Copyright (c) 2013 IEEE. Personal use is permitted.

For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
1
Adaptive Uncoupled Termination for Coupled
Arrays in MIMO Systems
Reza Mohammadkhani, and John S. Thompson
AbstractIn this paper, we present an adaptive uncoupled
matching network to compensate any performance degradation
of compact Multiple-Input Multiple-Output (MIMO) systems
due to antenna mutual coupling effects and channel variations.
This method varies the antenna terminal loads per symbol-
block, based on a random search algorithm, to nd an optimum
termination network that maximises the performance metric,
such as received power and capacity. It uses the received signals
(voltages across the resistances of the terminal loads) to estimate
or calculate the performance metric, in order to include the
total effects in the optimisation process. By applying the random
search algorithm, our method does not require any knowledge of
the array parameters such as impedance matrix, or derivatives of
the performance metric(s) unlike other optimisation techniques.
We demonstrate this scheme by performing simulations to
optimise the capacity of a 3 3 MIMO considering different
propagation scenarios. We observe signicant mean capacity
improvements (more than 2bits/s/Hz) for all assumed propagation
scenarios, when receive array antennas are spaced close as 0.05
and terminated non-identically.
Index TermsMIMO, mutual coupling, impedance matching,
adaptive termination.
I. INTRODUCTION
M
IMO systems by using multiple antennas at both trans-
mit and receive sides of the wireless link, provide
better link quality and higher data-rates [1], [2]. However,
applying MIMO technology to small wireless devices with
an array inter-element spacing less than half a wavelength
leads to antenna mutual coupling effects which degrade the
MIMO performance [3][5]. Many studies have presented
impedance matching techniques to compensate the electro-
magnetic coupling effects in the coupled array. Some ap-
plied a complex coupled matching network [6] known as the
Multiport-Conjugate Match (MCM) [4] which is shown to
be able to achieve optimal performance. Such a matching
network is complicated and difcult to construct practically
and it only offers a narrowband matching performance [7].
These problems motivate researchers to work on simpler
uncoupled termination approaches [8][12] which are easier
to implement and can achieve near-optimal performance with
a wider bandwidth [7], [12].
The major part of this work was performed at the University of Edinburgh
and it was nancially supported by the Ministry of Science, Research and
Technology, Iran . This paper is based in part on reference [31] which was
presented at the ISWCS 2010 conference.
R. Mohammadkhani is with the Department of Electrical Engineering, Uni-
versity of Kurdistan, Sanandaj, Iran. e-mail: (r.mohammadkhani@uok.ac.ir).
J. S. Thompson is with the Institute for Digital Communications
(IDCoM), University of Edinburgh, Edinburgh EH9 3JL, UK. e-mail:
(john.thompson@ed.ac.uk).
Existing uncoupled matching solutions consider
numerically-optimum loads to maximise the received
power [8], [13], or the capacity [9], [10] of compact MIMO
systems for a given propagation scenario. These methods
require a priori knowledge of the propagation channel and an
accurate mutual coupling model which is difcult to measure
in practice. Modelling of the mutual coupling is a challenging
problem for coupled array applications. Existing matching
network approaches apply a method suggested by [14] to
form an impedance matrix or an equivalent scattering matrix
in order to model the mutual coupling effects for the coupled
array in either transmit or receive mode. However, despite its
simplicity, the accuracy of this method for the receive array
is questioned by some other studies [15][18] and alternative
approaches [19][21] are suggested to model the mutual
coupling. Furthermore, [22] reveals the inuence of near-eld
scatterers on the mutual coupling, which is ignored by the
existing coupling compensation methods.
Assuming known mutual coupling and channel matrices, an
optimal uncoupled matching network is obtained in [10] by op-
timising the mean capacity with respect to the antenna terminal
loads for identical termination impedances. It is extended
to non-identical antenna terminations by [23], [24]. When
we have no knowledge of the mutual coupling matrix, these
studies are not able to nd the optimal termination network.
Furthermore, these methods are not designed to track the time
variations of the propagation channel. Therefore, we propose
an adaptive uncoupled termination method which compensates
the effects of the propagation channel and mutual coupling
together by directly dealing with the received signals. We
consider both identical and non-identical termination cases.
We do not use a known mutual coupling matrix, but it will be
included implicitly by using the voltages across the antenna
terminations to estimate the channel matrix. The algorithm
changes the receive array termination for each symbol-block
and calculates the MIMO performance metric (e.g. received
power or capacity). By comparing the current metric value
and the previously optimum one (the maximum value over
previous blocks), the algorithm selects the better impedance
network as the optimum termination until that iteration. It then
adds a random step to the terminal impedances to generate a
new candidate impedance network for the next block. This
process will be repeated until the algorithm converges to an
optimum termination network, in terms of received power or
capacity.
The rest of this paper is organised as follows. Section II
introduces our system model. Then, the proposed adaptive
termination approach is described in section III. After that,
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
2
section IV provides numerical results to verify the proposed
method, assuming either a perfectly known channel matrix,
or including imperfect channel state information (CSI) due to
estimation error and Doppler effects. Section V concludes the
paper.
Throughout this paper, boldface lowercase and uppercase
symbols refer to vectors and matrices, respectively and the
italic version of the symbols represent the entries of the
vector or matrix. The notations ()
H
, E{}, det() denote the
matrix Hermitian (conjugate-transpose), the expectation, and
the matrix determinant operators, respectively. The symbol I
M
also represents the identity matrix of size M M.
II. SYSTEM MODEL
In this section we provide a MIMO system model taking
into account the time-variation of the propagation channel,
antenna mutual coupling, and the channel estimation error
issues. We also address the capacity estimation techniques at
the end of this section.
Consider a MIMO system of M transmit and M receive
antennas, communicating through a at-fading Rayleigh chan-
nel. Using a discrete-time baseband model, the input-output
relationship at time instant k is given by
y[k] = H[k]x[k] +n[k] (1)
where x[k] C
M
is the transmit signal vector, y[k] C
M
is the receive signal vector, H[k] C
MM
denotes the
channel gain matrix including the effect of transmit and receive
antennas, and n[k] C
M
is the additive white complex
Gaussian noise vector at the receiver with zero-mean and
covariance matrix
2
n
I
M
, where I
M
is an M M identity
matrix.
To describe the time-variation of the channel, we use the
Gauss-Markov process model [25] as follows
H[k + 1] =

1 H[k] +

W[k] (2)
where W[k] C
MM
includes complex Gaussian entries
with zero-mean and unit-variance. The entries of W are
independent across rows, columns and time indices k. The
parameter R, 0 1 is introduced to control the
coherence time of the channel. In [25], some practical ranges
for have been calculated by tting the above Gauss-Markov
model into real systems measurements. For instance
3 10
7
10
4
, for a slow-fading indoor
environment with mobile speed 1-5 km/h and carrier
frequencies from 800 MHz to 5 GHz.
10
4
1.8 10
3
, for a slow-fading outdoor
environment with mobile speeds of the order of 5 km/h
and carrier frequencies from 800 MHz to 5 GHz.
= 1.8 10
2
, for a fast-fading outdoor environment
with mobile speed 50 km/h and carrier frequency 5 GHz.
A. Mutual Coupling
As mentioned in section I, the existing mutual coupling
models may not be exact for the coupled receive array.
However, they can reveal the total behaviour of the system
and the impact of terminations on the compact MIMO per-
formance. Therefore we apply a general model described in
[10]. We note that the proposed adaptive termination algorithm
is unaware of the system model and it works based on the
nal received signals (voltages across the resistances of the
termination loads), or any other parameters estimated from
them. Therefore, it does not strongly depend on the mutual
coupling model unlike [10] which uses an explicit matrix
model.
Let us begin with a case that array elements of both transmit
and receive sides are spaced sufciently far apart to avoid mu-
tual coupling (practically over half-wavelength inter-element
spacing). Assuming a Rayleigh-fading propagation channel
and using the popular Kronecker
1
structure, the channel matrix
can be expressed as [27]
H =
1/2
R
H
w

1/2
T
(3)
where H
w
is a M M matrix with independent identically
distributed (iid) and zero-mean unit-variance complex Gaus-
sian entries, and
T
and
R
are the transmit and receive
spatial channel correlation matrices, respectively.
Now consider compact receive wireless devices whose
element spacing is less than 0.5 and which possesses an
impedance matching network Z
L
. Assuming identical half-
wavelength dipole antennas for transmit and receive arrays,
the general MIMO channel matrix using Z-parameters can be
represented as [10]
H
mc
= 2r
11
R
1/2
L
(Z
R
+Z
L
)
1

1/2
R
H
w

1/2
T
. .
H
nc
R
1/2
T
(4)
where Z
R
and Z
T
are the receive and transmit ar-
ray impedance matrices, respectively with diagonal entries
Z
R,ii
, Z
T,ii
, (i = 1, , M) being the self impedances and
the off-diagonal entries Z
R,ij
, Z
T,ij
, (i = j, and i, j =
1, 2, , M) dening the mutual impedances. The real parts
of Z
L
and Z
T
are denoted by R
L
and R
T
, respectively and
r
11
= R
T,11
. For the special case of no mutual coupling
(Z
R,ij
= Z
T,ij
= 0) and matching all the receive and
transmit antennas to their self-impedances, the channel matrix
simplies to the term represented by H
nc
.
Throughout this work, we assume that the transmit antennas
are separated sufciently, self-impedance matched and that

T
= I
M
. Thus the channel matrix H
mc
from (4) can be
simplied as [10]
H
mc
= 2

r
11
R
1/2
L
(Z
R
+Z
L
)
1

1/2
R
H
w
. .
H
nc
(5)
This is the model we will use in the numerical studies in this
paper.
B. Channel Estimation
To exploit the capacity benets of the MIMO technology,
accurate channel knowledge at the receiver and/or transmitter
1
This model has deciencies for large number of antennas [26], but we use
it for 33 MIMO in our simulation results for which is still reasonably valid
.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
3
is usually required. One of the most popular channel estima-
tion approaches is the training-based estimator [28] which can
directly estimate the channel matrix including antenna mutual
coupling effects, from knowledge of the received signals and
(transmitted) training signals. In this paper, we do not go into
the details of the channel estimator, rather we assume that an
estimated channel matrix

H
mc
[k] is available. Considering the
channel estimation error, we can write [29]
H
mc
[k] =

H
mc
[k] +E[k] (6)
where E[k] is the channel estimation error matrix, whose
entries are zero-mean complex Gaussian random variables
with variance
2
E
. Throughout this work, we assume that each
block includes some training-symbols and at least one channel
estimate is provided per symbol-block from the knowledge of
transmitted training symbols.
C. Capacity
In general, the MIMO capacity is given by [30]
C = max
Q: Tr(Q)P
E
H
_
log
2
det
_
I
M
+
1

2
n
H
mc
QH
H
mc
__
(7)
where Q = E{xx
H
} is the input covariance matrix, and P is
the total transmit power. In this paper, it is assumed that the
transmitter has no knowledge of the channel, so the transmit
power will be divided equally among all the transmit antennas,
i.e., Q = (P/M)I
M
[1]. Thus, the resulting ergodic capacity
can be written as
C = E
H
_
log
2
det
_
I
M
+

M
H
mc
H
H
mc
__
(8)
where = P/
2
n
is the average signal-to-noise ratio at the
receiver. We note that the fading process {H
mc
[k]} is assumed
to be known for the receiver. For a system with a known
channel estimate

H
mc
and a known estimate variance
2
E
,
a lower bound of (8) is given by [29]
C
lower
= E
H
_
log
2

I
M
+
1
1 +
2
E
P

H
mc

H
H
mc

_
(9)
If we split pilot symbols into k
p
groups among each block, the
channel estimate

H
mc
can be expressed as an average of the
channel estimates

H
1
,

H
2
, ...,

H
k
p
from sub-blocks as follows
[28]

H
mc
=
1
k
p
k
p

i=1

H
i
. (10)
D. Received Power
Assume y(t) is a vector of the received voltages across the
resistance of the receive array terminal loads at time instant t.
Then the received power for ith antenna can be written as
P
r,i
= E
_
y
i
(t)y

i
(t)
r
Li
_
/P
0
, (i = 1, , M) (11)
where r
Li
represents the real part of the load terminal z
Li
for antenna i, and P
0
is the power received by a conjugate
matched isolated antenna which is used to normalise the
MIMO received power. An estimation of (11) for each symbol
block can be calculated by averaging time instances of the
received power over a block of L symbols as described in [31].
Fig. 1. Block diagram of a coupled receive/transmit array applying a match
network.
III. TERMINATION STRATEGIES
In the literature, applying a coupled or multiport matching
network [4], [6], [32] is introduced as a decoupling network
which may decouple the signals from closely spaced antennas
completely. Using network analysis, a matching network can
be placed between the coupled array and antenna loads (re-
ceive mode) or excitation sources (transmit mode) such that
it can be conjugate-matched from one side to the antenna
array and from the other side to the loads/sources. This
is shown in Fig. 1. However, constructing such a network
due to the required interconnections between all ports of
the matching network is complicated and it also offers a
narrowband matching performance [7].
Alternatively, uncoupled termination networks have been
widely investigated, either by terminating all receive anten-
nas identically [8][10] or individually applying non-identical
termination loads [23], [24]. These studies assuming a known
model comprising the channel matrix and the mutual coupling
matrix. Then they either numerically or mathematically seek
an optimum termination network which maximises the re-
ceived power or the capacity. There are two problems with the
existing studies. First, we need to know the array impedance
matrix Z
R
in equation (5). We would need to perform a
calibration step to do this, which is expensive and time-
consuming. Secondly, even if Z
R
is known, it may change
with time due to temperature, environment, and any other
parameter changes. This would lead to an incorrect solution
for the optimum termination network Z
L
.
Existing studies are model-based matching solutions which
assume that every parameter in the system is known except for
the matching impedance network, whereas for practical cases
the only known parameters are the termination impedance
network and the received signals (or any parameter estimate
derived from them, such as

H
mc
).
Recently, two studies [31], [33] appeared roughly at the
same time which suggest applying an adaptive matching
network for compact MIMO systems. The authors in [33]
present a general idea of having an adaptive (multiport or
uncoupled) impedance matching network to remedy any per-
formance degradation resulting from the antenna mutual cou-
pling and/or variations of the propagation channel. Although
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
4
several possible ideas are suggested to cover the practical
issues of implementing such matching network, no particular
approach is proposed to select the matching impedances. On
the other hand, [31] applies a combination of a performance
metric estimation (e.g., received power or capacity) from the
received signals, and a random search algorithm to control
an adaptive (identical) uncoupled matching network. It shows
that the proposed algorithm can optimise the performance
under different propagation scenarios while the optimisation
complexity is reduced by applying the random load selection
among all possible ranges of the antenna loads.
Here, we extend the work of [31] to the non-identical
matching impedance case and propose an adaptive uncoupled
termination which only relies on the knowledge of the re-
ceived signals and a training-based channel estimate

H
mc
per
symbol-block. We investigate the effect of different practical
issues such as estimation error, and time-variation of the
channel matrix on the matching performance.
In what follows, we rst review the conventional termi-
nations: characteristic impedance match, and self-impedance
conjugate match. Then, we describe the proposed adaptive
uncoupled termination which optimises the compact MIMO
performance (e.g. received power or capacity).
A. Characteristic Impedance Match
All receive antennas are terminated in a characteristic
impedance Z
0
. Therefore, the receive load network Z
L
at (5)
is set to Z
0
I
M
. This means we have no matching network for
this case.
B. Self-Impedance Conjugate Match
In this termination case, each receive antenna is terminated
in the conjugate of its self-impedance. In other words, Z
L
=
diag(Z

R
). This termination would result in maximum power
transfer to the load network when there is no mutual coupling
between array elements [7].
C. Adaptive Termination
The previous terminations in subsections III-A and III-B do
not consider the mutual coupling effects. Here, we propose
an adaptive, uncoupled termination network to counteract the
effects of mutual coupling and propagation channel changes,
and to optimise the MIMO performance. This technique uses a
diagonal matrix Z
L
and varies it per symbol-block based on a
random search algorithm, and then calculates the performance
metric(s) to examine the effect of the current termination on
the mutual coupling and therefore system performance.
A simplied schematic diagram of the proposed method is
shown in Fig. 2. This method comprises two major parts. One
is an estimator which provides instances of the channel matrix
H
mc
(or the statistical moments of the channel) including the
actual effects of the mutual coupling and the time-varying
propagation channel. The other part is an adaptive matching
network controller which selects the loads to counteract the
performance degradation due to the mutual coupling and/or
changes of the channel, based on the performance metric
estimates (capacity or received power).
Fig. 2. Schematic diagram of the proposed adaptive termination approach
for a compact MIMO system.
One may ask why we have not applied any other optimisa-
tion techniques such as Gradient based or Newton-Raphson
methods rather than a random load selection? Before we
answer this question, let us clarify our optimisation problem
for a M M MIMO system of M uncoupled transmit
antennas and M coupled receive antennas. We can express
the desired performance metric, either the received power or
the capacity, in our optimisation problem as a function of
received signals, i.e., f(y) = f(H
mc
x + n) (we drop the
time index k for simplicity). According to the results from
the previous matching solutions [3][11], the total channel
matrix H
mc
which includes the mutual coupling effects, and
therefore f(y), depend on the termination impedance matrix of
the coupled array, denoted by Z
L
. Denoting the channel matrix
with no mutual coupling by H, we can write our optimisation
problem as below
max
Z
L
f(H, Z
R
, Z
L
). (12)
In other words, we would like to nd a terminal impedance
network Z
L
which compensates any performance degradation
due to the mutual coupling or channel matrix changes. Here,
we are interested in uncoupled terminations where Z
L
is a
diagonal matrix written as
Z
L
=
_
_
_
_
_
z
L1
0 0
0 z
L2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 z
LM
_
_
_
_
_
. (13)
Suppose the channel matrix H and the array impedance
matrix Z
R
, during the optimisation period are xed or they
vary slowly. Then, the optimisation problem (12) can be
simplied as
max
z
L1
, ,z
LM
f(z
L1
, , z
LM
) (14)
where z
Li
= r
Li
+ jx
Li
for i = 1, , M with r
Li
R
+
and x
Li
R denoting the real (resistance) and imaginary
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
5
Fig. 3. An overview of the adaptive impedance matching algorithm.
(reactance) parts of z
Li
, respectively. We consider two dif-
ferent kinds of uncoupled terminations: (i) identical load-
ing, which all diagonal entries are equal to a load z
L
=
r
L
+ jx
L
, and (ii) non-identical loading in which terminal
loads z
L1
, z
L2
, , z
LM
are individually tuned to optimise
the compact MIMO performance.
We let z = [r
L1
, x
L1
, , r
LM
, x
LM
]
T
be the vector of
optimisation variables. Then, we try to nd an optimal z
opt
in which f(z
opt
) has the maximum possible value of f.
This could be obtained by producing a maximising sequence
z
(m)
, (m = 1, ) where f(z
(m+1)
) > f(z
(m)
), from the
following iterative equation [34]
z
(m+1)
= z
(m)
+
(m)
z
(m)
(15)
where m denotes the iteration number, the scalar
(m)
0
is the scale factor, and the vector z
(m)
determines the
direction of optimisation at iteration m. We need to specify
z in accordance with each optimisation technique. Fig. 3
shows an overview of the proposed idea of the adaptive
impedance matching technique. The algorithm starts from an
initial termination load such as characteristic impedance load,
and then estimates the total channel matrix H
mc
from the
knowledge of training/pilot symbols. Then the performance
metric, the capacity or the received power, is calculated to
predict the optimal load network Z
L
for the next symbol block.
This process continues until it converges to an optimal load
network which maximises the performance metric.
1) Gradient ascent method: Let us begin with the identical
loading case, and then we extend the result to non-identical
termination. Suppose that the function f is differentiable with
respect to the variables r
L
and x
L
. Using the Gradient ascent
method [34], we can nd a solution for (14) by substituting
z = f(z) into (15) as follows
z
(m+1)
= z
(m)
+
(m)
f(z
(m)
) (16)
where z
(m)
= [r
(m)
L
, x
(m)
L
]
T
is the value of the vector of
variables at iteration number m, and
(m)
> 0 here is called
step size. The term f(z
(m)
) represents the gradient of f with
respect to the entries of z at iteration m, given by
f(r
L
, x
L
) =
_
f
r
L
,
f
x
L
_
T
. (17)
As stated previously, usually no parameter is known for prac-
tical systems except the received signals, and the termination
network (which can be controlled by the optimisation algo-
rithm). Therefore the gradient term at (16) has to be estimated.
This estimate could be calculated from the following forward
differences [35]
f
r
L

f(r
L
+ r, x
L
) f(r
L
, x
L
)
r
(18a)
f
x
L

f(r
L
, x
L
+ x) f(r
L
, x
L
)
x
. (18b)
It means that in order to calculate an estimate of f(r
L
, x
L
)
at each iteration m, at least three channel estimates corre-
sponding to the load impedances z
L1
= r
L
+ jx
L
, z
L2
=
(r
L
+r)+jx
L
, and z
L3
= r
L
+j(x
L
+x) are required. In
general, this method requires 2M+1 channel estimates at each
iteration m, for the use of adaptive non-identical impedance
matching in a M M.
We numerically examine the Gradient-based method for the
capacity of a 3 3 MIMO system with different r and x
values in the next section. This method can converge to the
optimum load impedance if we choose a proper step size
(m)
for each channel realisation. Furthermore, the convergence rate
of the Gradient-based method depends on the quantisation
level of z
(m)
.
We also examined the Newton-Raphson method which uses
the second and rst order derivatives of f. We found that this
method does not converge to the optimum load network under
any circumstances. This might be due to complexity of the
second order derivative approximations.
2) Random Search: In this subsection, we explain how the
adaptive algorithm uses a random search (motivated by random
phase selection [36] and random walk [37] algorithms) for
the optimum load impedance. Assuming an identical loading
network, and describing the receive array load network (13)
as Z
L
= z
(m+1)
L
I
M
at iteration (m + 1), the impedance load
z
(m+1)
L
is obtained from the following equation:
z
(m+1)
L
= z
(m)
opt
+ z
(m)
, (m = 0, 1, , N
s
1) (19)
where z
(m)
opt
is the optimum load at mth iteration (initialised
by z
(0)
opt
= Z
0
= 50), the non-zero scalar z
(m)
de-
notes a complex step size randomly selected from the set
{r, jx, (r jx)} with an equal probability, and
N
s
is the number of load network variations. An overview
of the choice of step size z
(m)
at iteration m is shown in
Fig. 4. In practice, N
s
can be estimated from the experimental
data, or in a similar way the algorithm can stop after having
no change in the optimal load network for a specic number
of iterations while a small step size is used.
At each iteration (m+1), the capacity or received power is
calculated (or estimated from the knowledge of the received
signals) and compared to the previous value corresponds
to z
(m)
opt
. The impedance which corresponds to the higher
capacity/received power is held as the optimum load z
(m+1)
opt
for the (m + 1)th iteration. Fig. 5 illustrates the owchart of
the proposed algorithm for identical loading.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
6
Fig. 4. Complex step size z
(m)
at iteration m is selected randomly from
the set {r,jx,(r jx)} with an equal probability.
Fig. 5. Flowchart of the proposed adaptive termination approach.
In this work, we have considered several values for the step
sizes r, x from the range of 1 12. Larger values of
r and x result in a faster convergence but lower steady-
state performance compared to the smaller step sizes. So to
obtain improved performance, the algorithm can start with a
large step size and then decrease the step size after having
a completed specic number of load variations. We describe
the effect of choosing the step size value in more detail in
section IV-A.
The algorithm can also be extended to the non-identical
loading case. For any antenna#i, (i = 1, , M), the
load impedance z
Li
can be tuned individually according to
(19). This can be represented as (20) where z
(m+1)
Li
, (i =
1, , M), are the selected terminal impedances at (m+1)th
iteration, z
(m)
opt,i
, (i = 1, , M) are the optimum loads for the
mth iteration, and z
(m)
i
, (i = 1, , M) are independent
random complex step sizes at mth iteration. Let us summarise
the algorithm to optimise the capacity C as follows.
1) Initialise the array termination load z
(0)
L
= Z
0
(or Z
L
=
Z
0
I
M
);
2) Estimate the corresponding C from (8) for perfect CSI
or (9) for imperfect CSI;
3) Set C
opt
= C, and z
opt
= z
(0)
L
(or Z
opt
= Z
(0)
L
for
non-identical loading);
4) Calculate the next termination for the following symbol-
block from (19) or (20);
5) Estimate the corresponding C from (8) or (9);
6) If (C > C
opt
) then (z
opt
= z
L
, and C
opt
= C);
otherwise z
L
= z
opt
and go back to step 4.
Some matching solutions [8], [13] have found the optimum
match by maximising the received power. This algorithm
can also be applied to maximise the received power P
r
by
substituting P
r
from Section II-D for C, and P
r,opt
for C
opt
.
However, most commonly it is of interest to increase the data
rate and the capacity of MIMO wireless systems rather than
the received power.
IV. NUMERICAL RESULTS
In this section we provide a numerical study to evaluate the
matching performance of the proposed adaptive termination
approach for a 3 3 MIMO system with a coupled receive
array. Here, we consider the MIMO capacity optimisation,
but the result can similarly be extended to the received
power as well. We assume linear arrays of identical half-
wavelength dipoles are applied for both transmit and receive
sides. The receive array antenna element spacing is assumed
to be d = 0.05, where is the wavelength, and the transmit
antennas are considered to be placed far enough such that we
neglect the mutual coupling effect at the transmit side. Mutual
coupling impedances for the receive array are calculated using
the electromotive force (EMF) method [38]. Following our
assumptions in section II-A, we apply the channel model of (5)
including a time-variant term H
w
[k], under different scattering
distributions: 2D uniform, and a truncated 2D Laplacian
dened by the mean angle of incidence

and an angular
spread of 40

for a signal-to-noise-ratio (SNR) 20dB at the

receiver.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
7
_
_
_
_
_
_
z
(m+1)
L1
0 0
0 z
(m+1)
L2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 z
(m+1)
LM
_
_
_
_
_
_
=
_
_
_
_
_
_
z
(m)
opt1
0 0
0 z
(m)
opt2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 z
(m)
optM
_
_
_
_
_
_
+
_
_
_
_
_
_
z
(m)
1
0 0
0 z
(m)
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 z
(m)
M
_
_
_
_
_
_
(20)
0 20 40 60 80 100 120 140
100
50
0
50
6 7
8
9
1
0
1
0
r
L
()
x
L

(

)
10.27
9.90
Mean Capacity (bits/s/Hz)
10.48
Fig. 6. Contour plot of the mean capacity versus real and imaginary parts
of the antenna load impedance z
L
for uniform scattering distribution at the
receiver. Three points are marked for: self-impedance match z

11
(square),
Z
0
= 50 (triangle), and optimum load z
opt
(circle) for the identical loading.
0 50 100 150 200 250 300 350 400 450 500
0.99
1
1.01
1.02
1.03
1.04
1.05
1.06
1.07
Normalised capacity: C/C
z0
N
o
r
m
a
l
i
s
e
d

c
a
p
a
c
i
t
y
C
inst
C
z11
C
z0
= 0.3 = 0.003
= 3x10
5
Fig. 7. Normalised capacity of one channel realisation using the Gradient-
based adaptive identical impedance matching technique for different step size
values.
A. Perfect CSI at the receiver
Let us begin with a simple case that the channel matrix
is perfectly known at the receiver. We generate 200 random
time-invariant channel realisations assuming a uniform scat-
tering distribution such that the entries of the receive spatial
correlation matrix
R
are given by

R,ii
= 1 (i = 1, 2, , M) (21a)

R,ij
= J
0
(
2d

) (i = j), (i, j = 1, 2, , M). (21b)

Fig. 6 illustrates the average capacity as a function of real
Fig. 8. Normalised capacity values for 10 channel realisations using the
adaptive impedance matching by applying the Gradient algorithm with the
step size = 0.01.
0 500 1000 1500
1
1.05
1.1
1.15
1.2
1.25
1.3
(a)
M
e
a
n
(
C
Z
L
/
C
Z
0
)


0 50 100 150 200 250 300 350 400 450 500
1
1.01
1.02
1.03
1.04
1.05
1.06
1.07
Iteration, m
M
e
a
n
(
C
Z
L
/
C
Z
0
)
(b)
r = x = 1
r = x = 2
r = x = 4
r = x = 8
r = x = 12
Fig. 9. Normalised mean capacity versus iteration number for the adaptive (a)
non-identical, and (b) identical uncoupled terminations considering r = x
values from the set {1, 2, 4, 8, 12}().
and imaginary parts of the antenna load for a non-adaptive
identical termination case. It shows three different terminat-
ing cases of self-impedance match (marked by a square),
Z
0
= 50 match (marked by a triangle), and the numerically
optimum match (marked by a circle) and the corresponding
average capacity values. We observe that the average capacity
has a maxima for a specic identical termination called the
optimum load [9], [10]. The optimal load can be extended
to the non-identical case as well [23]. Although these papers
revealed the relevance of compact MIMO capacity/received
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
8
0 500 1000 1500 2000 2500 3000
9.5
10
10.5
11
11.5
12
12.5
iteration
M
e
a
n

C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)


Adaptive nonidentical
Adaptive identical
Selfconjugate match
50 match
10.27
9.901
12.39
10.5
Fig. 10. Mean capacity of a 33 MIMO system for the following matching
conditions: adaptive non-identical, adaptive identical, self-impedance conju-
gate, and Z
0
= 50 matched termination networks.
1.15 1.2 1.25 1.3 1.35 1.4
0
5
10
15
20
25
30
C/C
Z0
P
e
r
c
e
n
t
a
g
e
1 1.05 1.1 1.15 1.2
0
10
20
30
40
50
C/C
Z0
Identical
Loading
Nonidentical
Loading
(a) C/C
Z
0
instances
5 6 7
0
25
50
75
r
L1
()
P
e
r
c
e
n
t
a
g
e
37 36 35 34
0
25
50
75
100
x
L1
()
P
e
r
c
e
n
t
a
g
e
0 1 2
0
25
50
75
100
r
L2
()
0 1 2 3
0
25
50
75
100
x
L2
()
5 6 7
0
25
50
75
r
L3
()
36 35 34
0
25
50
75
100
x
L3
()
(b) Non-identical Loads: z
Li
= r
Li
+ jx
Li
, (i = 1, 2, 3)
10 20 30 40 50
0
10
20
30
40
50
60
P
e
r
c
e
n
t
a
g
e
r
L
()
36 35 34 33 32 31
0
10
20
30
40
50
60
x
L
()
(c) Identical Loads: z
Li
= r
L
+ jx
L
, (i = 1, 2, 3)
Fig. 11. Histogram plots of (a) C/C
Z
0
instances for both adaptive
termination cases, and the real and imaginary parts of the termination loads
for (b) non-identical, and (c) identical terminations at iteration number 3000.
5 dB
3 dB
1 dB
1 dB
3 dB
45

225

90

270

135

315

180

1
2
3
(a) z
0
= 50 match
5 dB
3 dB
1 dB
1 dB
3 dB
45

225

90

270

135

315

180

1
2
3
(b) Self-conjugate match
5 dB
3 dB
1 dB
1 dB
3 dB
45

225

90

270

135

315

180

1
2
3
(c) Adaptive identical match
5 dB
3 dB
1 dB
1 dB
3 dB
45

225

90

270

135

315

180

1
2
3
(d) Adaptive non-identical match
Fig. 12. Array element patterns in dB for (a) z
0
= 50 match, (b) self-
impedance conjugate match, (c) adaptive identical, and (d) adaptive non-
identical impedance matching networks. Zero degrees corresponds to the array
broadside.
2
4
6
8
10
E
i
g
e
n
v
a
l
u
e
#
1


0
0.5
1
1.5
E
i
g
e
n
v
a
l
u
e
#
2
0 500 1000 1500 2000 2500 3000
2
4
6
8
10
12
x 10
4
E
i
g
e
n
v
a
l
u
e
#
3
iteration
Nonidentical Loading
Identical Loading
Z
0
= 50
Selfconjugate match
no mutual coupling
Fig. 13. Eigenvalues of (H
mc
H
H
mc
) versus iteration number for different
antenna terminations.
power to the termination, they require prior knowledge of
the propagation channel and mutual coupling model for a
numerical search over all possible termination loads for the
optimal load. Furthermore, these studies perform the optimisa-
tion process over the mean values of the performance metric(s)
with respect to the termination load(s), whereas our proposed
algorithm seeks the optimum load network for each channel
realisation.
Now, we apply the adaptive uncoupled terminations for the
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
9
assumed propagation channel scenario. We rst investigate
using the Gradient algorithm for the adaptive impedance
matching technique. Fig. 7 shows the convergence behaviour
of this method for one channel realisation applying different
values of the step size . It can be seen in Fig. 7 that too large
a value of (e.g. 0.3) leads to instability in the algorithm per-
formance while making too low (e.g. 310
5
) leads to very
slow convergence performance. The value = 0.03 seems to
give the best compromise of stability and convergence
Since the mutual coupling model in our problem is un-
known, we can not separate the channel matrix H from the
total channel matrix H
mc
which includes the mutual coupling
effects and is measured/estimated from the received signals.
However, assuming H is known, we applied the following
normalisation
H
n
=

NM
H
||H||
F
, (22)
to further investigate the Gradient-based adaptive impedance
match. As we see from Fig. 8, even for the normalised H
n
,
there is a range of convergence characteristics, so the step size
could even be adjusted further according to the entries of the
channel matrix H- though such a solution is beyond the scope
of this paper. For practical implementation, the available set of
impedance values may be quantized to a nite set of distinct
impedances, which will limit the accuracy of the converged
solution and possibly reduce performance. We recall that for
identical impedance loads, at each iteration we need at least
3 channel estimates.
Alternatively, we use the random search algorithm for the
adaptive (identical) impedance matching technique. We inves-
tigate the effect of having different values of step size z by
choosing xed r = x values from the set {1, 2, 4, 8, 12}.
At each iteration, the algorithm applies an estimate of the
channel matrix including the realistic mutual coupling effects
and selected antenna loads. This is provided from the received
signal and knowledge of the training signals. Convergence
results of the normalised mean capacity for adaptive identical
and non-identical terminations versus the number of changes
of the termination network (mth iteration) are illustrated in
Fig. 9. The most signicant feature for both termination cases
is that applying a smaller step size leads to a higher steady-
state performance but requires a longer convergence time.
Assuming the same step size for both termination cases,
the identical adaptive termination rises sharply and reaches
its steady-state about ve to ten times faster than the non-
identical termination. However, as depicted in Fig. 9, the
adaptive non-identical termination can achieve much higher
steady performance albeit with a slower convergence time, for
all step sizes except for the case with r = x = 12.
In Fig. 10, the mean capacity of the system is shown for four
different matching networks: adaptive non-identical, adaptive
identical, self-impedance conjugate, and Z
0
= 50 matched
termination networks. Steady state values of the mean capacity
for all termination strategies are shown on the right hand side
y-axis. In order to achieve a higher steady-state performance
in a shorter convergence time, the adaptive algorithm applies
a variable step size for both termination cases according to
the results reported in Fig. 9. Therefore, initially r = x is
assumed to be 8 for both termination cases and then reduced
to 4, 2, and 1 at iterations 100, 400, and 1000 for the non-
identical case, and at iterations 30, 80, and 200 for the identical
case respectively. Any change in the propagation channel or
mutual coupling would result in a new optimal Z
L
which may
change with time. In order to track these possible changes,
the algorithm can also decide to update the selected optimum
values of C
(m)
opt
, Z
(m)
opt
and to increase the step size, whenever
(C
(m)
C
(m)
opt
) is not positive or C
(m)
opt
is not changed for a
large number of steps. These occurrences may indicate that
the optimal Z
L
has changed and the receiver should nd the
new best solution. As we observe from Fig. 10, both adaptive
terminations nd an optimum load network which results in a
higher mean capacity than the conventional terminations. The
adaptive identical case reaches to its steady state performance
just after 85 iterations while for the non-identical case it takes
about 1000 iterations to achieve to a point above 99% of
the steady-state. However, the latter case achieves about 2
(bits/s/Hz) higher capacity gain in the expense of a longer
convergence time.
Unlike earlier studies, the adaptive termination algorithm
performs the optimisation process over the capacity instances
rather than the mean capacity. For further investigation of the
adaptive algorithm behaviour, percentage histogram plots of
the normalised capacity instances C/C(Z
0
), and the real and
imaginary parts of the termination loads for both adaptive
termination cases at iteration number 3000 are shown in
Fig. 11. The y-axes are relative frequencies for the total of 200
channel realisations. Fig. 11a shows that the capacity instances
are improved for both adaptive terminations compared to the
conventional Z
0
= 50 match by at least 20% for the non-
identical case and by 5% for the identical loading case. As
can be seen from the results in Fig. 11b and 11c, more than
75% of the channel realisations are optimised by selecting a
non-identical terminal network as Z
L
= diag(6 j35, 1 +
j1, 6 j35), and about 50% of them by having an identical
network equals to Z
L
= (32j35)I
33
. The antenna pattern
of the array elements for the above optimal load networks
are plotted in Fig. 12 with = 0

as the array broadside.

The results conrm the beamforming behaviour [12], [23] and
decoupling effect of the antennas by having optimal uncoupled
terminations, which is more effective for the non-identical
case.
Furthermore, we examine the eigenvalues of H
mc
H
H
mc
using the above terminations in addition to the case with no
mutual coupling as shown in Fig. 13. The gure clearly shows
that the non-identical termination tends to improve stronger
eigenvalues to achieve higher capacity, while the identical case
has a similar trend for all eigenvalues.
Since the 3rd eigenvalue of the channel in Fig. 13 is
small in value, and the impedance load of the middle receive
antenna for the adaptive non-identical match is small, one
may argue that the middle receive antenna does not affect
the performance. In other words, the middle antenna can be
removed from the system with no change to the performance.
This is similar to antenna selection for the receive array in
the presence of mutual coupling in [39] when there is no
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
10
0 200 400 600 800 1000 1200 1400 1600 1800 2000
10
10.1
10.2
10.3
10.4
10.5
10.6
10.7
10.8
10.9
iteration, m
C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)


Adaptive nonidentical
Adaptive identical
Selfconjugate match
z
0
match
Fig. 14. Mean capacity of a 3 2 MIMO with the element spacing of
d = 0.1 at the receiver, for the following matching conditions: adaptive
non-identical, adaptive identical, self-impedance conjugate, and z
0
= 50
matched termination networks.
5 dB
3 dB
1 dB
1 dB
3 dB
45

225

90

270

135

315

180

1
2
Fig. 15. Element pattern of the receive antennas with d = 0.1 for the
assumed 3 2 MIMO using the adaptive impedance matching techniques.
Results for the non-identical and identical matching scenarios are similar .
impedance matching solution. To investigate this issue, we
simulate the performance of a 3 2 MIMO considering the
same propagation scenario as the above, except the element
spacing between the receive antennas which is doubled here,
i.e. 0.1. Fig. 14 shows the mean capacity of this system
over 200 channel realisations using the following matching
networks: adaptive non-identical match, adaptive identical
match, self-conjugate match and z
0
= 50 match. We see
that removing the middle receive antenna does not change
the performance improvement of the adaptive identical match
(about 6% above the capacity for z
0
= 50), while it reduces
the performance of the adaptive non-identical match. To fur-
ther the investigation, the element patterns for both adaptive
matching networks are shown in Fig. 15. Both identical and
non-identical match result in the same optimum loads and
consequently similar antenna pattern for such a receiver with
two antennas. We note that the 2nd antenna in Fig. 15 is in the
place of the 3rd antenna used in Fig. 12. Comparing Figures 12
and 15 reveals the impact of the middle antenna (which
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
10
12
14
16
18
C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
8
10
12
14
16
18
C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
8
9
10
11
12
13
14
Element Spacing (d/)
C
a
p
a
c
i
t
y

(
b
i
t
s
/
s
/
H
z
)


Adaptive Nonidentical
Adaptive Identical
Selfimpedance Match
50 Match
(a)
(b)
(c)
Fig. 16. Mean capacity as a function of element spacing for different
matching methods and three different propagation scenarios: (a) Uniform,
(b) Laplacian ( = 0

, = 40

), and (c) Laplacian( = 90

, = 40

).
is terminated in a small load impedance, i.e. roughly short
circuited) on antennas 1 and 3 such that the performance is
improved signicantly. It can be interpreted as a beamforming
process where antenna currents are weighted by employing
proper impedance loads.
To complete this part, we plot the mean capacity at itera-
tion number 4000 for the above terminations versus antenna
element spacing in Fig. 16, under the following scattering
distributions: (a) uniform, (b) Laplacian centred at

= 0

(broadside), and (c) Laplacian with

= 90

(endre). We
observe that for an element spacing d < 0.2, the non-
identical adaptive termination gives a much higher perfor-
mance improvement compared to other terminations but at
the cost of a longer convergence time, while the identical
adaptive termination has a similar performance to the self-
impedance conjugate match. For d > 0.2, both adaptive
impedance matching techniques achieve roughly the same
capacity performance as the self impedance conjugate match,
and higher than the characteristic impedance (Z
0
= 50).
However, the self conjugate match requires knowledge of the
self-impedances of antennas, whereas the adaptive algorithms
do not. As stated in [4], [40], [41], the self-impedances of
coupled antennas are different than the case when antennas
are isolated, and that measuring these impedances is difcult
in practice. Furthermore, antenna elements of practical arrays
may not be identical. For such cases, the adaptive non-identical
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
11
impedance matching technique can be applied to improve the
performance.
B. Imperfect CSI at the receiver: channel estimation error
We extend our investigation to the case that the channel
matrix is unknown for the receiver and it has to be estimated.
We assume a training-based estimation approach is applied and
the channel estimate

H
mc
is provided with an estimation error
variance
2
E
. The algorithm decides to select the termination
load(s) based on the capacity estimates calculated from the
knowledge of

H
mc
, either by applying (9) for block-at fading
channels or directly substituting

H
mc
into (8). In order to
assess the performance of the algorithm, we calculate and
plot the actual capacity corresponds to the selected terminal
network at each iteration. We include Doppler effects and time-
variations of the channel by applying Gauss-Markov channel
model described in section II.
We consider the block data transmission with a block
length of L = 100 symbols, including L
p
= 18 training
symbols per block which are split into 6 groups among the
block. For each block, a channel estimate

H
mc
is obtained by
averaging the channel estimates from sub-blocks according to
(10), where each sub-block channel estimate has an estimation
error
2
E
. We also assume a uniform scattering distribution
at the receiver. To investigate the effect of time-variations of
the channel, we use the Gauss-Markov channel model with
= 1.8 10
3
, and 1.8 10
2
for the slow- and fast-fading
scenarios, respectively.
The mean value of the normalised capacity instances for
the adaptive and non-adaptive uncoupled terminations are
depicted in Fig. 17 for slow-fading with solid lines and fast-
fading scenario with dashed lines. Capacity instances for all
termination cases have been normalised to the corresponding
instances for the 50 match. Mean capacity values for the
adaptive terminations are plotted for three different iterations
m = 100, 500, and 2000, in order to evaluate the convergence
behaviour of these termination cases. We observe that for both
slow and fast fading scenarios, better channel estimates (i.e.,
lower
2
E
) result in larger capacity improvements.
Furthermore, we can see that the adaptive identical termi-
nation achieves a slightly better performance than the self-
conjugate match for both fading scenarios after sufcient
iterations. It is noted that the self-conjugate match requires
the knowledge of the diagonal elements of the receive array
impedance matrix Z
R
, whereas the adaptive identical does not.
In comparison, the non-identical termination provides larger
capacity improvements even for less accurate channel estima-
tion scenarios (i.e. larger
2
E
). We recall that self-impedances
of coupled antennas are different than the case when they
are isolated [4], [10], and that measuring these impedances
is difcult in practice. We also note that practical antenna
arrays and consequently their impedance matrices may not be
Toeplitz in structure. The adaptive non-identical termination
would be a reliable solution to optimise the performance for
these corresponding scenarios.
0.95
1
1.05
1.1
1.15
1.2
1.25


M
e
a
n
(
C
/
C
Z
0
)
0 0.002 0.004 0.006 0.008 0.01
0.99
1
1.01
1.02
1.03
1.04
1.05
1.06
1.07


Channel Estimation Error,
2
E
M
e
a
n
(
C
/
C
Z
0
)
Adaptive Nonidentical
Selfconjugate
50 match
Adaptive Identical
Selfconjugate
50 match
(a)
(b)
m=2000
m=500
m=100
m=100
m=2000
m=500
Fig. 17. Mean value of the normalised capacity instances C/C(Z
0
) as a
function of estimation error
2
E
, applying adaptive (a) identical, and (b) non-
identical terminations for slow-fading and fast-fading scenarios at iterations
m = 100, 500, 2000.
V. CONCLUSION
Antenna mutual coupling and channel variations may de-
grade the performance of compact MIMO systems. This paper
presented single-port, adaptive, uncoupled matching networks
allowing both identical and non-identical component values
to optimise the performance of MIMO systems with coupled
arrays. The proposed method uses a random search algorithm
to change the termination network, in order to compensate the
performance degradation. It requires neither knowledge of the
array parameters nor derivatives of the performance metric, but
optimises performance by dealing with the received signals
at antenna terminals (voltages across the load resistances).
Simulation results for the capacity of a 3 3 MIMO under
different propagation scenarios indicated that these adaptive
networks are capable of optimising the performance in the
presence of strong mutual coupling and time-variations of
the channel. The adaptive non-identical termination gives a
signicant improvement in the mean capacity (about 2bits/s/Hz
for d = 0.05) at the expense of a longer convergence
time (more than ten times longer) compared to the identical
case. The latter approach can also be suggested as a reliable
candidate to the practical arrays whose antenna elements or
array parameters may vary along the array.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
12
REFERENCES
[1] I. E. Telatar, Capacity of multi-antenna Gaussian channels, European
Transactions on Telecommunications, vol. 10, pp. 585595, 1999.
[2] G. J. Foschini and M. J. Gans, On limits of wireless communications in
a fading environment when using multiple antennas, Wireless Personal
Communications, vol. 6, pp. 311335, 1998.
[3] P.-S. Kildal and K. Rosengren, Electromagnetic analysis of effective
and apparent diversity gain of two parallel dipoles, IEEE Antennas and
Wireless Propagation Letters, vol. 2, pp. 913, 2003.
[4] J. W. Wallace and M. A. Jensen, Mutual coupling in MIMO wireless
systems: a rigorous network theory analysis, IEEE Transactions on
Wireless Communications, vol. 3, no. 4, pp. 13171325, July 2004.
[5] Buon Kiong Lau and Jorgen Bach Andersen, On closely coupled
dipoles with load matching in a random eld, IEEE 17th International
Symposium on Personal, Indoor and Mobile Radio Communications
(PIMRC), Helsinki, Finland, pp. 15, Sept. 2006.
[6] J. Andersen and H. Rasmussen, Decoupling and descattering networks
for antennas, IEEE Transactions on Antennas and Propagation, vol.
24, no. 6, pp. 841 846, Nov. 1976.
[7] Buon Kiong Lau, J. B. Andersen, G. Kristensson, and A. F. Molisch,
Impact of matching network on bandwidth of compact antenna arrays,
IEEE Transactions on Antennas and Propagation, vol. 54, no. 11, pp.
32253238, Nov. 2006.
[8] J. B. Andersen and B. K. Lau, On closely coupled dipoles in a random
eld, IEEE Antennas and Wireless Propagation Letters, vol. 5, no. 1,
pp. 7375, Dec. 2006.
[9] Buon Kiong Lau, J.B. Andersen, A.F. Molisch, and G. Kristensson, An-
tenna matching for capacity maximization in compact MIMO systems,
in 3rd International Symposium on Wireless Communication Systems
(ISWCS), Valencia, Spain, Sept. 2006, pp. 253 257.
[10] Y. Fei, Y. Fan, B. K. Lau, and J.S. Thompson, Optimal single-
port matching impedance for capacity maximization in compact MIMO
arrays, IEEE Transactions on Antennas and Propagation, vol. 56, no.
11, pp. 3566 3575, Nov. 2008.
[11] Michael A. Jensen and Brett Booth, Optimal uncoupled impedance
matching for coupled MIMO arrays, in First European Conference on
Antennas and Propagation (EuCAP), Nov. 2006, pp. 1 4.
[12] M.A. Jensen and Buon Kiong Lau, Uncoupled matching for active
and passive impedances of coupled arrays in MIMO systems, IEEE
Transactions on Antennas and Propagation, vol. 58, no. 10, pp. 3336
3343, Oct. 2010.
[13] Yuanyuan Fei and J.S. Thompson, Continuous study of closely coupled
dipoles with matching networks, in The Second European Conference
on Antennas and Propagation (EuCAP), Nov. 2007, pp. 1 5.
[14] I. Gupta and A. Ksienski, Effect of mutual coupling on the performance
of adaptive arrays, IEEE Transactions on Antennas and Propagation,,
vol. 31, no. 5, pp. 785791, Sep 1983.
[15] R.S. Adve and T.K. Sarkar, Compensation for the effects of mutual
coupling on direct data domain adaptive algorithms, IEEE Transactions
on Antennas and Propagation, vol. 48, no. 1, pp. 86 94, jan 2000.
[16] H.T. Hui, Improved compensation for the mutual coupling effect in a
dipole array for direction nding, IEEE Transactions on Antennas and
Propagation, vol. 51, no. 9, pp. 24982503, Sep 2003.
[17] H.T. Hui, Compensating for the mutual coupling effect in direction
nding based on a new calculation method for mutual impedance, IEEE
Antennas and Wireless Propagation Letters, vol. 2, pp. 26 29, 2003.
[18] A.J. Kerkhoff and Hao Ling, A simplied method for reducing mutual
coupling effects in low frequency radio telescope phased arrays, IEEE
Transactions on Antennas and Propagation, vol. 59, no. 6, pp. 1838
1845, june 2011.
[19] D.F. Kelley and W.L. Stutzman, Array antenna pattern modeling
methods that include mutual coupling effects, IEEE Transactions on
Antennas and Propagation, vol. 41, no. 12, pp. 1625 1632, dec 1993.
[20] C.K.E. Lau, R.S. Adve, and T.K. Sarkar, Minimum norm mutual cou-
pling compensation with applications in direction of arrival estimation,
IEEE Transactions on Antennas and Propagation, vol. 52, no. 8, pp.
2034 2041, aug. 2004.
[21] H. T. Hui, A new denition of mutual impedance for application
in dipole receiving antenna arrays, IEEE Antennas and Wireless
Propagation Letters, vol. 3, pp. 364367, 2004.
[22] M. K. Ozdemir, H. Arslan, and E. Arvas, A mutual coupling model
for MIMO systems, in IEEE Topical Conference on Wireless Commu-
nication Technology, Oct. 2003, pp. 306 307.
[23] R. Tian and B. K. Lau, Uncoupled antenna matching for perfor-
mance optimization in compact MIMO systems using unbalanced load
impedance, IEEE Vehicular Technology Conference (VTC Spring 2008),
Singapore, pp. 299303, May 2008.
[24] Wen Fong Tsen and Hsueh-Jyh Li, Uncoupled impedance matching
for capacity maximization of compact MIMO arrays, IEEE Antennas
and Wireless Propagation Letters, vol. 8, pp. 1295 1298, 2009.
[25] R.H. Etkin and D.N.C. Tse, Degrees of freedom in some underspread
MIMO fading channels, IEEE Transactions on Information Theory,
vol. 52, no. 4, pp. 1576 1608, Apr. 2006.
[26] H. Ozcelik, M. Herdin, W. Weichselberger, J. Wallace, and E. Bonek,
Deciencies of Kronecker MIMO radio channel model, Electronics
Letters, vol. 39, no. 16, pp. 12091210, Aug. 2003.
[27] A. Molisch, Wireless Communications, John Wiley & Sons, 2005.
[28] M. Biguesh and A.B. Gershman, Training-based MIMO channel
estimation: a study of estimator tradeoffs and optimal training signals,
IEEE Transactions on Signal Processing, vol. 54, no. 3, pp. 884 893,
Mar. 2006.
[29] T. Yoo and A. Goldsmith, Capacity and power allocation for fading
MIMO channels with channel estimation error, IEEE Transactions on
Information Theory, vol. 52, no. 5, pp. 2203 2214, May. 2006.
[30] D. Tse and P. Viswanath, Fundamentals of Wireless Communication,
Cambridge University Press, 2005.
[31] R. Mohammadkhani and J.S. Thompson, Adaptive matching for
compact MIMO systems, in 7th International Symposium on Wireless
Communication Systems (ISWCS), Sept. 2010, pp. 107 111.
[32] J. W. Wallace and M. A. Jensen, Termination-dependent diversity
performance of coupled antennas: network theory analysis, IEEE
Transactions on Antennas and Propagation, vol. 52, no. 1, pp. 98105,
Jan. 2004.
[33] B. K. Lau and J. B. Andersen, Antenna system and method for
operating an antenna system, U.S. Patent 2010/0201598 A1, Aug. 2010.
[34] Stephen Boyd and Lieven Vandenberghe, Convex Optimization, Cam-
bridge University Press, 2009.
[35] C. T. Kelley, Iterative Methods for Optimization, Society for Industrial
and Applied Mathematics (SIAM), 1999.
[36] R. Mudumbai, D.R. Brown, U. Madhow, and H.V. Poor, Distributed
transmit beamforming: challenges and recent progress, IEEE Commu-
nications Magazine, vol. 47, no. 2, pp. 102 110, Feb. 2009.
[37] D. Aldous and J. Fill, Reversible Markov Chains and Random Walks
on Graphs, (Monograph in preparation.) [Online]. Available: http://stat-
www.berkeley.edu/users/aldous/RWG/book.html.
[38] C. A. Balanis, Antenna Theory: Analysis and Design, John Wiley &
Sonc, Inc., 2nd edition edition, 1997.
[39] Di Lu, D.K.C. So, and A.K. Brown, Receive antenna selection scheme
for v-blast with mutual coupling in correlated channels, in IEEE
19th International Symposium on Personal, Indoor and Mobile Radio
Communications, 2008. PIMRC 2008., Sep. 2008, pp. 1 5.
[40] Y. Fei, Compact MIMO Terminals with Matching Networks, Ph.D.
thesis, The University of Edinburgh, 2008.
[41] R. Mohammadkhani, Adaptive Impedance Matching to Compensate
Mutual Coupling Effects on Compact MIMO Systems, Ph.D. thesis,
The University of Edinburgh, 2012.
Reza Mohammadkhani received the BSc degree
in Electronics Engineering from Isfahan University
of Technology (IUT) in 2001, the MSc degree
in Communication Systems Engineering from Iran
University of Science and Technology (IUST) in
2004, and the PhD degree in Signal Processing from
The University of Edinburgh in 2012. From 2003
to 2005, he worked on several research projects
at Antenna and Microwave Research Center, IUST.
Currently, he is an assistant professor of Signal
Processing at the Electrical Engineering Department,
University of Kurdistan (UoK), Iran.
His research interests are in various elds of signal processing include:
wireless communication systems, Multiple-Input Multiple-Output (MIMO)
wireless systems, adaptive array processing, phased array radars, and passive
FM radars.
Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
13
John Thompson was appointed as a lecturer at what
is now the School of Engineering at the University
of Edinburgh in 1999. He was recently promoted
to a personal chair in Signal Processing and Com-
munications. His research interests currently include
signal processing, energy efcient communications
systems, and multi-hop wireless communications.
He has published over 200 papers to date including
a number of invited papers, book chapters and tuto-
rial talks, as well as co-authoring an undergraduate
textbook on digital signal processing.
During 2012-2014, Professor Thompson will serve as member-at-large
for the Board of Governors of the IEEE Communications Society. He was
technical programme co-chair for the IEEE Globecom Conference in Miami
in 2010 and will serve in the same role for the IEEE Vehicular Technology
Conference Spring in Dresden in 2013.