Antenna Selection for MIMO Systems with Closely Spaced Antennas.
ABSTRACT Physical size limitations in user equipment may force multiple antennas to be spaced closely, and this generates a considerable amount of mutual coupling between antenna elements whose effect cannot be neglected. Thus, the design and deployment of antenna selection schemes appropriate for next generation wireless standards such as 3GPP long term evolution (LTE) and LTE advanced needs to take these practical implementation issues into account. In this paper, we consider multipleinput multipleoutput (MIMO) systems where antenna elements are placed side by side in a limitedsize linear array, and we examine the performance of some typical antenna selection approaches in such systems and under various scenarios of antenna spacing and mutual coupling. These antenna selection schemes range from the conventional hard selection method where only part of the antennas are active, to some newly proposed methods where all the antennas are used, which are categorized as soft selection. For the cases we consider, our results indicate that, given the presence of mutual coupling, soft selection can always achieve superior performance as compared to hard selection, and the interelement spacing is closely related to the effectiveness of antenna selection. Our work further reveals that, when the effect ofmutual coupling is concerned, it is still possible to achieve better spectral efficiency by placing a few more than necessary antenna elements in user equipment and applying an appropriate antenna selection approach than plainly implementing the conventional MIMO system without antenna selection.

Article: Joint antenna selection and adaptive modulation in MIMO systems in the presence of CSI imperfection
[Show abstract] [Hide abstract]
ABSTRACT: This paper presents the strategy antenna selection with an adapt scheme for Multiple Input Multiple systems. The proposed scheme Average Spectral Efficiency (ASE) power for a given Bit Error Rate (B This system provided the advan antenna selection and adaptiv technique Also due to the practical estimates the channel perfectly , we effect of Imperfect Channel Sta (ICSI) on our system.01/2011;  SourceAvailable from: iet.ntnu.no
 [Show abstract] [Hide abstract]
ABSTRACT: It is well known that the capacity of spatial multiplexing multipleinput multipleoutput (SMMIMO) system employing optimal antenna selection can significantly outperform a system without selection for same number of costly radio frequency chains. However, it requires an exhaustive search for the optimal selection (OS) that grows exponentially with the available number of transmit (u) and receive (m) antennas. In this paper, a novel low complexity receive antenna selection (RAS) technique is proposed for SMMIMO to maximize the channel capacity over correlated Rayleigh fading environment. It is based on the Euclidean norms of channel matrix rows and the corresponding phase differences due to their direct impact on the capacity. Extensive analysis and simulations have shown near optimal performance for any signaltonoiseratio and correlation values with low complexity of ${\mathcal{O} \left({u^{2}m}\right)}$ vector calculations. This technique provides fast RAS to capture most of the capacity gain promised by multiple antenna systems over different channel conditions. Furthermore, it enables efficient spectrum utilization for next generation wireless communications.Wireless Personal Communications 06/2013; 70(4). · 0.98 Impact Factor
Page 1
Hindawi Publishing Corporation
EURASIP Journal on Wireless Communications and Networking
Volume 2009, Article ID 739828, 11 pages
doi:10.1155/2009/739828
Research Article
AntennaSelectionforMIMO Systemswith
CloselySpaced Antennas
YangYang,1Rick S.Blum,1and SanaSfar2
1Department of Electrical and Computer Engineering, Lehigh University, 19 Memorial Drive West, Bethlehem, PA 18015, USA
2CTO Office, InterDigital Communications, LLC, 781 Third Avenues, King of Prussia, PA 19406, USA
Correspondence should be addressed to Yang Yang, yay204@lehigh.edu
Received 1 February 2009; Revised 18 May 2009; Accepted 28 June 2009
Recommended by Angel Lozano
Physical size limitations in user equipment may force multiple antennas to be spaced closely, and this generates a considerable
amount of mutual coupling between antenna elements whose effect cannot be neglected. Thus, the design and deployment
of antenna selection schemes appropriate for next generation wireless standards such as 3GPP long term evolution (LTE) and
LTE advanced needs to take these practical implementation issues into account. In this paper, we consider multipleinput
multipleoutput (MIMO) systems where antenna elements are placed side by side in a limitedsize linear array, and we examine
the performance of some typical antenna selection approaches in such systems and under various scenarios of antenna spacing
and mutual coupling. These antenna selection schemes range from the conventional hard selection method where only part of the
antennas are active, to some newly proposed methods where all the antennas are used, which are categorized as soft selection. For
the cases we consider, our results indicate that, given the presence of mutual coupling, soft selection can always achieve superior
performanceas comparedtohardselection, andtheinterelementspacing isclosely relatedtotheeffectiveness ofantennaselection.
Ourworkfurtherrevealsthat,whentheeffectofmutualcouplingisconcerned,itisstillpossibletoachievebetterspectralefficiency
byplacingafewmorethannecessaryantennaelementsinuserequipmentandapplyinganappropriateantennaselectionapproach
than plainly implementing the conventional MIMO system without antenna selection.
Copyright © 2009 Yang Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1.Introduction
The multipleinput multipleoutput (MIMO) architecture
has been demonstrated to be an effective means to boost
the capacity of wireless communication systems [1], and
has evolved to become an inherent component of various
wireless standards, including the nextgeneration cellular
systems 3GPP long term evolution (LTE) and LTE advanced.
For example, the use of a MIMO scheme was proposed
in the LTE standard, with possibly up to four antennas at
the mobile side, and four antennas at the cell site [2]. In
MIMO systems, antenna arrays can be exploited in two
different ways, which are [3]: diversity transmission and
spatial multiplexing. However, in either case, one main
problem involved in the implementation of MIMO systems
is the increased complexity, and thus the cost. Even though
the cost for additional antenna elements is minimal, the
radio frequency (RF) elements required by each antenna,
which perform the microwave/baseband frequency transla
tion, analogtodigital conversion, and so forth, are usually
costly.
These complexity and cost concerns with MIMO have
motivated the recent popularity of antenna selection (AS)—
an attractive technique which can alleviate the hardware
complexity, and at the same time capture most of the
advantages of MIMO systems. In fact, for its low user
equipment(UE)complexity,AS(transmit)iscurrentlybeing
considered as a baseline of the singleuser transmit diversity
techniques in the LTE uplink which is a MIMO single carrier
frequency division multiple access (SCFDMA) system [4].
Further, when it comes to the RF processing manner, AS can
be categorized into two groups: (1) hard selection, where
only part of the antennas are active and the selection is
implemented in the RF domain by means of a set of switchs
(e.g., [5–7]); (2) soft selection, where all the antennas are
active and a certain form of transformation is performed
Page 2
2EURASIP Journal on Wireless Communications and Networking
in the RF domain upon the received signals across all the
antennas (e.g., [8–10]).
A considerable amount of research efforts have been
dedicated to the investigation of AS, and have solidly
demonstrated the theoretical benefits of AS (see [3] for a
tutorial treatment). However, previous works largely ignore
the hardware implementation issues related to AS. For
instance,thephysicalsizeofUEsuchasmobileterminalsand
mobile personal assistants, are usually small and invariable,
and the space allocated for an antenna array is limited. Such
limitationmakestheclosespacingbetweenantennaelements
a necessity, inevitably leading to mutual coupling [11], and
correlated signals. These issues have caught the interest of
some researchers, and the capacity of conventional MIMO
systems (without AS) under the described limitations and
circumstanceswasinvestigated,amongothers,in[12–17].To
give an example, the study in [12] shows that as the number
of receive antenna elements increases in a fixedlength array,
the system capacity firstly increases to saturate shortly after
the mutual coupling reaches a certain level of severeness; and
drops after that.
Form factors of UE limit the performance promised
by MIMO systems, and can further affect the proper
functionality of AS schemes. These practical implementation
issues merit our attention when designing and deploying AS
schemes for the 3GPP LTE and LTE advanced technologies.
There exist some interesting works, such as [18, 19] which
consider AS in size sensitive wireless devices to improve
the system performance. But in general, results, conclusions,
and ideas on the critical implementation aspects of AS
in MIMO systems still remain fragmented. In this paper,
through electromagnetic modeling of the antenna array and
theoreticalanalysis,weproposeacomprehensivestudyofthe
performance of AS, to seek more effective implementation of
ASinsizesensitiveUEemploying MIMOwherebothmutual
coupling and spatial correlation have a strong impact. In this
process, besides the hybrid selection [5–7], a conventional
yet popular hard AS approach, we are particularly interested
in examining the performance potential of some typical
soft AS schemes, including the FFTbased selection [9] and
the phaseshiftbased selection [10], that are very appealing
but seem not to have attracted much attention so far. At
the meantime, we also intend to identify the operational
regimes of these representative AS schemes in the compact
antenna array MIMO system. For the cases we consider, we
find that in the presence of mutual coupling, soft AS can
always achieve superior performance as compared to hard
AS. Moreover, effectiveness of these AS schemes is closely
related to the interelement spacing. For example, hard AS
workswellonlywhentheinterelementspacingisnolessthan
a half wavelength.
Additionally, another goal of our study is to address
a simple yet very practical question which deals with the
costperformance tradeoff in implementation: as far as
mutual coupling is concerned, can we achieve better spectral
efficiency by placing a few more than necessary antenna
elements in size sensitive UE and applying a certain adequate
AS approach than plainly implementing the conventional
MIMO system without AS? Further, if the answer is yes,
l
2r
dr
dr
Lr
···
Figure 1: Dipole elements in a sidebyside configuration (receiver
antenna array as an example).
how would we decide the number of antenna elements for
placement and the AS method for deployment? Our work
will provide answers to the above questions, and it turns out
the solution is closely related to identifying the saturation
point of the spectral efficiency.
This paper is organized as follows. In Section 2, we
introduce the network model for the compact MIMO system
andcharacterizetheinputoutputrelationshipbytakinginto
account the influence of mutual coupling. In Section 3, we
describethehardandsoftASschemesthatwillbeusedinour
study, and also estimate their computational complexity. In
Section 4, we present the simulation results. We discuss our
main findings in Section 5, and finally conclude this paper in
Section 6.
2.Network Model for Compact MIMO
We consider a MIMO system with M transmit and N receive
antennas (M, N > 1). We assume antenna elements are
placed in a sidebyside configuration along a fixed length
at each terminal (transmitter and receiver), as shown in
Figure 1. Other types of antenna configuration are also
possible, for example, circular arrays [11]. But it is noted
that, the sidebyside arrangement exhibits larger mutual
couplingeffectssincetheantennasareplacedinthedirection
of maximum radiation [11, page 474]. Thus, the sidebyside
configuration is more suitable to our study. We define Ltand
Lras the aperture lengths for transmitter and receiver sides,
respectively. In particular, we are more interested in the case
that Lr is fixed and small, which corresponds to the space
limitation of the UE. We denote l as the dipole length, r as
the dipole radius, and dr (dt) as the sidebyside distance
between the adjacent dipoles at the receiver (transmitter)
side. Thus, we have dr= Lr/(N − 1) and dt= Lt/(M −1).
A simplified network model (as compared to [13, 14],
e.g.) for transmitter and receiver sides is depicted in Figure 2.
Figure 3 illustrates a direct conversion receiver that connects
the output signals in Figure 2, where LNA denotes the low
noise amplifier, LO denotes the local oscillator, and ADC
denotes the analogtodigital converter. For the ease of the
following analysis, we assume that in the circuit setup, all the
antenna elements at the receiver side are grounded through
the load impedance ZLi, i = 1,...,N (cf. Figure 2), regardless
of whether they will be selected or not. In fact, ZLi,i =
1,...,N constitute a simple matching circuit. Such matching
circuit is necessary as it can enhance the efficiency of power
Page 3
EURASIP Journal on Wireless Communications and Networking3
transfer from the generator to the load [20, Chapter 11].
We also assume that the input impedance of each LNA in
Figure 3whichislocatedveryclosetotheantennaelementto
amplify weak received signals, is high enough such that it has
little measurable effect on the receive array’s output voltages.
This assumption is necessary to facilitate the analysis of the
network model. However, it is also very reasonable because
this ensures that the input of the amplifier will neither
overload the source of the signal nor reduce the strength of
the signal by a substantial amount [21].
Let us firstly consider the transmitter side, which can be
regarded as a coupled M port network with M terminals. We
define i = [i1,...,iM]Tand vt= [vt1,...,vtM]Tas the vectors
of terminal currents and voltages, respectively, and they are
related through
vt= ZTi,
(1)
where ZT denotes the impedance matrix at the transmitter
side. The (p,q)th entry of ZT(p,q), when p/ =q, denotes
the mutual impedance between two antenna elements, and
is given by [20, Chapter 21.2]:
ZT
?p,q?=
jη
4πsin2(kl/2)
?l/2
−l/2F(z) dz,
(2)
where
F(z) =
?e−jkR1
R1
+e−jkR2
R2
−2cos
??
?kl
2
?
·e−jkR0
R0
?
·sin
?
k
?l
2− z
.
(3)
In the above expression, η denotes the characteristic
impedance of the propagation medium, and can be calcu
lated by η =
?
μ/?, where μ and ? denote permittivity and
permeabilityofthemedium,respectively.Likewise,k denotes
the propagation wavenumber of an electromagnetic wave
propagating in a dielectric conducting medium, and can
be computed through k = ω√μ?, where ω is the angular
frequency. Finally R0, R1and R2are defined as
R0=
?
?
?
?
?
?
?
?
?
?p − q?2d2
t
(M −1)2+z2,
?p − q?2d2
R1=
?
?
t
(M −1)2+
?p − q?2d2
?
z −l
2
?2
,
R2=
?
t
(M −1)2+
?
z +l
2
?2
.
(4)
When p = q, ZT(p,q) is the selfimpedance of a single
antenna element, and can also be obtained from (2) by
simply redefining R0, R1and R2as follows:
R0=
?r2+z2,
?
?
R1=
r2+
?
?
z −l
2
?2
?2
,
R2=
r2+
z +l
2
.
(5)
Thus, the selfimpedance for an antenna element with l =
0.5λ and r = 0.001λ for example, is approximately
?p, p?= 73.08 +42.21jΩ.
Further, let us consider an example that M = 5 antenna
elements of such type are equally spaced over a linear array
of length Lt= 2λ. The impedance matrix ZTis given by
ZT
(6)
ZT=
⎛
⎜
⎜
⎝
⎜
⎜
⎜
⎜
⎜
⎜
⎜
73.08 +42.21j
−12.52 − 29.91j
73.08 +42.21j
4.01 +17.73j
−1.89 −12.30j
4.01 +17.73j
1.08 +9.36j
−12.52 −29.91j
4.01 +17.73j
−12.52 −29.91j
73.08 +42.21j
−1.89 − 12.30j
4.01 +17.73j
−12.52 − 29.91j
4.01 +17.73j
−12.52 −29.91j
73.08 +42.21j
−1.89 −12.30j
1.08 +9.36j
−12.52 −29.91j
4.01 +17.73j
−12.52 −29.91j
73.08 +42.21j
−1.89 −12.30j
−12.52 −29.91j
⎞
⎟
⎟
⎠
⎟
⎟
⎟
⎟
⎟
⎟
⎟
.
(7)
For i = 1,...,M, the terminal voltage vtican be related
to the source voltage xi via the source impedance Zsi by
vti
= xi − Zsiii. Define ZS
x = [x1,...,xM]. Then, from Figure 2, we can obtain the
following results: vt = x − ZSi and vt = ZTi. Therefore,
the relationship between terminal voltages vt and source
voltages x can be written in matrix form as vt = ZT(ZT+
ZS)−1x. Similar to [12], we choose Zsi = Z∗
roughly corresponds to a conjugate match in the presence of
mild coupling. In the case of uncoupling in the transmitter
= diag{Zs1,...,ZsM}, and
T(i,i), which
side, ZT is diagonal, and its diagonal elements are all the
same. Consequently, ZT(ZT+ ZS)−1is also diagonal, and its
diagonalelementcanbedenotedasδT= ZT(1,1)/[ZT(1,1)+
ZS(1,1)]. To accommodate the special case of zero mutual
coupling where vtis equal to x, in our model we modify the
relationship between vtand x into
vt= WTx,
(8)
where WT= δ−1
TZT(ZT+ZS)−1.
Page 4
4EURASIP Journal on Wireless Communications and Networking
H
············
x1
x2
Zs1
Zs2
ZsM
i1
i2
iM
vt1
vt2
vtM
vr1
vr2
vrN
ZL2
y1
y2
yN
ZT
ZR
xM
ZL1
ZLN
MIMO
propogation
channel
Overall transmitter side
impedance matrix
Overall receiver side
impedance matrix
Compound MIMO channel
Figure 2: Network model for a (M,N) compact MIMO system.
y1
y2
yN
LNA
LNA
ADC
ADC
I
Q
ADC
ADC
I
Q
90˚
LO
LO
RF chain
RF chain
···
······
90˚
Singal
processing
and
decoding
Figure 3: RF chains at the receiver side.
Denote vr= [vr1,...,vrN] as the vector of open circuited
voltages induced across the receiver side antenna array, and
y = [y1,..., yN] as the voltage vector across the output of
the receive array. Since we assumed highinput impedance of
these LNAs, a similar network analysis can be carried out at
the receiver side and will yield
y = WRvr,
(9)
where WR= δ−1
matrix at the receiver side, and ZL is a diagonal matrix
with its (i,i)th entry given by ZL(i,i) = ZLi = [ZR(i,i)]∗,
i = 1,...,N. δR is given by δR = [ZR(1,1)]∗/{ZR(1,1) +
[ZR(1,1)]∗}. It is noted that the approximate conjugate
match [12] is also assumed at the receiver side, so that the
load impedance matrix ZLis diagonal with its entry given by
Z∗
In frequencyselective fading channels, the effectiveness
of AS is considerably reduced [3], which in turn makes it
RZL(ZR+ ZL)−1. ZRis the mutual impedance
R(i,i), for i = 1,...,N.
difficult to observe the effect of mutual coupling. Therefore,
we focus our attention solely on flat fading MIMO channels.
The radiated signal vt is related to the received signal vr
through
vr= Hvt,
(10)
whereHisaN×M complexGaussianmatrixwithcorrelated
entries. To account for the spatial correlation effect and the
Rayleigh fading, we adopt the Kronecker model [22, 23].
This model uses an assumption that the correlation matrix,
obtained as Ψ = E{vec(H) vec(H)H} with vec(H) being the
operatorstackingthematrixHintoavectorcolumnwise,can
be written as a Kronecker product, that is, Ψ = ΨR⊗ ΨT,
where ΨRand ΨTare respectively, the receive and transmit
correlation matrices, and ⊗ denotes the Kronecker product.
This implies that the joint transmit and receive angle power
spectrum can be written as a product of two independent
Page 5
EURASIP Journal on Wireless Communications and Networking5
angle power spectrum at the transmitter and receiver. Thus,
the correlated channel matrix H can be expressed as
H = Ψ1/2
RHwΨ1/2
T,
(11)
where Hwis a N × M matrix whose entries are independent
identically distributed (i.i.d) circular symmetric complex
Gaussian random variables with zero mean and unit vari
ance. The (i, j)th entry of ΨRor ΨTis given by J0(2πdij/λ)
[24], where J0is the zeroth order Bessel function of the first
kind, and dijdenotes the distance between the i,jth antenna
elements.
Therefore, based on (8)–(11), the output signal vector y
at the receiver can be expressed in terms of the input signal x
at the transmitter through
y = WRΨ1/2
= WRΨ1/2
RHwΨ1/2
TWTx +n = Hx +n,
(12)
where H
compound channel matrix which takes into account both the
Rayleighfadinginwirelesschannelsandthemutualcoupling
effect at both transmitter and receiver sides, and n is the
thermal noise. For simplicity, we assume uncorrelated noise
at the receiving antenna element ports. For the case where
correlatednoiseisconsidered,readersarereferredto[16,17].
RHwΨ1/2
TWT can be regarded as a
3.HardandSoft AS for Compact MIMO
We describe here some typical hard and soft AS schemes
that we will investigate, assuming the compact antenna array
MIMO system described in Section 2. For hard AS, we focus
only on the hybrid selection method [5–7]. For soft AS, we
studytwotypicalschemes:theFFTbasedselection[9]which
embeds fast Fourier transform (FFT) operations in the RF
chains, and the phaseshiftbased selection [10] which uses
variable phase shifters adapted to the channel coefficients
in the RF chains. For simplicity, we only consider AS at
the receiver side with nRantennas being chosen out of the
N available ones, and we focus on a spatial multiplexing
transmission.
Weassumethatthepropagationchannelisflatfadingand
quasistatic, and is known at the receiver. We also assume that
the power is uniformly allocated across all the M transmit
antennas, that is, E{xxH} = P0IM/M. We denote the noise
power as σ2
ρ = P0/σ2
Shannon limit quite closely are used, the spectral efficiency
(in bits/s/Hz) of this (M,N) fullcomplexity (FC) compact
MIMO system without AS could be calculated through [1]
?
It is worth noting that the length limits of transmit and
receive arrays, Ltand Lr, enter into the compound channel
matrix H in a very complicated way. It is thus difficult to
find a closeform analytical relationship between CFC(M,N)
and Lt (Lr). Consequently, using Monte Carlo simulations
to evaluate the performance of spectral efficiency becomes a
necessity.
n, and the nominal signaltonoise ratio (SNR) as
n. Then assuming some codes that approach the
CFC(M,N) = log2
det
?
IM+ρ
MHHH
??
.
(13)
To avoid detailed system configurations and to make
the performance comparison as general and as consistent
as possible, we only use the spectral efficiency as the
performance of interest. Moreover, all these AS schemes we
study here are merely to optimize the spectral efficiency,
not other metrics. Since each channel realization renders a
spectral efficiency value, the ergodic spectral efficiency and
the cumulative distribution function (CDF) of the spectral
efficiency will be both meaningful. We will then consider
them as performance measures for our study.
3.1. Hybrid Selection. This selection scheme belongs to the
conventional hard selection, where nR out of N receive
antennas are chosen by means of a set of switches in the RF
domain (e.g., [5–7]). Figure 4(a) illustrates the architecture
of the hybrid selection at the receiver side. As all the antenna
elementsatthereceiversidearepresumedgroundedthrough
the load impedance ZLi, i = 1,...,N, the mutual coupling
effect will be always present at the receiver side. However,
this can facilitate the channel estimation and allow us to
extract rows from H for subset selection. Otherwise, the
mutual coupling effect will vary with respect to the selected
antenna subsets. For convenience, we define S as the nR× N
selection matrix, which extracts nR rows from H that are
associated with the selected subset of antennas. We further
define S as the collection of all possible selection matrices,
whose cardinality is given by S =
with hybrid selection delivers a spectral efficiency of
?N
nR
?
. Thus, the system
CHS= max
S∈S
log2
?
det
?
IM+ρ
M(SH)H(SH)
??
.
(14)
OptimalselectionthatleadstoCHSrequiresanexhaustive
search over all
nR
subsets of S, which is evident by (14).
Note that
?
?N
?
det
IM+ρ
M(SH)H(SH)
?
= det
?
InR+ρ
M(SH)(SH)H?
.
(15)
Then, the matrix multiplication in (14) has a complexity of
O(nRM · min(nR,M)). Calculating the matrix determinant
in (14) requires a complexity of O((min(nR,M))3). Thus, we
can conclude that optimal selection requires about O(S ·
nRM · min(nR,M)) complex additions/multiplications. This
estimated complexity for optimal selection can be deemed
as an upper bound of the complexity of any hybrid AS
scheme, since there exist some suboptimal but reduced
complexity algorithms, such as the incremental selection and
the decremental selection algorithms in [7].
3.2. FFTbased Selection. As for this soft selection scheme
(e.g., [9]), a Npoint FFT transformation (phaseshift
only) is performed in the RF domain firstly, as shown
in Figure 4(b), where information across all the receive
antennas will be utilized. After that, a hybridselectionlike
schemeisappliedtoextractnRoutofN informationstreams.
Page 6
6EURASIP Journal on Wireless Communications and Networking
RF
switches
1
y1
y2
yN
···
···
···
···
vr1
vr2
vrN
nR
Overall
receiver
side
impedance
matrix
RF chain
RF chain
(a) Hybrid selection
RF
switches
FFT
matrix
F
1
y1
y2
yN
···
···
···
···
···
vr1
vr2
vrN
nR
Overall
receiver
side
impedance
matrix
RF chain
RF chain
(b) FFTbased selection
1
···
···
···
vr1
vr2
vrN
y1
y2
yN
nR
Θ
Overall
receiver
side
impedance
matrix
Phase
shift
matrix
RF chain
RF chain
(c) Phaseshiftbased selection
Figure 4: AS at the receiver side for spatial multiplexing transimssions.
We denote F as the N × N unitary FFT matrix with its (k,l)
th entry given by:
F(k,l) =
1
√N
exp
?
−j2π(k −1)(l −1)
N
?
,
∀k,l ∈ [1,N].
(16)
Accordingly, this system delivers a spectral efficiency of
CFFTS= max
S∈Slog2
?
det
?
IM+ρ
M(SFH)H(SFH)
??
.
(17)
The only difference between (14) and (17) is the N
pointFFTtransformation.SuchFFTtransformationrequires
a computational complexity of O(MN logN). If we assume
N logN
≤
complexity of optimal selection that achieves CFFTScan be
estimated as O(S ·nRM ·min(nR,M)), which is the worst
case complexity.
nR · min(nR,M), then the computational
3.3. PhaseShift Based Selection. This is another type of soft
selection scheme (e.g., [10]) that we consider throughout
this study. Its architecture is illustrated in Figure 4(c). Let
us denote Θ as one nR × N matrix whose elements are
nonzero and restricted to be pure phaseshifters, that we
will fully define in what follows. There exists some other
work such as [25] that also considers the use of tunable
phaseshifterstoincreasethetotalcapacityofMIMOsystems.
However, in Figure 4(c), the matrix Θ that performs phase
shift implementation in the RF domain essentially serves
as a NtonR switch with nR output streams. Additionally,
unliketheFFTmatrix,Θmightnotbeunitary,andhencethe
resulting noise can be colored. Finally, this system’s spectral
efficiency can be calculated by [10]
CPSS= max
Θ
log2
?
det
?
IM+ρ
M(ΘH)H?
ΘΘH?−1(ΘH)
??
(18)
.
Let us define the singular value decomposition (SVD) of H
as H = UΛVH, where U and V are N × N, M × M unitary
matricesrepresentingtheleftandrightsingularvectorspaces
of H, respectively; Λ is a nonnegative and diagonal matrix,
consisting of all the singular values of H. In particular, we
denote λH,ias the ith largest singular value of H, and uH,ias
the left singular vector of H associated with λH,i. Thus one
solution to the phase shift matrix Θ can be expressed as [10,
Theorem 2]:
Θ = exp
?
j × angle
??uH,1,...,uH,nR
?H??
(19)
where angle{·} gives the phase angles, in radians, of a matrix
with complex elements, exp{·} denotes the elementby
element exponential of a matrix.
The overall cost for calculating the SVD of H is around
O(MN·min(M,N))[26,Lecture31].Computingthematrix
multiplication in (19) requires a complexity around the
order of O(MNnR). The matrix determinant has an order
of complexity of O((min(nR,M))3). Therefore, the phase
shiftbased selection requires around O(MN · max(nR,M))
complex additions/multiplications.
Page 7
EURASIP Journal on Wireless Communications and Networking7
1102030
N
4050 60
0
5
10
15
20
25
30
35
Uncorrelated without mutual coupling
Correlated without mutual coupling
Correlated with mutual coupling
Ergodic spectral efficiency (bits/s/Hz)
Figure 5: Ergodic spectral efficiency of a compact MIMO system
(M = 5) with mutual coupling at both transmitter and receiver
sides.
4.Simulations
Our simulations focus on the case when AS is implemented
only on the receiver side, but mutual coupling and spatial
correlation are accounted for at both terminals. However,
in order to examine the mutual coupling effect on AS at
the receiving antenna array, we further assume M = 5
equallyspaced antennas at the transmitter array, and the
interelement spacing dt is fixed at 10λ. This large spacing
is chosen to make the mutual coupling effect negligible at
the transmitting terminal. For the receiver terminal, we fix
the array length Lr at 2λ. We choose l = 0.5λ and r =
0.001λ for all the dipole elements. Each component in the
impedance matrices ZT and ZR is computed through (2)
which analytically expresses the self and mutual impedance
ofdipoleelementsinasidebysideconfiguration.Finally,we
fix the nominal SNR at ρ = 10dB.
As algorithm efficiency is not a focus in this paper, for
both hybrid and FFTbased selection methods, we use the
exhaustive search approach to find the best antenna subset.
For the phaseshiftbased selection, we compute the phase
shift matrix Θ through (19) given each H. For each scenario
of interest, we generate 5 ×104random channel realizations,
and study the performance in terms of the ergodic spectral
efficiency and the CDF of the spectral efficiency.
4.1. Ergodic Spectral Efficiency of Compact MIMO. In
Figure 5 we plot the ergodic spectral efficiency of a compact
MIMO system for various N. The solid line in Figure 5
depicts the ergodic spectral efficiency when mutual coupling
and spatial correlation is considered at both terminals. Also
for the purpose of comparison, we include a dashed line
whichdenotestheperformancewhenonlyspatialcorrelation
is considered at both sides, and a dashdot line which
123
nR
45
3
6
9
12
15
Hybrid selection
FFT based selection
Phaseshift based selection
Reduced full system w/o selection
Ergodic spectral efficiency (bits/s/Hz)
Figure 6: Ergodic spectral efficiency of a compact MIMO system
with AS, where M = 5 and N = 5.
corresponds to the case when only the simplest i.i.d Gaussian
propagation channel is assumed in the system. It is clearly
seen that mutual coupling in the compact MIMO system
seriously decreases the system’s spectral efficiency. Moreover,
in accord with the observation in [12], our results also
indicate that as the number of receive antenna elements
increases, the spectral efficiency will firstly increase, but after
reaching the maximum value (approximately around N = 8
in Figure 5), further increase in N would result in a decrease
of the achieved spectral efficiency. It is also worth noting that
when N = 5, the interelement spacing at the receiver side, dr,
is equal to λ/2, which probably is the most widely adopted
interelement spacing in practice. Thus, results in Figure 5
basically indicate that, by adding a few more elements and
squeezing the interelement spacing down from λ/2, it is
possible to achieve some increase in the spectral efficiency,
even in the presence of mutual coupling. But it is also
observed that such increase is limited and relatively slow as
comparedtothespatialcorrelatedonlycase,andthespectral
efficiency will saturate very shortly.
4.2. Ergodic Spectral Efficiency of Compact MIMO with AS.
To study the performance of the ergodic spectral efficiency
with regard to the number of selected antennas nR for a
compact MIMO system using AS, we consider three typical
scenarios, namely, N = 5 in Figure 6, N = 8 in Figure 7, and
N = 12 in Figure 8. In each figure, we plot the performance
of the hybrid selection, FFTbased selection, and phase
shiftbased selection. Additionally, we also depict in each
figure the ergodic spectral efficiency of the reduced full
complexity (RFC) MIMO, denoted as CRFC(M,nR), where
only nRreceive antennas are distributed in the linear array
and no AS is deployed. In Figure 6, it is observed that
Page 8
8EURASIP Journal on Wireless Communications and Networking
1
23
45678
3
6
9
12
15
Ergodic spectral efficiency (bits/s/Hz)
nR
Hybrid selection
FFT based selection
Phaseshift based selection
Reduced full system w/o selection
Figure 7: Ergodic spectral efficiency of a compact MIMO system
with AS, where M = 5 and N = 8.
CFC(M = 5,N = 5) > CPSS > CFFTS = CHS > CRFC(M =
5,nR), which in particular indicates the following:
(1) Soft AS always performs no worse than hard AS. The
phaseshiftbased selection performs strictly better
than the FFTbased selection.
(2) With the same number of RF chains, the system with
AS performs strictly better than the RFC system.
Interestingly, these conclusions that hold for this compact
antenna array case are also generally true for MIMO systems
without considering the mutual coupling effect (e.g., [10]).
But a crossreference to the results in Figure 5 can help
understand this phenomenon. In Figure 5, it is shown that
when N increases from 1 to 5, the ergodic spectral efficiency
ofthecompactMIMOsystembehavesnearlythesameasthat
of MIMO systems without considering the mutual coupling
effect.Therefore,itappearsnaturalthatwhenASisappliedto
the compact MIMO system with N ≤ 5, similar conclusions
can be obtained. It is also interesting that the FFTbased
selection performs almost exactly the same as the hybrid
selection.
Next we increase the number of placed antenna elements
to N = 8, at which the compact MIMO system achieves the
highest spectral efficiency (cf. Figure 5). We observe some
different results in Figure 7, which are CFC(M = 5,N = 8) >
CPSS> CFFTS> CHS. These results tell the following.
(1) SoftASalwaysoutperformshardAS.Thephaseshift
based selection delivers the best performance among
all these three AS schemes.
(2) The phaseshiftbased selection performs better than
the RFC system when nR
≤ 5. The FFTbased
1 2 3 4 5 6 7 8 9 1011 12
Ergodic spectral efficiency (bits/s/Hz)
3
6
9
12
15
nR
Hybrid selection
FFT based selection
Phaseshift based selection
Reduced full system w/o selection
Figure 8: Ergodic spectral efficiency of a compact MIMO system
with antenna selection, where M = 5 and N = 12.
1 2 3 4 5 6 7 8 9 101112
Ergodic spectral efficiency (bits/s/Hz)
3
6
9
12
15
nR
Phaseshift based selection (N = 5)
Phaseshift based selection (N = 8)
Phaseshift based selection (N = 12)
Reduced full system w/o selection
Figure 9: Ergodic spectral efficiency of a compact MIMO system
with the phaseshiftbased selection, where M = 5.
selection performs better than the RFC system when
nR≤ 4. The advantage of using the hybrid selection
is very limited.
We further increase the number of antennas to N =
12. Now the mutual coupling effect becomes more severe,
and different conclusions are demonstrated in Figure 8. It is
observed that CFC(M = 5,N = 12) > CPSS≥ CFFTS> CHS,
indicating the following.
Page 9
EURASIP Journal on Wireless Communications and Networking9
(1) Soft AS performs strictly better than hard AS.
(2) The phaseshiftbased selection performs better than
theFFTbasedselectionwhennR< 8.Afterthat,there
is not much performance difference between them.
Also, similar to what we have observed in Figures 6 and 7, in
terms of the ergodic spectral efficiency, none of the systems
with AS outperforms the FC system with N receive antennas
(andthusN RFchains).However,asfortheRFCsystemwith
only nR antennas (and thus nR RF chains), in Figure 8 we
observe the following.
(1) The RFC system always performs better than the
hybrid selection. The hybrid selection seems futile in
this case.
(2) The phaseshiftbased selection performs better than
the RFC system when nR< 5. The benefit of the FFT
basedselectionisverylimited,anditseemsnotworth
implementing.
This indicates that due to the strong impact of mutual cou
pling in this compact MIMO system, only the phaseshift
based selection is still effective, but only for a limited range
of numbers of the available RF chains. More specifically,
when nR< 5 it is best to use the phaseshiftbased selection,
otherwise the RFC system with nRantennas when 5 ≤ nR<
8. Further increase in the number of RF chains, however, will
notleadtoacorrespondingincreaseinthespectralefficiency,
as demonstrated in Figure 5.
For the purpose of comparison, we also plot the ergodic
spectral efficiency of the phaseshiftbased selection scheme
in Figure 9, by extracting the corresponding curves from
Figures 6–8. We find that by placing a few more antenna
elementsinthelimitedspacesothattheinterelementspacing
is less than λ/2, for example, N = 8 in Figure 9, the phase
shiftbased selection approach can help boost the system
spectral efficiency through selecting the best elements. In
fact, the achieved performance is better than that of the
conventional MIMO system without AS. This basically
answers the question we posed in Section 1 that is related to
the costperformance tradeoff in implementation. However,
further squeezing the interelement spacing will decrease the
performance and bring no performance gain, as can be seen
from the case of N = 12 in Figure 9.
4.3. Spectral Efficiency CDF of Compact MIMO with AS. In
Figure 10, we investigate the CDF of the spectral efficiency
for compact MIMO systems with N = 8 . We consider
the case of nR
= 4 in (Figure 10(a)) and nR
(Figure 10(b)). We use dotted lines to denote the compact
MIMO systems with AS, and dark solid lines for the FC
compact MIMO systems (without AS). We also depict the
spectral efficiency CDFcurves of the RFC systems of N =
4 and N = 6 in Figures 10(a) and 10(b), respectively in
gray solid lines. As can be seen in Figure 10(a), soft AS
schemes, that is, the phaseshiftbased and FFTbased AS
methods, perform pretty well as expected, but the hybrid
selection performs even worse than the RFC system with
N = nR= 4 without AS. When we increase nRto 6, as shown
= 6 in
68 10
Spectral efficiency (bits/s/Hz)
1214 16 1820
0
0.2
0.4
0.6
0.8
1
Hybrid
FFT
Phaseshift
FC (N = 8)
RFC (N = 4)
Empirical CDF of
spectral efficiency
(a) (nR= 4)
68 10
Spectral efficiency (bits/s/Hz)
12 14 161820
0
0.2
0.4
0.6
0.8
1
Hybrid
FFT
Phaseshift
FC (N = 8)
RFC (N = 6)
Empirical CDF of
spectral efficiency
(b) (nR= 6)
Figure 10: Empirical CDF of the spectral efficiency of a compact
MIMO system with AS. M = 5 and N = 8.
in Figure 10(b), the performance difference between hard
and soft AS schemes, or between the phaseshiftbased and
the FFTbased selection methods, is quite small. But none of
these systems with various AS schemes outperform the RFC
system of N = nR= 6 without AS, which is consistent with
what we have observed in Figure 7.
These results clearly indicate that when the mutual
coupling effect becomes severe, the advantage of using AS
can be greatly reduced, which however, is usually very
pronouncedinMIMOsystemswhereonlyspatialcorrelation
is considered at both terminals, as shown for example in
Figure 11. On the other hand, it is also found that the
spectral efficiency of a RFC system without AS, which is
usually the lower bound spectral efficiency to that of MIMO
systems with AS (as illustrated by an example of Figure 11),
can become even superior to the counterpart when mutual
coupling is taken into account (as shown in Figure 10 for
instance).However,itshouldbenotedthatthisphenomenon
is closely related to the network model that we adopt in
Section 2. In such model, we have assumed that all the
antenna elements are grounded through the impedance
ZLi,i = 1,...,N, regardless of whether they will be selected
ornot.Thus,forMIMOsystemswithN receiveelementsand
with a certain AS scheme, the mutual coupling impact at the
receiversidecomesfromalltheseN elements,andisstronger
than that of a RFC system with only nRreceive elements.
Page 10
10EURASIP Journal on Wireless Communications and Networking
68 1012141618202224
Spectral efficiency (bits/s/Hz)
0
0.2
0.4
0.6
0.8
1
Empirical CDF of
spectral efficiency
Hybrid
FFT
Phaseshift
FC (N = 8)
RFC (N = 4)
(a) (nR= 4)
810121416182022 24
Spectral efficiency (bits/s/Hz)
0
0.2
0.4
0.6
0.8
1
Empirical CDF of
spectral efficiency
Hybrid
FFT
Phaseshift
FC (N = 8)
RFC (N = 6)
(b) (nR= 6)
Figure 11: Empirical CDF of the spectral efficiency of a compact
MIMO system with AS. M = 5, N = 8, and mutual coupling is not
considered.
5.Discussions
In our study, we also test different scenarios by varying the
length of linear array Lr, for example, we choose Lr = 3λ,
4λ, and so forth. For brevity, we leave out these simulation
results here, but summarize our main findings as follows.
Suppose the ergodic spectral efficiency of a compact
antenna array MIMO system saturates at Nsat. Our simula
tion results (e.g., Figure 5) indicate that
Nsat>
?2Lr
λ
+1
?
, (20)
where ?·? rounds the number inside to the nearest integer
less than or equal to it. We also have nR< N for the sake of
deploying AS. Our simulations reveal that the interelement
spacing is closely related to the functionality of AS schemes.
For the cases we study, the conclusion is the following:
(1) Whendr≥ λ/2,bothsoftandhardselectionmethods
are effective, but the selection gains vary with respect
to nR. Particularly, the phaseshiftbased selection
delivers the best performance among these tested
schemes.PerformanceoftheFFTbasedselectionand
the hybrid selection appears undistinguishable.
(2) When dr< λ/2, there exist two situations:
(a) When nR ≤ ?2Lr/λ + 1?, the selection gain
of the phaseshiftbased selection still appears
pronounced, but tends to become smaller when
nR approaches ?2Lr/λ + 1?. The advantage of
using the FFTbased selection is quite limited.
The hybrid selection seems rather futile.
(b) When ?2Lr/λ + 1? < nR < Nsat, neither soft
nor hard selection seems effective. This suggests
that AS might be unnecessary. Instead, we can
simply use a RFC system with nR RF chains
by equally distributing the elements over the
limited space.
It is noted that in all these cases we examine, soft AS
always has a superior performance over hard selection. This
is because soft selection tends to use all the information
available, while hard selection loses some additional infor
mation by selecting only a subset of the antenna elements.
Our simulation results also suggest that, if hard selection is
to be used, it is necessary to maintain dr ≥ λ/2. Otherwise,
the strong mutual coupling effect could render this approach
useless. Further, if the best selection gain for system spectral
efficiency is desired, one can place Nsator so elements along
the limitedlength linear array, use ?2Lr/λ + 1? or less RF
chains, and apply the phaseshiftbased selection method.
Therefore, it becomes crucial to identify the saturation point
Nsat. This in turn requires the electromagnetic modeling of
the antenna array that can take into account the mutual
coupling effect.
6.Conclusion
In this paper, we proposed a study of some typical hard
and soft AS methods for MIMO systems with closely spaced
antennas. We assumed antenna elements are placed linearly
in a sidebyside fashion, and we examined the mutual
coupling effect through electromagnetic modeling of the
antenna array and theoretical analysis. Our results indicate
that, when the interelement spacing is larger or equal to
one half wavelength, selection gains of these tested soft and
hard AS schemes will be very pronounced. However, when
the number of antennas to be placed becomes larger and
the interelement spacing becomes smaller than a half wave
length, only the phaseshiftbased selection remains effective
and this is only true for a limited number of available RF
chains. The same conclusions however, are not observed
for the case of hard selection. Thus it seems necessary to
maintain the interelement spacing no less than one half
wavelength when the hard selection method is desired. On
the other hand, if the best selection gain for system spectral
efficiency is desired, one can employ a certain number of
elements for which the compact MIMO system attains its
maximum ergodic spectral efficiency, use ?2Lr/λ + 1? or
less RF chains, and deploy the phaseshiftbased selection
method. This essentially indicates, if the costperformance
tradeoff in implementation is concerned, by placing a few
more than necessary antenna elements so that the system
spectral efficiency reaches saturation and deploying the
phaseshiftbased selection approach, we can achieve better
Page 11
EURASIP Journal on Wireless Communications and Networking11
performance in terms of system spectral efficiency than the
conventional MIMO system without AS. Overall, our study
provides novel insight into the deployment of AS in future
generation wireless systems, including the 3GPP LTE and
LTE advanced technologies.
Acknowledgments
This material is based on research supported by the Air
Force Research Laboratory under agreement FA955009
10576, by the National Science Foundation under Grant
CCF0829958, and by the U.S. Army Research Office under
Grant W911NF0810449. The authors would like to thank
Dr. Dmitry Chizhik and Dr. Dragan Samardzija of Bell
Laboratories, AlcatelLucent for the helpful discussions on
the modeling and implementation issues.
References
[1] G. J. Foschini and M. J. Gans, “On limits of wireless com
munications in a fading environment when using multiple
antennas,” Wireless Personal Communications, vol. 6, no. 3, pp.
311–335, 1998.
[2] Overview of 3GPP Release 8 V0.0.3, November 2008,
http://www.3gpp.org/Release8.
[3] A. F. Molisch and M. Z. Win, “MIMO systems with antenna
selection,” IEEE Microwave Magazine, vol. 5, no. 1, pp. 46–56,
2004.
[4] J. Kotecha, “LTE:MIMO Techniques in 3GPPLTE,” Freescale
Semiconductor, June 2008.
[5] M. Z. Win and J. H. Winters, “Analysis of hybrid
selection/maximalratio combining in Rayleigh fading,” IEEE
Transactions on Communications, vol. 47, no. 12, pp. 1773–
1776, 1999.
[6] A. F. Molisch, M. Z. Win, and J. H. Winter, “Reduced
complexity transmit/receivediversity systems,” IEEE Transac
tions on Signal Processing, vol. 51, no. 11, pp. 2729–2738, 2003.
[7] A. Gorokhov, D. A. Gore, and A. J. Paulraj, “Receive antenna
selection for MIMO spatial multiplexing: theory and algo
rithms,” IEEE Transactions on Signal Processing, vol. 51, no. 11,
pp. 2796–2807, 2003.
[8] L. Collin, O. Berder, P. Rostaing, and G. Burel, “Soft vs. hard
antenna selection based on the minimum distance for MIMO
systems,” in Proceedings of the 38th Asilomar Conference on
Signals, Systems and Computers, vol. 2, pp. 1369–1373, Grove,
Calif, USA, November 2002.
[9] A. F. Molisch and X. Zhang, “FFTbased hybrid antenna
selection schemes for spatially correlated MIMO channels,”
IEEE Communications Letters, vol. 8, no. 1, pp. 36–38, 2004.
[10] X.Zhang,A.F.Molisch,andS.Y.Kung,“Variablephaseshift
based RFbaseband codesign for MIMO antenna selection,”
IEEE Transactions on Signal Processing, vol. 53, no. 11, pp.
4091–4103, 2005.
[11] C.A.Balanis,AntennaTheory:AnalysisandDesign,JohnWiley
& Sons, New York, NY, USA, 3rd edition, 2005.
[12] R. Janaswamy, “Effect of element mutual coupling on the
capacity of fixed length linear arrays,” IEEE Antennas and
Wireless Propagation Letters, vol. 1, pp. 157–160, 2002.
[13] C. Waldschmidt, S. Schulteis, and W. Wiesbeck, “Complete RF
system model for analysis of compact MIMO arrays,” IEEE
Transactions on Vehicular Technology, vol. 53, no. 3, pp. 579–
586, 2004.
[14] 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.
1317–1325, 2004.
[15] A. A. Abouda and S. G. H¨ aggman, “Effect of mutual coupling
oncapacityofMIMOwirelesschannelsinhighSNRscenario,”
Progress in Electromagnetics Research, vol. 65, pp. 27–40, 2006.
[16] M.J.Gans,“Channelcapacitybetweenantennaarrays—partI:
sky noise dominates,” IEEE Transactions on Communications,
vol. 54, no. 9, pp. 1586–1592, 2006.
[17] M. J. Gans, “Channel capacity between antenna arrays—
part II: amplifier noise dominates,” IEEE Transactions on
Communications, vol. 54, no. 11, pp. 1983–1992, 2006.
[18] Z.Xu,S.Sfar,andR.S.Blum,“Ontheimportanceofmodeling
the mutual coupling for antenna selection for closelyspaced
arrays,” in Proceedings of IEEE Conference on Information
Sciences and Systems (CISS ’06), pp. 1351–1355, Princeton, NJ,
USA, March 2006.
[19] D.Lu,D.K.C.So,andA.K.Brown,“Receiveantennaselection
scheme for VBLAST with mutual coupling in correlated
channels,” in Proceedings of IEEE International Symposium on
Personal, Indoor and Mobile Radio Communications (PIMRC
’08), pp. 1–5, Cannes, France, September 2008.
[20] S. J. Orfanidis, Electromagnetic Waves and Antennas, Rutgers
University, Piscataway, NJ, USA, 2004.
[21] J. Hewes, “Impedance and reactance,” The Electronics Club,
February 2008, http://www.kpsec.freeuk.com/imped.htm.
[22] J. P. Kermoal, L. Schumacher, K. I. Pedersen, P. E. Mogensen,
and F. Frederiksen, “A stochastic MIMO radio channel model
with experimental validation,” IEEE Journal on Selected Areas
in Communications, vol. 20, no. 6, pp. 1211–1226, 2002.
[23] C. Oestges, B. Clerckx, D. VanhoenackerJanvier, and A. J.
Paulraj, “Impact of fading correlations on MIMO communi
cation systems in geometrybased statistical channel models,”
IEEE Transactions on Wireless Communications, vol. 4, no. 3,
pp. 1112–1120, 2005.
[24] D.S. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, “Fading
correlation and its effect on the capacity of multielement
antenna systems,” IEEE Transactions on Communications, vol.
48, no. 3, pp. 502–513, 2000.
[25] Y. Nakaya, T. Toda, S. Hara, J.I. Takada, and Y. Oishi,
“Incorporation of RFadaptive array antenna into MIMO
receivers,” in Proceedings of IEEE Topical Conference on
Wireless Communication Technology (WCT ’03), pp. 297–298,
Honolulu, Hawaii, USA, October 2003.
[26] L. N. Trefethen and D. Bau III, Numerical Linear Algebra,
SIAM, Philadelphia, Pa, USA, 1997.
View other sources
Hide other sources
 Available from Sana Sfar · May 23, 2014
 Available from hindawi.com