieux, and P. Thitimajshima, “Near Shan
non Limit ErrorCorrecting Coding and Decoding: Turbo
Codes,” Proc. IEEE Int. Communications Conf. 2, Geneva,
23–26 May 1993, pp. 1064–1070.
36. L.R. Bahl, J. Cocke, F. Jelinek, and J. Raviv, “Optimal De
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL 125
coding of Linear Codes for Minimizing Symbol Error Rate,”
IEEE Trans. Inf. Theory 20 (2), 1974, pp. 284–287.
37. P. Robertson, E. Villebrun, and P. Hoeher, “Comparison of
Optimal and SubOptimal MAP Decoding Algorithms Op
erating in the Log Domain,” Proc. IEEE Int. Communications
Conf. 2, Seattle, 18–22 June 1995, pp. 1009–1013.
38. P. H.–Y. Wu and S.M. Pisuk, “Implementation of a Low
Complexity, Low Power, IntegerBased Turbo Decoder,”
Proc. IEEE Global Telecommunications Conf., San Antonio,
Tex., 25–29 Nov. 2001, pp. 946–951.
39. D.W. Bliss, P.H. Wu, and A.M. Chan, “Multichannel Mul
tiuser Detection of SpaceTime Turbo Codes: Experimental
Performance Results,” ThirtySixth Asilomar Conference on
Signals, Systems & Computers 2, Pacific Grove, Calif., 3–6
Nov. 2002, pp. 1343–1348.
40. D.W. Bliss, “Robust MIMO Wireless Communication in the
Presence of Interference Using Ad Hoc Arrays,” MILCOM
2003, 2, Oct. 2003, pp. 1382–1385.
41. C.M. Keller and D.W. Bliss, “Cellular and PCS Propagation
Measurements and Statistical Models for Urban Multipath
on an Antenna Array,” Proc. 2000 IEEE Sensor Array and
Multichannel Signal Processing Workshop, Cambridge, Mass.,
16–17 Mar. 2000, pp. 32–36.
42. B. Efron, The Jackknife, the Bootstrap and Other Resampling
Plans (Society for Industrial and Applied Mathematics, Phil
adelphia, 1982).
43. K.W. Forsythe, “Utilizing Waveform Features for Adaptive
Beamforming and Direction Finding with Narrowband Sig
nals,” Linc. Lab. J. 10 (2), 1997, pp. 99–126.
44. J. Pearl, Probabilistic Reasoning in Intelligent Systems: Net
works of Plausible Inference (Morgan Kaufmann, San Mateo,
Calif., 1988).
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
126 LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
naxiri w. niiss
is a a staff member in the
Advanced Sensor Techniques
group. He received M.S. and
Ph.D. degrees in physics from
the University of California
at San Diego, and a B.S.E.E.
in electrical engineering from
Arizona State University.
Previously employed by Gen
eral Dynamics, he designed
avionics for the AtlasCentaur
launch vehicle, and performed
research and development
of faulttolerant avionics. As
a member of the supercon
ducting magnet group, he
performed magnetic field
calculations and optimization
for highenergy particleaccel
erator superconducting mag
nets. His doctoral work was in
highenergy particle physics,
searching for bound states of
gluons, studying the twopho
ton production of hadronic
final states, and investigating
innovative techniques for
latticegaugetheory calcula
tions. At Lincoln Laboratory
he focuses on multiantenna
adaptive signal processing,
primarily for communication
systems, and on parameter
estimation techniques and
bounds, primarily for geolo
cation. His current research
includes ultrawide bandwidth
communication, geoloca
tion techniques using vector
sensor arrays, multipleinput
multipleoutput (MIMO)
radar concepts, algorithm
development for multichannel
multiuser detectors, MIMO
communication channel phe
nomenology, and information
theoretic bounds on MIMO
communication systems.
×rirn w. vonssrnr
is a senior staff member in the
Advanced Sensor Techniques
group. He received S.B. and
S.M. degrees, both in math
ematics, from MIT. In 1978
he joined Lincoln Laboratory,
where he has worked in the
areas of spreadspectrum com
munication, adaptive sensor
array processing, and syn
thetic aperture radar (SAR)
imaging. His work on spread
spectrum systems includes
electromagnetic modeling
(geometric theory of diffrac
tion) of antennas mounted on
an airframe, errorcorrection
coding, jamresistant synchro
nization techniques, and digi
tal matchedfilter design and
performance. In the area of
adaptive sensorarray process
ing he helped develop a num
ber of signalprocessing algo
rithms that exploit waveform
features to achieve levels of
performance (beamforming,
direction finding, geolocation,
and other forms of parameter
estimation) beyond those
attainable by nonexploitive
techniques. His work on SAR
imaging involves techniques
for resolution enhancement
and interference rejection for
foliagepenetration systems.
axaxna x. cnax
is an associate staff member
in the Advanced Sensor Tech
niques group. She received
MSEE and BSEE degrees in
electrical engineering from
the University of Michigan.
Her interests are in chan
nel phenomenology. She
has previously worked with
implementation of synthetic
aperture geolocation of cel
lular phones. Most recently,
she has worked on the imple
mentation of MIMO channel
parameterization.
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
The informationtheoretic bound on the spectral efficiency is a function of the total transmit power and the channel phenomenology. In implementing MIMO systems, we must decide whether channel estimation information will be fed back to the transmitter so that the transmitter can adapt. Most MIMO communication research has focused on systems without feedback. A MIMO system with an uninformed transmitter (without feedback) is simpler to implement, and at high SNR its spectralefficiency bound approaches that of an informed transmitter (with feedback). One of the environmental issues with which communication systems must contend is interference, either unintentional or intentional. Because MIMO systems use antenna arrays, localized interference can be mitigated naturally. The benefits extend beyond those achieved by singleinput multipleoutput systems, that is, a single transmitter and a multipleantenna receiver, because the transmit diversity nearly guarantees that nulling an interferer cannot unintentionally null a large fraction of the transmit signal energy. Phenomenology We discuss channel phenomenology and channel parameterization techniques in more detail in a later section. Aspects of the channel that affect MIMO system capacity, namely, channel complexity and channel stationarity, are addressed in this paper. The first aspect, channel complexity, is a function of the richness of scatterers. In general, capacity at high spectral efficiency increases as the singular values of the channel matrix increase. The distribution of singular values is a measure of the relative usefulness of various spatial paths through the channel. SpaceTime Coding and Receivers In order to implement a MIMO communication system, we must first select a particular coding scheme. Most spacetime coding schemes have a strong connection to wellknown singleinput singleoutput (SISO) coding approaches and assume an uninformed transmitter (UT). Later in the article we discuss spacetime lowdensity paritycheck codes, spacetime turbo codes, and their respective receivers. Spacetime coding can exploit the MIMO degrees of freedom to increase redundancy, spectral efficiency, or some combination
98
LINCOLN LABORATORY JOURNAL VOLUME 15, NUMBER 1, 2005
Spectral efficiency (bits/sec/Hz)
20 15 10 5 0 –5 0 5 10 M= M= M= M= 16 8 4 1
Eb /N0 (dB)
FIGURE 1. Spectralefficiency bound as a function of noise
spectraldensitynormalized energy per information bit (Eb /N0 ). The graph compares four different M × M multipleinput multipleoutput (MIMO) systems, assuming channel matrices with flat singularvalue distribution.
of these characteristics [11]. Preliminary ideas are discussed elsewhere [6]. A simple and elegant solution that maximizes diversity and enables simple decoupled detection is proposed in Reference 12. More generally, orthogonal spacetime block codes are discussed in References 13 and 14. A general discussion of distributing data across transmitters (linear dispersive codes) is given in Reference 15. High SNR design criteria and specific examples are given for spacetime trellis codes in Reference 16. Unitary codes optimized for operation in Rayleigh fading are presented in Reference 17. Spacetime coding without the requirement of channel estimation is also a common topic in the literature. Many differential coding schemes have been proposed [18]. Under various constraints at the transmitter and receiver, informationtheoretic capacity can be evaluated without conditioning on knowledge of the propagation channel [19, 20]. More recently, MIMO extensions of turbo coding have been suggested [21, 22]. Finally, coding techniques for informed transmitter systems have received some interest [23, 24]. Experimental Results Because informationtheoretic capacity and practical performance are dependent upon the channel phenomenology, a variety of experiments were performed. Both channel phenomenology and experimental procedures are discussed in later sections. Experiments were per
• BLISS, FORSYTHE, AND CHAN
MIMO Wireless Communication
WA T E R F I L L I N G
is a metaphor for the solution of several optimization problems related to channel capacity. The simplest physical example is perhaps the case of spectral allocation for maximal total capacity under a total power constraint. Let xk denote the power received in the kth frequency cell, which has interference (including thermal noise) denoted nk . If the total received power is constrained to be x, then the total capacity is maximized by solving
{ xk :∑ xk = x } k
k
W
Noise
max =
∑ log(1 + x
k /nk )
Power
Frequency
FIGURE A. Notional waterfilling example.
{ xk :∑ xk = x } k
k
max
∑ log(n
k
+ xk ) −
∑ log(n ).
k k
Use Lagrange multipliers and evaluate
∂ ∂x k
∑
j
log(n j + x j ) − µ
∑
j
x j − x
in each frequency cell. The volume of the water is the total received power of the signal. Note that cells with high levels of interference are not used at all. A similar solution results when the capacity is expressed by
to find a solution. The solution satisfies xk + nk = –1 µ for all nonzero xk . Figure A illustrates the solution graphically as an example of water filling. The difference between the water level (blue) and the noise level (red) is the power allocated to the signal
∑ log(1+ g x )
k k k −1 for gains gk . One can write the gains as g k = nk and use the waterfilling argument above. In this context, cells with low gains may not be used at all.
formed in an outdoor nonstationary environment in a mixed residential, industrial, and light urban settings. Intentional highpower interference was included. InformationTheoretic Capacity The informationtheoretic capacity of MIMO systems has been widely discussed [7, 25]. The development of the informed transmitter (“water filling”) and uninformed transmitter approaches is repeated in this section, along with a discussion of the relative performance of these approaches. (The concept of “water filling” is explained in the sidebar entitled “Water Filling.”) In addition, we introduce the topic of spectralefficiency bounds in the presence of interference, and we discuss
spectralefficiency bounds in frequencyselective environments. Finally, we summarize alternative channel performance metrics. Informed Transmitter For narrowband MIMO systems, the coupling between the transmitter and receiver for each sample in time can be modeled by using
z = Hx + n ,
(1)
where z is the complex receivearray output,
H ∈ nR × nT
is the nR × nT (number of receive by transmit antenna)
VOLUME 15, NUMBER 1, 2005 LINCOLN LABORATORY JOURNAL
99
The capacity is defined as the maximum of the mutual information [26] p(zx . and InR indicates an identity matrix of size nR . AssumT ing that the receiver can accurately estimate the channel. H) I (z. Here we assume that the fundamental limitation is the total power transmitted. In this discussion we assume that the environment is stationary over a period long enough for the error associated with channel estimation to vanish asymptotically. the capacity is given by C = sup log 2 x x† σ n InR + H x x σ n InR 2 2 † H† . This can be ob . (4) where the notation indicates determinant. H) = h(n ) = nR log 2 (π eσ n ). † indicates Hermitian conjugate. xH) = h(zH) − h(zx . but the transmitter does not attempt to optimize its output to compensate for the channel. Optimization over the nT × nT noisenormalized transmit covariance ma2 trix. capacity is averaged over an ensemble of quasistationary environments. NUMBER 1. n+ (6) (2) A waterfilling argument establishes that the entries dm in the diagonal matrix D ∈ n+ × n+ over the source conditional probability density p( xH) subject to various transmit constraints. is constrained by the total noisenormalized transmit power Po . (7) I (z. Noting that the mutual information can be expressed as the difference between two conditional entropies contain the n + topordered eigenvalues of HH†. 2005 This is a common transmit constraint. tr( P )= Po If Equation 7 is not satisfied for some dm . we impose the additional constraint P > 0 by using only a subset of channel modes. The sidebar entitled “Toy 2 × 2 Channel Model” discusses an example of IT and UT capacities for a simple lineofsight environment. the uninformed transmitter (UT) maximum spectral efficiency bound is given by C UT = log 2 InR + Po HH† . then an optimal transmission strategy is to transmit equal power from each antenna P = Po / nT In [7]. (5) To avoid radiating negative power. A variety of possible constraints exist for x x † . FORSYTHE. the mean capacity C IT is the channel capacity. where the expectation value is indicated by the notation . depending on the assumed transmitter limitations. AND CHAN MIMO Wireless Communication channel matrix. we can enforce this constraint by using the form tr{P} ≤ Po. Under the ergodic assumption (that is. H). CIT and CUT converge. it will not be satisfied for any smaller dm . The values dm must satisfy dm > Po + tr{D−1 } n+ . P = x x † /σ n . the ensemble average is equal to the time average). Uninformed Transmitter If the channel is not known at the transmitter. nT (8) sup log 2 InR + HPH† . and n is zeromeancomplex Gaussian noise.• BLISS. In order to study typical performance of quasistationary channels sampled from a given probability distribution. The resulting capacity is given by 100 LINCOLN LABORATORY JOURNAL VOLUME 15. as it may be difficult to provide the transmitter channel estimates. that 2 h(zx . By allowing different transmit powers at each antenna. (3) and that h(zH) is maximized for a zeromean Gaussian source x. Capacity Ratio At high SNR. xH) = log 2 . giving C IT = P. The informed transmitter (IT) channel capacity is achieved if the channel is known by both the transmitter and receiver. p(zH) C IT = log 2 Po + tr{D−1 } D. x is the transmitarray vector.
Given this asymptotic result. the theoretical capacity is not significantly affected as long as the number of antennas is much larger than the number of jammers. in addition to the spatiallywhite complex Gaussian noise n considered previously. C IT → P C UT log 2 Inmin + n o D T 2 2 h(zx . For the uninformed transmitter. such as in the case of a multipleinput singleoutput (MISO) system. AND CHAN MIMO Wireless Communication served in the large Po limit of the ratio of Equations 6 and 8. (13) H ≡ (I + R )−1/ 2 H . Nonetheless. the capacity in frequencyselective fading can be calculated by using an extension of Equations 6 and 8. int = p( z H ) p(η ) where the nmin diagonal entries in D contain all nonzero eigenvalues of H†H. the ratio CIT/CUT approaches one. Exploiting the orthogonality of frequency channels. in the special case in which H†H has a flat eigenvalue distribution. In the limit of high spectral efficiency for nJ infinite J/S jammers. the ratio is approximately equal to nT . This resistance to the effects of jammers is demonstrated experimentally later in the article. Equality is achieved if and only if the interference amplitudes have a Gaussian distribution. H) in the presence of the external interference becomes h(n + η ) . If the channel is rank one. this leads to the frequencyselective spectralefficiency bound VOLUME 15. (12) C int = sup inf I (z. Finally. the maximumminimum mutual information. The mutual information is again given by Equations 2 and 3. The transmitted noisenormalized power covariance matrix P is calcu lated by using H . Interference By extending the discussion in the previous section [8. NUMBER 1. H) ≤ log 2 π e σ n I + σ n R . † 1 n tr{HH } T † using Equation 6 with n + = 1 and Equation 8. tr( P )= Po † sup log 2  I + HPH . the optimal transmit covariance matrix is not unique. where † D contains the eigenvalues of HH .• BLISS. we can calculate capacity in the presence of uncooperative (worst case) external interference η . nR − n J ) min(nT . x H) . the channel matrix H(f ) is a function of frequency. then the convergence to one is logarithmically slow. nR ) C. becomes C IT. 27]. (14) = log(1 + Po maxeig{HH† }) tr log InR + n o HH† P T (10) ≈ maxeig{HH } . The spectralefficiency ratio is given by the maximum to the average eigenvalue ratio of H†H. FrequencySelective Channels In environments in which there is frequencyselective fading. where entropy h(zx . 2005 LINCOLN LABORATORY JOURNAL 101 . If nT > nR . (15) In general. 2 and σ n R is the spatialinterference covariance matrix. the uninformed transmitter spectralefficiency bound in the presence of noise is given by the same transformation of H → H . Thus the worstcase informed capacity. FORSYTHE. { } log 2 Po +tr{D−1 } D nmin (9) log 2 D nmin log 2 D nmin → log 2 ( Po ) − log 2 (nmin ) + log 2 ( Po ) − log 2 (nT ) + → 1. Gaussian interference corresponds to a saddle point of the mutual information at which the maximumminimum capacity is achieved. the loss in capacity approaches C int → min(nT . Similarly. At low SNR the ratio CIT /CUT is given by log 2[( Po + 1 / d max ) d max ] C IT → P C UT log 2 InR + n o HH† T using P. The capacity in the presence of Gaussian interference has a form identical to Equation 6 under the transformation D → D. we can make a few observations.
B beamwidths closely approximates many ad hoc definitions for physical arrays. 0 0. Consequently.2 0. The capacity of the 2 × 2 MIMO system is a function of the channel singular values and the total transmit power. H = 2 ( a1v 1 a2 v 2 ) . we employ a toy example for the purposes of introduction. 102 LINCOLN LABORATORY JOURNAL VOLUME 15. Eigenvalues of HH † are given by µ1. µ2 > where a1 and a2 are constants of proportionality (equal to the rootmeansquared transmittoreceive attenuation for transmit antennas 1 and 2 respectively) that take into account geometric attenuation and antenna gain effects. NUMBER 1.• BLISS. and we discuss the eigenvalue distribution of a 2 × 2 narrowband MIMO system in the absence of environmental scatterers. The separation between receive array responses can be described in a convenient form in terms of generalized beamwidths [40].6 0. FORSYTHE. one array’s individual elements can be resolved by the other array.4 0. Eigenvalues of HH for a 2 × 2 lineof† For small angular separations. indicated by a small separation in beamwidths. † v1 v 2 † a 2 1 − v 1 v 2 2 Po > 1 µ2 − 1 µ1 > .2 = 2a 2 2 1 Po + µ + µ1 1 2 . AND CHAN MIMO Wireless Communication TOY 2 × 2 C H A N N E L MODE L of channel matrix eigenvalues is essential to the effectiveness of multipleinput. Equations 6 and 7 in the main article are employed to determine the capacity for the 2 × 2 system. one vector for each transmit antenna.8 1. When the transmit and receive arrays are small. the larger eigenvalue decreases slightly. assuming µ1 > µ 2.0 Generalized beamwidth separation FIGURE A. To visualize the example. which is valid if the rectangle deformation does not significantly affect overall transmittertoreceiver distances. indicated by a larger separation. multipleoutput (MIMO) communication. For the purpose of this discussion. The ratio of channel matrix eigenvalues can be changed by varying the shape of the rectangle. As the array apertures become larger. one eigenvalue is dominant. then only the † v1 v 2 Eigenvalue/a 2 (dB) (1 ± ). The “waterfilling” technique (explained in a previous sidebar) must first determine if both modes in the channel are employed. Conversely. 2005 . 0 –10 –20 –30 where the absolute value is denoted by . Figure A displays the eigenvalues µ1 and µ 2 as a function of generalized beamwidth separation. the smaller eigenvalue increases. and v 1 and v 2 are unitnorm array response vectors. we can imagine two receive and two transmit antennas located at the corners of a rectangle. this definition of sight channel as a function of antenna separation. b12 = 2 π arccos { † v1 v 2 }. If the condition is not satisfied. Both modes are used if the following condition is satisfied. The columns of the channel matrix H (in Equation 1 in the main article) can be viewed as the receiverarray response vectors. we assume a = a1 = a2 .
The individual antenna elements can be resolved by the larger effective aperture produced by the scatterers. In complicated multipath environments. † = log 2 1 + 2a 2 1 + v 1 v 2 Spectral efficiency (b/sec/Hz) 10 5 2 1 0.1 (solid line) and 0. As demonstrated in Figure A.• BLISS. when using one mode.9 (dashed line). where Po is the total noisenormalized power. and small beamwidth separation increases the coherent gain. the ability to resolve antenna elements is related to the number of large singular values of the channel matrix and thus the capacity. AND CHAN MIMO Wireless Communication stronger channel mode is employed and the capacity. H( f )) ∑ ≈ ∫ df P ≈ P 1 log 2 I + o HH† . C UT.5 –10 –5 0 5 10 15 20 ( ) P . In both cases. NUMBER 1. The effect of the scatterers upon capacity depends on their number and distribution in the environment. from Equation 6. small arrays employ scatterers to create virtual arrays of a much larger effective aperture. 2a Po + 2 † 2 v1 v 2 2 2 Figure B displays the resulting capacity as a function of a 2Po (mean singleinput singleoutput SNR) for two beamwidth separations. 2 † v1 v 2 . 0. 2005 LINCOLN LABORATORY JOURNAL 103 . is given by C IT = log 2 (1 + µ1Po ) . At high values of a 2Po large beamwidth separation produces a higher capacity as the optimal MIMO system distributes the energy between modes. the total received power is much larger than a 2Po . The informed transmitter capacity of a 2 × 2 lineofsight channel. The total received power is given by † 2 v1 v 2 a 2 Po 1 − v † v 1 2 . µ1µ 2 2 † = 2 log 2 a 2 1 − v 1 v 2 − log 2 1 − { 2 Po +1 } FIGURE B.9.FS = ∫ df C nf n =1 UT ( Po .1 and 0. In this regime. FORSYTHE. nf nT (16) ∆f log 2 I + n o H( f n ) H† ( f n ) ∑ T nf n =1 ∆f where the distance between frequency samples is given by ∆f and the nf bin frequencypartitioned channel matrix is given by VOLUME 15. assuming antenna beamwidth separations of 0. both modes are used and the capacity is given by C IT 1 Po + µ + µ1 µ1 1 2 = log 2 0 2 2 0 . At low values of a 2Po the capacity associated with small beamwidth separation performs best. capacity is linear with receive power. µ 2 a 2P 0 (dB) µ µ P + µ + µ 1 = log 2 1 2 o 1 2 . and when using two modes. o otherwise.
a 2 is defined by the Frobenius norm squared of the channel matrix normalized by the number of transmitters and receivers. This assumption corresponds to an environment with complicated multipath scattering. and diagonal entries in Din Equation 7 are selected from the eigenvalues of HH† . The capacity can be expressed C IT. NUMBER 1. and characterizing MIMO channels. and channel parameterization and estimation techniques. channels can be characterized by a diversity order [30]. Other Performance Metrics The informationtheoretic capacity is not the only possible metric of performance. 28]. 2005 (19) The above model results in a (nT ⋅ nR ) × (nT ⋅ nR ) linkbylink covariance matrix of the Kronecker product form (ML M† ) ⊗ (MR M† )∗ for the entries in the chanL R nel matrix H. These topics are discussed in greater detail elsewhere [25. AND CHAN MIMO Wireless Communication 0 0 0 H( f1 ) H( f 2 ) 0 0 H≡ 0 0 0 H( f n f . normalized so that tr{P} ≤ nf Po . ) (17) For the informed transmitter channel capacity. Channel Phenomenology In this section we describe tools for modeling. First. in the case in which the system can resend frames of data. power is optimally distributed amongst both spatial modes and frequency channels. a2 = tr{HH† } . As an example. FORSYTHE. the (nT ⋅ n f ) × (nT ⋅ n f ) spacefrequency noisenormalized transmit covariance matrix H is a block diagonal matrix. This choice corresponds to the typical noisenormalized received power for a single receive 2 and single transmit antenna radiating power σ n Po . nf (18) all the entries in the channel matrix are sampled from identical independent complex Gaussians H G . Then we discuss the simplest channel characterization. In many practical situations this metric may be the best measure of performance. provided several additional conditions are met. which is maximized by Equation 6 with the appropriate substitutions for the frequencyselective channel. it may provide a channel eigenvalue distribution that is too flat. scatterers are concentrated around (but not too close to) the transmitter and receiver. multiple scattering of a particular kind (from transmitter element to transmitter scatterer to receiver scatterer to receiver element) dominates propagation. nT nR (20) The total received noisenormalized power produced by a set of orthogonal receive beamformers is given by . assuming a given probability of errorfree decoding of a frame. First we introduce the standard model and simple modifications to it. scatterers are sufficiently separated in angle when viewed by their associated array. followed by a description of channel estimation techniques. The simplest approach is to assume that 104 LINCOLN LABORATORY JOURNAL VOLUME 15. However. It can be employed to easily compare performance by using different constraints and environments. methods for determining how much channels have changed. FS ≈ max P 1 log 2 I + HP H† . While this approach is convenient from the perspective of performing analytic calculations. This product structure can arise from a spherical Green’s function model of propagation. Received Power It is often convenient to parameterize the incoming signal power in terms of a 2Po . estimating. or the achievable spectralefficiency bound. where a 2 is the meansquared link attenuation. another useful performance metric is the outage capacity [16]. Third. At the other extreme. be much larger than a 2Po . One approach uses the form H = ML G MR . Because of the block diagonal structure of H. A number of approaches that introduce spatial correlations have been suggested. Second. In general.• BLISS. for example. in general. which is mean receive power. which is used to indicate an effective cutoff in the eigenvalue distribution induced by spatial correlation. Standard Model A variety of techniques are used to simulate the channel matrix [29]. this choice can be mildly misleading because the total received power will.
statistical model of the ren ×n ceived signal Z. nR } a 2 Po . δntaps . T≡ T(δ ) ntaps (21) = Po 2 tr{D} tr{D−1 }tr{D} − n+ + . FORSYTHE. † (26) (22) The second term in Equation 22 is bounded from below by zero. Depending upon the issue being investigated. By assuming regular delay sampling. assuming T ∈ T s is the transmit sequence. no one metric will be useful for all situations. setting n + = 1. far from the nonlinear regime of the Shannon limit. 2 so that the transmit matrix has dimension (nT ⋅ ntaps ) × ns . (24) where R is the noiseplusinterference covariance ma 105 . However. ChannelDifference Metrics A variety of metrics are possible. signalinthemean. In this regime the total received power is given by tr{HPIT H† } → Po maxeig{HH† } . It is worth noting that P is not in general optimized by the informed transmitter to maximize received power but to maximize capacity. is −1 P + tr(D ) tr{HPIT H† } = tr o D − In+ n+ trix. given an arbitrary channel matrix. For very small a 2Po . the frequencyselective channels can be estimated by first estimating a finite impulseresponse MIMO channel. H. δ 2 . NUMBER 1. the optimal solution is to maximize received power. nR } n+ ≥ max{nT . we can use a discrete Fourier transform to construct the explicit frequencyselective form.• BLISS. T(δ1 ) T(δ ) 2 . T ) = e − tr[( Z − HT ) π ns nR † (28) where the npoint discrete Fourier transform is represented by n and the Kronecker product is represented by ⊗ . The total received power for the capacityoptimized informed transmitter. nR } a 2Po . which can be transformed to the frequency domain. In investigating channel variations. we may wish to think of these matriVOLUME 15. As an example. The resulting wideband channel matrix has the dimension (nT ⋅ ntaps ) × ns . which is achieved if there is only a single nontrivial mode in the channel. The previous channelestimation discussion explicitly assumed flat fading. ˆ ˆ ˆ [H(δ1 ) H(δ 2 ) H(δ ntaps )] = ZT † (TT † )−1. 2005 LINCOLN LABORATORY JOURNAL R −1 ( Z − HT )] R  ns . The maximumlikelihood estimate of H is given by ˆ (25) H = Z T† (TT† )−1 . A finite impulseresponse extension of Equation 1 is given by introducing delayed copies of T at delays δ1. Consequently. . ˆ ˆ ˆ [H( f1 ) H( f 2 ) H( f ntaps )] ˆ ˆ ˆ = [H(δ1 ) H(δ 2 ) H(δ ntaps )](ntaps⊗ InT ) . is given by p(Z  R . Channel Estimation The Gaussian probability density function for a multivariate. (27) (23) Using this form. AND CHAN MIMO Wireless Communication tr{HPH† }. two completely different channels can have the same capacity. an effective channel filter is associated with each transmittoreceive antenna link. assuming that the reference signals in T are known and TT† is nonsingular. This result is bounded from above by nT nR a Po . The uninformed transmitter rate is maximized by sending equal power to all transmit antennas so that tr{HPH† } becomes Po /nT tr{HH† } = nR a 2Po . n+ n+ The first term in Equation 21 is bounded from below by Po tr{HH } tr{D} ≥ Po min{nT . the total received power is greater than or equal to max{nT . This is done by transmitting the best mode only.
then signal ′ energy is lost. The second metric is sensitive to differences in both the singularvalue distribution and the channel eigenvector structure. capacity is invariant under channelmatrix transformations of the form † H → W1HW2 . The first metric measures differences in channel singularvalue distributions. FORSYTHE. MIMO capacity is only a function of the channel singular values. Hb ) 2 (32) † [log 2 λm (Ha H† ) − log 2 λm (Hb Hb )] . that is. 2005 ∑ ∑ = ∑ h ∑ ∑ h cos θ . The metric takes into account both the eigenvalue and eigenvector structure of the channels.nR ) m =1 (31) ∑ log † † 2 λm (H a H a ) − log 2 λm (Hb Hb ) . Equivalently. but motivations are provided. Consequently. the low SNR uninformed transmitter capacity approximation is given by P C = log 2 I + o HH† nT ≈ log 2 (e ) tr = log 2 (e ) Po nT P o nT HH† † ∑h m h m m (33) 2 P = log 2 (e ) o nT wm ≡ hm . The channel capacity is a function of HH †. It is motivated by the effect of receivebeamformer mismatch on capacity. the difference is a function of Po . Because capacity is a function of the structure of the channel singularvalue distribution. C ′ ≈ log 2 (e ) Po nT ∑ m † w ′m h m . Starting with Equation 8. One possible reason that a beamformer might use the wrong matched spatial filter is channel nonstationarity. = ∑ h m w ′m h m m 2 † 2 m hm † 2 h ′m h ′m † hm hm 2 m m hm 2 m m 2 2 (35) m m m 2 .• BLISS. the same a2Po . assuming that each eigenvalue is associated with an orthogonal dimension. NUMBER 1. the difference can be investigated in a high SNR limit. hm ∑ m † w mhm (30) However. and indicates the l2 norm. giving † ∆C UT ≈ log 2 H a H† − log 2 Hb Hb a where hm is the column of the channel matrix associated with transmitter m. the optimal receive beamformer is given by the matched response given in wm . δ (Ha . Here two metrics are discussed. ∆C UT = log 2 I + − log 2 I + a 2 Po † Ha Ha nT a Po † Hb Hb . AND CHAN MIMO Wireless Communication ces as being similar or very different. adversely affecting the capacity. Second. for some applications it is useful to employ a metric that is also invariant under this transformation. there are two problems with this definition. the metric is defined to be the Euclidean difference. If some other beamformer is employed. To increase the sensitivity to the shape of the eigenvalue distribution. First. In the low SNR limit.nR ) ∑ m =1 (29) where W1 and W2 are arbitrary unitary matrices. To address the first issue. giving 106 LINCOLN LABORATORY JOURNAL VOLUME 15. EigenvalueBased Metric As was mentioned earlier. The fractional capacity loss is given by C′ ≈ C where λm ( X ) indicates the mth largest eigenvalue of X. Both metrics are ad hoc. 2 (34) = min(nT . nT 2 Fractional Receiver Loss Metric In this section we introduce a powerweighted mean cos 2θ metric. A natural metric would employ the distance between the capacity for two channel matrices at the same average total received power. the metric should be sensitive to this structure. a ≡ min(nT . labeled w m . there is degeneracy in H singular values that gives a particular capacity.
. A α = A α L = A α R . As was previously discussed. but not the singularvector structure of the channel. α 1. and A α L and A α R are positivesemidefinite diagonal matrices. Channel Parameterization A commonly employed model assumes the channel is proportional to a matrix G. This section also discusses a simple application of LDPC to spacetime codes. F 2 where F indicates the Frobenius norm. the diagonal matrices can be set equal. a symmetric form is given by F = b UA α U † G ′ VA α V † = b UA α GA α V † (37) and Aα = n diag {α 0 . we can see that the channel capacity is a function of channel singular values. FORSYTHE. Singular Values The singularvalue distribution of H. are applications of techniques formulated for Bayesian belief networks. Using the previous discussion as motivation. It is worth noting that this approach does not necessarily provide an unbiased estimate of α. Furthermore. 2005 LINCOLN LABORATORY JOURNAL 107 . While the distribution is convenient. producing the new random channel matrix F. (38) ∑ γ (H. Data presented later in the article have sufficiently high SNR such that α can be estimated within ±0. (36) where the “powerweighted” expectation is evaluated over transmitters. Thus channel phenomenology can be investigated by studying the statistics of channel singularvalue distributions. Channel Parameter Estimation ˆ An estimate for α associated with particular transmit and receive locations is given by minimizing the meansquare metric given in Equation 32.. indicates averaging is over an ensemble of F for a given α and an ensemble of H for given transmit and receiver sites. H′ ) ≡ m ∑ 2 h m h ′m cos θm m h m h ′m . which are introduced below. ^ ^ (39) ˆ where X indicates the estimated value of X. Used here is the fact that arbitrary unitary transformations do not affect the statistics of the Gaussian matrix. 2 ˆ α = arg min δ [H. and as α → 1. where U and V are unitary matrices. It is generally desirable for metrics to be symmetric with respect to H and H′. are dependent upon the received SNR. α n −1 }2 ) . NUMBER 1. it produces simply more realistic channels than the uncorrelated Gaussian model. thus avoiding moral attributions with regard to channel matrices. While this approach is limited. From the discussion earlier. Estimates of α. or the related eigenvalue distribution of HH †. The spatial correlation matrices can be factored so that ML = UA α L U † and MR = V A α R V †. for LDPC codes.02. VOLUME 15. F(α )] . using the metric introduced here. α n −1 } tr(diag{α 0 . a spatially uncorrelated Gaussian matrix is produced. The significance of modern implementations of LDPC codes rests on iterative decoding algorithms that. is a useful tool for understanding the expected performance of MIMO communication systems. Here the expectation. it does suffer from a singularvalue distribution that is overly optimistic for many environments. n is given by the size of A α . AND CHAN MIMO Wireless Communication which is the powerweighted mean cos 2 θm estimate. a rankone channel matrix is produced. denoted by . empirically this model provides good fits to experimental distributions. α 1. one solution is to introduce spatial correlations using the transformation F = b ML GM† R [29]. where the entries are independently drawn from a unitnorm complex circular Gaussian distribution. where where b is used to set overall scale.G. which were studied extensively by R. The normalization for A α is chosen so that the expected value of F 2 is b 2nT nR . but has the satisfying characteristics that as α → 0 . Gallager [31]. SpaceTime LowDensity ParityCheck Codes This section of the article introduces lowdensity paritycheck (LDPC) codes.• BLISS. where cos θm is defined to be the inner product between the “good” and “bad” unitnorm array responses for the mth transmitter. Assuming that the number of transmit and receive antennas are equal and have similar spatial correlation characteristics. and U and V indicate random unitary matrices. The form of A α given here is somewhat arbitrary.
c) = (512. decoding complexity of LDPC codes scales linearly (with a fixed number of iterations) with the code length. denoted π (x). When a node is activated. Nodes are activated in any order. Messages that flow from P parent to child are denoted π k (uk ) and are treated as if they were priors. while messages that flow from child C to parent are denoted λk ( x ) and are treated as if they were likelihoods.…. can be interpreted as a likelihood function on the alphabet. 2005 involved in only a few paritycheck equations. graphical decoding techniques have motivated practical code design. FORSYTHE. NUMBER 1. In addition. we have rnR = cnC for regular LDPC codes. messages are received and transmitted from each node. resulting in renewed interest in these codes. making relatively long codes practical. denoted λ(x). if the dimensionality of the paritycheck matrix is r × c. in the context of machine learning. albeit at the cost of extremely long codewords. messages received from parents and children are used to update the internal (for that node) prior and likelihood functions π (x) and λ(x) for the node’s alphabet. ul ). LDPC Decoding Recently. LDPC codes are linear block codes defined by a paritycheck matrix. This function factors in the manner shown in the figure. Consequently. most entries in the paritycheck matrix are zero. consider the probability density function denoted p( x1. 1024) and (nR . x 3 . who studied their distance properties and decoding in a wellknown monograph [31]. subject only to the requirement that all incoming messages are available. More powerful codes that are not regular are also known [33]. each node in the graph maintains an alphabet (for example. Bayesian belief networks [34] can be used to formulate decoders for LDPC codes and turbo codes (the sidebar entitled “Bayesian Belief Net . G For decoding purposes. As an example. To evaluate a consistent set of distribution functions. expressing simpler variable dependencies than those allowed by the multivariate notation p( x1. AND CHAN MIMO Wireless Communication B AY E SI A N BE L I E F N E T WOR K S is often based on the Bayesian belief networks popularized by Pearl. Thus. LowDensity ParityCheck Codes LDPC codes were developed by Gallager. To appreciate the use of belief networks for decoding. The maximum of the belief function can be used as a decision on the value of the node’s alphabet. x l ) . An interpretation of various decoding algorithms in terms of Bayesian belief networks is presented elsewhere [2]. One probability distribution.…. Each symbol in the codeword is 108 LINCOLN LABORATORY JOURNAL VOLUME 15. Regular LDPC codes have nC paritycheck equations for each symbol. and each paritycheck equation involves nR symbols. the LDPC code used for some of the experiments described later satisfies (r. in a wellknown monograph [1]. With the advent of graphical decoding techniques. nC ) = (8. However. the symbol alphabet for coding applications) and several (probability) distributions over this alphabet. Subsequent developments in code design and decoding have led to codes that achieve levels of performance astonishingly close to the Shannon capacity [32]. Initial settings of the internal functions are provided (but not shown in the figure) to enable the process to start. it calculates its internal prior and likelihood functions and then makes its messages available to its parent and child nodes. x 4 ) in Figure A. 4). but not a normalized density) distribution. The factorization can be represented by a directed acyclic graph as shown. x 2 . with directed arrows expressing conditional probabilities of the more general form p( x  u1. can be interpreted as a prior density on the alphabet while another (nonnegative.• BLISS. each node keeps track of a belief function that is the product of priors and likelihoods: π (x) λ(x). At each node. softdecision decoding of LDPC codes became practical.
Initial states and a node update order must be chosen.J. so order of activation matters and convergence is not guaranteed. c 4. D. Only in special cases do the iterations converge to a Bayesian decision. However.C. us ) Π πkP(uk ) k≠j FIGURE A. FORSYTHE. Pearl. Calif. k ik which expresses the paritycheck relation. Unfortunately. and the paritycheck nodes.• BLISS.…. 2005 LINCOLN LABORATORY JOURNAL 109 . In this form. NUMBER 1. For LDPC codes. Cheng. the data nodes z 1. π (x) = Σ p(xu1. Each node has a symbol alphabet on which it maintains a belief function that factors as a product of a priorlike function and a likelihoodlike function. z 4. pp.…. x2. labeled by zeroes.uk :k≠j Σ λ (x)p(xu1. AND CHAN MIMO Wireless Communication Variable dependencies x1 x2 x3 x4 Parent nodes (alphabets uk ∈ Bk ) Update node (alphabet x ∈ A) Child nodes Node calculations and messages Messages passed Received Sent Loopless directed acyclic graph (DAG) Directed Markov field Bayesian belief network p(x1. the iterative technique is both practical and effective. MacKay. Each edge between a paritycheck node and a symbol node corresponds to a nonzero entry in the paritycheck matrix. Symbol decisions are based on the belief function. McEliece. J. San Mateo. we present only a concise summary. updating likelihood ratios lk for each nonzero entry in the matrix. and p(0  c i1 . each nonzero entry in the paritycheck matrix indicates the edge of a graph connecting a paritycheck node (row index) and a codeword symbol (column index). interesting graphs have loops. Figure 2 shows a graph illustrating data and paritycheck dependencies for the codewords. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference (Morgan Kaufmann.” IEEE J. and J. Typically. In general.us ) Π u k=1 s πkP(uk ) Belief λ(x) π (x) λ (x) = Π λkC (x) k π jC (x) = π (x) Π λ kC(x) k≠j λjP(uj) = x. 16 (2). References 1.F. which express the likelihood ratios. Areas Commun. for decoding applications. The resulting algorithm can be viewed as sweeping through the rows and columns of the paritycheck matrix. a thorough explanation of the steps in this algorithm is outside the scope of this article. beyond connecting the decoding algorithm of LDPC codes to Bayesian belief networks. 1998. Sel. but for many interesting applications. The notation below denotes l ij as the likelihood ratio stored with the ijth (nonzero) entry in a fixed (for the given step) row or column of the paritycheck matrix. Decoding occurs by treating the graph as a Bayesian belief network using the conditional probabilites p( z k  c k ). 2. Beliefs are updated by passing messages among nodes in a manner suggested by the terminology. The graph shows the symbol nodes c 1. Turbo codes and low density paritycheck codes have decoders based on this paradigm. c il ) = δ (∑ c ). x4) = p(x4x3)p(x3x1)p(x2x1)p(x1) likelihoods λC λP priors πP πC For loopless graphs. …. x3. the order of activation does not matter and the process converges. Bayesian belief networks provide a framework for representing conditional probabilities in a graphical manner.….. The example in Figure 2 is a single paritycheck code on four symbols.…. nodes are activated in a repetitive pattern for a certain number of iterations until a stopping criterion is met. “Turbo Decoding as an Instance of Pearl’s ‘Belief Propagation’ Algorithm. 140–152. 1988).J. R. works” provides more information). the iterative steps of the algorithm are summarized for the simple case of a binary symbol alphabet by the equations: VOLUME 15.
= 2n Un j ≠ k Un π ( βi j ) p(iold ) j 2. Column sweeps p(inew ) ∝ j ≠ k p(iold ) p(cLF ) k j For the code used in the experiments. .ci ) = δ ( 1 s c1 c2 c3 c4 Codeword component Parity checks Σ cik) k 0 0 0 Some initialization Flat priors for codeword nodes: π (ci ) Fixed likelihoods for evidentiary nodes: Node firing order: z 0 c z 0 c ··· Stopping rule: parity check satisfied λ k(x) = δ (x − zk) FIGURE 2. Some simplifications that are not possible for nonbinary symbol alphabets are involved in the binary case. all depedancies are expressed graphically through conditional probability densities as required for the formalism of Bayesian belief networks. NUMBER 1. . each row sweep involves eight l ij per row and each column sweep four l ij per column. which are assumed to be independent sampletosample. Both the redundancies in codewords ck and the relationship between the codewords and the data zk can be represented graphically. The datacodeword relationship is expressed through the probability densities p(zc). AND CHAN MIMO Wireless Communication Paritycheck matrix 1 1 0 0 0 1 1 0 0 0 1 1 Bayesian belief network z1 z2 z3 z4 Evidentiary nodes (observations) p(zc) p(0ci . . the row/column sweeps are expressed by: 1. c . Each of the row (column) operations is independent of any other row (column operation) and hence can be 110 LINCOLN LABORATORY JOURNAL VOLUME 15. This allows considerable acceleration of hardware decoders. . Column sweeps [column c with loglikelihood ratio l c(LLR ) ] l i(knew ) = ∑l j ≠k ( new ) ij + l c(LLR ) implemented in any order or in parallel. Symbol decisions ( p ) p m im ( LF ) . In this more general context. are involved in a parity check. Bit decisions (column c) sign ∑l k ik + l c(LLR ) . Row sweeps π ( βi ) pi k k ( new ) 3. Decoding can be halted after a fixed number of iterations or after the paritycheck equations are satisfied. cis) where the symbols cik . 2005 3. Row sweeps l i(new ) = tanh k 2 ∏ j ≠k l i(old ) j tanh 2 2. 1. FORSYTHE. Redundancies in the codewords are expressesed in a similar notation as p(0ci1. Softdecision decoding of low density paritycheck codes can be based on Bayesian belief networks. In this manner. Application of Bayesian belief networks to lowdensity paritycheck codes.…. 1 ≤ k ≤ s.• BLISS.
implementation of turbo decoders has been carried out and high datarate decoding is possible [38]. provides a feasible way to approach optimal performance.• BLISS. and therefore higher data rates. The first two transmitters send two bits (redundant in I and Q) of a symbol of an LDPC codeword over GF(16). vary in time and thus effectively probe the environment characterized by the channel matrix. Each spacetime channel transmits one of several possible quadrature phaseshift keying (QPSK) waveforms with slightly offset carrier frequencies. In this section we discuss the spacetime turbo code used in this example. We also introduce minimummeansquarederror extensions of the receiver. NUMBER 1. which are introduced and explained in the sidebar entitled “SpaceTime Codes. Decoding occurs by forming likelihood ratios based on channelmatrix estimates and then using the iterative decoder described above. practical MIMO performance requires the selection of a spacetime code and an appropriate matched receiver. which is derived from the MAP principle. 22]. FORSYTHE. Again. the coding provides spatial diversity. in the section on experiments. in which case channelmatrix estimates can vary sample to sample. 2005 LINCOLN LABORATORY JOURNAL 111 . respectively. provides a 2bit/ sec/Hz link for a 4 × 4 MIMO system with independent QPSK waveforms from each transmitter. The transmitter outputs. introduced elsewhere [35]. such as with interleavers and convolutional encoders. We develop a maximumlikelihood formulation of a multipleantenna multiuser receiver. The iterative decoding process. each of which is fully redundant yet different from any other subset. The approach used here. Modifications of the MAP algorithm include logMAP and maxlogMAP [37]. The remaining two transmitters send the other two bits of the same symbol. The LDPC spacetime code just described exhibits full spatial redundancy among all transmitters. The distribution of systematic bits is such VOLUME 15. SpaceTime Extension LDPC There are a variety of extensions of LDPC codes to spacetime codes. the I and Q components of a transmitter represent.” For the experiments described below. Less redundancy. The turbo encoder employs a rate1/3. For example. taking advantage of information exchange among component decoders. the permutation πk indicates the effect of a particular nonbinary coefficient in the paritycheck equation. For each component decoder. and the notation denotes the Hadamard (component by component) product. differential frequencies among transmitters enable spatial diversity. viewed collectively as a vector at any instant. the best decoding algorithm is the maximum a posteriori (MAP) algorithm or the BCJR algorithm [36]. Decoding is based on likelihood functions built over GF(16) using estimates of the channel matrices. and we discuss suboptimal implementations of the receiver. uses four transmitters. These differential frequencies are also large compared to the expected Doppler spreads and small compared to the signal bandwidth. only one type of extension was considered. illustrate that codes constructed with simple components. The differential frequencies are sufficiently large to effectively decorrelate the transmitted waveforms over the length of a codeword (1024 bits) even if the data sequences in each channel are identical. can be achieved by dividing the transmitters into subsets. the spacetime code discussed later. SpaceTime Turbo Code and Multichannel Multiuser Detectors While the theoretical performance is determined by the channel phenomenology. Since the transmitted vector varies significantly over the duration of a codeword. In the simplest example of such a code. which was introduced elsewhere [39]. combined with an iterative decoding process can achieve nearShannon capacity performance. Each transmitter sends the same complex baseband sequence (QPSK) shifted in frequency. two different LDPC codewords. A single data stream is turbo encoded and the encoded data stream is distributed redundantly amongst the transmitters. AND CHAN MIMO Wireless Communication The components of the vector pk express probabilities for the values of the kth symbol. Recently. Un is a WalshHadamard matrix. A number of spacetime extensions of turbo coding have been suggested [21. 16state convolutional encoder twice with two different 4096bit random interleavers. Note that the channel matrix can change during the codeword. SpaceTime Turbo Code Turbo codes. and we discuss the value and use of training data.
the transmitter can be informed of the propagation channel by the receiver and thus adjust its coding accordingly. 39.• BLISS. s1 S = S : S = s2 ∗ − s2 . is a minimummeansquarederror (MMSE) extension to an iterative implementation of a maximumlikelihood multipleantenna receiver. 40]. FORSYTHE. remodulation errors have a reduced effect on subtraction performance. A pair of bits ( I t1. This approach offers the largest informationtheoretic capacity but can be difficult to accomplish in a dynamic environment. A fixed mapping between these four integers and the quadrature phaseshift keying (QPSK) alphabet completes the coding and modulation. in principle improving the performance of the iterative multiuser detection for a given bit error rate. which is taken here. These are often chosen under one of several design criteria. Instead we briefly describe two important categories of spacetime codes that are not treated in the text. distributed randomly amongst the transmitters. s2 s1 SpaceTime Trellis Codes Figure A provides an example of a spacetime trellis code. consider a set of matrix symbols S contained in S. The information bits are encoded in matrices that are constrained to lie in the class S. NUMBER 1. These codes share transmitted power equally among all spatial channels. The MCMUD algorithm employed for this analysis iteratively combines a blind spacetimefrequency adaptive beamformer with a multiuser detector. The number of different types of spacetime codes is too large to provide a useful overview here. The second approach. consider the Alamouti code [1]. bkp }. For example. linearity of the likelihood function decouples decisions on the data symbols. This class is defined by the property that SS† is proportional to the identity matrix with a fixed (independent of S) proportionality constant. The parity bits are sent once. For some simple classes S. The likelihood function is linear in each sk . I t2 ) at time t enters a convolutional encoder with integer coeffip cients akp and bk at the pth lag in the kth channel. The probability of an error in deciding between two such = arg max Re tr(ZS†H† ) . There are two basic approaches to spacetime coding. The input bits are interpreted as the integers 0 or 1. Each codeword is a matrix symbol C. Because more energy is dedicated to systematic bits. AND CHAN MIMO Wireless Communication S PA C E T I M E C O D E S S  are used with multiple transmitters to provide spatial as well as temporal redundancy in the data received by an array of antennas. discussed elsewhere [3. In the first approach. 2005 Multichannel Multiuser Detector The multichannel multiuser detector (MCMUD) algorithm. Computations occur modulo 4 and result in an integer value between 0 and 3 for each channel. The maximumlikelihood decision for S is based on finding argmin Z − HS S∈S S∈S 2 S = [S0 S∗ ] with 0 S0 = s 1 s 2 s3 s 4 − s2 s1 − s4 s3 − s3 s4 s1 − s2 − s4 − s3 . which involves a linear function in the entries of S. 112 LINCOLN LABORATORY JOURNAL VOLUME 15. uses fixed codes of various rates that offer good performance on average (over all channels). The trellis code is defined by the coefficients {akp . Another example of an orthogonal matrix code is that each systematic bit is sent twice on two different transmitters. . also shown in the figure. Block Orthogonal Codes For data Z and channel matrix H. decoupling demodulation decisions. ∗ s1 The information symbols s1 and s2 are sent redundantly over both channels. The difference in weighting between the systematic and parity bits provides an effective puncturing of the code.
“A Simple Transmit Diversity Technique for Wireless Communications. For example. and N ∈ R s is additive Gaussian interference plus noise. T ∈ nT ×ns is the transmitted sen ×n quence. ns data matrix. Constrained searches over the code coefficients are commonly used to find codes with the smallest possible error between closest codewords under either criterion. When 4 ≥ rnR . an alphabet consisting of the integers modulo 4 is used for convolutional encoding at each transmitter. 1451–1458. 2. tr (Z − HT )(Z − HT )† R −1 . The resulting output symbols are mapped to a QPSK alphabet. Note that the spectral efficiency is 2 bits/sec/Hz.” IEEE J. S. (44) where R indicates the spatial covariance matrix of the interference plus noise.   indicates the determinant of a matrix. NUMBER 1. maximizing the Euclidean distance between the two codewords Ck becomes important. Code design often involves a pruned search over a class of codes based on a simple figure of merit. Substituting H . We present here the results of the maximum likelihood (ML) formulation of MCMUD. Areas Commun.. † ˆ H = ZT † (TT † )−1 . † The approximation H H ≈ nR I nT motivates one of the design criteria shown in the figure. it is important to ensure that the rank of the matrix difference is not too small. (44) where the matrix PT ≡ T † (TT † )−1 T projects onto the VOLUME 15. where the channel matrix H ∈ nR ×nT contains the complex attenuation between each transmit antenna and receive antenna. † indicates the Hermitian conjugate. Spacetime trellis codes introduce spatial as well as temporal redundancy in the transmitted data. The probability density for a multivariate signalinthemean model is given by p(Z  R . Sel. pp. Reference 1. 3}) kth transmitter (It1. and tr indicates the trace of a matrix. Integrating over H motivates the other. bk determine the code. 2005 LINCOLN LABORATORY JOURNAL 113 . When 4 < rnR . T ) = e − tr ( Z − HT ) R ( Z − HT ) { } (42) which is satisfied by (43) ˆ assuming TT is not rank deficient. Alamouti. bk ∈ {0. The coefficients ak. AND CHAN MIMO Wireless Communication Notation: rank r matrix codeword C symbols can be bounded by (Bhattacharyya bound) pe ≤ e 4 N 0 . The number of receive antennas nR by number of samples. Z ∈ nR ×ns . In the example shown. 1. 1998. p→ e { } (41) π ns nR R ns . Maximizing the probability density with respect to H is equivalent to minimizing the tr{} in Equation 41. 16 (8). is given by Z = HT + N . employing a quasistatic narrowband MIMOchannel model. − E ( C1 − C 2 )† H†H( C1 − C 2 ) nT transmitters nR receivers Design criteria for spacetime trellis codes (4 ≤ rnR or rnR ≥ 4) pe ≤ Π λk [(C1 – C2)(C1 – C2]† k=1 r –nR E 4N0  –rnR E † 1 pe ≤ – e –nR 4N0 tr[(C1– C2)(C1– C2) ] 4 Example of spacetime trellis code (ak . ⊥ − tr ZPT Z† R −1 { † −1 } π ns nR R ns . FORSYTHE.• BLISS. It2 ) data: bit pair p=0  Σ v1 k It1– p ap + q=0 Σ It2– q bqk mod4 v2 xtk i xt k trellis coding codeword QPSK symbol modulation FIGURE A. H. the minimum leastsquares distance between codewords (represented by spacetime matrices Ck) can be maximized. In both cases r denotes the rank of the matrix difference C1 – C2 .M.
such that nA = 1. and PT = Ins − PT projects onto the orthogonal complement of the row space of T. † w † Z X TX A .2 (M11 − M1. † † † † † ˆ H A = ZTA (TA TA − TA TB [TB TB ]−1 TB TA )−1 † † † − ZTB [TB TB ]−1 TB TA † † † † × (TA TA − TA TB [TB TB ]−1 TB TA )−1 ⊥ † ⊥ † = ZPTB TA (TA PTB TA )−1 . NUMBER 1. We define T ≡ (TA TB )† to be a partitioned form of T. ˆ H = ZT † (TT † )−1 = T T† † † A A Z TA TB † T T B A M 11 . Because the first term is free of TA . X The nR × 1 vector wA contains the receive beamforming ˆ weights. By factoring PTB = X † X . demodulation is performed by minimizing the second term. 2 ⋅ . then the sec114 LINCOLN LABORATORY JOURNAL VOLUME 15.2 M1. where the signal associated with each transmitter.2 M2. X 1 1 ⊥ ˆ R X ≡ Z X Z † = ZPTB Z † . (48) The determinant can be factored into terms with and without reference to TA . T ) = R . and H A is the channel estimate associated with TA . where the symmetric identity matrix has a dimension of ns minus the number of rows in TB. FORSYTHE.• BLISS. given in Equation 43.H ns ⊥ ZPT Z † − ns nR ⊥  ZPT Z † −ns . ns ⊥ † w † ZPTB TA A = 1− = 1− where R indicates the derivative of R with respect to some internal parameter. X ns ns † † ˆ H A ≡ Z X TX (TX TX )−1 ⊥ † ⊥ † = ZPTB TA (TA PTB TA )−1 .2 M2.2. where the nA × ns matrix TA contains the signals associated with a particular subset of nA transmit antennas and the (nT – nA) × ns matrix TB contains the signals associated with all other ⊥ transmit antennas. the ML statistic for estimating T is given by πe max p(Z  R .2 )−1 . assuming that R is not rank deficient. an iterative ap† † proach is much more practical. R (46) ns (50) .2 . † † Z TA TB † M1.2 M1.12 M1. AND CHAN MIMO Wireless Communication ⊥ row space spanned by T. (45) ond term can be simplified and interpreted in terms of a beamformer † † Ins − PTX PZ X = 1 − (TX TX )−1 TX PZ X TX .1. is considered to be user A and is demodulated by minimizing Ins − PTX PZ X . Using these results. −1 † −1 † −1 − M2. the rows of X form an orthonormal basis for the complement of the row space of TB such that X X † = I . R X is the interferencemitigated signalplusˆ noise covariance matrix estimate.2 ) . It is worth noting that ˆ ˆ the form for H A is simply the column of H . H. 2005 (53) . If TA is a row vector. we can show that ⊥ ⊥ ZPT Z † = Z X PTX Z † . ns where ˆ ˆ w A = R −1 H A .2 M−. By defining Z X ≡ ZX † and TX ≡ TA X † . X X (49) By focusing on the first column and substituting in for ˆ M1. This relationship is satisfied when ⊥ † ˆ = ZPT Z . in turn. This form suggests an iterative approach.2 −1 (52) † † = Z TA TB † (M11 − M1. (47) (51) The determinant of is minimized to demodulate the signals for all transmitters jointly. ⊥ Z X PTX Z † = Z X Z † Ins − PTX PZ X . we can find H A . Although it is theoretically possible to use the statis⊥ tic ZPT Z † directly for demodulation. and M2. Maximizing with respect to an internal parameter of R gives ⊥ tr{R −1 ZPT Z † R −1 R } − ns tr{R −1 R } = 0 .2. † TA TB † TBTB −1 ≡ M1. associated with TA . M1.
AND CHAN MIMO Wireless Communication which is the same form found in Equation 51. The new channel estimate has dimension (nR ⋅ nδ f ⋅ nδt ) × nT . codes require relatively long block lengths to be effective. (58) Figure 3 shows a diagram for this demodulator (MCMUD). The data matrix is replaced with Z STF ≡ [Z † (δ t1. the model given in Equation 40 for the received signal is incomplete for many environments. they are particularly sensitive to Doppler offsets. previous signal estimates can be used to temporally remove contributions from other transmitters. δ f 2 ) X X Z † (δ t1. these approximations become increasingly valid as the number of samples in the block increases.) The second approximation reduces the computation cost of the projection operator. With this approach. adversely affecting the performance of the spatialbeamformer interpretation of the ML demodulator. FORSYTHE. the spatialbeamformer interpretation presented in Equation 50 is formally the same. In general. delay. δ f1 ) Z † (δ t1. we can reduce the approximation error at the expense of additional computational complexity. but all projectors are extended to include delay and Doppler spread. which is a form of multiuser detection. and frequencyoffset correlations. The spacetime frequencyadaptive beamformer component compensates for spatial. δ f nδ f X )Z † (δ t nδt . Suboptimal Implementation A variety of suboptimal but computationally more efficient variants are possible. Because turbo FIGURE 3. By iteratively decoding the signal. 2005 LINCOLN LABORATORY JOURNAL 115 . The receiver iteratively estimates the channel and demodulates the signal. NUMBER 1. MMSE Extension Because of the effects of delay and Dopplerfrequency spread. Demodulation is performed by maximizing the magnitude of the inner product of the beamformer output w † Z and the interferencemitigated reference sigA ⊥ nal TA PTB . The operator that projects on the orthogonal complement of the row space of M is given by Info bits VOLUME 15. † (57) which is a (nR ⋅ nδ f ⋅ nδt ) × ns matrix that includes possible signal distortions. The first computational simplification is found by noting that the normalization term of the channel estimate in Equation 51 can be approximated by ⊥ † ⊥ † ˆ H A = Z PTB TA (TA PTB TA )−1 ⊥ † † = Z PTB TA (TA TA )−1// 2 † † † × (I − [TA TA ]−1/ 2 TA PTB TA [TA TA ]−1/ 2 )−1 † × (TA TA )−1/ 2 ⊥ † † ≈ Z PTB TA (TA TA )−1 . but T remains the same.• BLISS. (55) Channel estimation This operator can be approximated by ⊥ PM ≈ where m indicates the mth row in the matrix. Diagram of a multichannel multiuser detector (MCMUD) spacetime turbocode receiver. Turbo encoder m Spacetime multiplexer ∏ I − M † † −1 m ( Mm Mm ) Mm . (56) Turbo decoder (We did not assume that TA is a row vector in the previous discussion. The MMSE beamformer is given by w STF = argmin w † Z STF − TX STF = (Z STF Z † )−1 Z STF STF † TX 2 (54) . Block for each transmitter Temporal subtraction Spacetimefrequency adaptive beamformer Spacetime demultiplexer ⊥ PM = I − M† (MM† )−1 M . δ f nδ f X )] . Extending the beamformer to include delay and Doppler corrects this deficiency. By repeating the application of this approximate projection operator.
The error bars indicate a range of plus or minus one standard deviation of the estimates at a given site. Larger dimension spacetimefrequency processing is possible by using estimates of the data. 40]. In the experiment. Phenomenological Experiment This section presents channelcomplexity and channelstationarity experimental results for MIMO systems. Channelprobing sequences using both four and eight transmitters were employed. These transmitters can be used independently. NUMBER 1. The receive antenna array was placed on top of a tall onestory building (at Brookline Street and Henry Street). The receive array had a total aperture of less than 8 m. For reasonably stationary channels. Experimental System The employed experimental system is a slightly modified version of the system used previously at Lincoln Laboratory [3.3 MHz with a length of 1. Because the number of training samples is relatively small. A variety of coding and interference regimes were explored for both moving and stationary transmitters. highperformance sixteenchannel receiver system that can operate over a range of 20 MHz to 2 GHz. 41]. the use of training data competes directly with information bits. The transmit array was located on the top of a vehicle within two kilometers of the receive array. surrounded by two. knowledge of the encoded signal is used to provide that training data. Scatter plot of the peaknormalized meansquared singleinput singleoutput (SISO) link attenuation a 2 versus link range for the outdoor environment near the PCS frequency allocation. The transmitters periodically emitted 1. The receiver can be deployed in the laboratory or in a stationary “bread truck. arranged as three subapertures of less than 1. The spacetimecoding results are discussed later in the article [39. AND CHAN MIMO Wireless Communication 20 Training Data In principle.” MIT Campus Experiment The experiments were performed during July and August 2002 on and near the MIT campus in Cambridge.and threestory buildings. FORSYTHE. The transmit systems can be deployed in the laboratory or in vehicles. 2005 10 a 2 (dB) 0 –10 –20 –30 0 500 1000 1500 2000 Link range (m) FIGURE 4.7 msec repeated .7sec bursts containing a combination of channelprobing and spacetimecoding waveforms. it is useful to use a small number of temporal and frequency taps during the first iteration. or as a single coherent group of eight transmitters. When operating coherently as a multiantenna transmit system.• BLISS. This provides an initial channel estimate for the spacetimefrequency adaptive beamformer. because the channel and information can be estimated jointly. However. Synchronization between transmitters and receiver and transmitter geolocation is provided by GPS receivers in the transmitters and receivers. supporting a bandwidth up to 8 MHz. for more quickly moving channels some training data is useful. We then discuss the variation of MIMO channels as a function of time and as a function of frequency. Furthermore.5 m each.79 GHz). Massachusetts. as two groups of four coherent transmitters. The transmit array consists of up to eight arbitrary waveform transmitters. the individual transmitters can send independent sequences by using a common local oscillator. The Lincoln Laboratory array receiver system is a 116 LINCOLN LABORATORY JOURNAL VOLUME 15. The channelprobing sequence supported a bandwidth of 1. These outdoor experiments were performed in a frequency allocation near the PCS band (1. there is no need for training data. The transmitters can support up to a 2MHz bandwidth. a small amount of training data is introduced within a frame (20%). We introduce the experiments and then discuss channel mean attenuation and channel complexity. Here. Different four.or eightantenna subsets of the sixteenchannel receiver were used to improve statistical significance. the estimate for the previous frame can be employed as an initial estimate for the demodulator.
CDFs are displayed for narrowband SISO.59 ± 0.01 0.87 ± 0.76 ± 0. The systematic uncertainty in the estimation of α caused by estimation bias.85 ± 0.7 7.72 ± 0.69 ± 0. given the model. NUMBER 1.2 0. AND CHAN MIMO Wireless Communication ten times.1 5.02 VOLUME 15.85 ± 0. and α estimate CDFs. The table includes the distance (range) between transmitter and receiver. Attenuation Figure 4 displays the peaknormalized meansquared SISO attenuation averaged over transmit and receive antenna pairs for a given transmit site for the outdoor environment.02 0.0 0. the velocity of the transmitter. 4 × 4.71 ± 0.68 ± 0.78 ± 0.0 0.02 0. 2005 LINCOLN LABORATORY JOURNAL 117 .67 ± 0. and the estimated α for the transmit site. Table 1 is a list of transmit sites used for these results.02.6 9.86 ± 0. Because of the spatial diversity.04 0.01 0.01 0.03 0. List of Transmit Sites Site Location Range (m) 150 520 430 420 250 20 270 470 140 1560 270 1070 240 250 780 550 410 370 Velocity (m/sec) 0.7 0.02 0. as we would expect. All four or eight transmitters emitted nearly orthogonal signals simultaneously. This reduction in variation demonstrates one of the most important statistical effects that MIMO links exploit to improve communication link robustness. eigenvalue cumulative distribution functions (CDF).78 ± 0. The uncertainty in the estimate is evaluated by using a bootstrap technique. The CDF values reported here are evaluated over appropriate entries from Table 1. the number of transmit antennas.0 0. For example.1 0.80 ± 0. Channel Complexity We present channel complexity by using three different approaches: variation in a2 estimates.78 ± 0. Uncertainty in α is determined by using the bootstrap technique [42]. FORSYTHE.0 Number of antennas 8 8 8 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 α 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Henry and Hasting Brookline and Erie Boston University (BU) BU at Storrow Drive Glenwood and Pearl Parking lot Waverly and Chestnut Vassar and Amherst Chestnut and Brookline Harvard Bridge BU Bridge Vassar and Mass Ave Peters and Putnam Glenwood and Pearl Brookline and Pacific Pearl and Erie Storrow Drive and BU Bridge Glenwood and Magazine 0. the variation in mean antennapair received power decreases dramatically as the number of antenna pairs increases.02 0.2 0.04 0.83 ± 0. and 8 × 8 MIMO systems.• BLISS.2 0.03 0.02 0. Figure 5 displays CDFs of a 2 = tr{HH† }/(nT nR ) estimates normalized by mean a 2 for each transmit site.0 10.0 0.01 0.01 0. if we wanted to Table 1. is less than 0.02 0.1 11.6 2.05 0.1 9.01 0.2 7.70 ± 0.79 ± 0.
we would have to operate the SISO link with an excess SISO SNR (a 2Po ) margin of over 15 dB.2 0 –15 –10 –5 0 5 a 2 (dB) FIGURE 5.2 0 0. CDF of narrowband channel eigenvalue distributions for 4 × 4 MIMO systems. while the multiple antennas are driven with the same local oscillator.• BLISS. FORSYTHE. .2 0 –40 –30 –20 –10 0 10 20 CDF of α operate with a probability of 0.6 0. the environment does move.6 0.6 0. CDF of narrowband channel eigenvalue distributions for 8 × 8 MIMO systems.0 4×4 8×8 α FIGURE 8. there are always some small relativefrequency offsets. this effect may have been exaggerated in Figure 8 because of the limited number of 8 × 8 sites available in the experiment. the 118 LINCOLN LABORATORY JOURNAL VOLUME 15. 2005 Eigenvalue (dB) FIGURE 7.0 0.8 0.6 0.4 0. allowing for overall shifts in attenuation. AND CHAN MIMO Wireless Communication 1. While the transmitters and receivers are physically stationary.8 0. eigenvalues of the stationary transmitter do vary somewhat. Cumulative distribution function (CDF) of channel a 2 estimates. We might interpret this to indicate that optimized spacetime codes should operate with a relatively high probability of success.79 for 8 × 8 sites.5 0.8 0.8 0. given the commercial grade transmitters. Furthermore.0 0. The example variation is given for transmit sites 7 and 14 from Table 1. While we might expect smaller variation in the 8 × 8 systems because of the much larger number of paths. where the form x ± y indicates the estimated value x with statistical uncertainty y estimated by using a bootstrap uncertainty estimation technique.0 0.4 0. Figure 8 shows the CDFs for α estimates.4 0. which is not accounted for in the figure. CDF of α estimates for 4 × 4 and 8 × 8 MIMO sys tems. This effect is particularly noticeable near busy roads.9 1. However. and 8 × 8 MIMO systems. CDF of channel eigenvalues 1.2 0 –40 –30 –20 –10 0 10 20 CDF of a 2 0. normalized by the mean a 2 for each site.4 0.7 0. the steepness of the CDFs is remarkable.8 4×4 8×8 SISO CDF of channel eigenvalues 1. Nonetheless.0 0. 4 × 4. 1. The mean values of α for each environment are 0. for SISO. Channel Stationarity Figure 9 displays the temporal variation of eigenvalues of HH† for stationary and moving transmitters.76 for 4 × 4 sites and 0. the eigenvalues of the moving transmitter vary significantly more than those of the stationary environment.6 0.9 to close the link. As we would expect. Some care must be taken in interpreting these figures because eigenvalues are not independent. The CDFs are evaluated over all site lists. In this figure the normalization is fixed. eig{HH†}. Figures 6 and 7 present CDFs of eigenvalues for 4 × 4 and 8 × 8 meansquaredchannelmatrixelementnormalized narrowband channel matrices. The MIMO systems received the added benefit of array gain. Eigenvalue (dB) FIGURE 6. NUMBER 1.
γ {H(t0 ). 0. for a stationary 4 × 4 MIMO system. is used for all time samples. 0. CDF of time variation of powerweighted mean cos 2 θ .4. 0. for a moving 4 × 4 MIMO system. 0.05 0.6 0.0 γ{H(t0).8 0.4 0.1.10 0.04 γ{H(t 0 ). estimated at t = 0. Time (sec) cos 2 θ.6.3. 0.5. 2005 LINCOLN LABORATORY JOURNAL 119 .2.02 0.10 0.H(t)}. and 0.15 1. Conversely. FORSYTHE.3.02 0.2 0 0.5.9. CDF of time variation of powerweighted mean cos 2 θ.3 0.6 0.1. The variation is indicated by using the metric presented in Equation 36. 0. and the moving transmitter is located at site 14. This variation indicates that the eigenvector structure varies significantly. 0.8 0. 0.7 0.2. 0.06 Time (sec) FIGURE 12. NUMBER 1. FIGURE 10. Contours of CDF probabilities of 0.1 0. an example of the time variation of the powerweighted mean cos2θ metric (from Equation 36). the stationary transmitter is located at site 7. Because there is little variation.7.H(t)} . Over the same period the stationary transmitter is relatively stable. for stationary and moving 4 × 4 MIMO systems. The graph displays contours of CDF probabilities of 0. The significant variation of the moving transmitter is an indication that implementing an informed transmitter MIMO system would be very challenging for the moving transmitter.9 are displayed.0 1.1 While the movingtransmitter eigenvalues fluctuate more than those of the stationary transmitter.8 0. In the example. but might be viable for some stationary MIMO systems. γ {H(t0 ). Figures 11 and 12 display CDFs for stationary and moving transmitters.H(t)} . Eigenvalues (λ) of HH† as a function of time for (a) stationary and (b) moving transmitters.8.7. 0.04 0.8. H(t )} Time (sec) (b) FIGURE 9. FrequencySelective Fading Figure 13 gives an example of the frequency variation of the powerweighted mean cos 2θ.5 0. the values are remarkably stable in time. 0. γ {H(t0 ).15 Time (sec) (a) 20 10 eigenvalues tends to be more stable. Example time variation of powerweighted mean VOLUME 15. H(t )} 0.• BLISS. while the distribution of 1.4 0.0 Time (sec) FIGURE 11. H(t)} 0. displayed in Figure 10.2 Stationary Moving 0.6 0. 0.9 0.2 0 0. all curves are compressed near a γ value of 1.05 0.4. AND CHAN MIMO Wireless Communication 20 10 0 –10 –20 0 0.06 0 0.6. and 0. varies significantly for the moving transmitter within 10 msec.0 γ{H(t0). 0. 0.06 0. In the example. the stationary transmitter is located at site λ (dB) λ (dB) 0 –10 –20 0 0.2 0. 0. The same overall attenuation.4 0.04 0.02 0.
8 0.2.9 0. In other words. Figure 14 shows the CDF of the frequencyselective channel variation.4 0.4 0. 1. C(16. 0.1. four element subarrays are chosen at random to provide multichannel receivers. 0. and the resulting frequencyselective fading. to safely operate with the narrowband assumption. Example of frequencyselective variation of the powerweighted mean cos 2 θ . are both a function of environment and link length. Measured and simulated results in bit error rate probability for a spacetime lowdensity paritycheck (LDPC) code at a spectral efficiency of 2 bits/sec/Hz.2 –600 –400 –200 0 200 400 600 Frequency (kHz) FIGURE 13. 0. 0. The signal has a pulseshaped QPSK modulation and bandwidth of about 100 kHz.8.• BLISS.20 Simulated Biterror probability Frequency (kHz) FIGURE 14.5. CDF of frequencyselective variation of the 0.2 –600 –400 –200 0 0. 4) 4 × 4 0. About 4 to 5 dB excess Eb /N0 is required to reliably complete the link.05 0 0 2 4 6 8 10 12 Excess Eb /N0 (dB) FIGURE 15.25 γ{H(f0). bandwidths less than 100 kHz should be employed. The graph displays contours of CDF probabilities of 0. 2005 0.6 0.3.H(f )} . There is typically no identifiable lineofsight component in the propagation. Propagation delay spreads are typically several microseconds and Doppler spreads are at most a few hundred hertz. not necessarily pointing at the sources. Bit error rates are evaluated for an ensemble of 4 × 4 MIMO systems. This sensitivity indicates that there is significant resolved delay spread and that. We note that delay spread. The elements are oriented in various directions. Each estimated channel matrix suppports a theoretical capacity that can be in excess of 2 bits/sec/Hz.10 7.3 0. 0. AND CHAN MIMO Wireless Communication 1. γ {H(f0 ).9. The MIMO wireless link is realized with four cohered transmitters located on a stationary 120 LINCOLN LABORATORY JOURNAL VOLUME 15. some care must be taken in interpreting this result.and threestory residential buildings and some commercial buildings of similar heights in an urban setting. H( f )}. H(f )} Measured 0. γ {H(f0 ). Agreement between measured and simulated results are good to about 1 dB. FORSYTHE.5 0. Consequently. The code used is half rate with length 1024.15 powerweighted mean cos θ . The estimated channel matrices are used in the simulation to model propagation.6.4.6 0. For the example below. and 0. Coding provides a spectral efficiency of 2 bits/sec/Hz. SpaceTime LowDensity ParityCheckCode Experiments A lowdensity paritycheck code over GF(16) provides the basis of the example of experimental and simulated results shown in Figure 15. H(f)} 0. The resulting scale factor is used to evaluate the excess (beyond Shannon) Eb /N0 associated with the (unscaled) channel matrix.1 200 400 600 van several hundred meters away from an array of receivers situated on a onestory building. 0. The matrix is scaled until it supports a capacity of exactly 2 bits/sec/Hz.H(f )} .0 γ{H(f0 ). 0. 2 • Spectral efficiency 2 bits/sec/Hz 0. NUMBER 1. Relatively small frequency offsets induce significant changes in γ {H( f 0 ).7.7 0. . The environment consists predominantly of two.8 0.0 0. The receiver consists of sixteen channels fed by lowgain elements with wide azimuth beamshapes.
Channelprobing sequences and four. We discuss two examples with differ ent transmit locations. The receive antenna array was placed on top of a onestory building (at Brookline Street and Henry Street) surrounded by two. 2005 LINCOLN LABORATORY JOURNAL 121 . Experimental Parameters Outdoor experiments were performed in a frequency allocation near the PCS band. and then we summarize the performance of the MIMO system with stationary transmitter and receiver in a dynamic environment. Figure 18 shows the eigenvalues of the noisenormalized interferenceplusnoise spatial covariance matrix. The “noise” eigenvalues of the jammer spatial eigenvalue distribution Cambridge Receive array MIT Transmit array Boston University FIGURE 16. the channel transfer matrices are used both for the simulated results and for the computation of excess Eb /N0. Two wideband jammers were transmitting at a range of 100 m. For this example. Different fourantenna subsets of the sixteenchannel receiver were used to improve statistical significance. Example 2. AND CHAN MIMO Wireless Communication MIMO links can be evaluated. Total transmit power was approximately 100 mW. NUMBER 1.5 dB excess Eb /N0 is required to complete the link at 2 bits/sec/ Hz. We begin by discussing the experimental parameters. Figure 17 shows the geometry of the experiment. for an even more complicated environment. In both examples the link does not have line of sight. As the figure shows.and threestory buildings. which is related to the excess spectral efficiency (beyond 2 bits/sec/Hz) predicted by a capacity bound. we introduced artificial frequency offsets at the transmitters. we describe performance results with a mobile transmitter and multiple strong interferers. The channel transfer matrices have a random structure that varies from subarray to subarray. radiated from 0dBi antennas. Simulations agree to within about 1 dB. including the locations of the transmitter and receiver. Figure 16 shows the geometry of the experiment. quadraturephaseshiftkey (QPSK) signals were transmitted on four antennas at 123 × 103 chips per second for a total data rate of 246 kb/sec. The outdoor experiments were performed during July and August 2002 on and near the MIT campus. given the measured channel transfer matrix. A variety of coding and interference regimes were explored for both moving and stationary transmitters. using the spacetime code discussed earlier. Example 1. about 4. Example 1: map of MIMO communication experiment near the MIT campus. in a comparatively stationary environment. Additionally. Each jammer was received at a jammertonoise ratio (JNR) of approximately 25 dB. The transmitter was moving at 10 m/sec at a range of 500 m.and eighttransmitter spacetime codes were transmitted. While the transmitter is stationary. A 160kHz spectral limit was enforced by using a rootraisedcosine pulse shaping. These errors were within ±80 Hz. This section reports on the outdoor performance results of spacetime turbo codes for 4 × 4 MIMO systems. To simulate the effects of local oscillator errors. the transmitter was located in the parking lot at Boston University (University Road and Storrow Drive) with about a halfkilometer separation between the transmitter and receiver. VOLUME 15. In this example. SpaceTime TurboCode Experiments In this section we present the experimental performance of a spacetime turbo code. using a sixteenchannel receiver. Traffic on Storrow Drive is typically heavy and the posted speed limit of 45 mph is generally misinterpreted as the minimum allowed speed.• BLISS. Figure 15 shows symbolerror probability as a function of excess Eb /N0. the environment is nonstationary because of the traffic. For the examples discussed in this article. FORSYTHE.
AND CHAN MIMO Wireless Communication 40 Transmitter 25 mph Receive array Relative power (dB) 30 20 10 0 Jammers 25 dB JNR 100 m 1 2 3 4 FIGURE 17. Both the delay and the Doppler spread affect the design and performance of the receiver. a 2 is the meansquared link attenuation. Doppler taps: {–1. Starting with the highest complexity on the first iteration increases the probability of converging to the wrong solution. FORSYTHE. including the locations of the transmitter.5 5. 1}). however. 2/3.0 a Po (dB) FIGURE 19. in quiet regions (no nearby traffic). the Doppler spread was found to be within ±180 Hz. Because the channel contains significant Doppler spread. the experimental performance is 10 –1 Bit error rate 10 –2 10 –3 10 –4 Spatial beamforming Trainingdatabased SFAP STFAP MCMUD 3. a 2Po . the spatial beamformer performs poorly. coarse trainingdatabased spacefrequency adaptive beamforming (SFAP).0 5.5 6. 2/3. must bootstrap up in complexity iteratively. 0. 2bit/sec/Hz spacetime turbo code as a function of mean SISO signaltonoise ratio (SNR) (a 2Po ) for a Boston University transmit location. Performance improves with receiver complexity. 2. the algorithm. Doppler beamforming is required in this environment. Figure 19 displays the results for four detection variations: 1.0 3. Either of these explanations presents additional challenges to the receiver. are slightly higher than we would naively expect. This behavior is probably an indication of either delay spread or nonstationarity in the received jammer signal. and where Doppler taps are represented in resolution cells (60 Hz). using adaptive spatial beamforming. 0. 0. given the 0dB noise normalization. For the moving transmitter in experimental example 2.0 2 4. temporal taps: {–1/2. Eigenvalue distribution of the noisenormalized interferenceplusnoise spatial covariance matrix. Doppler taps: {–4/3. Eigenvalue number FIGURE 18. The number of delay and frequency taps in the adaptive processor depends upon the phenomenology. 0. and MCMUD with spacetimefrequency adaptive beamforming.5 4. 4. Doppler taps: {–4/3. Bit error rate of 4 × 4. temporal taps: {–1/2. –2/3. MCMUD with spacetimefrequency beamform122 LINCOLN LABORATORY JOURNAL VOLUME 15. 4/3}. Here. 1/2}). –2/3. Trainingdatabased adaptive spatial beamforming (three turbo iterations). 4/3}. Delay spread was found to be less than ±4 µsec. Spacetimefrequency beamforming employing decisiondirected channel estimation without multiuser detection (three turbo iterations. . For the stationary environment. 0. For the stationary transmitter near heavy traffic in experimental example 1. We note that experimental performance is essentially the same as was found in simulations. Experimental Example 1 Bit error rates for various detector alternatives are reported as a function of mean singleinput singleoutput (SISO) SNR. With the relatively long block lengths of the turbo code.• BLISS. 2005 ing (three turbo iterations. and jammers at a fixed jammertonoise ratio (JNR). 1/2}). spacetimefrequency adaptive beamforming (STFAP) employing decisiondirected channel estimation. Coarse trainingdatabased spacefrequency beamforming (one turbo iteration. Furthermore. the Doppler spread was found to be within ±150 Hz. Example 2: map of MIMO communication experiment near the MIT campus. receiver. NUMBER 1. Here a spacetimefrequency adaptive processor is employed. 3. no Doppler spread was detected.
Tony Tavilla. Finally. and Naveen Sunkavally of MIT and Nick Chang of the University of Michigan for their help with the experiment. Furthermore. Performance bounds assuming either an informed transmitter or an uninformed transmitter were presented for flatfading. Bit error rate VOLUME 15. A channel phenomenology parameterization was introduced. The authors would also like to thank the excellent Lincoln Laboratory staff involved in the MIMO experiment. phenomenological. assuming that all power of the transmit array is transmitted from a single transmit antenna. Lee Duter. Experimental demodulation performance results were presented for a variety of environments. This is the average SINR at a given receive antenna. and Dorothy Ryan of Lincoln Laboratory for her helpful comments. FORSYTHE. Bob Devine. Peter Priestner.• BLISS. and local oscillator errors. In the presence of the jammer. We note that this experimental system in this difficult environment operates at an SINR that is 25 dB better than the informationtheoretic SISO bound. Experimental Example 2 The experimental data includes the effects of two highpower wideband jammers. Bob Downing. Summary In this article we addressed informationtheoretic. Two spacetime coding techniques were introduced. 2bit/sec/Hz spacetime turbo code using the MCMUD receivver with spacetimefrequency beamforming as a function of mean SISO signaltointerferenceplusnoise ratio (SINR). and receiver issues for MIMO communication. Jeff Nowak. 2005 LINCOLN LABORATORY JOURNAL 123 . and Andy McKellips. The bit error rate is displayed in terms of the mean SISO signaltointerferenceplusnoise ratio (SINR). the results indicating that the observed channels can be typically characterized by high degrees of complexity. Bit error rate of 4 × 4. John Mann. Experimental performance of this spacetime turbo code for a stationary transmitter in the absence of interference is discussed elsewhere [39]. for environments with transmitters on moving vehicles. Figure 20 shows the bit error rate of the spacetime turbo code using the MCMUD receiver. one based on LDPC and the other on turbo codes. frequencyselective. The effectiveness of the receiver is due in part to its ability to compensate for delay and frequency spread. Furthermore. and operates probably at least 35 dB better than a practical SISO system. and jammed environments. Experimental phenomenological results were reported. NUMBER 1. The MCMUD employs a spacetimefrequency adaptive processor that uses a fourantenna receiver with temporal and frequency taps that cover a range of ±4 microseconds and ±200 Hz. there is only approximately a 3dB loss in a 2Po performance compared to the performance in an environment without jammers. coding. the channel varies significantly on a time scale less than 10 msec. similar to the simulated performance of the best spacetime codes. We also thank Ali Yegulalp of Lincoln Laboratory and Vahid Tarokh of Harvard University for their thoughtful comments. Acknowledgments The authors would like to thank Peter Wu of Lincoln Laboratory for his help developing the spacetime turbo code. a moving transmitter. including those with wideband jammers. in particular Sean Tobin. the authors would like to thank the MIT New Technology Initiative Committee for their support. the MIMO system (using the MCMUD receiver) operated dramatically better than SISO systems. AND CHAN MIMO Wireless Communication 10 –1 10 –2 10 –3 10 –4 –21 –20 –19 –18 –17 SISO SINR (dB) FIGURE 20.
Chicago. and A. 687–693. 2001. and A. Areas Commun. A. Inf. Bliss. 2001. 24. 27.O. 2041–2052. Theory 45 (5). Zheng and D. 7–11 May 2001.L. Utah. and A. 2003. on Signals.. 1999. 13. “Performance of SpaceTime Codes over a FlatFading Channel Using a SubspaceInvariant Detector.” IEEE Comm. Bliss. IEEE Int. D.. K. V.L. 3. Forney. A. “Design of CapacityApproaching Irregular LowDensity ParityCheck Codes. “Space–Time Block Codes: A Maximum SNR Approach. Theory 49 (9).C. 2002. Forsythe. Inf. G. Valenzuela. Communications Conf. Paulraj. Jelinek. Pacific Grove. “Performance Evaluation for Scattering MIMO Channel Models. Record ThirtyThird Asilomar Conf.” 34th Asilomar Conf. 140– 152. Microwave Mobile Communications (Wiley. pp. Ganesan and P.B. Glavieux. pp. 1451–1458. W. 1993. 1565–1569. Calif. Calif. pp.C. Louis. Sweldens. and R.” Eur. Wittneben. “MIMO Communication Capacity Using Infinite Dimension Random Matrix Eigenvalue Distributions. Trans. on Signals. Bliss. Farrokhi. Forsythe. Inf.–1 Nov. 41–59. Calif.” Proc. Raviv. Paulraj. Speech.” ThirtyFourth Asilomar Conference on Signals.W. A. Lett. “Optimal De LINCOLN LABORATORY JOURNAL VOLUME 15. A. Stefanov and T. “Capacity of FlatFading Channels Associated with a SubspaceInvariant Detector. Sel.” IEEE Commun. 23–26 May 1993. 9.” IEEE Trans. and J. Systems & Computers 1. L.–1 Nov. K. Marzetta and B. pp. Inf. pp. Pacific Grove. Chan. “Multichannel Adaptive Beamforming and Interference Mitigation in Multiuser CDMA Systems. 85–87. Bliss. on Signals.O. 124 pp. 15. Geraniotis. Areas Commun.R. Foschini. A. 2000. “Capacity of a Mobile MultipleAntenna Communication Link in Rayleigh Flat Fading. 1980). IEEE Vehicular Technology Conf. Fitz. 2461–2464. Duman. Calif. “Turbo Decoding as an Instance of Pearl’s ‘Belief Propagation’ Algorithm. 506–510. pp. D. Theory 47 (4). M. Telatar.” Proc.–1 Nov. 6.F. Areas Commun.M. 32. 5 (3). Sel. and A. Pacific Grove.W. Yegulalp. 24–27 Oct. 5 (2). Hochwald and W. Theory 45 (1). Bölcskei and A. 1998. Jafarkhani.P. Richardson.” IEEE Trans. Liu.M. pp. 20. Mo. 411–416. “Basestation Modulation Diversity for Digital SIMULCAST. R. 2000. Inf. Bahl. H. “LinkOptimal Space–Time Processing with Multiple Transmit and Receive Antennas.” IEEE Trans. Calderbank. 848–853. 23–28 Sept. AND CHAN MIMO Wireless Communication R EFER ENCE S 1. Sel.” IEEE Trans. LowDensity ParityCheck Codes (MIT Press. K. Amsterdam.” Proc. 5. T.” IEEE J. 29. 24–27 Oct. pp. 2001. “Full Rate Space–Time Turbo Codes. Systems & Computers 2. K. N.C. 1991). 58–60. and R. R. New York.R. 10 (6). 1999. “MIMO Wireless Communication Channel Phenomenology. 969–974. Thomas. 16. Forsythe. Bölcskei. Pacific Grove.W. 3. 1998. “Diversity and Freedom: A Fundamental Tradeoff in MultipleAntenna Channels. pp. Mass. Marzetta. 34.J. 16 (2). IEEE Int. 1775–1779. Conf. H.” 36th Asilomar Conf. McEliece.A. pp. 35. A. Forsythe.” Conf. pp.” IEEE Trans.” Bell Labs Tech. 215–219. pp. 7. pp.M. 2000. and N. IEEE Vehicular Technology Conf. T. Theory 46 (2). Hochwald. Chung.–1 Nov.W. B. Calif.D. pp. Calif. Cheng. 2000.0045 dB of the Shannon Limit. 139–157.” IEEE Trans. 2001.F. Signal Process.W.M. S. New York.” IEEE Trans. 8. D. and J. pp. 16 (8). 1974).J. 1996. 2001. 1999. pp. 19. Monzingo and T. Yegulalp. T. pp. Hochwald. Miller. Systems & Computers 1.W. 29 Oct.” Proc. 750–755.. 1999. Tarokh. and O. 29 Oct. Forsythe.Y.” ThirtyFifth Asilomar Conf. 52 (8). Cambridge. Alamouti. D. on Signals. “Diversity for DirectSequence Spread Spectrum Using Multiple Transmit Antennas. N. St. 1076–1093. 1456–1467.W. Foschini. Systems and Computers 1. “Unitary Space–Time Modulation for MultipleAntenna Communications in Rayleigh Flat Fading.. Lozano. 19 (5). Tse. Inf. L. Calderbank. “Environmental Issues for MIMO Capacity. “Differential Unitary SpaceTime Modulation. Jr.E. Paulraj. MacKay.J.. Hero. Forsythe. 585–595. 14. 543–564. Pacific Grove.1999. 10. 23. Cover and J. H. pp. “On the Design of LowDensity ParityCheck Codes within 0. Salt Lake City. New York. NUMBER 1. 48 (12) 2000.Y. Communications Conf. Calif. pp. and A. 19–22 May 1991. Tarokh. Gesbert. 2128–2142. pp. Telecommun. Pacific Grove. 4. Takeshita. Hassibi and B. Pacific Grove. Urbanke. F. V. “Layered SpaceTime Architecture for Wireless Communication in a Fading Environment When Using MultiElement Antennas. “Turbo Coded Modulation for Wireless Communications with Antenna Diversity.” Proc. 19–22 Sept. Wireless Communications and Networking Conf. pp. F. 1999. Systems & Computers 1. “MIMO Environmental Capacity Sensitivity. A. Hero. D. 764–768. M. 11.A. on Signals.. Hochwald and T. Systems & Computers 1. Systems and Computers 1.. “Space– Time Block Codes from Orthogonal Designs.W. 1 (2). Geneva. 28. 748–752. on Acoustics. and A. Urbanke. Stoica..W. Introduction to Adaptive Arrays (Wiley. Systems & Computers 1.” IEEE Trans. K. Shokrollahi. I. 12. 1963). 2001. 1650–1656. Com. D. “Performance of SpaceTime Codes in the Presence of Spatial Fading Correlation.• BLISS. 2005 . pp. Sharma and E. Weerackody. R. 3–6 Nov. 4–7 Nov.. S. 31.. 18.W. Cocke.L. FORSYTHE. “Capacity of MultiAntenna Gaussian Channels.A. G. 29 Oct.” ThirtyThird Asilomar Conf. 30. D.” IEEE Trans. 969–980. 2000.L. Gallager. pp.M. Swindlehurst.” IEEE Trans. T. pp. B. “Analyzing the Performance of the SpaceTime Block Codes with Partial Channel State Feedback. Jakes. 3. J.R. V. Y. 1064–1070. 36. C. “Joint Transmit and Receive Optimization for High Data Rate Wireless Communication Using Multiple Antennas. 3.J. 17. and Signal Processing 4.” Proc. pp. Geneva. Theory 47 (2). 2001.M. 619–637.J. Bliss.J. Inf. Richardson. “A Simple Transmit Diversity Technique for Wireless Communications. 25. pp.” ThirtyFourth Asilomar Conf. pp. pp. G. Keller.W.M. Lett. Antennas Propag. Elements of Information Theory (Wiley. Gore. G. H. 2073–2082. K. 22. Berrou.. R. 2.G. “Space–Time Codes for High Data Rate Wireless Communication: Performance Criterion and Code Construction.J. 23–26 May. 2. “Near Shannon Limit ErrorCorrecting Coding and Decoding: TurboCodes.N. 29 Oct. “HighRate Linear SpaceTime Codes. B.” IEEE J. pp. 50 (9). 21.M. Seshadri.” IEEE J.F. pp. Calif. 2004. on Signals.A..W. 1998. Pacific Grove. J.” ThirtyFourth Asilomar Conf.A. IEEE Int. D. 1362–1366. 744–765. J. and R. Thitimajshima.. and C.J. 26. Chang.. 33. Theory 44 (2). Sampath and A. on Signals. and P. 2002.
. 42. 38. Bliss. IEEE Int. J. D. Keller and D. D. 39. “Robust MIMO Wireless Communication in the Presence of Interference Using Ad Hoc Arrays. Inf.M. 946–951. 10 (2). 41. 1997. Seattle. E. 2001. 43. Systems & Computers 2. Communications Conf. Bliss. 2000. Calif. Efron. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference (Morgan Kaufmann..M. pp. P. IntegerBased Turbo Decoder. “Cellular and PCS Propagation Measurements and Statistical Models for Urban Multipath on an Antenna Array. pp. 2. 1382–1385. Robertson. J.• BLISS. 284–287. Pearl. pp. Lab. 25–29 Nov. Hoeher. 99–126. 40.” Linc. Villebrun. 2. 18–22 June 1995. pp. K. B. 44. 1974. Pisuk. 2005 LINCOLN LABORATORY JOURNAL 125 . Calif.” ThirtySixth Asilomar Conference on Signals. Forsythe. pp. pp. P. 32–36. San Antonio. 2002. Tex.W. 3–6 Nov. pp. AND CHAN MIMO Wireless Communication 37. H.” Proc. NUMBER 1. and A.” Proc. Bliss.M. Wu. 1009–1013. Wu and S. Pacific Grove. 1982). coding of Linear Codes for Minimizing Symbol Error Rate.” MILCOM 2003. P..” IEEE Trans. “Multichannel Multiuser Detection of SpaceTime Turbo Codes: Experimental Performance Results. Low Power. “Comparison of Optimal and SubOptimal MAP Decoding Algorithms Operating in the Log Domain. Mass. FORSYTHE.” Proc. “Implementation of a Low Complexity. Philadelphia.W.–Y. Cambridge.W. Oct. 1988). 2003.. 1343–1348. C. The Jackknife. 16–17 Mar.H.W. “Utilizing Waveform Features for Adaptive Beamforming and Direction Finding with Narrowband Signals. San Mateo. IEEE Global Telecommunications Conf. the Bootstrap and Other Resampling Plans (Society for Industrial and Applied Mathematics. VOLUME 15. 2000 IEEE Sensor Array and Multichannel Signal Processing Workshop. Chan.. and P. Theory 20 (2).
2005 .E. FORSYTHE. errorcorrection coding. As a member of the superconducting magnet group. degrees. He received S. degrees in physics from the University of California at San Diego. He received M. studying the twophoton production of hadronic final states.E. primarily for geolocation. is a senior staff member in the Advanced Sensor Techniques group. and a B.• BLISS. and on parameter estimation techniques and bounds. His work on SAR imaging involves techniques for resolution enhancement and interference rejection for foliagepenetration systems. and other forms of parameter estimation) beyond those attainable by nonexploitive techniques. She has previously worked with implementation of synthetic aperture geolocation of cellular phones. . and digital matchedfilter design and performance.S. jamresistant synchronization techniques. In 1978 he joined Lincoln Laboratory. is an associate staff member in the Advanced Sensor Techniques group. NUMBER 1. Previously employed by General Dynamics. he designed avionics for the AtlasCentaur launch vehicle.B. from MIT.M. adaptive sensorarray processing. She received MSEE and BSEE degrees in electrical engineering from the University of Michigan. 126 LINCOLN LABORATORY JOURNAL .S. primarily for communication systems.D. where he has worked in the areas of spreadspectrum communication. algorithm development for multichannel multiuser detectors. and information theoretic bounds on MIMO communication systems. is a a staff member in the Advanced Sensor Techniques group. His current research includes ultrawide bandwidth communication. Her interests are in channel phenomenology. and synthetic aperture radar (SAR) imaging. MIMO communication channel phenomenology. direction finding. she has worked on the implementation of MIMO channel parameterization. VOLUME 15. in electrical engineering from Arizona State University. His doctoral work was in highenergy particle physics. and investigating innovative techniques for latticegaugetheory calculations. and Ph. and performed research and development of faulttolerant avionics. and S. Most recently. In the area of adaptive sensorarray processing he helped develop a number of signalprocessing algorithms that exploit waveform features to achieve levels of performance (beamforming. searching for bound states of gluons. AND CHAN MIMO Wireless Communication . geolocation. geolocation techniques using vector sensor arrays. His work on spreadspectrum systems includes electromagnetic modeling (geometric theory of diffraction) of antennas mounted on an airframe. At Lincoln Laboratory he focuses on multiantenna adaptive signal processing. multipleinput multipleoutput (MIMO) radar concepts. both in mathematics. he performed magnetic field calculations and optimization for highenergy particleaccelerator superconducting magnets.