1416
3, MARCH 2012
Correspondence
A Comparison of Two MIMO Relaying Protocols
in Nakagami-m Fading
Maged Elkashlan, Phee Lep Yeoh, Nan Yang,
Trung Q. Duong, and Cyril Leung
Abstract—Transmit antenna selection with receive maximal-ratio com-
bining (TAS/MRC) and transmit antenna selection with receive selection
combining (TAS/SC) are two attractive multiple-input–multiple-output
(MIMO) protocols. In this paper, we present a framework for the compar-
ative analysis of TAS/MRC and TAS/SC in a two-hop amplify-and-forward
relay network. In doing so, we derive exact and asymptotic expressions for
the symbol error rate (SER) in Nakagami-m fading. Using the asymptotic
expressions, the SNR gap between the two protocols is quantified. Given
that the two protocols maintain the same diversity order, we show that
the SNR gap is entirely dependent on the array gain. Motivated by this,
we derive the SNR gap as a simple ratio of the respective array gains
of the two protocols. This ratio explicitly takes into account the impact
of the number of antennas and the fading severity parameter m. In
addition, we address the fundamental question of “How to allocate the
total transmit power between the source and the relay in such a way that
the SER is minimized?” Our answer is given in the form of new compact
expressions for the power allocation factor, which is a practical design tool
that optimally distributes the total transmit power in the network.
Index Terms—Multiple-input multiple-output (MIMO), Nakagami,
relays.
I. I NTRODUCTION
Wireless relaying is pivotal to extending network coverage and
providing connectivity in hard-to-reach areas [1], [2]. The question
is how to deliver high data rates over large areas at relatively low
overhead cost. To meet the demand for high-data-rate transmission,
multiple-input–multiple-output (MIMO) processing is fundamental
[3], [4]. MIMO relaying, where the source, relay, and destination
have multiple antennas, has garnered much research interest. From
a capacity perspective, the information theoretic analysis of MIMO
relaying has been reported in [5]–[7]. More recently, from an error
rate performance perspective, the design and analysis of such networks
have been presented in [8]–[10]. The advantages of MIMO relaying,
however, come at a cost of complex transceivers and power-intensive
signal-processing modules. In particular, the high power consumption
required to operate multiple radio-frequency (RF) chains presents a
Manuscript received July 9, 2011; revised November 14, 2011; accepted
December 22, 2011. Date of publication January 23, 2012; date of cur-
rent version March 21, 2012. The review of this paper was coordinated by
Prof. Y. Ma.
M. Elkashlan is with the School of Electronic Engineering and Computer
Science, Queen Mary, University of London, E1 4NS London, U.K. (e-mail:
maged.elkashlan@eecs.qmul.ac.uk).
P. L. Yeoh is with the Department of Electrical and Electronic Engineer-
ing, The University of Melbourne, Melbourne, Vic. 3010, Australia (e-mail:
phee.yeoh@unimelb.edu.au).
N. Yang is with the Wireless Networking Technologies Laboratory, CSIRO
ICT Center, Marsfield, NSW 2122, Australia (e-mail: jonas.yang@csiro.au).
T. Q. Duong is with Blekinge Institute of Technology, SE-371 79 Karlskrona,
Sweden (e-mail: dqt@bth.se).
C. Leung is with the Department of Electrical and Computer Engineering,
University of British Columbia, Vancouver, BC V6T 1Z4, Canada (e-mail:
cleung@ece.ubc.ca).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2012.2185259
0018-9545/$31.00 © 2012 IEEE
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO.
major challenge to the direct implementation of MIMO relaying in
battery-operated wireless ad hoc and sensor networks.
An attractive solution to the aforementioned challenge is to apply
MIMO selection protocols in MIMO relaying. Among them, transmit
antenna selection with receive maximal-ratio combining (TAS/MRC)
in MIMO relaying was proposed in [11]. In TAS/MRC relaying, a
single antenna is selected at the transmitters, and all the antennas at
the receivers are MRC combined. To further reduce the number of RF
chains in the network, transmit antenna selection with receive selection
combining (TAS/SC) in MIMO relaying was considered in [12]. In
TAS/SC relaying, a single antenna at the transmitters and a single
antenna at the receivers are jointly selected. Several transmit antenna
selection strategies for MIMO relaying were introduced in [13]. While
the aforementioned works stand on their own merits, the purpose
of this paper is to present a unified framework for the comparative
analysis of TAS/MRC and TAS/SC in a two-hop amplify-and-forward
relay network.
We begin with the question: “What performance improvement does
TAS/MRC relaying offer relative to TAS/SC relaying?” In tackling
this, we derive exact and asymptotic expressions for the symbol error
rate (SER) in Nakagami-m fading, where m1 and m2 are the fading
parameters in the source-to-relay link and the relay-to-destination link,
respectively. Our asymptotic expressions demonstrate that the two pro-
tocols offer the same diversity order of NR × min(m1 NS , m2 ND ),
where NS , NR , and ND are the number of antennas at the source, relay,
and destination, respectively. This leads to the conclusion that the SNR
gap between the two protocols is explicitly dependent on the array
gain. Motivated by this, we pursue the question: “Which distinct para-
meters determine this improvement in Nakagami-m fading?” In doing
so, we formulate the SNR gap as a simple ratio of the array gains of the
two protocols. This ratio captures the impact of the number of antennas
and the impact of the m fading parameters on the SNR gap. To gain
further insights, we examine the important question: “What is the op-
timal transmit power allocation that minimizes the SER in each MIMO
relaying protocol?” In addressing this, we derive easy-to-calculate
expressions for the power allocation factor to optimally allocate the
total transmit power between the source and the relay. The power
allocation factor is a simple yet powerful design tool that maximizes
the network performance without expending additional resources.
II. P ROTOCOL D ESCRIPTION
Consider the distributed MIMO relay network shown in Fig. 1,
where the source, relay, and destination are equipped with NS , NR ,
and ND antennas, respectively. The source has no direct link to the
destination. We detail two distributed MIMO relaying protocols. In
each protocol, the transmission from the source to the destination via
the relay is facilitated by a two-step process. In the following, ES
is the source transmit power, ER is the relay transmit power, N0 is
the variance of the additive white Gaussian noise (AWGN),  ·  is
the Euclidean norm, | · | is the absolute value, (·)† is the conjugate
transpose, and IN is the N × N identity matrix.
A. TAS/MRC Relaying
Step 1: In the source-to-relay link, a single antenna from NS
antennas at the source is selected to maximize the received SNR at the
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012 1417

Step 2: In the relay-to-destination link, a single antenna from NR

antennas at the relay and a single antenna from ND antennas at the
destination are jointly selected to maximize the received SNR at the

destination. In this protocol, the relay applies a variable gain of G =
1/ ES |hmax
1 |2 + τ N0 . The received signal vector at the destination
is given by
Fig. 1. MIMO relaying with NS , NR , and ND antennas.

relay. In doing so, all the NR antennas at the relay are MRC combined. yRD = ER hmax
2 GySR + nD (4)
The received signal vector at the relay is given by
where nD is the AWGN with E[nD ] = N0 . In (4), the magnitude
 of the channel coefficient in the relay-to-destination link is |hmax |=
ySR = ES hmax x + nR (1) 2
1
max1≤i≤NR ,1≤j≤ND {|hi,j 2 |}, where h2
i,j
is the i.i.d. Nakagami-m
fading channel coefficient between the ith transmit antenna at the relay
where x is the transmit symbol, and nR is the NR × 1 AWGN
and the jth receive antenna at the destination.
vector satisfying E[nR n†R ] = INR N0 . In (1), we define hmax 1 =
The end-to-end SNR is expressed as γTAS/SC = Υ1 Υ2 /(Υ1 +
max1≤i≤NS {hi1 }, where hi1 is the NR × 1 complex channel vector
Υ2 + τ ), where Υ1 = (ES /N0 )|hmax 1 |2 and Υ2 = (ER /N0 )|hmax 2 |2
between the ith transmit antenna at the source and the NR receive
are the instantaneous SNRs in the source-to-relay link and the relay-
antennas at the relay, with independent and identically distributed
to-destination link, respectively. As in TAS/MRC relaying, CSI is
(i.i.d.) Nakagami-m fading entries.
estimated in each link by the receiver using pilot symbols sent by the
Step 2: In the relay-to-destination link, a single antenna from NR
transmitter.
antennas at the relay is selected to maximize the received SNR at
the destination. In doing so, all the ND antennas at the destination
areMRC combined. Prior to forwarding, a variable gain of G = III. S YMBOL E RROR R ATES
1/ ES hmax 1 4 + τ N0 hmax
1 2 is applied to counteract the fading
in the source-to-relay link. We consider two variable gain protocols, In this section, we present new exact and asymptotic expressions
i.e., channel-noise-based (CNB) relaying that takes into account the for the SER of TAS/MRC relaying and TAS/SC relaying. In deriving
noise variance with τ = 1 and channel-based (CB) relaying that does the SER, we focus on CB relaying with τ = 0. CB relaying is
not account for the noise variance with τ = 0. analytically tractable, and its performance asymptotically approaches
The received signal vector at the destination is given by that of CNB relaying at high SNRs. We begin by presenting the exact
cdf of the end-to-end SNRs, which is denoted by FγTAS/MRC (γ)

yRD = ER hmax 
GySR + nD (2) and FγTAS/SC (γ). Based on these, the SER is directly evaluated
2
according to

where ySR = (hmax
1 )† ySR , and nD is the ND × 1 AWGN vector  ∞
†
satisfying E[nD nD ] = IND N0 . In (2), the channel vector norm in a b e−bγ
the relay-to-destination link is hmax  = max1≤i≤NR {hi2 }, where
Pχ = √ Fγχ (γ) dγ (5)
2 2 π γ
h2 is the ND × 1 complex channel vector between the ith transmit
i 0
antenna at the relay and the ND receive antennas at the destination,
with i.i.d. Nakagami-m fading entries. where χ ∈ {TAS/MRC, TAS/SC}. This applies
√ to all general modu-
The end-to-end SNR is written as γTAS/MRC = Γ1 Γ2 /(Γ1 + Γ2 + lations with SER of the form P = E[aQ( 2bγ)], where a and b are
τ ), where Γ1 = (ES /N0 )hmax 1 2 and Γ2 = (ER /N0 )hmax 2 2 are constellation-specific constants [8].
the instantaneous SNRs in the source-to-relay link and the relay-to-
destination link, respectively. In each link, we assume that channel A. TAS/MRC Relaying
state information (CSI) is estimated by the receiver using pilot symbols
sent by the transmitter. In the source-to-relay link, the source transmits Exact: The exact cdf of γTAS/MRC is evaluated by applying alge-
pilot symbols, and the relay feeds back the index of the strongest braic manipulation along similar lines to [14], according to
transmit antenna to the source. In the relay-to-destination link, the
∞  
relay broadcasts pilot symbols, and the destination feeds back the γ2
index of the strongest transmit antenna to the relay. FγTAS/MRC (γ) = 1 − F̃Γ1 γ+ fΓ2 (ω + γ) dω (6)
ω
0

B. TAS/SC Relaying
Step 1: In the source-to-relay link, a single antenna from NS
antennas at the source and a single antenna from NR antennas at the
relay are jointly selected to maximize the received SNR at the relay.
The received signal at the relay is given by
where F̃Γ1 (γ) = 1 − FΓ1 (γ) is the complementary cdf of Γ1 , fΓ2 (·)
is the pdf of Γ2 , and ω = Γ2 − γ.
In the source-to-relay link, the channel vector entries in hi1 follow a
Nakagami-m distribution with fading parameter m1 . As such, the cdf
of Γ1 is written as [4]
 ⎡  k ⎤NS
ySR = ES hmax
1 x + nR (3)
⎢ −γ
m1

m1 NR −1 γm
γ1
1
⎥
FΓ1 (γ) = ⎣1 − e γ1
⎦ (7)
where nR is the AWGN, with E[nR ] = N0 . In (3), the magnitude k!
k=0
of the channel coefficient in the source-to-relay link is |hmax 1 |=
max1≤i≤NS ,1≤j≤NR {|hi,j 1 |}, where h1
i,j
is the i.i.d. Nakagami-m
fading channel coefficient between the ith transmit antenna at the where γ 1 = ES /N0 is the average SNR between each antenna pair in
source and the jth receive antenna at the relay. the source-to-relay link. In the relay-to-destination link, the channel
1418 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012

vector entries in hi2 follow a Nakagami-m distribution with fading  j−i−σ−ς−m2 ND + 12

q 0 m1 (r0 + 1)m2
parameter m2 . Accordingly, the pdf of Γ2 is written as × b+ +
γ1 γ2
 m2 ND −γ
m2
m2 γ m2 ND −1 e γ 2 × (r0 + 1)i−j−1 2i−j−σ−ς−m2 ND − 2
3

fΓ2 (γ) = NR
γ2 Γ(m2 ND )  
Γ σ + ς + m 2 ND − 1
−j+i
⎡  l ⎤NR −1 × 2
Γ(σ + ς + m2 ND + 1)
⎢ −γ
m2

m2 ND −1 γm
γ
2
2 ⎥  
× ⎣1 − e γ2
⎦ (8) 3
l! × Γ σ + ς + m 2 ND + +j−i
l=0 2

σ + ς + m 2 ND − 1
−j+i
where γ 2 = ER /N0 is the average SNR between each antenna pair in × 2 F1 2
,
2
the relay-to-destination link.
Substituting (7) and (8) into (6) and then applying the identity σ + ς + m 2 ND + 1
−j+i
× 2
;
[15, eq. (3.471.9)], the cdf of γTAS/MRC is presented in 2
 
q0 (r0 + 1)m1 m2
FγTAS/MRC (γ) σ + ς + m2 ND + 1; 1 − 4
γ1γ2
  
2NR

NS
NS NR −1
NR − 1  −2 
=1− (−1)q0 +r0 −1 q 0 m1 (r0 + 1)m2
Γ(m2 ND ) q0 r0 × b+ + (10)
q0 =1 r0 =0 γ1 γ2
qk−1  

m1 NR −1

qk−1   1 qk −qk+1 by substituting (9) into (5) and applying identities [15, eq. (3.326.2)]
×
qk k! and [17, eq. (2.16.6.3)] to solve the resulting integrals after some
k=1 qk =0
algebraic manipulations. It is important to note that (10) consists of

m2 ND −1 rl−1 


rl−1   1 rl −rl+1 finite summations of the hypergeometric function 2 F1 [a, b, c, z] [16,
× eq. (15.1.1)], which is available in commonly used numerical software
rl l!
l=1 rl =0 packages such as Matlab and Mathematica.
2 ND −1    Asymptotic: To evaluate the asymptotic SER, we begin by deriving

σ ς+m

σ ς + m 2 ND − 1 the first-order expansion of our cdf expression in (9). A key observa-

×
i j tion is that the first-order expansion of the cdf must be evaluated for
i=0 j=0
three separable cases reflecting the fading severities in the source-to-
 ς+m2 ND − j−i+1
2
 σ+ j−i+1
2  q  j−i+1 relay link and the relay-to-destination link. In each case, we denote
m2 m1 0 2
× γ 2 = ργ 1 and apply the Maclaurin series expansion for Kν (z) from
γ2 γ1 r0 + 1 [15, eq. (8.446)] as γ 1 → ∞. Our first-order expansion of the cdf is
 q0 m1 (r +1)m2
 given by
−γ + 0
× γ σ+ς+m2 ND e γ1 γ2
 NR ×min(m1 NS ,m2 ND )
   γ
q0 (r0 + 1)m1 m2 Fγ∞ (γ) =Ψ
× Kj−i+1 2γ (9) TAS/MRC γ1
γ1γ2  
−NR ×min(m1 NS ,m2 ND )
+ o γ1 (11)
where Kν (·) is the νth-order modified Bessel function of the second
kind [16, eq. (9.6.2)]. In (9), we define σ = q1 + q2 + . . . + qm1 NR −1 where Ψ captures the three cases expressed as
and ς = r1 + r2 + . . . + rm2 ND −1 . ⎧ mm1 NS NR
Finally, we derive the exact SER of TAS/MRC relaying shown in ⎪
⎪ 1
, Case 1 : m1 NS < m2 ND
⎪
⎪ (m1 NR )!NS
⎪
⎪ m2 NR ND
PTAS/MRC ⎨ m N N2
m
, Case 2 : m1 NS > m2 ND
ρ 2 R D (m2 ND )!NR
Ψ=
√


NS NR − 1 ⎪
m N N
NS NR −1 m1 1 S R
a a bNR ⎪
⎪
= − (−1)q0 +r0 −1 ⎪
⎪ (m1 NR )! S m2 NR ND
N

2 Γ(m2 ND ) q0 r0 ⎪
⎩ m2
q0 =1 r0 =0 + ρm2 NR ND (m NR , Case 3 : m1 NS = m2 ND

m1 NR −1 qk−1 
 2 ND )!


qk−1   1 qk −qk+1 (12)
×
qk k! and o(·) represents the higher order terms (i.e., we write f (x) =
k=1 qk =0
o[g(x)] as x → x0 if limx→x0 (f (x)/g(x)) = 0).
rl−1  

m2 ND −1

rl−1   1 rl −rl+1 The asymptotic SER of TAS/MRC relaying is obtained in closed
× form by substituting (11) into (5) and solving the resultant integral
rl l! using [15, eq. (3.326.2)], which results in
l=1 rl =0
1 

2 ND −1  
σ ς+m
 aΨΓ +q  
σ ς + m 2 ND − 1 ∞
PTAS/MRC = √2 (bγ 1 )−q + o γ −q (13)
× 2 π 1
i j
i=0 j=0
 σ  ς+m2 ND −j+i−1 where q = NR × min(m1 NS , m2 ND ). From (13), we find that TAS/
m1 m2 MRC achieves a diversity order of Gd = NR × min(m1 NS , m2 ND ),
×
γ1 γ2 which represents the slope of the asymptotic SER curve defined as
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012 1419

Δ
Gd = limγ 1 →∞ (− log PTAS/MRC / log γ 1 ) [18]. We further derive which is found by substituting (16) and (17) into (15), invoking
the array gain from (13) as [15, eq. (3.324.1)] to expand the incomplete gamma function, and
 √  N ×min(m N ,m N ) 1 applying the identity in [15, eq. (3.471.9)] to solve the resulting
2 π R 1 S 2 D
integrals. In (18), we define σ = q1 + q2 + · · · + qm1 −1 and ς =
Ga,TAS/MRC = b 1  (14) r1 + r2 + . . . + rm2 −1 .
aΨΓ 2 + q
We derive the exact SER in closed form by substituting (18) into (5).
in which the SNR advantage of the asymptotic SER curve relative to a After applying identities [15, eq. (3.326.2)] and [17, eq. (2.16.6.3)] to
reference curve of γ −G
1
d
[18] is presented. simplify the integrals and performing some algebraic manipulation, we
obtain the exact SER in

B. TAS/SC Relaying PTAS/SC

Exact: The exact cdf of γTAS/SC is evaluated according to √ NS NR NR ND −1   
a a bNR ND

NS NR NR ND − 1
∞   = −
γ 2 2 Γ(m2 ) q0 r0
FγTAS/SC (γ) = 1 − F̃Υ1 γ+ fΥ2 (ω + γ) dω (15) q0 =1 r0 =0
ω  k−1    q −q 
0  q

m1 −1
qk−1 1 k k+1
× (−1)q0 +r0 −1
qk k!
where F̃Υ1 (γ) = 1 − FΥ1 (γ) is the complementary cdf of Υ1 , fΥ2 (·) k=1 qk =0

m2 −1 rl−1 


rl−1   1 rl −rl+1
is the pdf of Υ2 , and ω = Υ2 − γ.
In the source-to-relay link, the cdf of Υ1 is given by ×
⎛   ⎞k rl l!

 NS NR 
l=1 rl =0
NS NR Γ m1 , γ m 1

2 −1   
γ
(−1)k−1 ⎝ ⎠

1
FΥ1 (γ) = 1 − σ ς+m
σ ς + m2 − 1
k=1
k Γ(m1 ) ×
i j
(16) i=0 j=0
 σ  ς+m2 −j+i−1
where Γ(·, ·) is the incomplete gamma function [16, eq. (6.5.3)]. In the m1 m2 3
× 2i−j−σ−ς−m2 − 2
relay-to-destination link, the pdf of Υ2 is given by [19] γ1 γ2
 m2 −γ
m2  j−i−σ−ς−m2 + 12
m2 γ m2 −1 e γ2
× b+
q 0 m1
+
(r0 + 1)m2
(r0 + 1)i−j−1
fΥ2 (γ) = NR ND γ1 γ2
γ2 Γ(m2 )
 
⎛   ⎞l Γ(σ + ς + m2 − 1
− j + i)Γ σ + ς + m2 + 3
+j−i

NR ND −1  Γ  m2 , γ m2 × 2 2
NR ND − 1 γ2 Γ(σ + ς + m2 + 1)
× (−1)l ⎝ ⎠ . (17) 
l Γ(m2 ) σ + ς + m2 − 1
− j + i σ + ς + m2 + 1
−j+i
l=0
× 2 F1 2
, 2
;
2 2
The cdf of γTAS/SC is presented in  
q0 (r0 + 1)m1 m2
FγTAS/SC (γ) σ + ς + m2 + 1; 1 − 4
γ1γ2
2NR ND  −2 
=1− q 0 m1 (r0 + 1)m2
Γ(m2 ) × b+ + . (19)
γ1 γ2

NS NR NR ND −1   
NS NR NR ND − 1
× (−1)q0 +r0 −1 Asymptotic: Following similar steps to the derivation of (11), we
q0 r0
q0 =1 r0 =0 evaluate the first-order expansion of the cdf in (18) as γ 1 → ∞
qk−1  

qk−1   1 qk −qk+1
according to three separable cases. The result is

m1 −1

×  NR ×min(m1 NS ,m2 ND )

qk k! γ
k=1 qk =0 Fγ∞ (γ) =Φ
TAS/SC γ1

m2 −1 rl−1 


rl−1   1 rl −rl+1 
−NR ×min(m1 NS ,m2 ND )

× + o γ1 (20)
rl l!
l=1 rl =0

2 −1    where

σ ς+m
ς + m2 − 1
×
σ ⎧ mm1 NS NR
i j ⎪
⎪ 1
, Case 1 : m1 NS < m2 ND
i=0 j=0 ⎪
⎪ (m1 !)NS NR
⎪
⎪ m2 NR ND
 σ+ j−i+1  ς+m2 − j−i+1 ⎨ m NmN2 , Case 2 : m1 NS > m2 ND
2 2 ρ 2 R D (m2 !) R ND
N
m1 m2 Φ=
× ⎪
m N N
m1 1 S R
γ1 γ2 ⎪
⎪
⎪
⎪ (m1 !)NS NR
 q  j−i+1 q  ⎪
⎩ m N N
m 2 R D
2 −γ 0 m1 +
(r0 +1)m2 2
+ ρm2 NR N , Case 3 : m1 NS = m2 ND.
0
× γ σ+ς+m2 e γ1 γ2 D (m 2 !)
NR ND

r0 + 1 (21)
  
q0 (r0 + 1)m1 m2 Inserting (20) into (5) and using [15, eq. (3.326.2)] to solve the
× Kj−i+1 2γ (18)
γ1γ2 resultant integral, we derive the asymptotic SER of TAS/SC relaying
1420 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012

in closed form as m1 NS = m2 ND or the higher order expansion of the SER for

1  Case 1: m1 NS < m2 ND and Case 2: m1 NS > m2 ND .
aΦΓ +q   The optimal power allocation is determined by solving
∞
PTAS/SC = √2 (bγ 1 )−q + o γ −q
1 (22) ∞
2 π ∂PTAS/MRC /∂η = 0. After some mathematical manipulations,
we find that the optimal power allocation is the value of η, which
where q = NR × min(m1 NS , m2 ND ). As indicated in (22), TAS/SC satisfies
relaying retains a diversity order of Gd = NR × min(m1 NS ,  m2 NR ND −m1 NS NR
m2 ND ). This diversity order is the same as that of TAS/MRC relaying. (1 − η)m2 NR ND +1 ET b
From (22), the array gain of TAS/SC relaying is given by η m1 NS NR +1 N0
 √  N ×min(m N ,m N )1
1 
m2m2 NR ND +1 ND Γ
R 1 S 2 D
2 π + m2 NR ND (m1 NR )!NS
Ga,TAS/SC = b 1  . (23) =  2
 . (26)
aΦΓ 2 + q mm
1
1 NS NR +1
NS Γ 1
2
+ m1 NS NR (m2 ND )!NR

Comparing (14) and (23), it is clear that TAS/MRC relaying and For the special case of m1 NS = m2 ND , (26) reduces to
TAS/SC relaying exhibit different array gains. ⎛ m1 NS NR NS
⎞−1
m1 NS NR +1
m2 (m1 NR )! m1 NS NR +1 ⎠
η = ⎝1 + m1 NS NR
. (27)
NR
C. Comparison of TAS/MRC Relaying and TAS/SC Relaying m1 NS NR +1
m1 (m2 ND )! m1 NS NR +1
Noting that TAS/MRC relaying and TAS/SC relaying offer the
same diversity order, we focus on their respective array gains. The Considering Rayleigh fading (i.e., m1 = m2 = 1) and equal number
SNR gap between TAS/MRC relaying and TAS/SC relaying is defined of antennas at the source, relay, and destination (i.e., NS = NR =
as the ratio of the array gains of TAS/MRC relaying in (14) and ND ), (27) reduces to η = 0.5.
TAS/SC relaying in (23). Using (12) and (21) and performing some
mathematical manipulations, the ratio is given as (24), shown at the
B. TAS/SC Relaying
bottom of the page.
Following the same procedure as outlined in the previous subsec-
tion, we optimize both ES and ER by jointly considering the source-to-
IV. O PTIMAL P OWER A LLOCATION relay link and the relay-to-destination link. The objective function for
the optimal power allocation that minimizes the SER is derived based
Power allocation is an important design criterion that allows net-
on (22). The objective function is given by
work service providers to maximize performance without squandering
 1 
valuable resources. In this section, we apply our new asymptotic solu- (m1 N0 )m1 NS NR Γ + m 1 NS NR
∞ a
tions to efficiently derive the optimal power allocation that minimizes PTAS/SC = √ 2
2 π (ηET b)m1 NS NR (m1 )!NS NR
the SER. To this end, we consider a general weight for the power
allocation factor η such that ES = ηET and ER = (1 − η)ET , where 1 
0 < η < 1 and ET is the total power constraint. (m2 N0 )m2 NR ND Γ 2
+ m 2 NR ND
+ (28)
((1 − η)ET b) m2 NR ND
(m2 )!NR ND
A. TAS/MRC Relaying
which is formulated by examining all three cases of Φ in (21). Solving
To evaluate the optimal power allocation, we consider both ES and ∞
∂PTAS/SC /∂η = 0, the optimal power allocation is the value of η that
ER by taking into account the joint contributions of the source-to-relay satisfies
link and the relay-to-destination link. The objective function for the
 m2 NR ND −m1 NS NR
optimal power allocation that minimizes the SER is derived based on (1 − η)m2 NR ND +1 ET b
(13). As such, it is given by η m1 NS NR +1 N0
 1 
a (m1 N0 )m1 NS NR Γ + m 1 NS NR 1 
∞
PTAS/MRC = √ 2 m2m2 NR ND +1 ND Γ + m2 NR ND (m1 )!NS NR
2 π (ηET b)m1 NS NR (m 1 NR )!NS =  12  . (29)
mm 1 NS NR +1
NS Γ + m1 NS NR (m2 )!NR ND

1 2
1
(m2 N0 )m2 NR ND Γ 2
+ m 2 NR ND For the special case of m1 NS = m2 ND , (29) reduces to
+ (25)
((1 − η)ET b) m2 NR ND
(m2 ND )!NR ⎛ m2 NR ND
⎞−1
NS NR
m1 NS NR +1
m2 (m1 )! m1 NS NR +1
which is formulated by examining all three cases of Ψ in (12). Indeed, η = ⎝1 + m1 NS NR
⎠ . (30)
NR ND
m1 NS NR +1
(25) can be viewed as the first-order expansion of the SER for Case 3: m1 (m2 )! m1 NS NR +1

⎧  m N1 N
⎪
⎪ (m1 NR )!NS
⎪
1 S R
⎪ (m1 !) S R , Case 1 : m1 NS < m2 ND
⎪
⎨
N N
 1
Ga,TAS/MRC N m N N
(m2 ND )! R 2 D R
= , Case 2 : m1 NS > m2 ND (24)
Ga,TAS/SC ⎪
⎪ (m2 !)ND NR
⎪
⎪  m 1
⎪
⎩ (m1 !)−NS NR +(m2 !)−NR ND (m2 /(m1 ρ))m1 NR NS 1 NR NS
, Case 3 : m1 NS = m2 ND.
((m1 NR )!)−NS +((m1 NS )!)−NR (m2 /(m1 ρ))m1 NR NS
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012 1421

Fig. 2. SER of TAS/MRC relaying and TAS/SC relaying. Case 1: Relay-to- Fig. 3. SER of TAS/MRC relaying and TAS/SC relaying. Case 2: Source-to-
destination link is the stronger link. We set m1 = 1, NS = 1, m2 = 2, and relay link is the stronger link. We set m1 = 2, NS = 3, m2 = 1, and ND = 1.
ND = 3.

For Rayleigh fading (i.e., m1 = m2 = 1) and equal number of anten-

nas at the source and the destination (i.e., NS = ND ), (30) reduces to
η = 0.5, which is independent of NR .

V. R ESULTS AND D ISCUSSIONS

In this section, we highlight the impacts of the number of antennas
and the fading parameters on the performance gap between TAS/MRC
relaying and TAS/SC relaying. In the examples, “•” marks the Monte
Carlo simulations for TAS/MRC relaying, and “◦” marks the Monte
Carlo simulations for TAS/SC relaying. For both protocols, we con-
sider CNB relaying (τ = 1) in the Monte Carlo simulations. The solid
curves represent the exact SER in (10) and (19). The dashed lines
represent the asymptotic SER in (13) and (22). We see an excellent
agreement between the simulations and the exact analytical curves in
the medium-to-high-SNR regime. Furthermore, the asymptotic lines
accurately predict the diversity orders and array gains of both proto-
cols.
Figs. 2–4 consider binary phase-shift keying (BPSK) modulation
Fig. 4. SER of TAS/MRC relaying and TAS/SC relaying. Case 3: Source-
and plot the SER (which is equivalent to the BER for BPSK) of to-relay link and relay-to-destination link are in equilibrium. We set m1 = 1,
TAS/MRC relaying and TAS/SC relaying with equal average energies NS = 2, m2 = 1, and ND = 2.
at the source and the relay (i.e., ES = ER ). We address the following
three cases based on m1 NS and m2 ND . there is no SNR gap between the two protocols at high SNRs. This
observation is confirmed using (24).
A. Case 1: m1 NS < m2 ND
In this case, the relay-to-destination link is stronger than the source- C. Case 3: m1 NS = m2 ND
to-relay link. In Fig. 2, we set m1 = 1, NS = 1, m2 = 2, and ND = 3.
In this case, the source-to-relay link and the relay-to-destination
Given Gd = NR × min(m1 NS , m2 ND ), we identify that the diver-
link are in equilibrium. In Fig. 4, we set m1 = 1, NS = 2, m2 = 1,
sity order, which is reflected in the slope of the asymptotes, is directly
and ND = 2. Using (24), the SNR gap between the two protocols is
proportional to NR . As such, we see that the slope of the asymptotes
Ga,TAS/MRC /Ga,TAS/SC = 0.6247 dB, 1.5051 dB, and 1.8616 dB
increases according to Gd = 1, 2, and 3 for NR = 1, 2, and 3, respec-
for NR = 1, 2, and 3, respectively.
tively. We also see that, for the same diversity order, the SNR gap be-
Fig. 5 plots the SNR gap between TAS/MRC relaying and TAS/SC
tween the two protocols increases with increasing NR . Using (24), this
relaying using (24). The figure highlights the impact of m1 and m2
SNR gap is conveniently calculated as Ga,TAS/MRC /Ga,TAS/SC =
on the SNR gap. The source-to-relay link and relay-to-destination
0 dB, 1.5051 dB, and 2.5938 dB for NR = 1, 2, and 3, respectively.
link are in equilibrium with NS = NR = ND and m1 = m2 . Setting
NS = NR = ND = 2 and m1 = m2 = 1 provides an SNR gap of
B. Case 2: m1 NS > m2 ND Ga,TAS/MRC /Ga,TAS/SC = 1.5051 dB. We find that this gap in-
In this case, the source-to-relay link is stronger than the relay-to- creases with increasing m1 and m2 . For example, setting NS =
destination link. In Fig. 3, we set m1 = 2, NS = 3, m2 = 1, and NR = ND = 2 and m1 = m2 = 3 results in a larger SNR gap of
ND = 1. We observe from the figure that, for the same diversity order, Ga,TAS/MRC /Ga,TAS/SC = 2.1684 dB.
1422
Fig. 5. Ratio of the array gains of TAS/MRC relaying and TAS/SC relaying.
The source-to-relay link and relay-to-destination link are in equilibrium, with
NS = NR = ND and m1 = m2 .
Fig. 6. SER of TAS/MRC relaying and TAS/SC relaying. The source-to-relay
link and relay-to-destination link are in equilibrium, with ET = 20 dB, m1 =
2, NS = 2, NR = 2, m2 = 1, and ND = 4.
Fig. 6 plots the SER of TAS/MRC relaying and TAS/SC relaying
with BPSK modulation. The figure examines the impact of η on the
SER, where we set ET = 20 dB, m1 = 2, NS = 2, NR = 2, m2 =
1, and ND = 4. Based on (27), the SER for TAS/MRC relaying is
minimized at η = 0.65. As such, optimal power allocation is achieved
by assigning 65% of the total transmit power to the source and 35% to
the relay. Based on (30), the SER for TAS/SC relaying is minimized at
η = 0.58. The values of η for optimal power allocation are confirmed
in Fig. 6.
VI. C ONCLUSION
This paper has provided SER performance comparison between
TAS/MRC and TAS/SC relaying. To facilitate this, we have derived
new exact and asymptotic expressions for the SER in Nakagami-m
fading. Based on the asymptotic expressions, we have shown that
both protocols offer the same diversity order of NR × min(m1 NS ,
m2 ND ). This means that the SNR gap between the two protocols
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 61, NO. 3, MARCH 2012
is entirely dependent on the array gain. Motivated by this, we have
analyzed the SNR gap as a simple ratio of their respective array gains.
We have then examined the fundamental question of how to allocate
the total transmit power between the source and the relay such that
the SER is minimized. In addressing this question, we have derived a
simple and practical design rule for the power allocation factor of the
two protocols.
R EFERENCES
[1] D. Soldani and S. Dixit, “Wireless relays for broadband access,” IEEE
Commun. Mag., vol. 46, no. 3, pp. 58–66, Mar. 2008.
[2] S. Berger, M. Kuhn, and A. Wittneben, “Recent advances in amplify-
and-forward two-hop relaying,” IEEE Commun. Mag., vol. 47, no. 7,
pp. 50–56, Jul. 2009.
[3] Q. Li, G. Li, W. Lee, M.-I. Lee, D. Mazzarese, B. Clerckx, and Z. Li,
“MIMO techniques in WiMAX and LTE: A future overview,” IEEE Com-
mun. Mag., vol. 48, no. 5, pp. 86–92, May 2010.
[4] Z. Chen, J. Yuan, and B. Vucetic, “Analysis of transmit antenna
selection/maximal-ratio combining in Rayleigh fading channels,” IEEE
Trans. Veh. Technol., vol. 54, no. 4, pp. 1312–1321, Jul. 2005.
[5] B. Wang, J. Zhang, and A. H. Madsen, “On the capacity of MIMO relay
channels,” IEEE Trans. Inf. Theory, vol. 51, no. 1, pp. 29–43, Jan. 2005.
[6] H. Bolcskei, R. U. Nabar, O. Oyman, and A. J. Paulraj, “Capacity scaling
laws in MIMO relay networks,” IEEE Trans. Wireless Commun., vol. 5,
no. 6, pp. 1433–1444, Jun. 2006.
[7] Y. Fan and J. Thompson, “MIMO configurations for relay channels:
Theory and practice,” IEEE Trans. Wireless Commun., vol. 6, no. 5,
pp. 1774–1786, May 2007.
[8] H. A. Suraweera, P. J. Smith, A. Nallanathan, and J. S. Thompson,
“Amplify-and-forward relaying with optimal and suboptimal transmit
antenna selection,” IEEE Trans. Wireless Commun., vol. 10, no. 6,
pp. 1874–1885, Jun. 2011.
[9] C.-K. Wen, K.-K. Wong, and C. T. K. Ng, “On the asymptotic properties
of amplify-and-forward MIMO relay channels,” IEEE Trans. Commun.,
vol. 59, no. 2, pp. 590–602, Feb. 2011.
[10] Z. Zhou and B. Vucetic, “A cooperative beamforming scheme in MIMO
relay broadcast channels,” IEEE Trans. Wireless Commun., vol. 10, no. 3,
pp. 940–947, Mar. 2011.
[11] P. L. Yeoh, M. Elkashlan, and I. B. Collings, “Exact and asymptotic SER
of distributed TAS/MRC in MIMO relay networks,” IEEE Trans. Wireless
Commun., vol. 10, no. 3, pp. 751–756, Mar. 2011.
[12] H. A. Suraweera, G. K. Karagiannidis, Y. Li, H. K. Garg, A. Nallanathan,
and B. Vucetic, “Amplify-and-forward relay transmission with end-to-end
antenna selection,” in Proc. IEEE WCNC, Sydney, Australia, Apr. 2010,
pp. 1–6.
[13] X. Cai and G. B. Giannakis, “Performance analysis of combined transmit
selection diversity and receive generalized selection combining in rayleigh
fading channels,” IEEE Trans. Wireless Commun., vol. 3, no. 6, pp. 1980–
1983, Nov. 2004.
[14] T. A. Tsiftsis, G. K. Karagiannidis, P. T. Mathiopoulos, and
S. A. Kotsopoulos, “Nonregenerative dual-hop cooperative links with se-
lection diversity,” EURASIP J. Wireless Commun. Netw., vol. 2006, no. 2,
pp. 17862-1–17862-8, Apr. 2006.
[15] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and
Products, 6th ed. New York: Academic, 2000.
[16] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
with Formulas, Graphs, and Mathematical Tables, 10th ed. New York:
Dover, 1972.
[17] A. P. Prudnikov, Y. A. Brychkov, and O. I. Marichev, Integrals and Series
Volume 2: Special Functions, 1st ed. New York: Gordon and Breach,
1986.
[18] Z. Wang and G. B. Giannakis, “A simple and general parameterization
quantifying performance in fading channels,” IEEE Trans. Commun.,
vol. 51, no. 8, pp. 1389–1398, Aug. 2003.
[19] T. Gucluoglu and T. M. Duman, “Performance analysis of transmit and
receive antenna selection over flat fading channels,” IEEE Trans. Wireless
Commun., vol. 7, no. 8, pp. 3056–3065, Aug. 2008.