On the Maximum and Minimum of Double Generalized

Gamma Variates with Applications to the Performance

of Free-space Optical Communication Systems

Item Type Article

Authors Al-Quwaiee, Hessa; Ansari, Imran Shafique; Alouini, Mohamed-

Slim

Citation On the Maximum and Minimum of Double Generalized Gamma

Variates with Applications to the Performance of Free-space
Optical Communication Systems 2016:1 IEEE Transactions on
Vehicular Technology

Eprint version Post-print

DOI 10.1109/TVT.2016.2515662

Publisher Institute of Electrical and Electronics Engineers (IEEE)

Journal IEEE Transactions on Vehicular Technology

Rights (c) 2015 IEEE. Personal use of this material is permitted.

Permission from IEEE must be obtained for all other users,
including reprinting/ republishing this material for advertising or
promotional purposes, creating new collective works for resale
or redistribution to servers or lists, or reuse of any copyrighted
components of this work in other works.

Download date 07/03/2019 09:10:03

Link to Item http://hdl.handle.net/10754/593137

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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

On the Maximum and Minimum of Double

Generalized Gamma Variates with Applications to
the Performance of Free-space Optical
Communication Systems
Hessa AlQuwaiee, Student Member, IEEE, Imran Shafique Ansari, Member, IEEE,
and Mohamed-Slim Alouini, Fellow, IEEE

Abstract—In this work, we derive the exact statistical char- interference and efficient employment time [7], [8]. This
acteristics of the maximum and the minimum of two modified1 makes FSO applicable for broadband wireless technologies
double generalized gamma variates in closed-form in terms of such as optical fiber backup, metropolitan area network, and
Meijer’s G-function, Fox’s H-function, the extended generalized
bivariate Meijer’s G-function and H-function in addition to last mile access [9]. On the other hand, FSO still have some
simple closed-form asymptotic results in terms of elementary challenges that can affect link range and system reliability
functions. Then, we rely on these new results to present the such as atmospheric turbulence conditions (scintillation) that
performance analysis of (i) a dual-branch free-space optical is comparable to the fading phenomena in the RF systems.
selection combining diversity and of (ii) a dual-hop free-space Weather conditions can also affect signal strength as the air
optical relay transmission system over double generalized gamma
fading channels with the impact of pointing errors. In addition, being the medium of the transmission. This can be solved by
we provide asymptotic results of the bit error rate of the employing network redundancy along the transmission path.
two systems at high SNR regime. Computer-based Monte-Carlo Moreover, thermal expansion, weak earth quakes, and dynamic
simulations verify our new analytical results. wind loads can result in misalignment between the transmitter
Index Terms—Free-space optical, double generalized Gamma, and the receiver introducing the pointing errors [2], [10]–[12].
turbulence, pointing errors, dual-branch diversity, selection com- Diversity, in which multiple copies of the same data is
bining, dual-hop relay system, hybrid RF/FSO. transmitted to the receiver such that each copy experiences
different path/channel, has shown a great deal in mitigating
I. I NTRODUCTION fading effects in RF systems in addition to improving both
Recently, radio frequency spectrum for wireless communi- performance and capacity [13]–[15]. For diversity combining
cations is facing a crunch. Research is driven to find solutions techniques, selection combining (SC) is one of the most
such as utilizing the spectrum more efficiently or using un- efficient and simple combining schemes. In SC, the receiver
regulated one such as the optical spectrum under which free- aims to process the branch with maximum signal-to-noise
space optical (FSO) communications operates [5], [6]. FSO ratio (SNR). Diversity schemes in FSO systems was first
can complement or substitute radio frequency (RF) systems introduced by Ibrahim [16] followed by great number of
due to their higher bandwidth, cost effective, resistivity to research instigating diversity over weak and strong turbulence
conditions. Some investigated the spatial diversity and com-
1 Under the impact of pointing error (i.e. I = I I where I is the
p a a bining techniques of correlated and independent log-normal
irradiance modeled as double generalized gamma random variable [1] and turbulence channels [17]–[20]. Also, spatial diversity was
Ip is the pointing error whose probability density function (PDF) can be
found in [2, Eq. (10)]). The PDF of I can be found in [3, Eq. (8)]
studied over K-distributed channels for different combining
Copyright (c) 2015 IEEE. Personal use of this material is permitted. schemes in [21]. Moreover, SC over Gamma-Gamma under
However, permission to use this material for any other purposes must be the impact of pointing errors has been investigated [4].
obtained from the IEEE by sending a request to [email protected]. Relaying also has been an interesting area of research as
The research reported in this publication was supported by King Abdulaziz
City of Sciences and Technology (KACST) under the Grant Agreement it enhances the capacity of the system and provides energy
number AT-34-145 and by the King Abdullah University of Science and efficient coverage. In this technique, a relay node is positioned
Technology (KAUST). Furthermore, it is important to note that our work in the way between transmitter and receiver to support direct
in [4] is a special case of this paper.
H. Alquwaiee, and M.-S.Alouini are with King Abdullah University of data transmission. Multiple research works have been reported
Science and Technology (KAUST), Computer, Electrical and Mathematical to analyze the relay system on both symmetric and asym-
Science and Engineering Division (CEMSE), Thuwal 23955-6900, Saudi metric links (i.e. symmetric links when source-relay (S~R)
Arabia (e-mail: hessa.alquwaiee, [email protected]).
I. S. Ansari is currently with the Department of Electrical and Com-
and relay-destination (R~D) links fall under the same fading
puter Engineering (ECEN), Texas A&M University at Qatar (TAMUQ), model) [22]–[25]. Moreover, relaying over FSO system was
Education City, Doha, Qatar. Please note when this work was initially introduced and then followed by several research work con-
submitted I. S. Ansari was with King Abdullah University of Science and
Technology (KAUST), Computer, Electrical and Mathematical Science and
cerning symmetric multi-hop FSO systems over K-distributed,
Engineering Division (CEMSE), Thuwal 23955-6900, Saudi Arabia (email: Gaussian and Gamma-Gamma turbulence channel [26]–[28].
[email protected], [email protected]). On the other hand, several studies have been conducted to

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

analyze asymmetric fixed and variable gain FSO relay systems of γi can be given as [37, Eq. (14)]
as such integrating RF links in FSO systems. Specifically, β − 12 β −1 ri (λi +σi )

assuming that RF and FSO links are, respectively, subject ξi2 σi 1i

λi 2i 2 (2π)1− 2 riβ1i +β2i −2
Fγi (γ) =
to Rayleigh and Gamma-Gamma or M-distributed turbulence α λ Γ(β1i )Γ(β2i )
 2i i vi  1,κ 
channels [29]–[31]. Analyzing variable gain relay system
× Gurii+1,u
,1
i +1
C γ
i µr ,i
 3i , (2)
analytically might not be tractable and mathematically not i
 κ4i ,0
feasible. Throughout the years, the end-to-end SNR has been we
where ξi = 2σ is the ratio between the equivalent beam width
upper bounded by the minimum SNR among the sublinks [32] s
at the receiver and the pointing error displacement standard de-
and harmonic mean of the each link SNR [33]. Few research
viation (jitter), αx , βx , and Ωx are identified using the variance
work have been reported focusing on variable gain FSO relay
of the small and large scale fluctuations of the laser beam from
systems over Gamma-Gamma with pointing errors [4], [34],
[38, Eq. (18)-(20)] inserted in [1, Eq. (8a), (8b), (9)], λi and σi
[35].
are positive integers such as λσii = α α
1i
, ui = ri (1 + λi + σi ),
Double generalized gamma channel model for free-space  vi
2i
ri
A2,i hi A1i B1i
optical communication systems developed by [1] covers a very vi = α2i λi , Ci = (λi +σi ) , hi = 2 1/vi ,
ri (1+ξi )A2i
wide range of turbulence conditions and makes it generic to β −1 β −1 (λi +σi )
σ λ
describe the FSO channel. Also, pointing errors was integrated ξi2 σi 1i 2 λi 2i 2 (2π) 1−
2 β1ii β2ii
A1i = Γ(β1i )Γ(β2i ) , A2i = λi σi σi λ ,
λi σi Ω1i Ω2ii
to this model to better show the impact of pointing error Qσi +λi  
1
impairments on the FSO system [3]. Hence, in this work, B1i = j=1 Γ vi + κ1i,j , where κi,j is the jth-term
we study the dual-branch FSO selection combining and dual- ξ 2 +v
of κi , κ1i = ∆(σi : β1i ), ∆(λi : β2i ), κ3i = ∆(ri : i vi i )
hop variable gain FSO relay operating on such channels with comprising of ri terms, and κ4i = [∆(ri : κ2i )] comprising of ui
the impact of pointing errors to show diversity enhancement terms such that κ2i = ξi2 /vi , κ1i , ∆(x : y) , xy , y+1 . . . , y+x−1
x x
on the system performance and capacity. First, we express and [∆(z : am )] , ∆(z : a1 ), ∆(z : a2 ) . . . , ∆(z : am ).
the statistical properties of the maximum of the minimum of By setting σi = λi = Ωi = 1,β1i = αi , β2i = βi , hi =
ri
double Generalized Gamma random variables under the impact ξi2
ξi2 +1
, ξi >> 1 then hi = 1, Ci = αri2βi , ui = 3ri , vi = 1,
of pointing errors in terms of H-function and G-function. In 2 2
i

particular, we find the cumulative distribution function (CDF), κ3i = ∆(ri : ξi + 1), κ4i = ∆(ri : ξi ), ∆(ri : αi ), ∆(ri : βi )
the probability density function (PDF), the moment generating we have the special case of Gamma-Gamma (GG) turbulence
function (MGF), and the moments in closed-form. We then use perturbed by pointing errors [4].
these results to evaluate performance measures such as the III. C LOSED -F ORM S TATISTICAL C HARACTERISTICS
average bit error rate (BER) and the ergodic capacity (EC)
A. Maximum of Two Variates
of both FSO systems. Finally, we multiplex two independent
FSO links with desired priority on the data rate of each links, Here, we study the statistical characteristic of the maximum
thereby, utilizing the resources efficiently and effectively. of two modified generalized gamma random variates γM which
is given by
II. C HANNEL AND S YSTEM M ODELS γM = max(γa , γb ), (3)

In this work, we consider a single FSO branch with two assuming that γa and γb are independent not necessarily
types of detection techniques, heterodyne and intensity modu- identically distributed (i.n.i.d) random variables (RVs).
lation/direct detection (IM/DD). Data transmission is affected 1) Cumulative Distribution Function: The CDF can be
by path loss, atmospheric turbulence conditions, pointing easily derived as
errors, and additive white Gaussian noise (AWGN) that can ua ,1
 
γ
va  1,κ 
FγM (γ) = A3a A3b Gra +1,ua +1 Ca µr ,a  3a
be modeled as a  κ4a ,0
yi = ηIi x + wi , (1)   vb  1,κ 
× Gurbb+1,u
,1
b +1
Cb µrγ ,b  3b , (4)
where η is the effective photoelectric conversion ratio, wi  b κ4b ,0
refers to the AWGN sample with power spectral density equals β −1 β −1 ri (λi +σi )
β1i +β2i −2
ξi2 σi 1i 2 λi 2i 2 (2π)1− 2 r
to N0 , and Ii is the receiver irradiance that is defined as where A3i = i
. For
α2i λi Γ(β1i )Γ(β2i )
Ii = Ia,i Ip where Ia,i and Ip reflect the turbulence-induced the case of Gamma-Gamma turbulence model, we have A3i =
α +β1 −2
fading and the pointing error effect, r respectively [36]. The ξi2 ri 1
Γ(αi ) Γ(βi )(2π)ri −1
.
I i
end-to-end SNR is defined as γi = E[Iii ]ri µri ,i where ri refers Moreover, an asymptotic expression can be obtained via
to the detection method (i.e. ri = 1 heterodyne detection and the expansion of the Meijer’s G-function [39, Eq. (26)] as
ri = 2 IM/DD), µri ,i is the average SNR defined in [37, Ci ( µrγ ,i )vi → 0 (i.e high SNR) as
i
Eq. (13)], and E[.] is the expectation operator. If we assume !κ4b,x
ua X u b  va κ
that the turbulence channel is modeled as double generalized X γ Ca 4a,k γ vb Cb
Gamma under the assumption of pointing errors, then the CDF F̃γM (γ) u A3a A3b
µr ,i >>1
i
µvraa,a µvrbb,b
k=1 x=1
Qua Qub
l=1;l6=k Γ(κ4a,l − κ4a,k ) l=1;l6=x Γ(κ4b,l − κ4b,x )
× Q ra Qrb .
κ4a,k κ4b,x l=1 Γ(κ3a,l − κ4a,k ) l=1 Γ(κ3b,l − κ4b,k )
(5)

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

The asymptotic expression in (5) is simple and is in terms of can be written as

elementary functions that can be evaluated using any computer "
1/va 
#
A3a A3b ua ,1 Ca γ  (κ5a ,1/va)
software. FγM (γ) = Hra +1,ua +1
2) Probability Density Function: The PDF is then obtained va vb µra ,a  (κ6a ,1/va)
" 1/vb  #
by differentiating (4) with respect to γ yielding ub ,1 Cb γ  (κ5b ,1/vb)
   vb  κ  × Hrb +1,ub +1 , (9)
A3a A3b ub ,0 γ  3b µrb ,b  (κ6b ,1/vb)
fγM (γ) = vb Grb ,ub Cb µr ,b
γ b
 κ4b
   va   where κ5i = 1, κ3i and κ6i = κ4i , 0. Afterward, we apply the
×Gruaa+1,u
,1
Ca µrγ ,a  1,κ3a identity [42, Eq. (2.2)], after inserting (9) in (8) to obtain
a +1 a  κ4a ,0
A3a A3b
va  κ  MγM (s) =
 
+ va Gruaa,u
,0
a
Ca µrγ ,a  3a
 κ4a va vb
a    
  vb  1,κ  0,1 (0;1,1)
 
×Gurbb+1,u
,1
b +1
Cb µrγ ,b  3b .
 κ4b ,0 (6) 
 1,0 (−;−,−) Ca1/va 

b   
−1
sµra ,a 
ua ,1 (κ5a ,[va ])
× H ,
   
An asymptotic expression of the PDF can be obtained by the ra +1,ua +1 (κ6a ,[va−1
]) 1/v

Cb b
 
same way done earlier for the CDF. In other words, by using 


ub ,1

(κ5b ,[vb−1 ]) sµrb ,b


the expansion of the Meijer’s G-function, we can reach to an 
rb +1,ub +1

(κ6b ,[vb−1 ])
asymptotic result as (10)
ua X ub  κ4a,k !κ4b,x
˜
X Ca Cb where H[.] is the bivariate H-function defined in [42]. It
fγM (γ) u A3a A3b
µri ,i >>1 µvraa,a µvrbb,b can be evaluated efficiently using MATHEMATICAr imple-
k=1 x=1
Qua Qub mentation in [43] or MATLABr implementation in [44]. In
l=1;l6=k Γ(κ4a,l − κ4a,k ) =x Γ(κ4b,l − κ4b,x )
× Qrb Qrl=1;l6 this work, we have implemented the bivariate H-function in
l=1 Γ(κ3b,l− κ4b,x ) l=1 Γ(κ3a,l − κ4a,k )
a

 MATHEMATICAr and easily evaluated the expressions.

vb va Simplified formula can be obtained if solving (8) but using
× + γ va κ4a,k +vb κ4b,x −1 . (7)
κ4a,k κ4b,x the asymptotic CDF (5) to reach to the following result
!κ4b,x
Derived PDFs were verified via Monte-Carlo simulations as Xua Xub 
Ca
κ4a,k
Cb
M̃γM (s) u A3a A3b
µr ,i >>1
i
µvraa,a µvrbb,b
k=1 x=1
0.4 Qua Qub
l=1;l6=k Γ(κ4a,l − κ4a,k ) =x Γ(κ4b,l − κ4b,x )
Monte-Carlo simulation
Exact analytical result: Eq. (6)
× Qra Ql=1;l6
rb
l=1 Γ(κ3a,l − κ4a,k ) l=1 Γ(κ3b,l − κ4b,k )
0.35
Probability Density Function (PDF), f γ (γ)

Exact asymptotic result: Eq. (7)

M

0.3 Γ(va κ4a,k + vb κ4b,x + 1) s−(va κ4a,k +vb κ4b,x )

× . (11)
κ4a,k κ4b,x
0.25
Setting the same special parameters listed in Sec. II for the
0.2
Gamma-Gamma case, the MGF can be expressed as
0.15
M∗γM (s) = A3a A3b
    
0.1 1 κ5a  κ5b  Ca Cb
× G1,0:3ra ,1:3rb ,1
1,0:ra +1,3ra +1:rb +1,3rb +1 −
  
 κ6a  κ6b  sµra ,a , sµrb ,b ,
0.05
(12)
0
−,−:−,−:−,−
1 2 3 4 5 6 7 8 9 10 in agreement with [4, Eq. (8)] where G−,−:−,−:−,− [.] is the ex-
γ
tended generalized bivariate Meijer’s G-function (EGBMGF)
Figure 1: PDF validation for the maximum of two double GG variates that can be evaluated using the code given in [43, Table II].
4) Moments: The moments is defined as
shown in Fig. 1. Z ∞
n
3) Moment Generating Function: The MGF defined as E[γ ] , γ n fγ (γ)dγ. (13)
MγM (s) , E[e−γM s ] can be expressed in terms of CDF using 0

integration by parts as [40]

Z ∞
MγM (s) = s exp(−γs)FγM (γ)dγ . (8)
0
     
1/C  a
1 m,n
 m,n
 (a,C)
Using the property C Gp,q z = Hp,q z  , we

 b (b,C)
transform each term to the Fox’s H-function that can be
evaluated using Mathematica [41]. As such, the CDF in (4)

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

Placing (6) in (13) and utilizing equation [45, Eq. (1.7)], we obtained as
obtain the following A3a
  va  κ 
ua ,0 γ  3a

µra ,a
n fγm (γ) = va Gra ,ua Ca µr ,a
n
E[γM ] = A3a A3b vb γ a  κ4a
1/va    
C
a A3b  vb  κ
1/vb
v v 
a b 
 + vb Gurbb,u
,0
Cb µrγ ,b  3b − fγ (γ) ,
M
u +1,ua
×Huba +rb +1,u
C
 b
µra ,a  (κ7a ,vb ),(κ3b ,va ) γ b
b
 κ4b
b +ra +1

1/v
Ca a µr ,b
 (κ4b ,va ),(κ8a ,vb )
!n b (20)
µrb ,b
+ 1/v
va where fγM (γ) is defined in (6). Similarly, an asymptotic
Cb b expression of the PDF can be obtained when expanding the
 1/v v v  
a b 
u +1,ub
×Huab +ra +1,u
C
 a
aµ
rb ,b   (κ7b ,va ),(κ3a ,vb ) . Meijer’s G-function leading to
a +r b +1 1/vb
C µra ,a
 (κ4a ,vb ),(κ8b ,va ) ua  κ4a,k
b
(14)
X Ca
f˜γm (γ) u A3a va γ va κ4a,k −1
n n µri ,i >>1 µvraa,a
where κ7i = 1 − κ6i − and κ8i = 1 − κ5i −
vi Setting the vi . Qua k=1
same parameters listed in Sec. II, (14) simplifies to l=1;l6=k Γ(κ4a,l − κ 4a,k )
n ∗
× Qra
E[γM ] = A3a A3b l=1 Γ(κ3a,l − κ4a,k )
 n    ub
!κ4b,x
µra ,a 3r +1,3ra Cb µra ,a  κ7a ,κ3b

Cb
× G3rba +rb +1,3r
X
Ca b +r a +1 C µ
a r ,b  κ ,κ
b 4b 8a
+ A3b vb vb γ vb κ4b,x −1
n µ r ,b
Qub x=1
    b
µrb ,b 3r +1,3rb Ca µr ,b
 κ7b ,κ3a ,

+ G3rab +ra +1,3r b (15)
l=1;l6=x Γ(κ4b,l − κ4b,x )
a +r +1 Cb µra ,a 
Cb b κ4a ,κ8b
× Qrb − f˜γM (γ). (21)
in agreement with [4, Eq. (10)]. l=1 Γ(κ3b,l − κ4b,x )
Monte-Carlo simulations were utilized again to verify the
B. Minimum of Two Variates
In this section, we list the statistical properties of the min- 6
imum of two modified generalized gamma random variables Monte-Carlo simulation
Exact analytical result: Eq. (20)
γm given by
Probability Density Function (PDF), f γ (γ)

5 Exact asymptotic result: Eq. (21)

m

γm = min(γ1 , γ2 ) . (16)
4
1) Cumulative Distribution Function: The CDF is given by
Fγm (γ) = 1 − Pr(min(γ1 , γ2 ) > γ) 3

= Fγ1 (γ) + Fγ2 (γ) − Fγ1 (γ)Fγ2 (γ), (17)

2
under the assumption that γ1 and γ2 are i.n.i.d. Substituting
(2) in (17), we obtain
1
  va  
ua ,1 γ 1,κ3a
Fγm (γ) = A3a Gra +1,ua +1 Ca µr ,a 
 κ4a ,0
a 0
   0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
vb  1,κ 
+ A3b Gurbb+1,u
,1 γ  3b γ
b +1
Cb µr ,b  κ4b ,0
b
Figure 2: PDF validation for the minimum of two double GG variates
  va  
ua ,1
− A3a A3b Gra +1,ua +1 Ca µr ,a γ  1,κ3a
a  κ4a ,0 PDFs as in Fig. 2.
  vb  1,κ  3) Moment Generating Function: Placing (18) in the def-
× Gurbb+1,u
,1
b +1
Cb µrγ ,b  3b . (18) inition of the MGF, and utilizing Eq. (10) and [46, Eq.
 b κ4b ,0
(07.34.21.0088.01)], the MGF can be expressed as
In addition, an asymptotic expression can be obtained via the √   va  ∆(v ,1),κ 
expansion of the Meijer’s G function as A3a va ua ,va va a 3a
Mγm (s) = va −1 G ra +va ,ua C a sµr ,a

ua  va κ (2π) 2 a  κ4a
X γ Ca 4a,k √
F̃γm (γ) u A3a A3b vb ub ,vb
  vb  ∆(v ,1),κ 
µri ,i >>1 µvraa,a + G C vb b 3b
− MγM (s).
k=1 b sµr ,b
vb −1 rb +vb ,ub

κ
(2π) 2 b
!κ4b,x  4b
Qua ub
l=1;l6=k Γ(κ4a,l − κ4a,k ) γ vb Cb
X
× Qra + A3b v (22)
κ4a,k l=1 Γ(κ3a,l − κ4a,k ) x=1
µrbb,b
Qub
Γ(κ − κ )
More simplified expression for the MGF can be obtained
l=1;l6=x 4b,l 4b,x
× Qrb − F̃γM (γ). (19) if the expansion of the Meijer’s G-function were utilized in a
κ4b,x l=1 Γ(κ3b,l − κ4b,x )
2) Probability Density Function: Differentiating (18) with
respect to γ, a closed-form expression of the PDF of γm is

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

similar way to find the MGF of the maximum 2) Average Bit Error Rate: Using the definition of the
ua  κ4a,k average BER [43, Eq. (12)]
X Ca
M̃γm (s) u A3a va Γ(va κ4a,k ) qp
Z ∞
µri ,i >>1 (sµra ,a )va P = exp(−qγ) γ p−1 Fγ (γ) dγ (28)
k=1
Qua ub  κ4b,x 2Γ(p) 0
l=1;l6=k Γ(κ4a,l − κ4a,k ) Cb
X
× Qra + A3b vb and placing (4) into (28) then utilizing the identity [42, Eq.
l=1 Γ(κ3a,l − κ4a,k ) x=1
(sµrb ,b )vb (2.2)], we obtain P SC as
Qub
Γ(vb κ4b,x ) l=1;l6=x Γ(κ4b,l − κ4b,x ) A3a A3b
× Qrb − M̃γM (s). (23) P SC =
l=1 Γ(κ3b,l − κ4b,x ) 2Γ(p)va vb
   
The MGF for Gamma-Gamma model can be obtained by 0,1
  (1−p;1,1)
setting the same parameters listed in Sec. II 
 1,0 (−;−,−) Ca1/va 

     
−1
qµra ,a 
ua ,1 (κ5a ,[va ])
∗ 3ra ,1 Ca  1,κ3a × H , (29)
  
Mγm (s) = A3a Gra +1,3ra sµr ,a 
 
−1
κ4a ra +1,ua +1 (κ6a ,[va ]) 1/v
a Cb b
 
(κ5b ,[vb−1 ])
   
qµrb ,b
    ub ,1 
Cb  1,κ3b
+ A3b G3r b ,1

− A3a A3b (κ6b ,[vb−1 ])
 
rb +1,3rb sµr ,b  κ rb +1,ub +1
b 4b
    
1  κ5a  κ5b  Ca Cb where p and q indicate different modulation schemes param-
× G1,0:3r a ,1:3rb ,1 
1,0:ra +1,3ra +1:rb +1,3rb +1 −  κ  κ 
  sµ ,
ra ,a sµrb ,b
.
6a 6b eters [40, Table I].
(24) An asymptotic expression of P at high SNR can be derived
as
in agreement with [4, Eq. (15)]. Z ∞
4) Moments: Substituting (20) in the definition of the qp
moments (13) and utilizing [46, Eq. (2.25.2.1)] and Eq. (14), P u exp(−qγ) γ p−1 F̃ (γ) dγ. (30)
µri ,i >>1 2Γ(p) 0
we obtain the moments as
 n   !n Solving the integration leads to the following:
n µra ,a κ4a + vn µrb ,b
E[γm ] = A3a Γ a + A3b ua Xub  κ4a,k !κ4b,x
1/va κ3a + vn 1/vb
A3a A3b X Ca Cb
Ca a Cb
P SC u
µri ,i >>1 2Γ(p) µvraa,a µvrbb,b
 n

κ4b + v n
×Γ b − E[γM ] (25) k=1 x=1
κ3b + vn Qua Qub
l=1;l6=k Γ(κ4a,l − κ4a,k ) =x Γ(κ4b,l − κ4b,x )
b
  m n × Qra Ql=1;l6
rb
x Q Q Γ(κ3a,l − κ4a,k ) Γ(κ 3b,l − κ4b,k )
where Γ = Γ(xi )/ Γ(yi ) such that m and n are l=1 l=1
y i=1 i=1 Γ(va κ4a,k + vb κ4b,x + p) q −(va κ4a,k +vb κ4b,x )
the lengths of x and y, respectively. × . (31)
In the Gamma-Gamma case, (25) simplifies to κ4a,k κ4b,x

n ∗

µra ,a
n 
κ4a +n
 Diversity and coding gain can be extracted from (31) if the
E[γm ] = A3a 1/va
Γ expression consists of only one term. In particular when the
κ3a +n
!Cna   BER is the form P ≈ (Gc µ)−Gd [47] where Gc and Gd are
µrb ,b κ4b +n
+ A3b 1/v
Γ the diversity and coding gain respectively, and µ is the average
κ3b +n
Cb b SNR. Upon our observation in other study [37], this is possible
n
ξ2
   
− A3a A3b
µra ,a 3rb +1,3ra Cb µra ,a  κ7a ,κ3b
G3ra +rb +1,3rb +ra +1 Ca µr ,b 

if the absolute difference between { viiri , rβi1i β2i
σi , ri λi } ∈ κ4i is
Ca b κ4b ,κ8a
significance. If so, then the summation in (31) reduces toa
n 
ξ2
  
µrb ,b C µ
single dominant term results from the min viiri , rβi1i β2i
σi , ri λi .
3r +1,3rb a r ,b  κ7b ,κ3a

+ G3rba+ra +1,3r b
a +rb +1 Cb µra ,a 
. (26)
Cb κ4a ,κ8b
In this case, the diversity and coding gain can be written as
as found in [4, Eq. (16)].
Gd = va κ4a,k + vb κ4b,x ,
A3a A3b Ca κ4a,k Cb κ4b,x Γ(va κ4a,k + vb κ4b,x + p)

IV. A PPLICATIONS Gc = q
A. Performance of Dual-Branch SC Systems 2Γ(p)κ4a,k κ4b,x
Qua Qub !−1/Gd
Based on our results in Section III-A, we now study the l=1;l6=k Γ(κ4a,l − κ4a,k ) l=1;l6=x Γ(κ4b,l − κ4b,x )
× Qra Qrb ,
performance of the selection combining scheme over dual- l=1 Γ(κ3a,l − κ4a,k ) l=1 Γ(κ3b,l − κ4b,k )
branch FSO links. (32)
1) Outage Probability: The outage probability Pout is
defined as follows: where k ∈ {1, ra + 1, ra + ra σa + 1} and x ∈ {1, rb + 1, rb +
rb σb + 1}.
Pout = Pr[γ < γth ] = Fγ (γth ). (27)
Accordingly, the outage probability for the maximum SNR
of two links is obtained by substituting (4) into (27).

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

For the Gamma-Gamma case, we have 2) Average Bit Error Rate: P DH is obtained by placing (18)
into (28), and utilizing (29) and [43, 2.24.3.2] yielding
∗ A3a A3b
P SC = p−1/2
2Γ(p) A3a va
     P DH = va −1
1,0:3ra ,1:3rb ,1 p  κ5a  κ5b  Ca Cb 2Γ(p)(2π) 2
× G1,0:ra +1,3ra +1:rb +1,3rb +1  qµra ,a , qµr ,b ,
     va  
− κ6a  κ6b  b ∆(va ,1−p),κ5a
× Guraa+v
,va +1
a +1,ua +1
C a
va 
(33) qµr ,a a
 κ6a
p−1/2  vb  
in agreement with [4, Eq. (19)] A3b vb ub ,vb +1

vb  ∆(vb ,1−p),κ5b
+ vb −1 Grb +vb +1,ub +1 Cb qµr ,b κ6b
3) Ergodic Capacity: The ergodic capacity in FSO systems 2Γ(p)(2π) 2 b

is defined as − P SC . (37)
Z ∞
C , E[log2 (1 + cγ)] = log2 (1 + cγ)fγ (γ) dγ, (34) Similar to the SC, we can obtain an asymptotic expression of
0
the BER via Meijer’s G-function expansion by solving (30) as
where c = 1 for heterodyne detection r = 1 and c = e/2π for follows:
IM/DD r = 2. It is very important to note that the expression ua  κ4a,k
in (34) is exact for r = 1 while it is a lower bound for the case A3a X Ca
P DH u
of r = 2 [48, Eq. (26)] [49, Eq. (7.43)]. log2 (1 + cγ) is ex- µri ,i >>1 2Γ(p) (qµra ,a )va
pressed as ln(1+γ)
ln(2) . C SC is obtained by using (34) along with Quk=1
a
Γ(va κ4a,k + p) l=1;l6 =k Γ(κ4a,l − κ4a,k )
     
the identities ln(1 + x) = G1,2 x  1,1 = H1,2 x  (1,1),(1,1)
  × Qra
2,2  1,0 2,2  (1,1),(0,1) κ4a,k l=1 Γ(κ3a,l − κ4a,k )
ub  κ4b,x
and [42, Eq. (2.3)] yielding A3b X Cb
+
A3a A3b 2Γ(p) x=1 qµrb ,b vb
C SC =
va vb ln(2) Qub
  Γ(vb κ4b,x + p) l=1;l6 =x Γ(κ4b,l − κ4b,x )
× − P SC . (38)
  
0,ub
(1−κ4b ;[vb−1 ],[vb−1 ])
Qrb

ub ,rb
 cµr ,b
b κ4b,x l=1 Γ(κ3b,l − κ4b,x )
(1−κ3b ;[vb−1 ],[vb−1 ]) 1/vb
  
    Cb 

×v H
 1,2 (1,1;1,1)  In this case, diversity or coding can not be extracted due
b
   
  2,2 (1,0;1,1) Ca
1/va
µr ,b
 to the fact that the expression consist of 3 terms and can not
    −1 b 
(κ5a ,[va ])
ua ,1 1/vb
be reduced to the form of P ≈ (Gc .µ)−Gd . For the case of
  
−1
Cb µra ,a
(κ6a ,[va ])
 
ra +1,ua +1
   
Gamma-Gamma, we have
0,ua −1 −1
  
  (1−κ4a ;[va ],[va ]) cµra ,a ∗ A3a 3ra ,2 Ca  1−p,κ5a

 ua ,ra
(1−κ ;[v −1
],[v −1
a ]) 1/va
 P DH = G



1,2
 3a a
(1,1;1,1)
Ca 
 2Γ(p) ra +2,3a +1 qµra ,a  κ6a
+ va H   
1/vb
 .
A3b 3rb ,2
 
 1−p,κ5b

∗
2,2 (1,0;1,1) Cb µra ,a
Grb +2,rb ub +1 qµCrb ,b 
 
  
(κ5b ,[vb−1 ])
 + − P SC , (39)
 ub ,1 Ca
1/va
µr ,b
b
 2Γ(p) b κ6b
(κ6b ,[vb−1 ])
 
rb +1,ub +1
(35)
which agrees with [4, Eq. (22)]
  Using
3) Ergodic Capacity:  (34) along with the identities
For the Gamma-Gamma case, we have 1,2
 1,1
ln(1 + x) = G2,2 x  , [50, Eq. (12)], and [46, Eq.
∗ A3a A3b 1,0
C SC =
ln(2) (07.34.21.0011.01)], we obtain C DH in closed- form as
        
3rb ,0:3ra ,1:1,2 κ4b  κ5a  1,1  Ca µr ,b cµr ,b A3b  κ9b ,κ3b
× G3rb ,rb :ra +1,3ra +1:2,2    b
, Cbb
C DH = G ub +2vb ,vb Cb
κ3b  κ6a  1,0  Cb µra ,a v

     ln(2)(2π)vb −1 rb +2vb ,ub +2vb (cµrb ,b ) b  κ4b ,κ10b
3r ,0:3r ,1:1,2 κ  κ  1,1  Bb µra ,a cµra ,a   
+ G3raa ,ra :rbb +1,3rb +1:2,2 4a  5b   B µ , B . A3a ua +2va ,va Ca
 κ9a ,κ3a
κ3a  a r ,b
κ6b 1,0 a b + G 
ln(2)(2π)va −1 ra +2va ,ua +2va (cµra ,a ) a  κ4a ,κ10a
v
(36)
in agreement with [4, Eq. (21)]. − C SC , (40)
where κ9i = ∆(vi , 0), ∆(vi , 1) and κ10i = ∆(vi , 0), ∆(vi , 0).
B. Performance of Dual-Hop Relay Systems For the special case of Gamma-Gamma, we have
  
In this section, we present the performance analysis of a ∗ A3b 3rb +2,1  0,1,κ3b
C DH = Grb +2,3rb +2 cµCrb ,b 
dual-hop (DH) FSO variable gain relay transmission systems ln(2) b κ4b ,0,0
whose end-to-end SNR is known to be given by γDH = A3a 3ra +2,1
 
0,1,κ

∗
Ca 
 3a
(γ1 γ2 )/(1 + γ1 + γ2 ) and which is typically approximated by + Gra +2,3ra +2 cµr ,a  − C SC (41)
ln(2) a κ 4a ,0,0
γDH ' min(γ1 , γ2 ) [32]. Thus, results obtained in Section III-B
facilitate the following analysis. in agreement with [4, Eq. (23)].
1) Outage Probability: The outage probability for the min-
imum SNR of two links is obtained by substituting (18) into C. Two Data Streams Multiplexing
(27). Assuming two FSO links that transmit independent data to
the same destination. The first link requires high data rate

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

D1 compared to the second FSO link with a lower data rate single FSO link and dual-branch SC of FSO/FSO and RF/FSO.
D2 . An initial thought is to use the results obtained from We can notice instantly that SC helps to improve the perfor-
the maximum and the minimum in Sections IV-A and IV-B, mance comparing to the single link with expense of utilizing
respectively. The link providing the maximum capacity C M more hardware at the receive side. Also, dual-branch RF/FSO
can be identified and D1 data rate can be assigned to it. SC performance better than symmetric branches FSO/FSO.
Similarly, the link providing the minimum capacity C m can be This encourages the integration of RF into FSO systems or
identified and D2 data rate can be assigned to it. In this work, vice verse. Regarding the asymptotic results, they converge
we try to accommodate each link’s need and at the same time fast to exact ones even for low SNR (i.e starting at 30 dB).
try to utilize the network resources efficiently and intelligently. Interestingly, the asymptotic results representing of all terms in
We introduce the parameter  ∈ [0, 1] which consists of the the summation matches perfectly the exact results which give
percentage of time the first data stream is assigned to the our expression more value instead of using special functions.
channel with the best SNR. Thus, D1 and D2 can be given as In the same fashion, two dominants terms asymptotic results
provide better match than single dominant term.
D1 =  C M + (1 − ) C m , (42)
D2 = (1 − ) C M +  C m . (43)

10 0
V. R ESULTS AND D ISCUSSION
FSO
In this section, we verify and validate our analytical results
10 -2

Average Bit Error Rate (BER)

of hybrid FSO communication systems, assuming IM/DD r = a
2, using computer-based simulations. We also consider RF link -4 b
10
experiencing the same channel fading model. a FSO/FSO

-6
10 b
RF/FSO
A. Dual-Branch SC Systems a

First, we investigate the BER performance of dual-branch 10 -8

Exact analytical result: Eq. (29)
Exact result by Monte-Carlo Simulation
FSO/FSO (ra = rb ∈ {1, 2}) and RF/FSO (ra = 1 and Asymptotic Result (All Terms): Eq. (31) b
rb ∈ {1, 2}) SC systems under differentially coherent binary 10 -10
Asymptotic Result (Two Terms)
Asymptotic Result (Single Term)
phase shift keying (DPSK) modulation in which p = 1 and
0 10 20 30 40 50 60 70
q = 1. In this work, we consider two scenarios of atmospheric Average Signal-to-Noise Ratio (SNR) per branch (dB)
turbulence conditions, strong (consider the following set of
Figure 4: The effect of turbulence conditions ((a) Strong turbulence and (b)
parameters: α1,i = 1.8621, α2,i = 1, β1,i = 0.5, β2,i = 1.8, Moderate conditions) on the average BER of DPSK over single FSO link,
Ω1,i = 1.5074, and Ω2,i = 0.9280 such that λi = 17 dual-branch FSO/FSO and RF/FSO SC with severe pointing error, ξ = 1.2
and σi = 9) and moderate (consider the following set of
parameters:α1,i = 2.1690, α2,i = 1, β1,i = 0.55, β2,i = 2.35, Moreover, Fig. 4 shows the impact of the turbulence condi-
Ω1,i = 1.5793, and Ω2,i = 0.9671 where λi = 28 and tions on the performance of FSO systems. Clearly, they lead
σi = 13). to some degradation. Again, the asymptotic results including
all terms in the summation provide an excellent match with
the exact result. It is noticeable that for strong conditions
10 0 the asymptotic expressions including single dominant terms is
better than for the moderate conditions and that is due to the
ξ2
fact that the differences between { viiri , rβi1i β2i
σi , ri λi }, i ∈ {a, b},
10 -2
for moderate conditions is not significance.
Average Bit Error Rate (BER)

a
Secondly, we evaluate another performance metric that is
10 -4
the ergodic capacity of dual-branch SC FSO/FSO and RF/FSO
b
SC as shown in Fig. 5. Generally, SC systems provides higher
data rate than a single link system. Also, we can notice that
10 -6 c behavior of RF/FSO and FSO/FSO capacity are not in big
Exact analytical result : Eq. (29)
Exact result by Monte-Carlo simulation
difference in contrast with BER.
Asymptotic Result (All Terms): Eq. (31)
10 -8 Asymptotic Result (Two Terms)
Asymptotic Result (Single Term)
B. Dual-Hop Transmission Systems
0 10 20 30 40 50 60 70 We evaluate BER under DPSK modulation and ergodic
Average Signal-to-Noise Ratio (SNR) per branch (dB) capacity of FSO relay systems including dual-hop RF-FSO
Figure 3: The impact of pointing error (ξ = 1.2) on the average BER of and FSO-FSO links as shown in Figs. 6 and 7. We had to
DPSK over (a) single FSO link, (b) dual-branch FSO/FSO and (c) RF/FSO compare the performance of a FSO link that suffers strong
SC in strong turbulence conditions. turbulence conditions in addition to severe pointing errors with
relay assisted link of RF-FSO and FSO-FSO experiencing
First, Fig. 3 shows the impact of pointing errors (i.e strong moderate turbulence conditions with ξ >> 1. Clearly, relay
effect of pointing errors as ξ → 0) on the performance of links outperform the single link in addition to providing higher

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

for low SNR values (i.e starting at 20 dB).

16

FSO
14 FSO/FSO C. Two Data Streams Multiplexing
RF/FSO
Fig. 8 shows the ergodic capacity of our proposed scheme
Ergodic Capacity (Bits/Sec/Hz)

12 Simulation
with different values of . As observed, as  increases, D1
10 increases while D2 decreases and vice versa. By this model,
8
D1 can get data rate that is close to the maximum capacity
the system can provide and at the same time D2 gets higher
6 data rate than expected.
4

16
2
D1
14
0 D2
ε=1
0 10 20 30 40 50
Average Signal−to−Noise Ratio (SNR) per branch (dB)

Ergodic Capacity (Bits/Sec/Hz)

12
ε = 0.75
Figure 5: Ergodic capacity of single FSO link, dual-branch FSO/FSO, RF/FSO 10
SC in strong turbulence conditions and under severe pointing errors, ξ = 1.2. ε = 0.5
8

6
10 0

2
10 -1
Average Bit Error Rate (BER)

0
0 5 10 15 20 25 30 35 40 45 50
Average Signal-to-Noise Ratio (SNR) per branch (dB)
10 -2
Figure 8: Capacity Multiplexing of two i.i.d FSO link with ξ = 6.5 in strong
a
turbulence conditions

b
10 -3 Exact analytical result: Eq. (37)
Exact result by Monte-Carlo simulation c
Asymptotic Result (All Terms): Eq. (38) VI. C ONCLUSION
Asymptotic Result (3 Terms)
Closed-form expressions for the CDF, the PDF, the MGF,
10 -4
0 20 40 60 and the moments of the maximum and the minimum of two
Average Signal-to-Noise Ratio (SNR) per hop (dB)
modified double generalized gamma variates were derived.
Figure 6: Average BER of DPSK over (a) single FSO link (b) dual-hop FSO- Furthermore, we derived analytical expressions for average bit
FSO (c) RF-FSO. error rate and ergodic capacity and then applied them to find
the performance of both dual-branch hybrid FSO SC scheme
and dual-hop hybrid FSO relay system. Monte-Carlo computer
14
FSO simulations were carried out to validate our results. Finally,
12
FSO−FSO based on our results for the maximum and minimum, we
RF−FSO
Simulation
have proposed a data streams multiplexing scheme for two
Ergodic Capacity (Bits/Sec/Hz)

10 independent FSO links that compromises each link’s data rate

judiciously in order to utilize the network available capacity
8
in a flexible fashion.
6
R EFERENCES
4 [1] M. A. Kashani, M. Uysal, and M. Kavehrad, “A novel statistical channel
model for turbulence-induced fading in free-space optical systems,”
2
IEEE/OSA J. Lightwave Technol., vol. 33, no. 11, pp. 2303–2312, March
2015.
[2] A. Farid and S. Hranilovic, “Outage capacity optimization for free-space
0 optical links with pointing errors,” Journal of Lightwave Technology,
0 5 10 15 20 25 30 35 40 45 50
Average Signal−to−Noise Ratio (SNR) per hop (dB) vol. 25, no. 7, pp. 1702–1710, July 2007.
[3] H. AlQuwaiee, I. S. Ansari, and M.-S. Alouini, “On the performance of
Figure 7: Ergodic capacity of single FSO link and dual-hop FSO-FSO and free space optical wireless communication systems over double general-
RF-FSO. ized Gamma fading channel,” in Proceedings of 4th IEEE International
Conference on Communications and Networking (COMNET’ 2014),
Hammamet, Tunisia, Mar. 2014.
[4] ——, “On the maximum and minimum of two modified Gamma-
data rate than a single FSO link. Moreover, the asymptotic Gamma variates with applications,” in Proceedings of IEEE Wireless
analysis shows an excellent match with the exact results even Communications and Networking Conference (WCNC’ 2014), Istanbul,
Turkey, Apr. 2014, pp. 269–274.

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

[5] W. O. Popoola and Z. Ghassemlooy, “BPSK subcarrier intensity [26] T. Tsiftsis, H. Sandalidis, G. Karagiannidis, and N. Sagias, “Multihop
modulated free-space optical communications in atmospheric free-space optical communications over strong turbulence channels,”
turbulence,” IEEE/OSA Journal of Lightwave Technology, vol. 27, in Proceedings of IEEE International Conference on Communications,
no. 8, pp. 967–973, Apr. 2009. (ICC ’06), Istanbul, Turkey, June 2006, pp. 2755–2759.
[6] J. Park, E. Lee, and G. Yoon, “Average bit-error rate of the Alamouti [27] G. Karagiannidis, T. Tsiftsis, and H. Sandalidis, “Outage probability of
scheme in Gamma-Gamma fading channels,” IEEE Photonics Technol- relayed free space optical communication systems,” Electronics Letters,
ogy Letters, vol. 23, no. 4, pp. 269–271, Feb. 2011. vol. 42, no. 17, pp. 994–995, August 2006.
[7] D. Kedar and S. Arnon, “Urban optical wireless communication net- [28] M. Safari and M. Uysal, “Relay-assisted free-space optical communica-
works: the main challenges and possible solutions,” IEEE Communica- tion,” IEEE Transactions on Wireless Communications, vol. 7, no. 12,
tions Magazine, vol. 42, no. 5, pp. S2–S7, May 2004. pp. 5441–5449, December 2008.
[8] S. Arnon, J. Barry, G. Karagiannidis, R. Schober, and M. Uysal, [29] E. Lee, J. Park, D. Han, and G. Yoon, “Performance analysis of the
Advanced Optical Wireless Communication Systems. Cambridge Uni- asymmetric dual-hop relay transmission with mixed RF/FSO links,”
versity Press, 2012. IEEE Photonics Technology Letters,, vol. 23, no. 21, pp. 1642–1644,
[9] M. Uysal, J. Li, and M. Yu, “Error rate performance analysis of coded Nov. 2011.
free-space optical links over Gamma-Gamma atmospheric turbulence [30] I. S. Ansari, M.-S. Alouini, and F. Yilmaz, “On the performance of
channels,” IEEE Transactions on Wireless Communications, vol. 5, no. 6, hybrid RF and RF/FSO fixed gain dual-hop transmission systems,” in
pp. 1229–1233, Jun. 2006. Proceedings of 2013 Saudi International Electronics, Communications
[10] H. Sandalidis, T. Tsiftsis, G. Karagiannidis, and M. Uysal, “BER and Photonics Conference (SIECPC), Riyadh, SA, April 2013, pp. 1–6.
performance of FSO links over strong atmospheric turbulence channels [31] H. Samimi and M. Uysal, “End-to-end performance of mixed RF/FSO
with pointing errors,” IEEE Communications Letters, vol. 12, no. 1, pp. transmission systems,” IEEE/OSA Journal of Optical Communications
44–46, Jan. 2008. and Networking, vol. 5, no. 11, pp. 1139–1144, Nov 2013.
[11] H. Sandalidis, T. Tsiftsis, and G. Karagiannidis, “Optical wireless [32] S. Ikki and M. Ahmed, “Performance analysis of dual-hop relaying
communications with heterodyne detection over turbulence channels communications over generalized Gamma fading channels,” in Global
with pointing errors,” IEEE/OSA Journal of Lightwave Technology, Telecommunications Conference (GLOBECOM ’07), Nov. 2007, pp.
vol. 27, no. 20, pp. 4440–4445, Oct. 2009. 3888–3893.
[12] X. Tang, Z. Ghassemlooy, W. O. Popoola, and C. Lee, “Coherent [33] M. Hasna and M.-S. Alouini, “End-to-end performance of transmission
polarization shift keying modulated free space optical links over a systems with relays over Rayleigh-fading channels,” IEEE Transactions
Gamma-Gamma turbulence channel,” American Journal of Engineering on Wireless Communications, vol. 2, no. 6, pp. 1126–1131, Nov 2003.
and Applied Sciences, vol. 4, no. 4, pp. 520–530, 2011. [34] I. Ansari, F. Yilmaz, and M.-S. Alouini, “On the performance of mixed
[13] M. K. Simon and M.-S. Alouini, Digital communication over fading RF/FSO variable gain dual-hop transmission systems with pointing
channels. New York: Wiley, 2000. errors,” in Proceedings of 78th IEEE Vehicular Technology Conference
[14] H.-C. Yang, “New results on ordered statistics and analysis of minimum- (VTC Fall), Las Vegas, USA, Sept 2013, pp. 1–5.
selection generalized selection combining (GSC),” IEEE Transactions on [35] M. Aggarwal, P. Garg, and P. Puri, “Dual-hop optical wireless relaying
Wireless Communications, vol. 5, no. 7, pp. 1876–1885, July 2006. over turbulence channels with pointing error impairments,” IEEE/OSA
[15] Y.-C. Ko, M.-S. Alouini, and M. K. Simon, “Analysis and optimization Journal of Lightwave Technology, vol. 32, no. 9, pp. 1821–1828, May
of switched diversity systems,” IEEE Transactions on Vehicular Tech- 2014.
nology, vol. 49, no. 5, pp. 1813–1831, Sep. 2000. [36] W. Gappmair, “Further results on the capacity of free-space optical
[16] M. Ibrahim and A. Ibrahim, “Performance analysis of optical receivers channels in turbulent atmosphere,” IET Communications, vol. 5, no. 9,
with space diversity reception,” IEEE Transactions on Communications, pp. 1262–1267, Jun. 2011.
vol. 134, no. 6, pp. 369–372, Dec. 1996. [37] H. AlQuwaiee, I. S. Ansari, and M.-S. Alouini, “On the performance
[17] E. J. Shin and V. W. Chan, “Optical communication over the turbulent of free-space optical communication systems over double generalized
atmospheric channel using spatial diversity,” in Proceedings of IEEE gamma channel,” To appear in IEEE Journal on Selected Areas in Com-
Global Telecommunications Conference, 2002 (GLOBECOM’02), vol. 3, munications - 2014 Special Issue on Optical Wireless Communication.
Nov. 2002, pp. 2055–2060. [38] M. Al-Habash, L. C. Andrews, and R. L. Phillips, “Mathematical
[18] A. Ibrahim, M.M.; Ibrahim, “Part 1: Optical communication over the model for the irradiance probability density function of a laser beam
clear turbulent atmospheric channel using diversity,” IEEE Journal on propagating through turbulent media,” Optical Engineering, vol. 40,
Selected Areas in Communications, vol. 22, no. 9, pp. 1896–1906, Nov. no. 8, pp. 1554–1562, Feb. 2001.
2004. [39] I. S. Ansari, F. Yilmaz, and M.-S. Alouini, “A unified performance of
[19] H. Moradi, M. Falahpour, H. Refai, and P. LoPresti, “A diversity free-space optical links over Gamma-Gamma turbulence channels with
combining approach for MIMO FSO nodes with misaligned receivers,” pointing errors,” in Proceedings of IEEE 81st Vehicular Technology
in Proceedings of Global Telecommunications Conference (GLOBE- Conference (VTC Spring’ 2015), Glasgow, Scotland, May 2015.
COM’11), Houston, TX, USA, Apr. 2011, pp. 1–5. [40] ——, “Impact of pointing errors on the performance of mixed RF/FSO
[20] H. Kazemi and M. Uysal, “Performance analysis of MIMO free-space dual-hop transmission systems,” IEEE Wireless Communications Letters,
optical communication systems with selection combining,” in Proceed- vol. 2, no. 3, pp. 351–354, Jun. 2013.
ings of the 21st Signal Processing and Communications Applications [41] F. Yilmaz and M.-S. Alouini, “Product of the powers of generalized
Conference (SIU), Haspolat, Cyprus, April. 2013, pp. 1–4. Nakagami-m variates and performance of cascaded fading channels,” in
[21] T. Tsiftsis, H. Sandalidis, G. Karagiannidis, and M. Uysal, “Optical Proceedings of IEEE Global Telecommunications Conference, (GLOBE-
wireless links with spatial diversity over strong atmospheric turbulence COM’09), Honolulu, USA, Nov 2009, pp. 1–8.
channels,” IEEE Transactions on Communications, vol. 8, no. 2, pp. [42] P. Mittal and K. Gupta, “An integral involving generalized function
951–957, Feb. 2009. of two variables,” Proceedings of the Indian Academy of Sciences -
[22] M. O. Hasna and M.-S. Alouini, “A performance study of dual-hop trans- Section A, vol. 75, no. 3, pp. 117–123, Mar. 1972.
missions with fixed gain relays,” in Proceedings of IEEE International [43] I. S. Ansari, S. Al-Ahmadi, F. Yilmaz, M.-S. Alouini, and
Conference on Acoustics, Speech, and Signal Processing (ICASSP ’03), H. Yanikomeroglu, “A new formula for the BER of binary modula-
vol. 4, Apr. 2003, pp. IV–189–92. tions with dual-branch selection over generalized-K composite fading
[23] F. Yilmaz, O. Kucur, and M.-S. Alouini, “A novel framework on channels,” IEEE Transactions on Communications, vol. 59, no. 10, pp.
exact average symbol error probabilities of multihop transmission over 2654–2658, Oct. 2011.
amplify-and-forward relay fading channels,” in Proceedings of the 7th [44] K. Peppas, “A new formula for the average bit error probability of dual-
International Symposium on Wireless Communication Systems (ISWCS), hop amplify-and-forward relaying systems over generalized shadowed
York, UK, Sep. 2010, pp. 546–550. fading channels,” Wireless Communications Letters, IEEE, vol. 1, no. 2,
[24] Y. Zhu, Y. Xin, and P.-Y. Kam, “Outage probability of Rician fading pp. 85–88, April 2012.
relay channels,” in Proceedings of Military Communications Conference [45] K. Gupta and P. Mittal, “The H-function transform II,” Journal of the
(MILCOM), Washington, DC, Sep. 2006, pp. 1–6. Australian Mathematical Society, vol. 12, no. 04, pp. 444–450, Nov.
[25] S. N. Datta, S. Chakrabarti, and R. Roy, “Error analysis of noncoherent 1971.
FSK with variable gain relaying in dual-hop Nakagami-m relay fading [46] I. Wolfram, Mathematica Edition: Version 8.0. Champaign, Illinois:
channel,” in Proceedings of International Conference on Signal Process- Wolfram Research Inc., 2010.
ing and Communications (SPCOM’ 2010), Bangalore, India, July 2010,
pp. 1–5.

0018-9545 (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/TVT.2016.2515662, IEEE
Transactions on Vehicular Technology

10

[47] Z. Wang and G. Giannakis, “A simple and general parameterization Mohamed-Slim Alouini (S’94, M’98, SM’03, F’09)
quantifying performance in fading channels,” IEEE Transactions on was born in Tunis, Tunisia. He received the Ph.D.
Communications, vol. 51, no. 8, pp. 1389–1398, Aug 2003. degree in Electrical Engineering from the California
[48] A. Lapidoth, S. Moser, and M. Wigger, “On the capacity of free-space Institute of Technology (Caltech), Pasadena, CA,
optical intensity channels,” IEEE Transactions on Information Theory, USA, in 1998. He served as a faculty member
vol. 55, no. 10, pp. 4449–4461, Oct 2009. in the University of Minnesota, Minneapolis, MN,
[49] S. Arnon, J. Barry, G. Karagiannidis, R. Schober, and M. Uysal, Ad- USA, then in the Texas A&M University at Qatar,
vanced optical wireless communication systems. Cambridge University Education City, Doha, Qatar before joining King
Press, 2012. Abdullah University of Science and Technology
[50] S. Gupta, “Integrals involving products of G-function,” Proceedings of (KAUST), Thuwal, Makkah Province, Saudi Arabia
the National Academy of Sciences, India Section A: Physical Sciences, as a Professor of Electrical Engineering in 2009.
vol. 39, no. 2, pp. 193–200, 1969. His current research interests include the modeling, design, and performance
analysis of wireless communication systems.

Hessa AlQuwaiee (S’09) received the B.S degree

in Computer Engineering from Prince Mohamed
University (PMU), Alkhobar, Saudi Arabia in 2011.
In 2008, she was awarded the discovery scholar-
ship to complete graduate studies in King Abdullah
University of Science and Technology (KAUST).
In 2013, she earned her M.S. degree in Electrical
Engineering. She is now toward her PhD degree at
KAUST. Her current research interests include, but
not limited to, channel characterization and perfor-
mance analysis of optical wireless communications,
body-to-body communications, diversity combining techniques and MIMO
systems.

Imran Shafique Ansari (S’07, M’15) was born in

India in 1987. He completed his B.Sc. in Com-
puter Engineering from King Fahd University of
Petroleum and Minerals (KFUPM) in 2009 (with
First Honors) and M.Sc. and PhD from King Abdul-
lah University of Science and Technology (KAUST)
in 2010 and 2015, respectively. Currently, he is a
Postdoctoral Research Associate (PRA) with Texas
A&M University at Qatar (TAMUQ). From May
2009 through Aug. 2009, he was a visiting scholar
with Michigan State University (MSU), East lansing,
MI, USA, and from Jun. 2010 through Aug. 2010, he was a research intern
with Carleton University, Ottawa, ON, Canada. He has been a student member
of IEEE and IET since 2007 and has served in various capacities. He has
served on IET CC-EMEA (Communities Committee-Europe, Middle-East
and Africa) for a complete term from Oct. 2010-Sep. 2013 and has been
re-elected to serve for another term from Oct. 2015-Sep. 2018. He is an
active reviewer for various IEEE Transactions and various other journals.
He has served as a TPC for various IEEE conferences. He is a recipient of
recognized reviewer certificate by Elsevier Optics Communications in 2015,
a recipient of recognized reviewer certificate by OSA Publishing in 2014,
a recipient of appreciation for an exemplary reviewer for IEEE Wireless
Communications Letters (WCL) in 2014, a recipient of post-doctoral research
award (PDRA) (first cycle) with Qatar national research foundation (QNRF)
in 2014, a recipient of KAUST academic excellence award (AEA) in 2014,
and a recipient of IEEE Richard E. Merwin student scholarship award in
Jul. 2013. Imran has authored/co-authored around 34 journal and conference
publications. His current research interests include free-space optics (FSO),
channel modeling/signal propagation issues, relay/multihop communications,
physical layer secrecy issues, and diversity reception techniques among others.

0018-9545 (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.