1

The Effect of Macrodiversity on the Performance of

Maximal Ratio Combining in Flat Rayleigh Fading

arXiv:1208.4899v3 [cs.IT] 24 Jan 2013

Dushyantha A. Basnayaka, Student Member, IEEE, Peter J. Smith, Senior Member, IEEE and Philippa A.
Martin, Senior Member, IEEE

AbstractThe performance of maximal ratio combining

(MRC) in Rayleigh channels with co-channel interference (CCI)
is well-known for receive arrays which are co-located. Recent
work in network MIMO, edge-excited cells and base station
collaboration is increasing interest in macrodiversity systems.
Hence, in this paper we consider the effect of macrodiversity
on MRC performance in Rayleigh fading channels with CCI.
We consider the uncoded symbol error rate (SER) as our
performance measure of interest and investigate how different
macrodiversity power profiles affect SER performance. This is
the first analytical work in this area. We derive approximate and
exact symbol error rate results for M -QAM/BPSK modulations
and use the analysis to provide a simple power metric. Numerical
results, verified by simulations, are used in conjunction with the
analysis to gain insight into the effects of the link powers on
performance.
Index TermsMacrodiversity, MRC, symbol error rate,
Rayleigh fading, Network MIMO, CoMP.

I. I NTRODUCTION
Maximum ratio combining (MRC) is a well-known linear combining technique that maximizes the signal-to-noise
ratio (SNR) in noise limited systems [1]. In the presence
of co-channel interferers, MRC is sub-optimal compared to
minimum mean squared error (MMSE) combining. However,
MMSE combining requires instantaneous channel knowledge
of both the desired source and interfering sources. In contrast,
MRC only requires a knowledge of the desired source and
hence is simpler to implement. For this reason, there is still
interest in MRC processing in the presence of interference. In
[2], MRC is investigated for large systems where it was shown
that in the limit as the number of antennas increases, intercell
interference effects disappear. In [3], a switched MRC/MMSE
receiver is proposed where the simplicity of MRC is preferred
when the interference levels drop below a threshold. Here,
MRC performance in the presence of small but non-zero
interference is important. There are well-known methods to
estimate the interference level in comparison with the signal
level as described in [4].
The performance of MRC systems with co-located antenna arrays is well known for Rayleigh fading channels with multiple
co-channel interferers [5], [6]. Recently, interest in distributed
combining has grown due to research in cooperative systems,
base-station collaboration [7, pp. 69], edge-excited cells [8],
D. A. Basnayaka, P. J. Smith and P. A. Martin are with the Department of
Electrical and Computer Engineering, University of Canterbury, Christchurch,
New Zealand. E-mail:{dush, p.smith, p.martin}@elec.canterbury.ac.nz.
D. A. Basnayaka is supported by a University of Canterbury International
Doctoral Scholarship.

[9] and network MIMO [10], [11]. In the standards, distributed

processing is part of coordinated multipoint transmission
(CoMP) in LTE Advanced. For these macrodiversity systems,
every link may have a different average SNR since the sources
and the receive antennas are all in different locations. This
variation in SNR makes performance analysis more complex
and to the best of our knowledge no analytical results are
currently available for such systems.
Hence, in this paper we analyze the symbol-error-rate (SER)
of macrodiversity MRC systems. Note that the system is
not new. Standard MRC processing is considered and so
the general form of the receiver output and the initial steps
in the performance evaluation are well-known. However the
macrodiversity layout creates a new channel structure which
is far more complex than the microdiversity channel. Hence,
the MRC output has a completely new statistical distribution
and a novel, more advanced analysis is required for system
performance evaluation. In particular, we consider a distributed
antenna array performing MRC combining for a single antenna
desired source in the presence of an arbitrary number of single
antenna co-channel interferers. The analysis also covers the
case where both the desired and interfering sources may have
multiple antennas. Since the sources and the receive antennas
are not co-located, the channels are normally independent and
so the focus is on independent Rayleigh fading channels where
each link has a different SNR. In this paper, we evaluate the
SER over Rayleigh fading for fixed values of the long term
link SNRs. Hence, the SER is computed over fast fading while
path loss effects and shadowing are held constant. Looking at
the joint effects of the slow fading (see, for example, [12],
[13]) would be an interesting topic for future work. In the
scenario where some sources have multiple antennas, there
may be spatial correlation in the channels corresponding to the
antennas at that source. However, this is beyond the scope of
the current work where independent channels are considered.
We provide specific results for BPSK and QPSK modulations,
but the analysis can be applied to M -QAM and a wide range
of modulations where the SER can be written in terms of an
expected value of the Gaussian Q-function and Q2 -function.
The general analytical approach follows the techniques in
[14]. The novelty in the analysis is the identification of
a representation for the interference and noise term in the
combiner output and the use of this representation in exact
SER calculations. We then use the SER results to analyze the
effect of macrodiversity on MRC performance.
The rest of the paper is organized as follows. In Sec. II,
we give the system model and in Sec. III the performance

results and SER are derived. Sec. IV gives numerical results

where the analysis is verified by simulation and conclusions
are presented in Sec. V.

1
Source 1
(desired)

II. S YSTEM M ODEL

= Hs + n ,
T

(1)

where P = (Pik ), P 2 is the element-wise square root of P ,

the operator, , represents Hadamard multiplication and the
elements, H w,ik , of H w satisfy H w,ik CN (0, 1) i, k.
The matrix, P , is the global power matrix for the system
and for the k th source,
power matrix is also
n an individual
o
H
defined by P k = E h kh k = diag (P1k , P2k , . . . , PnR k ),
for k = 1, 2, . . . , N . In the microdiversity case, P k I .
In macrodiversity scenarios, P k is no longer proportional to
the identity and these more general power matrices make the
analysis more complex. Assume, without loss of generality,
that user 1 is the desired user. For the purpose of decoding
user 1, (1) can be rewritten as
s + n
r = h 1 s1 + H
= h 1 s1 + i ,

(3)
(4)

is all
where h 1 is the first column of H , H
of H ,
 columns

, and s =
excluding the first column, meaning H = h 1 , H

T
(s2 , . . . , sN ) . The nR 1 vector, i is the interference and
noise vector. With MRC processing, the output of the combiner
is given by [14]

hH
hH
1 i
1 r
=
s
+
.
1
H
H
h1 h1
h1 h1

(5)

MRC
combiner

Source 2

nR

nR 1

receive vecwhere r = (r1 , r2 , . . . , rnR ) is the C

tor, H = (hik ) is the C nR N channel matrix, s =
T
(s1 , s2 , . . . , sN ) is the C N 1 signal vector and n =
T
(n1 , n2 , . . . , nnR ) is the C nR 1 additive-white-Gaussiannoise (AWGN)

vector at the receive antennas such that n
CN 0 , 2I . The signals are
to be zero-mean, unit
 normalized

power variables so that E |si |2 = 1 for i = 1, 2, . . . , N .
The channel matrix, H , has independent
zero-mean,
complex


H ik |2 = Pik . Hence,
Gaussian elements such that E |H
equation (1) can be rewritten as


1
r =
H w P 2 s + n,
(2)

r =

2
Interfering sources

Consider N single-antenna distributed users communicating

with nR distributed transmission points (TP) [16] each with a
single receive antenna over an independent flat fading Rayleigh
channel. The system diagram is given in Fig.1. The received
signal is given by

Source N
Fig. 1.
shown.

System diagram. To reduce clutter, only paths from source 2 are

follows that Z is also zero-mean Gaussian conditioned on h 1

and s. The conditional variance of Z is given by
 


H +n H h 1 
s+n sH H
h H

n
o
H
1
 h 1 , s ,
E |Z|2 |h 1 , s = E



2


hH
1 h1

(7)

n
o
sH H
H + 2I h 1
hH
1 E Hs
,
=
2

hH
h
1 1
P

N
2
2
hH
1
k=2 P k |sk | + I h 1
.
=
2

hH
h
1 1

(8)

(9)

Hence, since Z is a conditional Gaussian with variance given

by (9), it follows that Z has the exact representation
r
P

N
2
2
hH
1
k=2 P k |sk | + I h 1
Z=
U,
(10)
hH
1 h1
where U CN (0, 1). Using this representation in (5) gives
the combiner output in simplified signal plus noise form as

Y
U,
(11)
r = s1 +
X
where X = h H
1 h1, Y
P
N
2
P
|s
|
+
2I .
k
k
k=2

= hH
s ) h 1 and D (
s) =
1 D (

The interference and noise term in (5) can be written as



s + n
hH
H
1
hH
i
.
(6)
Z = H1 =
h1 h1
hH
1 h1

A. A Simple SER Analysis

Following the standard approach [14], we develop a condi and

tional Gaussian representation for Z as follows. Since H
n are zero-mean Gaussian and independent of h 1 and s, it

With the combiner output given by (11), SERs for many

modulations can be obtained using standard methodology [14].

III. P ERFORMANCE A NALYSIS

As an example, for BPSK, we have the SER

!

Y
Ps = Pr 1 +
Re (U ) > 0 ,
X
!)
(
r
2X 2
,
=E Q
Y
R

and
P nR 2 ik nR 3
,
ik = QnR i1
l6=i,k (ik il ik il )
ik
,
ik =
ik
ik = Pi1 Qk Qi Pk1 ,
ik = Pi1 Pk1 ,
Q = D (
s ) P 1 = diag (Q1 , Q2 , . . . , QnR ) .

(12)
(13)

2
t2

where Q (x) = 12 x e
dt is the Gaussian Q-function
2 1
defined
gives the BPSK SER
 in[17].
Defining = X Y
as E Q 2 . Note that in general is a function of s but
this dependence is not shown for convenience. For BPSK, each
element of s has unit modulus and so there is no dependence
on s and the SER in (13) is valid for any values of s. For many
modulations [18], [19], SERs are constructed from similar
functions of the form
n p o
b ,
(14)
W1 (a, b, s) = E aQ
Z
p 
b f () d,
(15)
=
aQ
0

where f () is the probability density function (pdf) of .

Hence, our approach involves averaging the Q-function in
(15) over . There are alternative routes to the same result.
For example, the Q-function in (13) could be averaged over
the joint distribution of X and Y . For some modulations,
such as BPSK, the SER can be given exactly in terms of
W1 (a, b, s), whereas for other modulations it will provide an
approximation. Using integration by parts on (15) gives
 2
Z
2
w
a
w2
F
dw,
(16)
e
W1 (a, b, s) =
b
2 0

where F (.) is the cumulative distribution function (cdf) of .

Hence, SER performance for MRC relies on the evaluation of
(16) which in turn relies on the cdf of .
In the microdiversity case, all the P i matrices are proportional
to the identity and reduces to a simplified expression,
2 , where 2 is a chi-squared random variable [20]. In
the macrodiversity case, this reduction does not occur and
is proportional, not to a simple chi-squared random variable,
but to a ratio of powers of correlated quadratic forms. This
is the novel analytical challenge posed by the macrodiversity
scenario. The derivation of the cdf is based on the joint
distribution of X, Y . From [21], the joint distribution of X
and Y becomes


nR
nR X
Q x
X
y Pi
x
i1
ik e Pi1 e ik
fX,Y (x, y) =
i=1 k6=i



u y Qi x u (x)
for ik > 0
P
i1



u Qi x y u (x) for ik < 0,

Pi1

(17)

where u (x) is the standard unit step function defined as

(
0 x<0
u (x) =
(18)
1 x > 0,

(19)
(20)
(21)
(22)
(23)

Note that each term in the summation of (17) has its own
region of validity depending on the algebraic sign of ik . For
example, when ik > 0, the region of validity becomes the
infinite region bounded below by the x = 0 and y = PQi1i x
curves. The ik = 0 condition has been ignored since the
case of distributed users with a single antenna always yields
ik 6= 0. The cdf of is defined by

 2
X
<r ,
(24)
F (r) = Pr ( < r) = Pr
Y


= Pr X 2 rY < 0 ,
(25)
ZZ
fX,Y (x, y) dxdy,
(26)
=
D

where
the domain of integration is defined by D =

x, y : x 0, y 0 and x2 ry < 0 . In Appendix A, the
integral in (26) is computed giving
F (r) =

nR
nR X
X

Fik (r) ,

(27)

i=1 k6=i

1
2
where Fik (r) = Fik
(r) for ik > 0 and Fik (r) = Fik
(r) for
1
2
ik < 0, where Fik (r) and Fik (r) are given in (28) and (29)
and ik is given in (20).


r
2
rQ
ik
2i
ik
r r
Pi1 ik
1
1 e Pi1 +
e 4ik
(r) =
Fik
ik
2ik ik

p
 (28)
1
rik ik ,



r
2
rQ
ik
2i
ik
r r
Pi1 ik
1 e Pi1
e 4ik
ik
2ik ik

 
 r

p
r
1
,
ik
erfi
rik ik + erfi
2 ik
(29)

2
Fik
(r) =

where
1 ik Qi
Qi
ik
,
ik =
+
,
(30a)
Pi1
Pi1
2ik
Rx
2
and (x) = 2 0 et dt in (35) and (36) is the standard

 
error function [17]. Furthermore, Q (x) = 0.5 1 x2 .
For convenience, we expand ik in (20) and also give the result
here as
ik =

ik = PN

u=2

1
Pi1

1
Pk1

(Pku Piu ) |su |

(31)

Note that the case of ik = 0 is not considered as this is the

case when Pi1 = Pk1 , an event which occurs with probability

zero when the receive antennas are not co-located. As for the
cdf, the SER analysis is performed separately according to the
algebraic sign of ik . Therefore, substituting (27) into (16), the
final result is
nR X
nR
X
s

W1 (a, b, ) =
Psik ,
(32)
i=1 k6=i

where Psik = Ps1ik for ik > 0 and Psik = Ps2ik for ik < 0,
and Ps1ik and Ps2ik are given in (33) and (34). The results in
(33) and (34) are obtained using the following three standard
integral identities [17]
r

Z



1

x
e
1 x dx =
1 ,

0
for Re () > 0; Re () < Re () , (35)
Z

ja
p
,
2 a2


Re () > 0; Re () > Re a2 , (36)

xex (jax) dx =

for
Z

eq

x2

dx =

2q

for

q > 0.

(37)

For multi-level constellations, the values of s affect D (

s ) and
therefore . Hence, SER results must average (32) over all
possible values of s. This gives
X
W1 (a, b) =
W1 (a, b, s) Pr (
s) ,
(38)
s
where (38) may be an exact or approximate SER result, the
summation is over all possible s and Pr (
s ) is the probability
of a particular s value. Finally, for BPSK modulation, the SER
in (13) becomes
Ps = W1 (1, 2, s) .

(39)

B. Extended SER Analysis

For M -QAM, first order SER approximations can be found
via expressions of the form in (14). Exact results involve
expectation over the Q2 (.) function in addition to (14). As
an example, consider 4-QAM where the SER is given by

1
1
Ps = Pr + Re (U ) > 0

2

(40)
1
1
or + Im (U ) > 0

2
r 2


Ps = 1 Pr Re (U ) <
(41)
2
o
n
2
(42)
= 1 E (1 Q ( ))
 2

(43)
= 2E {Q ( )} E Q ( ) .


Here, the 2E Q
= W1 (2, 1, s) term in (43) is a
good
approximation
to
P
s [14] and the remaining term,



, makes only a small adjustment. However, in
E Q2
other variations of M -QAM modulation schemes the contribution from Q2 (.) is not negligible [14]. Therefore, for general

M -QAM, the exact SER is useful and this can be

in
written
terms of W1 (a, b, s) and W2 (a, b, s) = E aQ2
b . The
first expectation is found in (32). The second expectation can
be derived as follows. Let,
Z
p 
W2 (a, b, s) = a
Q2
b f () d.
(44)
0

Using integration by parts on (44) gives

r Z
 2
w2
2
w
dw.
e 2 Q (w) F
W2 (a, b, s) = a
0
b

(45)

In order to facilitate our analysis we need two fundamental

probability integrals. Therefore, we derive both integrals in
Appendix B along with their regions of convergence, since
they may have applications in other communication problems.
Note that similar results may be found in [15], but these are for
restricted ranges of the parameter values. The macrodiversity
integrals require a wider range of values and the analysis in
Appendix B enables us to evaluate both the integral values and
the precise region of validity. As for the simple SER analysis,
the extended analysis is also performed separately according
to the algebraic sign of ik . Therefore, substituting (27) into
(45), the final result is derived in Appendix B as
W2 (a, b, s) =

nR X
nR
X

Psik ,

(46)

i=1 k6=i

where Psik = Ps1ik for ik > 0 and Psik = Ps2ik for ik < 0,
where Ps1ik and Ps2ik are given in (47) and (48), respectively.
Hence, the exact SERs are computable using (32) and (46)
for any M -QAM modulation. As in Sec. III-A, for multilevel constellations the SER results depend on W1 (a, b) and
W2 (a, b) results where W1 (a, b) is given in (38) and
X
W2 (a, b) =
W2 (a, b, s) Pr (
s) .
(49)
s
For QPSK modulation the SER in (43) becomes
Ps = W1 (2, 1, s) W2 (1, 1, s) .

(50)

C. A Simple Power Metric

The SER and any other performance metrics are functions
of the power matrices P 1 , P 2 , . . . , P N . Although (32) and
(46) give the exact SER as a function of these powers, the
result is complex and does not offer any simple insights into
the relationship between performance and the powers. Hence,
we consider (5) and (9) which give the mean SINR of the
combiner as

2


hH
1 h1
P

m
P , E
,
(51)
N
H

2I h
hH

h
h
+

k
1
1
k
k=2

where m
P is a performance
ometric based on the link powers
n
2
= 1. Exact evaluation of (51)
and we have used E |si |
is possible but it is rather involved and produces complex
expressions. Hence, we prefer the compact approximation
based on the first order delta method, similar to the Laplace

Ps1ik

Ps2ik

a Pi1 ik
1
1 r
=

2 ik 2
1
2
+
2

a Pi1 ik
1
1 r
=


ik
2
2
1
2
+
2

a Pi1 ik
Ps1ik =
2 ik

Qi
2
bPi1

Qi
2
bPi1

 +

 +

v
u
ik
b
u  ik

 ik 1
t
.
2
Qi
1
ik (ik
2bik )
b 2 + bP 2
i1

v
u
b

ik
u 

ik
 ik + ik
t
.
2
ik (ik
2bik )
2bik
b 1 + Qi2
2


q
2Qi
1
1
+
tan
2
1
bPi1
ik

r
 + b I2
8
ik ik
Qi
1

+ bP
2
2

ik

q


1
ik

 j b
ik ik

(34)

bPi1

2
ik

1
ik
,

b 4bik
2

i1

(47)

2Qi
2
bPi1

tan
1+
a Pi1 ik
1
r
Ps2ik =

ik
2
8
Qi
1

+ bP
2
2
1

(33)

i1

I1 ik

approximation [22], given by


2 
H
h
h
E
1 1
P
 o.
mP = n
N
H
2
E hH
1
k=2 h kh k + I h 1

(48)

2
ik 1
, ik
b
2
4bik

+I1

ik
2
1

, ik
4bik
2 bik 2

!!

D. Remarks on Systems with Multiple Co-located Receive

Antennas at TPs
(52)

Using established results for the moments of quadratic forms

[20, pp. 119], we obtain
 
P 1 )2 + Tr P 21
Tr (P
.
(53)
mP = P
N
2
Tr
i=2 P 1P i + P 1

From (53), we observe that while a P 1 matrix with large

entries boosts the numerator, hence improving performance,
it also interacts with the interferers in the denominator. Since
MRC is based on weighting the strongest signal, the most
advantageous interference profile is for the stronger interferers
to line up with the weaker desired signals and vice-versa. This
intuitive result is precisely captured by (53) which increases
P 1 ) and also increases as Tr (P
P 1P i ) decreases, for
with Tr (P
i = 2, 3, . . . , N .
Although mP captures some of the important relationships
between performance and the power matrices, it is not always
an accurate predictor of performance. As the SINR grows,
the mean becomes further from the lower tail which governs
error rate performance. Hence, we expect the mean to carry
less information about SER at high SINR. This is discussed
in more detail in Sec. IV.

The analysis in Sec. III is restricted to situations where the

TPs have a single antenna each. However, if the receiver, for
example has two co-located antennas at any TP, the system
analysis still can be handled by the same method, but will
result in a different joint distribution for (17). Thus, every new
scenario for co-located antennas gives a new joint distribution
and in turn this gives a different error rate expression. A pragmatic solution is to use a perturbation approach. If Pi1 = Pr1
(corresponding to receive antennas i and r of the desired user,
being co-located at ith TP) then we can use Pi1 and Pi1 + for
the two powers where is a small perturbation. This approach
provides stable and accurate results as will be shown in Sec.
IV.
IV. N UMERICAL

AND

S IMULATION R ESULTS

For the numerical results, we consider a system with three

distributed receive antennas and also a larger system with 6
receive antennas deployed in three sets of co-located pairs.
Hence, there are three positions at which one or two antennas
are deployed and these are refereed to as locations. Note that
the number of interferers in the system
since,
P is irrelevant,
2
from (10), their effect is governed by N
P
|s
|
.
Hence,
k
k
k=2 P
N
2
one interferer with a power matrix equal to
k=2 P k |sk |
is equivalent to N 1 interferers with power matrices
2
2
P 2 |s2 | , . . . , P N |sN | . Hence, we consider a single interferer
throughout. In this section we consider BPSK and 4-QAM
results where |si |2 = 1 i. Hence, for both systems, we
parameterize the performance by three parameters which are
independent of the transmit symbols. The average received sigP 1 ) /nR 2 . The total
nal to noise ratio is defined by = Tr (P

TABLE I
PARAMETERS FOR F IGURES 2 AND 3

TABLE III
PARAMETERS FOR F IGURE 5

Decay Parameter

Decay Parameter
mP

Sc. No.
S1
S2
S3
S4
S5
S6
S7
S8
S9
S10

1
1
1
1
1
10
10
10
10
10

Desired
1
= 65
1
= 65
1
= 65
=1
=1
1
= 65
1
= 65
1
= 65
=1
=1

(dB)
3.06
7.68
28.64
5.97
5.97
12.93
17.30
27.62
15.60
15.60

Interfering
1
= 65
=1
= 65
=1
1
= 65
1
= 65
=1
= 65
=1
1
= 65

mP

Err. Floor
1.36e-1
6.26e-2
1.80e-3
2.49e-2
2.76e-2
1.42e-2
4.90e-3
1.68e-4
1.21e-4
2.57e-4

Sc. No.
S16
S17
S18
S19
S20

(dB)
17.42
21.34
27.68
19.65
19.64

Interfering
1
= 65
=1
= 65
=1
1
= 65

S1

Pik = Kk ()

S2
S5

SER

Err. Floor
1.54e-2
5.20e-3
1.99e-4
7.68e-5
1.72e-4

P 1 ) /Tr (P
P 2 ).
signal to interference ratio is defined by = Tr (P
The spread of the signal power across the three locations is
assumed to follow an exponential profile, as in [23], so that a
range of possibilities can be covered with only one parameter.
The exponential profile is defined by
i1

Err. Floor
1.50e-3
1.61e-4
5.51e-7
1.54e-7
1.04e-6

mP

Desired
1
= 65
1
= 65
1
= 65
=1
=1

(dB)
17.57
21.69
29.32
20.57
19.96

Interfering
1
= 65
=1
= 65
=1
1
= 65

10

Decay Parameter

30
30
30
30
30

Desired
1
= 65
1
= 65
1
= 65
=1
=1

10

TABLE II
PARAMETERS FOR F IGURE 4

Sc. No.
S11
S12
S13
S14
S15

20
20
20
20
20

for receive location i and source k where

P k ) / 1 + + 2 , k = 1, 2,
Kk () = Tr (P

(54)

(55)

and > 0 is the parameter controlling the uniformity of the

powers across the antennas. Note that as 0 the received
power is dominant at the first location, as becomes large
( 1) the third location is dominant and as 1 there is
an even spread, as in the standard microdiversity scenario. In
Figs. 2-3 we show SER results for the ten scenarios (S1-S10)
given in
Table I. Note that an error floor occurs as
2 0 for fixed . The value of the error floor is obtained
by letting 2 0 in (32). In Table I we report the values of
mP and the error floor for the ten scenarios considered. Note
that mp is given for a 2 value corresponding to = 20 dB.
Figures 2 and 3 verify the analytical results in (32) for BPSK
modulation with simulations and also explore the effect of
different power profiles. In Fig. 2, a low SIR is considered
with = 1. Here, S1 is the worst case since the desired
signal profile is aligned with the interferer and the profile is
rapidly decaying giving little diversity. S3 is the best since the
profiles are opposing and the best desired signal aligns with
the weakest interference. Since Fig. 2 has a low SIR the major
impact on performance is caused by the presence or absence

S4

10

S3

10

Simulation
Analytical (exact)
0

10

[dB]

15

20

25

30

Fig. 2. Analytical and simulated SER values for a MRC receiver with BPSK
modulation in flat Rayleigh fading for scenarios S1-S5 with parameters: nR =
3 and = 1.

of a high SIR or low SIR at each antenna. In Fig. 3, the same

power profiles are considered but at higher SIR, = 10, the
order is changed. S6 is still the worst as this scenario has high
interference at all antennas and little diversity. In contrast S8
is no longer the best with S9 now giving better performance.
Note that S9 has greater diversity with an even spread of power
across the antennas and this becomes more important at high
SIR.
Another comparison between scenarios can be seen in Table
I. Note that in Fig. 2 the ordering based on mP correctly
identifies the best and worst scenarios whereas in Fig. 3 the
mp metric suggest that S8 is best whereas S9 is better. The
mp metric gives some intuition about macrodiversity MRC
performance, especially at low SIR, but it doesnt accurately
capture diversity effects (seen in the lower tail of the combiner
output) which are needed for accurate performance prediction.
In Fig. 4, the same power profiles are considered for QPSK
transmission. Here, the exact results from Sec. III-B are
verified by simulation. In particular, the SER expression in
(43) for QPSK modulation is used along with (32) and (46).
The relative performance provided by the 5 scenarios is the
same as in Fig. 3 except that the cross over of S13 and S15
in Fig. 4 (equivalent to the cross over of S8 and S10 in Fig.
3) does not occur until SNR > 30 dB.
Finally, in Fig. 5 we consider the six antenna receiver where
antennas 1,2 are co-located, antennas 3,4 are co-located else-

10

10

Simulation
Analytical (exact)

Simulation
Analytical (exact)

10

10

S16

10

S6
SER

SER

10

10

10

S8

S18

S7

S17

10

10

S10

S20
6

10

S9
4

10

10

[dB]

15

20

25

30

Fig. 3. Analytical and simulated SER values for a MRC receiver with BPSK
modulation in flat Rayleigh fading for scenarios S6-S10 with parameters:
nR = 3 and = 10.

S19

10

10

[dB]

15

20

25

30

Fig. 5. Analytical and simulated SER values for a MRC receiver with QPSK
modulation in flat Rayleigh fading for scenarios S16-S20 with parameters:
nR = 6 and = 20.

10

10

S11
2

SER

10

S13
S12

10

10

S15

Simulation
Analytical (exact)

S14

10

10

[dB]

15

20

25

30

Fig. 4. Analytical and simulated SER values for a MRC receiver with QPSK
modulation in flat Rayleigh fading for scenarios S11-S15 with parameters:
nR = 3 and = 30.

where and antennas 5,6 are also co-located and separated from
antennas 1-4. Here, the long term receive SNR of a source
at antennas 1 and 2 will be the same. Hence, we use the
perturbation approach of Sec. III-D to obtain results. Fig. 5
validates the perturbation approach by simulation and shows
a large performance improvement relative to Fig. 4 due to the
increased number of antennas. Again, the results due to the
five scenarios follow the same order as in Figs. 3 and 4. Note
that when = 1 for both desired and interfering sources, the
system layout is microdiversity. Hence, scenarios S4, S9, S14
and S19 provide microdiversity results.
V. C ONCLUSION
Exact SER results are derived for BPSK and M -QAM modulations in a Rayleigh fading macrodiversity system employing

MRC. The results have applications to several systems of

current interest in communications including network MIMO
and cooperative communications. The analysis is used to study
the effects of the macrodiversity power profiles on MRC
performance. It is shown that simple power metrics may
capture several features of MRC performance but the impact
of diversity in a distributed system is important at realistic
SINR values. Here, the exact results are necessary to provide
an accurate performance measure. In general, performance
improves as the desired signal dominates the interferer at some
antennas and as the desired power is spread more evenly over
the receive antennas. The exact balance between these two key
features is difficult to obtain in a simple form but is provided
by the exact solutions given.

A PPENDIX A
C ALCULATION OF THE CDF

OF

Since each term in the summation of (17) depends on the

algebraic sign of ik , the final cdf has two parts as below
F (r) =

nR
nR X
X

Fik (r) ,

(56)

i=1 k6=i

1
2
where Fik (r) = Fik
(r) for ik > 0 and Fik (r) = Fik
(r) for
1
ik < 0. In subsection A-A we derive Fik (r) followed by the
2
derivation of Fik
(r) in subsection A-B.

1
A. Derivation of Fik
(r)
1
From the joint pdf in (17), when ik > 0, Fik
(r) is given
by
ZZ
1
fX,Y (x, y) dxdy,
(57)
Fik (r) =
F1

n
o
2
where F1 = x, y : x, y 0, y PQi1i x 0, y xr 0 . By
using standard methods for 2-D integrals we arrive at
Z Z xr2


Q x
Pi1 ik
y Pi
x
1
i1
dydx.
Fik (r) =
e Pi1 e ik
ik
xQi
rQi
ik
Pi1
Pi1
(58)
The final result then becomes


r
2
rQ
ik
2i
r r
ik
Pi1 ik
1
Fik
(r) =
e 4ik
1 e Pi1 +
ik
2ik ik

p
 (59)
1
rik ik ,
where

1 ik Qi
,
Pi1
ik
Qi
+
.
=
Pi1
2ik

ik =

(60a)

ik

(60b)

The expression in (28) follows using standard methods

of integration in (58) and employing the following integral
identity [17] where necessary:
r
Z
p 
2
1 
ex dx =
.
(61)
1
2

2
(r)
B. Derivation of Fik
2
From the joint pdf in (17), when ik < 0, Fik
(r) is given
by
ZZ
2
fX,Y (x, y) dxdy,
(62)
Fik
(r) =
F2

n
o
2
where F2 = x, y : x, y 0, y PQi1i x 0, y xr 0 . By
using standard methods for 2-D integrals we arrive at
Z rQ
i Z xQi


Qi x
Pi1
Pi1
Px ik y Pi1
2
i1
e
Fik (r) = ik
e
dydx. (63)
2
0

A PPENDIX B
D ERIVATION OF THE EXACT SER
The integral in (45)
is required for the exact SER analysis.
Substituting F w2 /b from
(27) into (45)

gives two new
1
2
integrals involving Fik
w2 /b or Fik
w2 /b , which are given
in (28) and (29). These two integrals can be written in terms
of known functions and two fundamental probability integrals
that we denote I1 (, ) and I2 (, ). These integrals are
computed below.
A. Integral Form I
Consider the integral,
Z
2
xex Q (x) (jx) dx.
I1 (, ) =

Applying the integral forms of Q(.) and (.) gives

Z Z Z
2j x x2 t21 +t22
xe
dt2 dt1 dx. (68)
I1 (, ) =
x
0

0
2

Using the substitutions, t1 = r cos and t2 = r sin , the

integral then becomes
Z Z Z
2
2
2j r2
I1 (, ) =
xrex r cos 2 drddx, (69)
0
r1
0

where tan = 2, r1 = x/ 2 cos and r2 = x/ sin .

Using standard methods of integration with some simplifications we obtain
Z 2
dt
j
I1 (, ) =
2
2 0
t 1 2
(70)
Z
dt
j2 2

.
2 0
(2 ) t2 2
Defining

I11 (, ) =

where
erfi (x) =

(jx)
.
j

(65)

x
r

The final result then becomes



r
2
rQ
ik
2i
ik
r r
Pi1 ik
2
1 e Pi1
e 4ik
Fik
(r) =
ik
2ik ik
 r



p
1
r
rik ik + erfi
,
ik
erfi
2 ik
(64)

(67)

I12 (, ) =

allows (70) to be rewritten as

I1 (, ) =

dt
,
t2 1 2
(2

dt
,
) t2 2

j
j2
I11 (, )
I12 (, ) .
2
2

(71)
(72)

(73)

The integral in (71) and (72) can be solved in closed form to

give
!
1
2

I11 (, ) =
tanh1
(74)
2 + 1
2 + 1

The function erfi (.) is the error function with a complex ar- and

gument defined in [17]. Note that the square roots appearing in
2
= 2

p

(29) are the positive square root of ik . The expression in (29) I12 (, )=
otherwise.
1 2 tan1 2 ( 2 )
follows using standard methods of integration and employing

the following integral identity [17] where necessary:
(75)
r
Z

2
1
Note that some intermediate steps in the derivation show that
erfi ax .
(66)
eax dx =
2 a
1 + 2 > 22 is required for the existence of (67). This
constraint is satisfied by the current problem. This can easily

be seen by substituting the arguments of both I1 (., .) functions

in (48) in to 1 + 2 > 22 followed by simplifications using
(30).
B. Integral Form II
Consider the integral,
Z
2
xex Q (x) (1 (x)) dx.
I2 (, ) =

(76)

Applying the integral forms of Q(.) and (.) we obtain

Z Z Z
2 x2 t21 t22
I2 (, ) =
xe
dt2 dt1 dx. (77)
x
0

x
2

Following the same procedure as in Appendix B-A and with

some simplifications we arrive at
Z
d
2
I2 (, ) =
2 0 2 sin2
(78)
Z /2
1
d
1

,
+
2
1 2 cos2 4

where tan = 2. Making another substitution as t = tan

in (78) gives
Z 2
2
dt
I2 (, ) =
2
2 0
( ) t2 + 2
(79)
Z
dt
1
1
+

.
2 2 t2 + 1 2
4
Defining
I21 (, ) =
I22 (, ) =

Z0

dt
,
) t2 + 2
dt
,
t2 + 1 2
(2

(80)
(81)

allows (79) to be rewritten as

I2 (, ) =

2
1
1
I21 (, ) +
I22 (, )
.
2
2
4

(82)

The integrals in (80) and (81) can be solved in closed form to

give

2
2 =

p

I21 (, )=
otherwise
12 tan1 2 (2 )

(83)

and

I22 (, ) =

2
1
21

coth

2
21

1 = 2


otherwise.

(84)

As for I1 (, ) there is an associated region of validity, 1 +

22 > 2, which is satisfied by the problem.
R EFERENCES
[1] M. K. Simon and M. S. Alouini, Digital Communications over Fading
Channels: A Unified Approach to Performance Analysis, New York, NY,
USA: Wiley, 2000.

[2] Q. H. Ngo, E. G. Larsson and T. L. Marzetta, Uplink power efficiency of

multiuser MIMO with very large antenna arrays, Proc. Allerton Conf. on
Communication, Control, and computing, Illinois, USA, pp. 12721279,
2011.
[3] J. Zhu, Qiang Li and Qinghua Li, Pragmatic adaptive MRC and MMSE
MIMO-OFDM receiver algorithum, U.S. Patent 20080310486, Dec. 18,
2008.
[4] D. R. Pauluzzi and N. C. Beaulieu, A Comparison of SNR Estimation
Techniques for the AWGN Channel, IEEE Trans. Commun., vol. 48, No.
10, pp. 16811691, Oct. 2000.
[5] J. Cui and A. U. H. Sheikh, Outage probability of cellular radio systems
using maximal ratio combining in the presence of multiple interferers,
IEEE Trans. Commun., vol. 47, No. 47, pp. 11211124, Aug. 1999.
[6] Y. Tokgoz and B. D. Rao, The effect of imperfect channel estimation
on the performance of maximum ratio combining in the presence of
cochannel interference, IEEE Trans. Veh. Technol., vol. 55, no. 5, pp.
15271534, Sep. 2006.
[7] E. Biglieri, R. Calderbank, A. Constantinides, A. Goldsmith, A. Paulraj
and H. V. Poor, MIMO Wireless Communication, 1st ed, Cambridge:
Cambridge University Press, 2007.
[8] S. Catreux, R. L. Kirlin and P. F. Driessen, Capacity and performance
of multiple-input multiple-output wireless systems in a cellular context,
IEEE PACRIM, Victoria, BC, Canada, pp. 516519, 1999.
[9] J. Zhang, X. Bi and Y. Wang, Antenna pairing for space-frequency
block codes in edge-excited distributed antenna systems IEEE PIMRC,
Istanbul, Turkey, pp. 117122, 2010.
[10] S. Venkatesan, A. Lozano and R. Valenzuela, Network MIMO: Overcoming intercell interference in indoor wireless systems, IEEE ACSSC,
pp. 8387, Jul. 2007.
[11] M. K. Karakayali, G. J. Foschini, and R. A. Valenzuela, Network
coordination for spectrally efficient communications in cellular systems,
IEEE Trans. on Wireless Commun. Mag., vol. 13, no. 4, pp. 5661, Aug.
2006.
[12] M. Matthaiou, N. D. Chatzidiamantis, G. K. Karagiannidis, and J. A.
Nossek, On the capacity of generalized-K fading MIMO channels, IEEE
Trans. Signal Processing, vol. 58, no. 11, pp. 5939-5944, November 2010.
[13] C. Zhong, K.-K. Wong, and S. Jin, Capacity bounds for MIMO
Nakagami-m fading channels, IEEE Trans. Signal Processing, vol. 57,
no. 9, pp. 3613-3623, Sept. 2009.
[14] J. G. Proakis, Digital Communications, 4th ed, New York: McGraw-Hill,
2001.
[15] A. Prudnikov, Y. Brychkov and O. Marichev, Tables and Integrals,
2nd ed, Gordon and Breach, New York, 1986.
[16] D. Lee, B. Clerckx, E. Hardouin, D. Mazzarese, S. Nagata, and K.
Sayana, Coordinated multipoint transmission and reception in LTEAdvanced: Deployment scenarios and operational challenges, IEEE
Commun. Mag., vol. 50, no. 2, pp. 148155, 2012.
[17] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 7th ed, Boston: Academic Press, 2000.
[18] A. Goldsmith, Wireless Communication, 4th ed, New York: McGrawHill, 2000.
[19] P. J. Smith, Exact performance analysis of optimum combining with
multiple interferers in flat Rayleigh fading, IEEE Trans. Commun., vol.
55, no. 9, pp. 16741677, Sep. 2007.
[20] K. S. Miller, Multidimensional Gaussian Distributions, 1st ed, New
York: John Wiley & Sons, 1964.
[21] A. Firag, P. J. Smith, H. Suraweera and A. Nallanathan, Beamforming
in correlated MISO systems with channel estimation error and feedback
delay, IEEE Trans. on Wireless Commun., vol. 10, no. 8, pp. 25922602,
2011.
[22] O. Lieberman, A Laplace approximation to the moments of a ratio of
quadratic forms, Biometrika, vol. 81, no. 4, pp. 681690, Dec 1994.
[23] H. Gao, P. J. Smith and M. V. Clark, Theoretical reliability of MMSE
linear diversity combining in Rayleigh-fading additive interference channels, IEEE Trans. Commun., vol. 46, no. 5, pp. 666672, May. 1998.

10

Dushyantha Basnayaka (S11-M12) was born

in 1982 in Colombo, Sri Lanka. He received the
B.Sc.Eng degree with 1st class honors from the
University of Peradeniya, Sri Lanka, in Jan 2006.
PLACE
He is currently working towards for his PhD degree
PHOTO
in Electrical and Computer Engineering at the UniHERE
versity of Canterbury, Christchurch, New Zealand.
He was an instructor in the Department of Electrical
and Electronics Engineering at the University of
Peradeniya from Jan 2006 to May 2006. He was
a system engineer at MillenniumIT (a member company of London Stock Exchange group) from May 2006 to Jun. 2009. Since
Jul. 2009 he is with the communication research group at the University of
Canterbury, New Zealand.
D. A. Basnayaka is a recipient of University of Canterbury International Doctoral Scholarship for his doctoral studies at UC. His current research interest
includes all the areas of digital communication, especially macrodiversity
wireless systems. He holds one pending US patent as a result of his doctoral
studies at UC.

Peter Smith (M93-SM01) received the B.Sc degree in Mathematics and the Ph.D degree in Statistics from the University of London, London, U.K.,
in 1983 and 1988, respectively. From 1983 to 1986
PLACE
he was with the Telecommunications Laboratories
PHOTO
at GEC Hirst Research Centre. From 1988 to 2001
HERE
he was a lecturer in statistics at Victoria University,
Wellington, New Zealand. Since 2001 he has been a
Senior Lecturer and Associate Professor in Electrical and Computer Engineering at the University of
Canterbury in New Zealand. Currently, he is a full
Professor at the same department.
His research interests include the statistical aspects of design, modeling
and analysis for communication systems, especially antenna arrays, MIMO,
cognitive radio and relays.

Philippa Martin (S95-M01-SM06) received the

B.E. (Hons. 1) and Ph.D. degrees in electrical
and electronic engineering from the University of
Canterbury, Christchurch, New Zealand, in 1997
PLACE
and 2001, respectively. From 2001 to 2004, she
PHOTO
was a postdoctoral fellow, funded in part by the
HERE
New Zealand Foundation for Research, Science and
Technology (FRST), in the Department of Electrical and Computer Engineering at the University
of Canterbury. In 2002, she spent 5 months as a
visiting researcher in the Department of Electrical
Engineering at the University of Hawaii at Manoa, Honolulu, Hawaii, USA.
Since 2004 she has been working at the University of Canterbury as a lecturer
and then as a senior lecturer. Currently, she is an Associate Professor at the
same department. In 2007, she was awarded the University of Canterbury,
College of Engineering young researcher award. She served as an Editor for
the IEEE Transactions on Wireless Communications 2005-2008 and regularly
serves on technical program committees for IEEE conferences.
Her current research interests include multilevel coding, error correction
coding, iterative decoding and equalization, space-time coding and detection,
cognitive radio and cooperative communications in particular for wireless
communications