This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

Performance Analysis of Gamma-Gamma Fading FSO MIMO

Links with Pointing Errors
Manav R. Bhatnagar, Senior Member, IEEE and Zabih Ghassemlooy, Senior Member, IEEE

Abstract—The bit error rate (BER) performance of the free major drawback of using FSO technology is its dependence on
space optical (FSO) link suffers from the atmospheric turbulence. weather and atmospheric conditions, which makes it difficult
By employing additional transmit and receive apertures at the to realize an all-time communication link availability. In a
transmitter and receiver, respectively, the error rate of the FSO
communication system can be significantly improved. However, clear atmosphere (i.e., attenuation of 0.5 db/km), the major
the pointing errors (PEs), generated because of the building challenge faced by the FSO links is the atmospheric turbulence
sway, have the potential to eradicate the benefits of the multiple (AT), which results in irradiance fluctuation. For achieving
transmit and/or receive apertures based FSO communication satisfactory error performance in OOK based FSO systems, the
system. Therefore, for a general and realistic study of the FSO detection threshold has to be adaptively varied in accordance
multiple-input multiple-output (MIMO) system, the effect of PEs
in the Gamma-Gamma (GG) fading atmospheric fluctuations is to the AT. However, with OOK systems there is a problem
considered in this paper. We study two schemes for the FSO in selecting an optimal detection threshold level in presence
MIMO systems: i) equal gain combining (EGC), and ii) maximal of the AT [5]; this situation may sometime lead to failure of
ratio combining (MRC). A new power series based representation the FSO link. The subcarrier intensity modulation (SIM) [5],
is proposed for the probability density function of the GG fading [6] is an alternative for the OOK technology, which does not
FSO links with PEs. This new series representation contains
only the terms with exponent of the random variable (RV) as require the adaptive threshold detection. Hence, SIM is more
compared to the closed-form representation, which contains the rugged to the ATs and provides satisfactory error performance.
Meijer-G function of the RV. Then we derive the average BER The SIM leverages on advances made in signal processing as
for both combining schemes over the GG fading FSO links well as the maturity of radio frequency (RF) devices such
with PEs. By using the derived BER expressions, we derive as highly selective filters and stable oscillators, which then
the asymptotic BER for both schemes. The analytical diversity
order and combining gains for both systems are also obtained. permits the use of modulation techniques such as phase-
The effect of PEs over the performance of the schemes is shift keying (PSK) [6] and quadrature amplitude modulation
analyzed under different scenarios and it is observed that the (QAM) [5].
PEs significantly degrade the diversity of the FSO MIMO system. Multiple-input multiple-output (MIMO) wireless commu-
It is deduced by simulation and analysis that though the EGC nication system is a well established technology nowadays,
scheme is simpler to implement in practice but the MRC scheme
is more rugged to the large PEs. known for improved capacity, better diversity gain, and in-
creased coverage area as compared to the conventional sin-
Index Terms—Bit error rate, EGC, free space optical links, gle antenna based technology. The virtues of the MIMO
Gamma-Gamma fading, MIMO system, MRC, subcarrier inten-
sity modulation. technology have encouraged the researchers to incorporate
multiple transmit and receive apertures in FSO communication
I. I NTRODUCTION systems [7]–[16]. The Gamma-Gamma (GG) distribution has
At present, free space optical (FSO) wireless communica- been widely adopted for study of behavior of the FSO links
tions is attracting lots of attention from the research com- under wide range of AT conditions (weak to strong) because
munity, because of very large bandwidth, low implementation it fits well to the experimental results [3], [5], [6], [17]. The
cost, and excellent security [2], [3]. The fast and low cost GG distribution is represented in terms of the modified Bessel
implementation properties of the FSO link makes it suitable function of the second kind, and therefore, it is very challeng-
for the disaster relief applications [4]. The intensity modulation ing and complicated to analyze the performance of a general
and direct detection (IM/DD) with on/off keying (OOK) are FSO MIMO system under the GG fading. In [7], by using a
famously used in the FSO communications; this technology power series representation of the modified Bessel function,
has a virtue of simple practical implementation. However, the approximate error rate of the binary modulation is derived for
equal gain combining (EGC) and maximal ratio combining
Manav R. Bhatnagar is with the Department of Electrical Engineering and (MRC) over the GG fading AT. Later on by using the power
Bharti School of Telecommunication Technology and Management, Indian series representation of the GG probability density function
Institute of Technology - Delhi, Hauz Khas, New Delhi 110016, India (e-
mail: manav@ee.iitd.ac.in). (PDF), the average bit error rate (BER) of the Alamouti space-
Zabih Ghassemlooy is with the Faculty of Engineering and Environ- time block coded [18] 2×1 FSO system is derived in the form
ment, Northumbria University, Newcastle-upon-Tyne NE1 8ST, U.K. (e-mail: of an infinite series in [10]. In [19], analytical approach has
z.ghassemlooy@northumbria.ac.uk).
This paper was presented in part at the IEEE International Conference on been adopted to eventuate the performance of space-diversity
Communications (ICC 2015) at London, U.K., June 2015 as [1]. FSO links over correlated GG fading channels.
This work was supported in part by the Department of Science and Apart from scintillation effects, the performance of the FSO
Technology, Government of India for the Project “Path Selection in Multi-Hop
Cooperative Wireless Networks: Algorithm Design and Hardware Implemen- link degrades because of the building sway, which introduces
tation” (Project Ref. No. SB/S3/EECE/0179/2013). a pointing error (PE) between transmitter (Tx) and receiver

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

(Rx), representing a particular problem in urban areas, where analysis of combining schemes under PEs. Numerical results
FSO equipment is placed on high-rise structures [20], [21]. In are discussed in Section V and some useful conclusions for
most of the existing literature of the MIMO FSO system [7]– the work presented are drawn in Section VI.
[16], the effect of the PE has been ignored. However, for a
more realistic study of the FSO system, the PE along with II. P RELIMINARIES
the atmospheric fluctuations must be considered [21]. Joint A. System Model
optimization of a partially coherent Gaussian beam for FSO
Let us consider a SIM based FSO system [6] with N -
communication over turbulent channels with PEs is presented
transmit and M -receive apertures, where an MPSK symbol
in [22]. The PE can occur along with a boresight error, which
x is transmitted by all transmit apertures over a transmis-
causes a deterministic displacement of the laser beam at the
sion time interval. The received electrical signal at the i-th,
receiving aperture. The boresight errors are more pronounced
i = 1, .., M receive aperture, following optical-to-electrical
in the longer distance FSO communication systems like in
(O-E) conversion, is given as:
earth-to-satellite links [23]. PE can be mitigated by means of
N
i) broadening the optical beam but at the cost of received X
power level (i.e., reduced signal-to-noise ratio (SNR) and yi = η Ii,j x + ei , (1)
j=1
increased error rate) and the security breach; and ii) using
a narrow optical beam and multiple receive apertures (i.e., where ei is the zero-mean complex-valued additive white
receive diversity). Narrowing of the optical beam reduces the Gaussian noise (AWGN) of σ 2 variance; Ii,j is the real-
geometrical losses and increases the FSO link security but valued irradiance of the link between the the j-th transmit
at the cost of increased misalignment. One could also use aperture and i-th receive aperture, containing the effects of
tracking to overcome the PE. A statistical channel model is GG distributed AT and pointing loss [21]; and η denotes the
proposed for FSO MIMO communication systems over weak O-E conversion coefficient. It is assumed that E[|x|2 ] = Ex ,
atmospheric fading channels with PEs in [24]; the outage where E[·] stands for the expectation.
probability and diversity order of the FSO MIMO system The Rx can combine the received electrical signals in two
are studied, and it is deduced that the diversity order is ways: i) MRC, and ii) EGC.
determined by PE effects other than the number of Txs or 1) EGC: The EGC is performed as follows:
Rxs. The probability density function (PDF) of GG distributed M
X X N
M X M
X
AT with PEs and zero boresight errors is derived in the form y= yi = η Ii,j x + ei . (2)
of Meijer-G function in [21]. Recently, the average error rate i=1 i=1 j=1 i=1
and ergodic capacity of Alamouti code [18] in 2 × 1 FSO From (2), the received SNR will be:
system with PEs are derived in [25] by following the channel P 2
M PN
model of [21]. Due to presence of the Meijer-G function, to η 2 Ex i=1 j=1 Ii,j
derive the average BER of a generalized FSO MIMO system γEGC =
M σ2
with PEs under GG fading is a challenging problem. In [26], P 2
M P N
asymptotic BER performance for FSO communication systems γ̄ i=1 j=1 I i,j
using transmit laser selection over GG fading AT channels with = , (3)
M
PEs is analyzed, and some useful insight of the diversity of
where γ̄ = η 2 Ex /σ 2 denotes the average SNR.
the system is obtained.
For the EGC, a maximal-likelihood (ML) Rx is defined as:
Motivated by aforementioned facts, in this work, we in-
N
M X
vestigate the performance analysis of the EGC and MRC X
schemes in the GG fading FSO links with PEs. The SIM x̂ , min |y − η Ii,j x̃|2 . (4)
x̃∈A
i=1 j=1
scheme with M -ary PSK (MPSK) modulation is adopted for
the performance analysis. First, we develop a power series The MPSK constellation is denoted by A.
expansion of the Meijer-G function, then by using the series 2) MRC: The combined electrical signal is written as:
manipulations we get the average BER of the EGC and MRC M X
X N
schemes. The proposed analysis sheds substantial light on the y=η Ii,j yi
asymptotic properties of the FSO MIMO system in presence i=1 j=1
of the PEs; it demonstrates that the asymptotic attributes–  2
M N N
M X
diversity and combining gain–are dominated by PEs if they X X X
= η2  Ii,j  x + η Ii,j ei . (5)
are more hostile than the AT. Moreover, a useful comparison
i=1 j=1 i=1 j=1
of both schemes in the presence of PEs is performed by using
the derived BER results. From (5), the received SNR is given by:
The rest of the paper is organized as follows. Section P M  PN 2
II introduces the system and channel model; a new power η 2 Ex i=1 j=1 Ii,j
γMRC =
series based representation of the PDF of GG distributed σ2
2
atmospheric fluctuation with PEs is developed in this section.

XM N
X
In Section III, the average BER for EGC and MRC schemes = γ̄  Ii,j  . (6)
is derived. Section IV deals with asymptotic performance i=1 j=1

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

" ξ2 −1
αβξ 2
  
2 2 αβx 2 2 αβx
fIi,j (x) = Γ(α − ξ )Γ(β − ξ ) 1 F2 0; 1 + ξ − α, 1 + ξ − β;
A0 Γ(α)Γ(β) A0 A0
α−1
Γ(ξ 2 − α)Γ(β − α) αβx
   
2 2 αβx
+ 2 1 F2 α − ξ ; 1 + α − ξ , 1 + α − β;
Γ(1 + ξ − α) A0 A0
2
 β−1  #
Γ(ξ − β)Γ(α − β) αβx 2 2 αβx
+ 1 F2 β − ξ ; 1 + β − α, 1 + β − ξ ; (11)
Γ(1 + ξ 2 − β) A0 A0

ML Rx for the MRC is given by: C. Power Series Representation

 2 Based on the Slater’s theorem [28], the Meijer-G function
M
X N
X in (8) can be expressed as:
x̂ = min |y − η 2  Ii,j  x̃|2 . (7) 3 Q3 ∗
j=1 Γ(bj − bh )
x̃∈A   a  X
3,0 1
i=1 j=1
G1,3 z  = z bh

b 1 , b2 , b 3 Γ(a1 − bh )
h=1
∗

×1 F2 1 + bh − a1 ; [1 + bh − b] ; z , (10)
B. Channel Model
where [·] indicates a row vector, b = [b1 , b2 , b3 ], (·)∗ indicates
Here we consider the GG fading due to the AT with PEs. to ignore the terms with bj = bh , Γ(·) is the Gamma func-
When both the AT and the PEs are considered, the distribution tion [27], and 1 F2 (·; ·, ·; ·) is the generalized Hypergeometric
of Ii,j is given by [21, Eq. (8)]: function [27]. Note that the expansion in (10) is valid only if
   bj − bh ∈/ Z and a1 − bh ∈ / Z.
αβξ 2 3,0 αβx  ξ2 From (8) and (10), we get (11). Let us use the following
fIi,j (x) = A0 Γ(α)Γ(β) G1,3 A0  ξ 2 −1,α−1,β−1 , (8)
series representation in (11):
··· ∞
where Gm,n
p,q (·| ··· ) is the Meijer-G function [27, Eq. (9.301)].
X (a1 )n z n
F (a ;
1 2 1 1 2 b , b ; z) = , (12)
Assuming that the optical radiation is a plane wave, the two
n=0
(b1 )n (b2 )n n!
parameters α and β that characterize the irradiance fluctuation
PDF are related to the atmospheric conditions by [6, Eq. (3)]: where (a1 )n , (b1 )n , and (b2 )n are the Pochhammer symbols.
The Pochhammer symbol (a1 )n is defined as:
1 (a1 )0 = 1,
α = 
12/5 7/6
 ,
0.49σl2 / 1+1.11σl
e −1 (a1 )n = a1 (a1 + 1)(a1 + 2)...(a1 + n − 1), n ≥ 1,(13)
1
β = , (9) similarly, (b1 )n and (b2 )n can be defined.
12/5 5/6
 
0.51σl2 / 1+0.69σl After some algebra, we get the PDF of Ii,j in series form
e −1
as:
where σl2 = 1.23Cn2 k 7/6 L11/6 is the log irradiance variance, 2 X∞

k = 2π/λ, λ denotes the wavelength, L is the FSO prop- fIi,j (x) = X0 xξ −1 + Yn xn+α−1
agation path length, and Cn2 represents the refractive index ∞
n=0

structure parameter. X
+ Zn xn+β−1 , (14)
The effect of the PEs is characterized by the following n=0
2
p
parameters: A0 = [erf(ν)] , ν = π/2R/wb , ξ 2 = we2 /4σs2 ,
where
R is the radius of the Rx aperture, wb is the received beam ξ2
h√
ξ2
i1/2 
2 αβ
size, we = πerf(ν)wb2 /(2νe−ν ) is the equivalent X0 = 2
Γ(α − ξ )Γ(β − ξ )2

2 Γ(α)Γ(β) A0
beamwaist, σs is the variance of the Gaussian distributions
for both horizontal and vertical buildings’ sway (i.e., PE ξ2 Γ(ξ 2 − α)Γ(β − α)
Yn =
displacement variance), and erf(·) denotes the error function. Γ(α)Γ(β) Γ(1 + ξ 2 − α)
n+α
(α − ξ 2 )n

It is rather challenging to directly deal with the PDF of the αβ
×
FSO channel with PEs due to the presence of the Meijer-G (1 + α − ξ 2 )n (1 + α − β)n n! A0
function in (8). Specially, for FSO MIMO systems, it is not ξ2 Γ(ξ 2 − β)Γ(α − β)
easy to obtain analytical results by using the PDF given in (8). Zn =
Γ(α)Γ(β) Γ(1 + ξ 2 − β)
Therefore, we need an alternative representation of the PDF of n+β
(β − ξ 2 )n

the GG fading channels with PEs, which can render simplified αβ
× . (15)
analysis. Moreover, an alternative form can allow us to find (1 + β − α)n (1 + β − ξ 2 )n n! A0
asymptotic characteristics like diversity gain and combining Let us make the following remarks for the proposed series
gain of the FSO MIMO system with PEs. representation.

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

P N PM
• The proposed series based representation of the PDF The MGF of j=1 i=1 Ii,j can be written by using the
of GG channels with PEs, cf. (14), contains two power independence among Ii,j s as:
series. Each series contains summation terms with only M
N Y

M N
exponents of x. Therefore, it is easy to calculate an
Y
M PN PM
Ii,j (s) = MIi,j (s) = MIi,j (s) . (21)
j=1 i=1
integral containing the proposed series representation as j=1 i=1
compared to a complicated function based representation
From (20) and (21) and after some algebra, we get
in (8).  
• It can be easily shown by applying the ratio test for each
X MN
M PN PM
i=1 Ii,j
(s) =
power series in (14) that the new series representation j=1
k +k +k =M N
k1 , k2 , k3
1 2 3
0≤k1 ,k2 ,k3 ≤M N
contains converging power series with infinite radius of !k 2 !k 3
convergence.  k 1 ∞ ∞
2 X X
× X̃0 s−ξ Ỹn s−n−α Z̃n s−n−β
n=0 n=0
D. BER Calculation
 
X MN
=
For MPSK, from the signal-space concept, the instantaneous k1 +k2 +k3 =M N
k1 , k2 , k3
0≤k1 ,k2 ,k3 ≤M N
BER is given by [29]:
∞
X 2
τM
X √ × Cn (k1 , k2 , k3 )s−n−k1 ξ −k2 α−k3 β
. (22)
P e(γ) = ζM Q (ap γ) , (16) n=0
p=1
The coefficients Cn (k1 , k2 , k3 ) in (22) can be calculated
where Q(·) is the q-function, ζM = ap = 2 as [27, Eq. (0.316)]:
max(log2 M,2) ,
√ (2p−1)π M

 
2 sin M , and τM = max 4 , 1 are the modulation Cn (k1 , k2 , k3 ) = X̃0k1 Ỹn(k2 ) ∗ Z̃n(k3 ) , (23)
dependent parameters of an MPSK constellation containing
(k)
M-points. Therefore, the average BER of the SIM will be: where ∗ denotes the convolution and the subscript hn means
(3)
τM Z ∞ that hn is convolved k −P1 times with itself, i.e., hn = hn ∗
√
hn ∗ hn . The PDF of
N PM
i=1 Ii,j can be obtained by
X
P e(γ̄) = ζM Q ap x fγ (x)dx. (17) j=1
p=1 0 taking the inverse Laplace transform of (22)
√
 
X MN
By observing that Q(x) = (1/2)erfc(x/ 2), where erfc(·) is fPN PM Ii,j (x) =
j=1 i=1 k1 , k2 , k3
the complimentary error function, we can rewrite (17) as: k +k +k =M N 1 2 3
0≤k1 ,k2 ,k3 ≤M N

τM Z ∞  √  ∞
ζM X ap x X Cn (k1 , k2 , k3 ) 2
P e(γ̄) = erfc √ fγ (x)dx. (18) × 2
xn+k1 ξ +k2 α+k3 β−1 . (24)
2 p=1 0 2 n=0
Γ(n + k1 ξ + k2 α + k3 β)
P 2
N PM
From (24), the PDF of j=1 i=1 Ii,j will be:
III. P ERFORMANCE E VALUATION  
1 X MN
In this section, we will conduct the performance analysis of f(PN PM I )2 (x) =
the EGC and MRC schemes. We follow the standard moment j=1 i=1 i,j 2 k +k +k =M N k1 , k2 , k3
1 2 3
0≤k1 ,k2 ,k3 ≤M N
generating function (MGF) based approach to find the BER ∞
of the FSO MIMO system. Expressions of the BER for both
X Cn (k1 , k2 , k3 ) n+k1 ξ2 +k2 α+k3 β
−1
× 2 + k α + k β)
x 2 . (25)
schemes are obtained in the form of power series. By using n=0
Γ(n + k 1 ξ 2 3
the BER expressions, the diversity order and combining gain
By using the method of change of random variables, the PDF
for both schemes are obtained.
of the received SNR of the EGC can be easily derived as:
 
X MN
A. Average BER Analysis of EGC Scheme fγ (x) =
EGC k1 , k2 , k3
k +k +k =M N 1 2 3
0≤k1 ,k2 ,k3 ≤M N
The MGF of Ii,j is given as:
∞
Z ∞ X ΞEGC
n (k1 , k2 , k3 ) n+k1 ξ2 +k2 α+k3 β
−1
× n+k1 ξ2 +k2 α+k3 β
x 2 , (26)
MIi,j (s) = e−sx fIi,j (x)dx. (19) n=0 γ̄ 2
0
where
From (14) and (19), the MGF can be expressed in the form n+k1 ξ2 +k2 α+k3 β
of power series by: M Cn (k1 , k2 , k3 )
2
ΞEGC
n (k1 , k2 , k3 ) = . (27)
∞ ∞ 2Γ(n + k1 + k2 α + k3 β) ξ2
2 X X
MIi,j (s) = X̃0 s−ξ + Ỹn s−n−α + Z̃n s−n−β , (20) The instantaneous BER of MPSK constellation for the EGC
n=0 n=0 scheme will be:
τM Z ∞  √ 
where X̃0 = X0 Γ(ξ 2 ), Ỹn = Yn Γ(n+α), and Z̃n = Zn Γ(n+ ζM X ap x
P e(γ̄) = erfc √ fγ (x)dx. (28)
β). 2 p=1 0 2 EGC

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

!
X M
fPM (PN I )2 (x) = 0≤k ≤N
i=1 j=1 i,j
PN PN −k1
[jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
j =M
k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2
0≤jk ,k ,N −k −k ≤M
1 2 1 2
∞ 0≤k ≤N
X Jn ([jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 )
×  PN PN −k1 
n+ k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ 2 +k2 α+(N −k1 −k2 )β)
n=0 Γ 2
PN −k1
n+
PN
k1 =0 j
k2 =0 k1 ,k2 ,N −k1 −k2 (
k1 ξ2 +k2 α+(N −k1 −k2 )β )
−1
×x 2 (37)

P 2
Following substitution of x = y 2 in (28) and then using (29): The MGF of
N
j=1 Ii,j is obtained by taking the Laplace
Z ∞
1

a+1
 transform of (32):
y a−1 erfc(cy)dy = a √ Γ , (29) N NX
∞ X −k1
0 ac π 2 X
M( PN (s) =2 Pn (k1 , k2 , N − k1 − k2 )
j=1 Ii,j )
we can solve the integral in (28) in a closed-form and obtain n=0 k1 =0 k2 =0
the following: n+k1 ξ2 +k2 α+(N −k1 −k2 )β

  ×s− 2 , (33)
X MN
P e(γ̄) = ζM with
k +k +k =M N
k1 , k2 , k3
1 2 3  
0≤k1 ,k2 ,k3 ≤M N 1 N
∞ τM √ 2 Pn (k1 , k2 , N − k1 − k2 ) =
X ΞEGC
n (k1 , k2 , k3 ) X ( 2/ap )n+k1 ξ +k2 α+k3 β 2 k 1 , k2 , N − k 1 − k 2
× √
Γ( n + k1 ξ 2 + k2 α + (N − k1 − k2 )β /2)
2


n=0 γ̄
n+k1 ξ +k2 α+k3 β
(n + k1 ξ 2 + k2 α + k3 β) π
2 p=1 ×

n + k1 ξ 2 + k2 α + k3 β + 1
 Γ(n + k1 ξ 2 + k2 α + (N − k1 − k2 )β)
×Γ . (30) ×Cn (k1 , k2 , N − k1 − k2 ). (34)
2
P M PN   2
For a FSO system with a single transmit and a single receive The MGF of i=1 j=1 Ii,j is given
aperture, i.e., N = M = 1, the expression of BER can be  M
obtained from (30) as: by MPM (PN I )2 (s) = M(PN I )2 (s) and can be
i=1 j=1 i,j j=1 i,j
conveniently expressed as:
τM √ 2
X̃0 X ( 2/ap )ξ X
P e(γ̄) = ζM ξ2 2
√ MPM (PN I )2 (s) =
2Γ(ξ 2 )γ̄ 2 p=1 (ξ ) π i=1 j=1 i,j
PN PN −k1
j =M
k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2
 2  ∞
ξ +1 X Ỹn 0≤jk ,k ,N −k −k ≤M
1 2 1 2
×Γ + ζM n+α
!
2 n=0 2Γ(n + α)γ̄
2 M
τ √ × 0≤k ≤N
X ( 2/ap )n+α
M 
n+α+1
 [jk1 ,k2 ,N −k1 −k2 ]0≤k12 ≤N −k1
× √ Γ ∞
p=1
(n + α) π 2 X 0≤k ≤N
√ × Jn ([jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 )
∞ τ
X Z̃n X M
( 2/ap )n+β n=0
+ζM √ PN −k1
n+β
(n + β) π n+
PN
j (
k1 ξ2 +k2 α+(N −k1 −k2 )β )
n=0 2Γ(n + β)γ̄
2 k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2

 
p=1 ×s− 2 ,(35)
n+β+1 0≤k ≤N
×Γ . (31) where [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 is a row vector containing
2
all coefficients corresponding to the distribution of k1 and k2
satisfying the constraints 0 ≤ k1 ≤ N , 0 ≤ k2 ≤ N − k1 . For
0≤k ≤2
example, for N = 2, [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤2−k1 contains the
B. Average BER Analysis of MRC Scheme following elements: [j0,2,0 , j0,1,1 , j1,1,0 , j1,0,1 , j0,0,2 , j2,0,0 ];
P 2 each of these elements can take values between 0 and M but
N
The PDF of j=1 I i,j can be written by using (25) as: the sum of all the elements is equal to M . Further, in (35),
we have:
N N −k 
1 X X1

N 0≤k ≤N
Jn ([jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 ) = Pn(j0,N,0 ) (0, N, 0) ∗
f ( PN (x)
2 =
j=1 Ii,j ) 2 k1 , k2 , N − k1 − k2
k1 =0 k2 =0 Pn(j1,N −1,0 ) (1, N − 1, 0) ∗ .. ∗ .. ∗ Pn(jN,0,0 ) (N, 0, 0). (36)
∞
X Cn (k1 , k2 , N − k1 − k2 )
× 2
By taking the 
inverse Laplace
2 transform of (35), we get the
n=0
Γ(n + k 1 ξ + k2 α + (N − k1 − k2 )β) PM P N
PDF of i=1 I
j=1 i,j given in (37) (given on the top
n+k1 ξ2 +k2 α+(N −k1 −k2 )β
−1
×x 2 . (32) of previous page).

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

0≤k ≤N
!
Jn ([jk1 ,k2 ,N −k1 −k2 ]0≤k12 ≤N −k1 )
ΞMRC
0≤k ≤N
n [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 =  PN PN −k1  (39)
n+ k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ 2 +k2 α+(N −k1 −k2 )β)
Γ 2

!
X M
P e(γ̄) = ζM 0≤k ≤N
PN PN −k1
[jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
j =M
k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2
0≤jk ,k ,N −k −k ≤M
1 2 1 2
!
0≤k ≤N
∞
ΞMRC
n [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
X
× PN PN −k1
k ξ2 +k2 α+(N −k1 −k2 )β )
k2 =0 k1 ,k2 ,N −k1 −k2 ( 1
n+ j
k1 =0
n=0
γ̄ 2

√ n+ N
PN −k1 2
k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ +k2 α+(N −k1 −k2 )β )
τM
P
X ( 2/ap ) k1 =0
× PN PN −k1 √
2
p=1 (n + k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ + k2 α + (N − k1 − k2 )β)) π
PN PN −k !
n + k1 =0 k2 =01 jk1 ,k2 ,N −k1 −k2 k1 ξ 2 + k2 α + (N − k1 − k2 )β + 1


×Γ (41)
2

From (6) and (37), the PDF of the instantaneous received of a RF system can be characterized by two parameters:
SNR of the MRC scheme will be: combining gain (Gc ) and diversity gain (Gd ). The combining
X gain specifies the relative horizontal shift of the BER versus
fγ (x) =
MRC
PN PN −k1
SNR plots on the log-log scale; whereas, the diversity gain
j =M
k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2 shows the slope of the decay of these plots, at asymptotically
0≤jk ,k ,N −k −k ≤M
1 2 1 2
! high values of the SNR. The asymptotic value of the BER can
M be expressed as:
× 0≤k ≤N
[jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
! P e(γ̄) ≈ (Gc γ̄)−Gd . (42)
0≤k ≤N
∞
ΞMRC
n [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
X
× PN PN −k1
k ξ2 +k2 α+(N −k1 −k2 )β )
k2 =0 k1 ,k2 ,N −k1 −k2 ( 1
n+ j
k1 =0
n=0
γ̄ 2
A. Asymptotic BER Analysis of EGC
PN −k1
n+
PN
k1 =0 j
k2 =0 k1 ,k2 ,N −k1 −k2 (
k1 ξ2 +k2 α+(N −k1 −k2 )β )
×x 2 −1
, (38) In the case of EGC, the term corresponding to the smallest
! exponent of the average SNR γ̄ in the power series in
0≤k ≤N (30) dominates the BER, at high SNR. Further, we can put
with ΞMRC
n [jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1 given in (39).
p = 1 for simplifying the asymptotic BER expression in (30).
The BER of the MRC scheme can be written as: Therefore, after substituting n = 0 and p = 1 in (30), we get:
τM Z ∞
X √
X 
MN

P e(γ̄) = ζM Q ap x f γ (x)dx. (40) P e(γ̄) = ζM
0
MRC k1 , k2 , k3
p=1 k +k +k =M N 1 2 3
0≤k1 ,k2 ,k3 ≤M N
√ p √
After substituting Q (ap x) = (1/2)erfc(ap x/2) and x = ΞEGC
2
(k1 , k2 , k3 ) ( 2/a1 )k1 ξ +k2 α+k3 β
0
y 2 in (40), then exploiting (29), we can solve the integral in × k ξ2 +k α+k β √
γ̄
1 2
2
3 (k1 ξ 2 + k2 α + k3 β) π
(40) in closed-form and obtain (41).
k1 ξ 2 + k2 α + k3 β + 1
 
×Γ . (43)
IV. A SYMPTOTIC BER A NALYSIS 2
The performance of a RF wireless communication can be In (43), we have the following constraints over the values of
understood easily by considering very high values of the SNR. k1 , k2 , k3 :
In the considered SIM FSO systems, we deal with the received
signal, which is converted into the electrical form from the k1 + k2 + k3 = M N
optical state; hence, the SIM scheme is analogous to the 0 ≤ k1 , k2 , k3 ≤ M N. (44)
RF modulation. Therefore, the conventions used for a RF
Further, the exponent of γ̄ is given by k1 ξ 2 + k2 α + k3 β /2.


wireless communication systems can be used in the case of
SIM FSO MIMO system. At high SNR, the BER performance Since the diversity of the scheme is dominated by the lowest

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

!
X M
P e(γ̄) = ζM 0≤k ≤N
PN PN −k1
[jk1 ,k2 ,N −k1 −k2 ]0≤k21 ≤N −k1
j =M
k1 =0 k2 =0 k1 ,k2 ,N −k1 −k2
0≤jk ,k ,N −k −k ≤M
1 2 1 2
!
0≤k ≤N
ΞMRC
0 [jk1 ,k2 ,N −k1 −k2 ]0≤k12 ≤N −k1
× PN PN −k1
k ξ2 +k2 α+(N −k1 −k2 )β )
k2 =0 k1 ,k2 ,N −k1 −k2 ( 1
j
k1 =0
γ̄ 2

√ PN PN −k1 2
( 2/a1 ) k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ +k2 α+(N −k1 −k2 )β )
× PN PN −k1 √
( k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 (k1 ξ 2 + k2 α + (N − k1 − k2 )β)) π
PN PN −k1 2

!
k1 =0 k2 =0 jk1 ,k2 ,N −k1 −k2 k1 ξ + k2 α + (N − k1 − k2 )β + 1
×Γ (51)
2

value of the exponent of γ̄, we have the following minimiza- TABLE I

tion problem: VALUE OF C0 (kGd ) FOR DIFFERENT VALUES OF kGd .
Gd kGd C0 (kGd )
min k1 ξ 2 + k2 α + k3 β

Find    2 M N
ξ 2
Γ(α−ξ 2
)Γ(β−ξ 2
)Γ(ξ2 ) αβ ξ
M N ξ 2 /2 [M N, 0, 0] Γ(α)Γ(β) A0
such that k1 + k2 + k3 = M N   M N
ξ Γ(ξ −α)Γ(β−α) αβ α
 2 2
and 0 ≤ k1 , k2 , k3 ≤ M N. (45) M N α/2 [0, M N, 0] Γ(β)Γ(1+ξ2 −α) A0
   M N
ξ2 Γ(ξ2 −β)Γ(α−β) αβ β
The solution of this minimization problem is given as: M N β/2 [0, 0, M N ] Γ(α)Γ(1+ξ2 −β) A 0

M N · min ξ 2 , α, β .

(46)
Therefore, the diversity gain of the scheme is given by: • If ξ 2 > min {α, β}, then Gd = (M N/2) min {α, β}.
MN Therefore, if PEs are less effective than the worst ATs,
· min ξ 2 , α, β .

Gd = (47)
2 then the ATs dominate the high SNR decay of the BER
For the calculation of the asymptotic BER, we keep only plots.
the dominating term (having the lowest exponent of γ̄) and • If ξ 2 < min {α, β}, then Gd = (M N/2) ξ 2 . Hence, if
ignore all other non-dominating terms. Let us denote the PEs are more severe than the worst ATs, then the PEs
values of k1 , k2 , k3 corresponding to the dominating term by a dominate the high SNR decay of the BER plots.
vector kGd = [k1 , k2 , k3 ]. It can be seen from (47) that there
are three possible distributions of kGd : kGd = [M N, 0, 0], B. Asymptotic BER Analysis of MRC
kGd = [0, M N, 0], and kGd = [0, 0, M N ]; one of these three It can be seen from (41) that the average probability of
vectors is used for the calculation of the asymptotic BER error of the MRC scheme contains summation terms with
depending upon the lowest-valued parameter out of ξ 2 , α, β, different exponents of the average SNR γ̄. Since the diversity
respectively. With these observations, the asymptotic BER in is dominated by the smallest exponent of the SNR, we put
(43) gets simplified as: n = 0 and p = 1 to get the asymptotic BER as given in (51).
√ For finding the diversity order from (51), we need to solve
ΞEGC (kGd ) ( 2/a1 )2Gd
 
0 2Gd + 1 the following minimization problem:
P e(γ̄) = ζM √ Γ , (48)
γ̄ Gd 2Gd π 2 ( N N −k
X X1
where Find min jk1 ,k2 ,N −k1 −k2
M Gd C0 (kGd ) k1 =0 k2 =0
ΞEGC
0 (kGd ) = . (49) )
2Γ(2Gd )
× k1 ξ 2 + k2 α + (N − k1 − k2 )β


The value of C0 (kGd ) for different values of kGd is given in
Table I. From (42) and (48), the combining gain of the EGC N NX
X −k1
scheme will be: such that jk1 ,k2 ,N −k1 −k2 = M
√
!−1/Gd k1 =0 k2 =0
EGC ζM ΞEGC
0 (kGd )( 2/a1 )2Gd Γ 2Gd2+1
Gc = √ . (50) and 0 ≤ jk1 ,k2 ,N −k1 −k2 ≤ M . (52)
2Gd π
A straightforward
 solution
of the minimization problem in (52)
Let us make some comments for the diversity order of the is M N min ξ 2 , α, β . Therefore, the diversity gain of the
EGC scheme. scheme is given by:
• The diversity of the EGC scheme depends not only upon
MN
· min ξ 2 , α, β .

the AT but also on the PE severity. Gd = (53)
2

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

TABLE II schemes achieve the same diversity order. The combining gain
VALUE OF ΞMRC
0 (jGd ) FOR DIFFERENT VALUES OF Gd . advantage of the MRC over EGC is
Gd ΞMRC (jGd ) 1/Gd
0
GMRC

 2 M c
M −1 Γ(Gd )
M N ξ 2 /2 1 Γ(N ξ /2) ∆G(M, N ) = = M2 Gd


Γ(M N ξ2 /2) 2Γ(N ξ2 )
ξ2 M N GEGC
c
Γ(2Gd )
ξ2 Γ(α−ξ2 )Γ(β−ξ2 )Γ(ξ2 )

αβ  M/Gd
× Γ(α)Γ(β) A0 Γ(2Gd /M )
Γ(N α/2) M
× , (56)
Γ(Gd /M )
 
1
M N α/2 Γ(M N α/2)2Γ(N α)
  M N
ξ Γ(ξ2 −α)Γ(β−α) αβ α
2
i.e., for high SNR on a log-log scale the SNR required to

× Γ(β)Γ(1+ξ2 −α) A0
1

Γ(N β/2) M
 attain a certain performance is 10log(∆G(N, M )) dB lower
M N β/2

Γ(M N β/2) 2Γ(N β)
  M N
for the MRC scheme than for the EGC scheme. Further, for
ξ2 Γ(ξ2 −β)Γ(α−β) αβ β M = 1, ∆G(N, 1) = 1, as can be seen from (56) that both
× Γ(α)Γ(1+ξ2 −β) A0
EGC and MRC are equivalent for a single receive aperture
based reception. Further, let us denote the diversity of a FSO
system with a single transmit  and a single receive aperture
From the solution of the minimization problem in (52), it can (N = M = 1) as G′d = min ξ 2 , α, pβ /2. Then by using the
be deduced that there are only three possible distributions of Stirling’s formula: Γ(z) → e−z z z 2π/z for z → ∞ [30, Eq.
0≤k ≤N
the row vector [jk1 ,k2 ,N −k1 −k2 ]0≤k12 ≤N −k1 corresponding to (6.1.37)] in (56), it can be shown that for a large number of
0≤k1 ≤N
receive apertures, we get the following asymptotic combining
the dominating term: jGd = [jk1 ,k2 ,N −k1 −k2 ] j0≤k 2 ≤N −k1
k1 ,k2 ,N −k1 −k2 =0
,
jN,0,0 =M
gain for the MRC:
0≤k1 ≤N
1/(N G′d )
0≤k2 ≤N −k1
2Γ(2N G′d )

j Gd = [j ]
k1 ,k2 ,N −k1 −k2 jk1 ,k2 ,N −k1 −k2 =0 , e
lim ∆G(M, N ) = . (57)
0≤k1 ≤N
j0,N,0 =M
M →∞ 4N G′d Γ(N G′d )
0≤k2 ≤N −k1
and jGd = [j ]
k1 ,k2 ,N −k1 −k2 jk1 ,k2 ,N −k1 −k2 =0 depending upon Therefore, for N = 2, the MRC renders combining gains of
j0,0,N =M
2 1.3325 dB, 0.985 dB, and 0.7813 dB for G′d = 0.5, 0.7, and
the minimum-valued parameter ξ , α, and β, respectively.
0≤k1 ≤N 0.9, respectively. Moreover, by using the Stirling’s formula in
The notation [jk1 ,k2 ,N −k1 −k2 ] j0≤k 2 ≤N −k1
k1 ,k2 ,N −k1 −k2 =0
denotes a row (56), it can be shown that:
jN,0,0 =M
vector with all zero entries, except jN,0,0 = M . lim ∆G(M, N ) = 1, ∀M ≥ 1, (58)
N G′d →∞
Following these observations, we can only keep the term
dominating the decay of the BER at high SNR in (51) and get i.e., MRC does not provide any combining gain if the FSO
the simplified asymptotic BER as: channel with PEs has large diversity order or a large number
! of transmit apertures are used.
ζM ΞMRC jGd √ By using the Euler’s formula [30, Eq. (6.1.2)]
0
( 2/a1 )2Gd
 
2Gd + 1 n!nz
P e(γ̄) = √ Γ .(54) Γ(z) = lim , z 6= 0, −1, .. (59)
γ̄ Gd 2Gd π 2 n→∞ z(z + 1)(z + 2)...(z + n)
!
it can be shown that
The values of ΞMRC
0 jGd for different values of Gd are 1
lim Γ(z) =
, z 6= 0, −1, .. (60)
listed in Table II. The combining gain of the MRC scheme is z→0 z
given by: After employing (60) in (56), we have the following result for
! √ a very small value of the diversity gain:
!−1/Gd
( 2/a1 )2Gd

MRC MRC 2Gd + 1 lim ∆G(M, N ) = M, Gd 6= 0, −1, .. (61)
Gc = ζM Ξ 0 jGd √ Γ . (55) Gd →0
2Gd π 2
i.e., irrespective of the number of transmit apertures, the MRC
All comments for the diversity order of the EGC scheme scheme provides a relative performance gain of 10logM , for
are as it is applicable to the MRC scheme as well. Now we very small value of Gd → 0. For example, for M = 3, 6, and
will use the diversity order and combining gains to compare 9, the combining gain of the MRC scheme is 11 dB, 18 dB,
the EGC and MRC in GG fading with PEs in next subsection. and 22 dB, respectively, for Gd → 0.

C. Comparison of EGC and MRC V. N UMERICAL R ESULTS

The MRC scheme is more complex than EGC scheme, cf. In this section, we provide a detailed discussion upon the
(2), (5). In particular, proposed characterization of the EGC and MRC schemes. The
PMwhile PN for EGC only the gain of the analytical plots are obtained by using the derived theoretical
combined channel i=1 Ii,j has to be estimated, M
P 2 j=1 results in the previous sections; the simulation results are
N
channel gains j=1 Ii,j , 1 ≤ i ≤ M , have to be estimated obtained by simulating the GG fading channels with PEs
to perform the MRC. Hence, from complexity point of view, through MATLABTM . The power series based expressions in
EGC is preferable. It can be seen from (47) and (53) that both (30), (31), and (41) are truncated to a finite number of terms

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

0
10

−1
10
−1
10
−2
10

−3
10
A0=1, ξ=1.56,
−2 α=4.2, β=1.4
BER

10

BER
−4
10
ξ=10.45, 2.15, 1.45,
1.15, 0.95,0.65,0.45 Analysis, K=1
−5
10 Analysis, K=2
Analysis, K=5
−6 Analysis, K=7
−3
10 10
Analysis, K=10
Analysis, K=12 A0=1, ξ=9.56,
−7
10 Analysis, K=15
Simulation α=4.0, β=1.9
Simulation
Analysis
−8
0 5 10 15 20 25 30 35 40 45 10
SNR [dB] 0 5 10 15 20 25 30 35 40 45
SNR [dB]

Fig. 1. BER versus SNR performance of the FSO system with N = Fig. 3. Convergence behavior of the proposed power series based BER expres-
M = 1, QPSK, GG fading with PEs, α = 2.1, β = 2, ξ = sions of the EGC and MRC schemes in a FSO system with N = 1, M = 2,
10.45, 2.15, 1.45, 1.15, 0.95, 0.65, 0.45, and A0 = 1. QPSK, GG fading moderate (α = 4.0, β = 1.9) and strong (α = 4.2,
β = 1.4) AT with PEs, A0 = 1, and ξ = 9.56, 1.56.

10
−1
for a large change in PE from ξ = 10.45 to ξ = 2.15 in
ξ=0.73 this region, there is no significant degradation in the BER
10
−2
performance. For example, at BER=10−3 , there is a SNR loss
of approximately 2 dB at ξ = 2.15 than that at ξ = 10.45.
−3
10 Fig. 2 shows the analytical and simulated BERs versus SNR
ξ=0.93
plots for the EGC and MRC schemes for QPSK constellation,
BER

−4
ξ=1.23
10 N = 1, and M = 2. The PE is characterized by the following
parameters: A0 = 0.75 and ξ = 11.23, 1.23, 0.93, 0.73. The
−5
10 AT is considered to be moderate [6], i.e., α = 4.0 and β = 1.9.
ξ=11.23
A close matching of the analytical and simulated BER plots
−6
10 is evident from the figure. For the considered AT, the MRC
Simulation
Analysis, EGC
scheme provides slight performance gain as compared to the
−7
10
Analysis, MRC EGC scheme. For example, at BER=10−3 , the relative SNR
0 5 10 15 20 25 30 35 40 45 gain is approximately 0.5 dB and 1 dB at PEs ξ = 11.23 and
SNR [dB]
ξ = 0.93, respectively, for the MRC scheme. Further, the PEs
Fig. 2. Performance plots of the EGC and MRC schemes in a FSO system indistinguishably affect the performance of the MRC and EGC
with N = 1, M = 2, QPSK, GG fading moderate AT with PEs, α = 4.0 schemes. Significant reduction in the performance is observed
and β = 1.9, A0 = 0.75, and ξ = 11.23, 1.23, 0.93, 0.73.
for ξ = 1.23, 0.93, 0.73, as compared to ξ = 11.23 case. For
K for obtaining the analytical BER results. The asymptotic example, at BER=10−5 , the EGC scheme requires 11.5 dB
BER performance of the EGC and MRC schemes is obtained more transmit power at ξ = 1.23 to achieve its performance
by using (42), (48), and (51). at ξ = 11.23. For plotting the analytical BER values, the
In Fig. 1, we consider a FSO system with N = M = 1. This truncated power series with 20 terms are used.
system uses the QPSK constellation based SIM scheme with The convergence of the proposed power series based BER
the AT parameters α = 2.1 and β = 2. The PE in the FSO link expressions is shown in Fig. 3 for MRC scheme with M = 2,
is varied as ξ = 10.45, 2.15, 1.45, 1.15, 0.95, 0.65, 0.45 with N = 1, and QPSK constellation over the moderate (α = 4.0,
A0 = 1. The simulated BER versus SNR plots are also shown β = 1.9) and strong (α = 4.2, β = 1.4) AT regimes [6].
in the figure. The simulated BER values closely match with the The PE is varied from low value ξ = 9.56 to high value
analytical BER values for all the SNR values and PEs, as seen ξ = 1.56 with A0 = 1. The power series are truncated to
from the figure. This observation justifies the correctness of K = 1, 2, 5, 7, 10, 12, 15 terms, where K denotes number of
the proposed analysis. Further, it should be noted that only 20 terms used for truncation. It can be seen from the figure that
terms are used for plotting the analytical curves. The effect for K = 15 the proposed BER expressions closely converge
of PEs can be observed from figure; decreasing value of ξ to the simulated BER values at all values of the SNR, for both
denotes increasing severity of PEs. For large PEs, i.e., ξ = kind of ATs and all considered values of the PEs. However,
0.45, the FSO system becomes almost useless as the value of the convergence becomes slower for moderate turbulence and
BER remains more than 10−1 for all SNR values. Moreover, low PEs, whereas, it is faster for strong AT and large PEs. For
for the considered AT, the PE seems tolerable for ξ > 1; example, for K = 12 the BER values converge satisfactorily

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

10

0
10 2
ξ=0.54
N=1, M=8
diversity=MNξ2/2=0.2916
−1 1.8 N=1, M=4
10
N=2, M=8
1.6 N=1, M=2
−2
10 N=2, M=4
1.4 N=2, M=2
N=4, M=2
−3
10 1.2 N=8, M=2

∆ G(M,N) [dB]
N=1, M=1
ξ=0.74
BER

1
diversity=MNξ2/2=0.5476
−4
10
0.8
−5
10
Asymptotic BER 0.6
EGC
−6
10 MRC 0.4
Reference diversity
−7 ξ=5.54 0.2
10
diversity=MNβ/2=1.4
0
0 10 20 30 40 50 3 3.5 4 4.5 5 5.5 6 6.5 7
SNR [dB] L [km]

Fig. 4. Diversity performance the EGC and MRC schemes in a FSO system Fig. 5. Performance gain ∆G(M, N ) of the MRC over EGC versus link
with N = 1, M = 2, QPSK, GG fading strong AT with PEs, α = 4.2 and 2 = 1.7 × 10−14 ,
distance L plots in a FSO system with λ = 1550 nm, Cn
β = 1.4, A0 = 1, and ξ = 0.54, 0.74, 5.54. wb /R = 10, and σs /R = 0.1.

even at low SNR values for ξ = 1.56. For SNR> 40 dB even 7

K = 1 provides a sufficiently tight convergence for the low
PE case, i.e., ξ = 9.56; on the other hand K = 1 is sufficient 6
N=1, M=8
N=2, M=8
for the BER calculation for SNR> 30 dB in the case of large N=1, M=4
PEs, i.e., ξ = 1.56. 5
N=2, M=4
N=1, M=2
In Fig. 4, the diversity performance of the MRC and EGC N=2, M=2
schemes is shown for QPSK, N = 1, M = 2, over the strong N=1, M=1
∆ G(M,N) [dB]

4
AT regimes, α = 4.2, β = 1.4. The PEs are characterized by
the following parameters: A0 = 1 and ξ = 0.54, 0.74, 5.54. 3
The derived asymptotic BER is sufficiently tight with the
analytical BER at high SNR, for all values of the PEs, as 2
seen in the figure. However, for high values of the PEs, the
asymptotic BER is able to characterize the performance of 1
both schemes from moderate to high SNR values. For example,
for ξ = 0.54, the asymptotic BER overlaps with the analytical 0
BER for SNR>21 dB, whereas, for ξ = 5.54 such overlapping 0 2 4 6 8
σs/R
10 12 14 16 18

is seen for SNR> 42 dB. Further, it can be seen from the figure
that the diversity gain of both schemes is adversely affected Fig. 6. Performance gain ∆G(M, N ) of the MRC over EGC versus
normalized standard deviation of the building sway σs /R plots in a FSO
due to the presence of large PEs, however, in the case of system with L = 4.25 km, λ = 1550 nm, Cn 2 = 1.7 × 10−14 , and
small PEs, the diversity gain is dominantly decided by the wb /R = 10.
AT. In order to check the diversity order in each scenario,
the reference diversity plots are also shown in the figure. The For a fixed value of N (number of transmit apertures) the
proposed diversity analysis given in Section IV is verified by combining gain of MRC improves with the increasing value
the figure. of M . For example, at L = 7.25 the MRC scheme provides
The asymptotic performance gain of the MRC scheme over 0.6 dB and 1.1 dB gains for N = M = 2 and N = 2, M = 8,
the EGC scheme, i.e., ∆G(M, N ) is shown as a function of respectively. In Fig. 6, ∆G(M, N ) is plotted for a fixed link
the distance L between the Tx and Rx assuming a plane wave length L = 4.25 km, normalized beamwaist wb /R = 10, and
prorogation, in Fig. 5. It is assumed that the FSO system varying the normalized values of the standard deviation of the
operates at λ = 1550 nm with Cn2 = 1.7 × 10−14 . The building sway σs /R. It can be seen from the figure that for
normalized values of the beamwaist and standard deviation small values of σs /R, the asymptotic performance gain of the
of the building sway are wb /R = 10 and σs /R = 0.1, MRC scheme remains constant, however, for σs /R > 4.6,
respectively. The values of α, β, and ξ 2 are calculated by using the relative gain of the MRC scheme monotonically increases
the expressions given in the Subsection II.B. It can be seen for all considered configurations of the transmit and receive
from the figure that the relative combining gain of the MRC apertures, except for M = N = 1. It appears from figure
monotonically increases with the link distance. For example, that the MRC scheme offers a significant relative performance
for N = 1 and M = 8, MRC renders performance gains of 1.7 gain for large PEs. For example, for N = 1, M = 4, the MRC
dB and 1.9 dB for L = 4.25 km and 7.25 km, respectively. scheme renders an approximately 5 dB gain at σs /R = 18;

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

11

and for N = 1, M = 8 the gain is approximately 7 dB at [14] X. Song and J. Cheng, “Subcarrier intensity modulated MIMO optical
σs /R = 18. Further, for a fixed number of transmit apertures communications in atmospheric turbulence,” IEEE/OSA J. Opt. Com-
mun. and Net., vol. 5, no. 9, pp. 1001–1009, Sept. 2013.
the combining gain of MRC improves with increasing number [15] M. R. Bhatnagar, “A one bit feedback based beamforming scheme
of receive apertures, for all values of the PE. for FSO MISO system over Gamma-Gamma fading,” IEEE Trans.
Commun., vol. 63, no. 4, pp. 1306 – 1318, Apr. 2015.
VI. C ONCLUSIONS [16] M. M. Abadi, Z. Ghassemlooy, M.-A. Khalighi, S. Zvanovec, and
M. R. Bhatnagar, “FSO detection using differential signaling in outdoor
The Meijer-G function based PDF expression of the GG correlated-channels condition,” IEEE Photon. Technol. Lett., vol. 28,
fading FSO links with PEs poses a significant challenge for no. 1, pp. 55–58, Jan. 2016.
[17] M. Uysal, J. Li, and M. Yu, “Error rate performance analys is of coded
deriving the analytical performance of a generalized FSO free-space optical links over Gamma-Gamma atmospheric turbulence
MIMO system. Due to this problem, it is difficult to theo- channels,” IEEE Trans. Wireless Commun., vol. 5, no. 6, pp. 1229–1233,
retically study the effect of the PEs in a FSO MIMO over June 2006.
[18] S. M. Alamouti, “A simple transmit diversity technique for wireless
different SNR range. We have derived a novel power series communications,” IEEE J. Select. Areas Commun., vol. 16, no. 8, pp.
based expression for the PDF of GG fading with PEs. The 1451–1458, Oct. 1998.
proposed series based expression has been shown to render a [19] G. Yang, M. K. Khalighi, Z. Ghassemlooy, and S. Bourennane, “Perfor-
mance evaluation of receive-diversity free-space optical communications
fast and accurate performance evaluation of two well-known over correlated Gamma-Gamma fading channels,” OSA Applied optics,
combining schemes. Further, the proposed analysis has shed vol. 52, no. 24, pp. 5903–5911, Aug. 2013.
a substantial light over the asymptotic behavior of the FSO [20] X. Liu, “Free-space optics optimization models for building sway
and atmospheric interference using variable wavelength,” IEEE Trans.
MIMO system in presence of the GG AT and PEs. Closed- Commun., vol. 57, no. 2, pp. 492–498, Feb. 2009.
form expressions of the combining gains of the EGC and [21] W. Gappmair, “Further results on the capacity of free-space optical
MRC schemes have been derived. It has been shown by using channels in turbulent atmosphere,” IET Commun., vol. 5, no. 9, pp.
1262–1267, June 2011.
these derived combining gains that the MRC scheme is more [22] I. E. Lee, Z. Ghassemlooy, W. P. Ng, and M.-A. Khalighi, “Joint
impregnable for the large PEs than EGC scheme. However, optimization of a partially coherent Gaussian beam for free-space optical
in low to moderate PEs EGC is preferable due to its lower communication over turbulent channels with pointing errors,” Optics
Lett., vol. 38, no. 3, pp. 350–352, Feb. 2013.
implementation complexity and slight loss in the performance [23] F. Yang, J. Cheng, and T. A. Tsiftsis, “Free-space optical communication
compared to MRC. with nonzero boresight pointing errors,” IEEE Trans. Commun., vol. 62,
no. 2, pp. 713–725, Feb. 2014.
R EFERENCES [24] A. A. Farid and S. Hranilovic, “Diversity gain and outage probability
for MIMO free-space optical links with misalignment,” IEEE Trans.
[1] M. R. Bhatnagar and Z. Ghassemlooy, “Performance evaluation of FSO Commun., vol. 60, no. 2, pp. 479–487, Feb. 2012.
MIMO links in Gamma-Gamma fading with pointing error,” in Proc. [25] M. R. Bhatnagar and S. Anees, “On the performance of Alamouti
IEEE International Conference on Communications (ICC), pp. 1–6, June scheme in Gamma-Gamma fading FSO links with pointing errors,” IEEE
2015, London, U.K. Wireless Commun. Lett., vol. 4, no. 1, pp. 94–97, Feb. 2015.
[2] D. Kedar and S. Arnon, “Urban optical wireless communication net- [26] A. Garca-Zambrana, B. Castillo-Vzquez, and C. Castillo-Vzquez,
works: The main challenges and possible solutions,” IEEE Commun. “Asymptotic error-rate analysis of FSO links using transmit laser
Mag., vol. 42, no. 5, pp. S2–S7, Jan 2004. selection over Gamma-Gamma atmospheric turbulence channels with
[3] L. Andrews, R. Phillips, and C. Hopen, Laser Beam Scintillation With pointing errors,” Optics Express, vol. 20, no. 3, pp. 2096–2109, Jan.
Applications. New York: USA: SPIE Press, 2001. 2012.
[4] F. Ahdi and S. Subramaniam, “Optimal placement of FSO relays for [27] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
network disaster recovery,” in Proc. IEEE International Conference on Products, 6th ed. San Diego, CA, USA: Academic Press, 2000.
Communications (ICC), pp. 3921 – 3926, June 2013, Budapest, Hungary. [28] K. Roach, “Meijer-G function representations,” in Proc. ACM Interna-
[5] K. P. Peppas and C. K. Datsikas, “Average symbol error probability tional Conf. Symbolic Algebraic Computation, pp. 205–211, July 1997.
of general-order rectangular quadrature amplitude modulation of optical [29] J. Lu, K. B. Letaief, J. C.-I. Chuang, and M. L. Liou, “M -PSK and
wireless communication systems over atmospheric turbulence channels,” M -QAM BER computation using signal-space concepts,” IEEE Trans.
J. Opt. Commun. Netw., vol. 2, no. 2, pp. 102–110, Feb. 2010. Commun., vol. 47, no. 2, pp. 181–184, Feb. 1999.
[6] W. O. Popoola and Z. Ghassemlooy, “BPSK subcarrier intensity modu- [30] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions.
lated free-space optical communications in atmospheric turbulence,” J. New York, USA: Dover Publications, Inc., 1972.
Lightw. Technol., vol. 27, no. 8, pp. 967–973, Apr. 2009.
[7] E. Bayaki, R. Schober, and R. K. Mallik, “Performance analysis of
MIMO free-space optical systems in Gamma-Gamma fading,” IEEE
Trans. Commun., vol. 57, no. 11, pp. 3415–3424,, Nov. 2009.
[8] N. Letzepis and A. G. Fabregas, “Outage probability of the Gaussian
MIMO free-space optical channel with PPM,” IEEE Trans. Commun.,
vol. 57, no. 12, pp. 3682–3690, Dec. 2009.
[9] T. A. Tsiftsis, H. G. Sandalidis, G. K. Karagiannidis, and M. Uysal,
“Optical wireless links with spatial diversity over strong atmospheric
turbulence channels,” IEEE Trans. Wireless Commun., vol. 8, no. 2, pp.
951–957, Feb. 2009.
[10] J. Park, E. Lee, and G. Yoon, “Average bit-error rate of the Alamouti
scheme in Gamma-Gamma fading channels,” IEEE Photon. Technol.
Lett., vol. 23, no. 4, pp. 269–271, Feb. 2011.
[11] C. Abou-Rjeily, “On the optimality of the selection transmit diversity for
MIMO-FSO links with feedback,” IEEE Commun. Lett., vol. 15, no. 6,
pp. 641–643, Jun. 2011.
[12] M. R. Bhatnagar, “Differential decoding of SIM DPSK over FSO MIMO
links,” IEEE Commun. Lett., vol. 17, no. 1, pp. 79 – 82, Jan. 2013.
[13] W. Popoola, Z. Ghassemlooy, J. Allen, E. Leitgeb, and S. Gao, “Free-
space optical communication employing subcarrier modulation and
spatial diversity in atmospheric turbulence channel,” IET Opt., vol. 2,
no. 1, pp. 16–23, Feb. 2008.

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JLT.2016.2526053, Journal of
Lightwave Technology

12

Manav R. Bhatnagar SM-IEEE received his Ph.D.

in Wireless Communications from Department of
Informatics, University of Oslo, Norway in 2008
and M. Tech. in Communications Engineering from
Indian Institute of Technology Delhi, New Delhi,
India in 2005.
From 2008 to 2009, he was a Postdoctoral
Research Fellow with University Graduate Center
(UNIK), University of Oslo, Kjeller, Norway. He
held visiting appointments with the Wireless Re-
search Group, Indian Institute of Technology Delhi,
New Delhi, India; the Signal Processing in Networking and Communications
(SPiNCOM) Group, University of Minnesota Twin Cities, Minneapolis; the
Alcatel-Lucent Chair, SUPLEC, France; the Department of Electrical Com-
puter Engineering, Indian Institute of Science at Bangalore, India; UNIK,
University of Oslo, Kjeller, Norway; the Department of Communications and
Networking, Aalto University, Finland; and INRIA/IRISA Lab, University of
Rennes, Lannion, France.
He is currently an Associate Professor with the Department of Electrical
Engineering, Indian Institute of Technology Delhi, New Delhi, India. His
research interests include signal processing for multiple-input-multiple-output
systems, cooperative communications, non-coherent communication systems,
distributed signal processing for cooperative networks, multiuser communica-
tions, ultrawideband-based communications, free-space optical communica-
tion, cognitive radio, software defined radio, power line communications, and
satellite communications.
Dr. Bhatnagar was selected as an ”Exemplary Reviewer” of the IEEE COM-
MUNICATIONS LETTERS for 2010 and 2012. He has served as an Editor for
the IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS during
2011-2014.

Zabih (Fary) Ghassemlooy Professor Zabih (Fary)

Ghassemlooy CEng, Fellow of IET, SM-IEEE: Re-
ceived BSc (Hons) degree in Electrical and Electron-
ics Engineering from the Manchester Metropolitan
University in 1981, and MSc and PhD from UMIST
in 1984 and 1987, respectively. 1987-88 he was a
Post-doctoral Research Fellow at City University,
London. 1988 joined Sheffield Hallam University
as a Lecturer, becoming a Reader in 1995 and a
Professor in Optical Communications in 1997. 2004
joined the University of Northumbria at Newcastle
(UNN) as an Associate Dean (AD) for Research in the School of Computing,
Engineering and Information Sciences. 2012-2014 he was AD for research
and innovation, Faculty of Engineering, at UNN, UK. He now heads the
Northumbria Communications Research Laboratories within the Faculty. His
research interests are in optical wireless communications, free space optics,
visible light communications. He has published over 550 papers (over 215
Journals, and 4 books) and have supervised 48 PhD students. He is the Vice-
Chair of EU Cost Action IC1101. He is chief editor of the British Journal of
Applied Science and Technology and the International Journal of Optics and
Applications.

0733-8724 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.