Vous êtes sur la page 1sur 17

This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Available online at www.sciencedirect.com
Journal of the Franklin Institute 349 (2012) 845860
Moments generating function of the harmonic mean
of two non-identical gamma random variables and its
applications in wireless communications
Kostas Peppas
n
National Center of Scientic Research Demokritos, Wireless Communications Laboratory, 15310 Athens, Greece
Received 19 May 2011; received in revised form 2 June 2011; accepted 18 November 2011
Available online 6 December 2011
Abstract
In this paper, a closed-form expression for the moments generating function of the half-harmonic
mean of two independent, not necessarily identically distributed gamma random variables with
arbitrary parameters is presented. This statistical result is useful to the performance analysis of dual-
hop wireless communication systems with amplify-and-forward relays in a Nakagami-m fading
environment. The proposed mathematical analysis is substantiated by various numerically evaluated
and computer simulation results.
& 2011 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
1. Introduction
The performance of wireless communication systems is highly susceptible to fading
conditions, due either to multi-path wave propagation or to shadowing from physical or
man-made obstacles that affect wave propagation. Such phenomena result in variations of the
channel strength over both time and frequency, thus impairing the performance of wireless
receivers. Of particular interest is the performance analysis of digital communications systems
operating over such environments. The mathematical methodologies used to address this
problem, rely on different statistical distributions of the signal amplitude or power variations.
Among the statistical properties of fading distribution functions, the moments generating
function (MGF) plays an important role in the performance evaluation of wireless
www.elsevier.com/locate/jfranklin
0016-0032/$32.00 & 2011 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.jfranklin.2011.11.012
n
Tel.: 30 2106503208.
E-mail addresses: kpeppas@iit.demokritos.gr, kppeppas@gmail.com.
Author's personal copy
communication systems. More specically, in Simon and Alouini [30] it is shown that the
so-called MGF-based approach signicantly simplies the mathematical analysis required for
the evaluation of important performance metrics, such as the average bit error probability
(ABEP) and allows unication under a common framework in a large variety of digital
communications systems, covering all known modulation techniques and fading models.
Diversity reception is an efcient technique to mitigate the deleterious effects of multi-path
and shadow fading and improve the performance of a wireless communication system
[30,3,20,6]. A signicant breakthrough in the wireless transmission technologies are the
so-called multi-hop systems. Such systems have become an important topic in the research
community due to their ability to extend the coverage without using large power at the
transmitter and their possible application in cellular, ad hoc networks and military
communications [28,29,17,16,5,35,25,4]. A special class of multi-hop communication systems
are the collaborative/cooperative ones [17,16] where relaying is used as a form of spatial
diversity to overcome highly shadowed or deeply faded links. In a cooperative diversity
network, the destination terminal combines the signals received from the source as well as
from relay terminals. As a result, the destination terminal can be beneted from the multiple
signal replicas available, by employing a variety of diversity techniques, such as maximal
ratio combining (MRC), equal gain combining (EGC) or selection combining (SC).
Various protocols have been proposed to achieve the benets of multi-hop transmission.
One of them is the amplify-and-forward (AF) which uses relays that amplify the received
signal from the source and then retransmit towards the destination. Therefore, at the relay
no decoding is performed. Another popular protocol is the so-called decode-and-forward
[19] which uses relays that decode the signal and then transmit the detected version to the
next node. In the open technical literature there are several works that have studied the
performance of AF relaying systems. Representative examples can be found in Hasna and
Alouini [1214], Anghel and Kaveh [2], Renzo et al. [24], Peppas et al. [21], Maham and
Hjorungnes [18], Yang and Chen [32], Zhao et al. [35], Zhang et al. [34], Senaratne and
Tellambura [27], Karagiannidis [15], Efthymoglou et al. [9], Hadzi-Velkov et al. [11]. In
these works it is shown that the performance of digital communication systems over AF
relay channels is strongly related to the harmonic mean of the instantaneous signal-to-
noise ratio (SNR) of hops. Moreover, in Hadzi-Velkov et al. [11], motivated by the fact
that the multi-hop AF Rayleigh fading channel is a cascaded one [26], a general analytical
framework for the evaluation of second order statistical parameters was presented.
The statistical properties of the harmonic mean of two independent, not necessarily
identically distributed exponential random variables are studied in Hasna and Alouini [12] and
Anghel and Kaveh [2]. Similar results for identically distributed gamma random variables are
presented in Hasna and Alouini [14]. Based on these statistical formulas, the error and outage
performance of dual-hop communication systems operating over Nakagami-m fading
channels was addressed. In Senaratne and Tellambura [27], the statistical properties of the
harmonic means of non-identically distributed gamma random variables with integer-valued
parameters were studied. Furthermore, the authors in Renzo et al. [24] proposed a unied
framework for the performance analysis of AF systems. In that work, a very generic integral
representation for the MGF of the harmonic mean of N arbitrarily distributed random
variables was derived. Also, in Yang and Chen [32] an innite series representation of the
probability density function (PDF) of the half harmonic mean of two gamma RVs is given.
The main contribution of this paper is to derive a closed-form expression for the MGF
of the half harmonic mean of two independent, not necessarily identically distributed
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 846
Author's personal copy
(i.n.i.d.) gamma random variables (RVs) with arbitrary parameters. To the best of our
knowledge, this result is available only in the case of independent, identically distributed
(i.i.d.) gamma RVs [14]. Given that the instantaneous power of a Nakagami-m envelope
follows a gamma distribution, this formula can be efciently used to study the performance
of dual-hop wireless communication systems over Nakagami-m fading channels. The
proposed mathematical analysis is substantiated by means of numerically evaluating
results and Monte-Carlo simulations.
The remainder of this paper is structured as follows: Section 2 provides a closed form
expression for the MGF of the half harmonic mean of two i.n.i.d. gamma RVs. Section 3
addresses some special cases of interest, namely the i.i.d. case as well as the case of positive
integer values of the distribution parameters. In Section 4 the previously derived statistical
formula is used to evaluate in closed-form integrals involving the product of Bessel
functions of different arguments. Such integrals are frequently encountered in the
performance analysis of wireless communication systems with relays. In Section 5 we apply
our results in the performance of AF communication systems. Numerical and computer
simulation results are presented in Section 6 while Section 7 concludes the paper.
2. Moment generating function of the half harmonic mean of i.n.i.d. gamma variates
In this section, we rst recall the denition of a gamma RV and then present some key
results on the MGF of the half harmonic mean of two gamma variates.
Denition 1. The RV X follows a gamma distribution with parameters a40 and b40 if its
PDF is given by
f
X
x9
x
a1
e
x=b
b
a
Ga
Ux 1
where Gz9
_
1
0
e
t
t
z1
dt is the Gamma function [10, Eq. (8.310.1)] and U(x) is the unit
step function.
Hereafter, we will use the notation XGa,b to denote that the RV X follows the
gamma distribution with parameters a and b.
Denition 2 (Harmonic mean of two RVs). Let X
1
, X
2
be two not necessarily identically
distributed random variables. We dene the half harmonic mean H
X
1
,X
2
of the RVs X
1
and
X
2
as
H
X
1
,X
2
9
1
X
1

1
X
2
_ _
1

X
1
X
2
X
1
X
2
2
In the following we will derive a closed form expression for the MGF of H
X
1
,X
2
which is
the main result of our paper. We rst prove the following useful lemma:
Lemma 1 (Probability density function of H
X
1
,X
2
). An integral representation for the PDF of
the half harmonic mean of two i.n.i.d. gamma distributed RVs, f
H
X
1
,X
2
z, is given by
f
H
X
1
,X
2
z
z
a
1
a
2
1

2
i 1
b
a
i
i
Ga
i

_
1
0
exp z
1
b
1
t

1
b
2
1t
_ _ _ _
dt
t
a
1
1
1t
a
2
1
3
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 847
Author's personal copy
Proof. Let us dene the RVs U 1=X
1
and V 1=X
2
. The PDF of the sum W U V is
obtained using the convolution theorem, namely
f
W
w
_
1
0
f
U
tf
V
wt dt
_
w
0
f
U
tf
V
wt dt 4
Performing a simple RV transformation we nd that the PDFs of U and V are related to
the PDFs of X
1
and X
2
, respectively, as
f
U
u
1
u
2
f
X
1
1
u
_ _
5a
f
V
v
1
v
2
f
X
2
1
v
_ _
5b
Substituting the previously derived equations to Eq. (4), f
W
(w) is obtained as
f
W
w
_
w
0
f
X
1
1
t
_ _
f
X
2
1
wt
_ _
dt
t
2
wt
2
6
The PDF of H
X
1
,X
2
is given by f
H
X
1
,X
2
z 1=z
2
f
W
1=z. Using Eq. (6), f
H
X
1
,X
2
z is written as
f
H
X
1
,X
2
z
_
1=z
0
f
X
1
1
t
_ _
f
X
2
z
1zt
_ _
dt
t
2
1zt
2
7
Performing the change of variables t zt, f
H
X
1
,X
2
z is expressed as
f
H
X
1
,X
2
z z
_
1
0
f
X
1
z
t
_ _
f
X
2
z
1t
_ _
dt
t
2
1t
2
8
Substituting the analytical expressions of f
X
1
x and f
X
2
x to Eq. (8), Eq. (3) is obtained. &
The MGF of H
X
1
,X
2
is now obtained from the following theorem:
Theorem 1 (Moments generating function of H
X
1
,X
2
). The MGF of the half harmonic mean
of two i.n.i.d. gamma distributed RVs, M
H
X
1
,X
2
s, is given in closed-form as
M
H
X
1
,X
2
s
b
1
b
2
_ _
a
2
F
1
a
2
,a
1
a
2
,a
1
a
2
; a
1
a
2
; z
1
s,z
2
s 9
where F
1
is the Appell hypergeometric function [22, Eq. (7.2.1.41)], and z
1;2
s are given by
z
1;2
s
2sb
1
b
2
b
1
b
2
sb
1
b
2
7

b
1
b
2

2
sb
1
b
2
2b
1
2b
2
sb
1
b
2

_ 10
Proof. The MGF of H
X
1
,X
2
is dened as the Laplace transform of H
X
1
,X
2
, namely
M
H
X
1
,X
2
s9E/expsH
X
1
,X
2
S
_
1
0
e
sz
f
H
X
1
,X
2
z dz 11
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 848
Author's personal copy
where E/ S denotes expectation. Using Eq. (3), H
X
1
,X
2
is expressed as
M
H
X
1
,X
2
s
1

2
i 1
b
a
i
i
Ga
i

_
1
0
_
1
0
exp z s
1
b
1
t

1
b
2
1t
_ _ _ _

z
a
1
a
2
1
t
a
1
1
1t
a
2
1
dt dz 12
Changing the order of the integration, M
H
X
1
,X
2
s can be expressed as
M
H
X
1
,X
2
s
1

2
i 1
b
a
i
i
Ga
i

_
1
0
_
1
0
exp z s
1
b
1
t

1
b
2
1t
_ _ _ _
z
a
1
a
2
1
dz
_ _

dt
t
a
1
1
1t
a
2
1
13
We dene the inner integral (with respect to z) as
I
1
9
_
1
0
exp z s
1
b
1
t

1
b
2
1t
_ _ _ _
z
a
1
a
2
1
dz 14
Using the denition of the gamma function, I
1
can be easily evaluated as
I
1

Ga
1
a
2
b
1
b
2
t1t
a
1
a
2
sb
1
b
2
t
2
sb
1
b
2
b
1
b
2
t b
2

a
1
a
2
15
Consequently, M
H
X
1
,X
2
s is expressed as
M
H
X
1
,X
2
s
b
a
2
1
b
a
1
2
Ga
1
a
2

2
i 1
Ga
i

_
1
0
t
a
2
1
1t
a
1
1
sb
1
b
2
t
2
sb
1
b
2
b
1
b
2
t b
2

a
1
a
2
dt 16
To the best of our knowledge, this integral is not included in tables of classical reference
books such as Gradshteyn and Ryzhik [10] and Prudnikov et al. [22]. To obtain a closed-
form solution of M
H
X
1
,X
2
s, we proceed as follows: We dene
Db
1
,b
2
; s9sb
1
b
2
t
2
sb
1
b
2
b
1
b
2
t b
2
17
the denominator of Eq. (16). One can easily observe that Db
1
,b
2
; s has two real roots r
1
s
and r
2
s, 8sZ0 and 8b
1
,b
2
40. This is because of the fact that r
1
sr
2
s 1=sb
1
o0.
Moreover, r
1
s and r
2
s have opposite signs. After some straightforward algebraic
manipulations, r
1
s and r
2
s can be expressed as the inverse of z
1
s and z
2
s,
respectively, where z
1;2
s are given in Eq. (10). Thus, Db
1
,b
2
; s can be written as
Db
1
,b
2
; s b
a
1
a
2
2
1
t
r
1
s
_ _
a
1
a
2
1
t
r
2
s
_ _
a
1
a
2
18
Consequently, Eq. (16) is expressed as
M
H
X
1
,X
2
s
b
1
b
2
_ _
a
2
Ga
1
a
2

2
i 1
Ga
i

_
1
0
t
a
2
1
1t
a
1
1
1
t
r
1
s
_ _
a
1
a
2
1
t
r
2
s
_ _
a
1
a
2
dt 19
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 849
Author's personal copy
The integral in Eq. (19) can be solved in closed form by making use of the denition of the
Appell hypergeometric function [22, Eq. (7.2.1.41)], i.e.
F
1
a,b
1
,b
2
,c; y
1
,y
2

1
i
1
,i
2
0
a
i
1
i
2
b
1

i
1
b
2

i
2
c
i
1
i
2
y
i
1
1
i
1
!
y
i
1
1
i
1
!
, maxfjy
1
j,jy
2
jgo1

Gc
GaGca
_
1
0
t
a1
1t
ca1

2
i 1
1y
i
t
b
i
dt,
Rc4Ra40 20
where x
y
Gx y=Gx denotes the Pochhammer symbol and R denotes the real
part. Consequently, Eq. (9) is readily obtained. As it is obvious, the condition for the
convergence of the integral representation of the Appell hypergeometric function is always
satised. Regarding the convergence of the innite series representation, we can observe
that z
2
s is negative 8b
1
,b
2
,s40. Moreover, we can easily show that 0oz
1
so1. Indeed,
the latter inequality can be written as
b
1
b
2
sb
1
b
2

b
1
b
2

2
sb
1
b
2
2b
1
2b
2
sb
1
b
2

_
2sb
1
b
2
41
3

b
1
b
2

2
sb
1
b
2
2b
1
2b
2
sb
1
b
2

_
4b
2
b
1
sb
1
b
2
21
If the quantity b
2
b
1
sb
1
b
2
is negative, the last inequality is valid 8b
1
,b
2
,s40. Else, by
taking the squares of both parts and after some straightforward algebraic manipulations,
the inequality can be reduced to sb
2
1
b
2
40 which is again true. Finally, using the identities
[22, Eqs. (7.2.4.25), (7.2.4.27)]
F
1
a,b
1
,b
2
,c; y
1
,y
2
1y
1

b
1
1y
2

b
2
F
1
ca,b
1
,b
2
,c;
y
1
y
1
1
,
y
2
y
2
1
_ _
1y
1

cab
1
1y
2

b
2
F
1
ca,cb
1
b
2
,b
2
,c; y
1
,
y
1
y
2
1y
2
_ _
22
one can also guarantee the convergence of the corresponding innite series representation. &
Note that the Appell F
1
hypergeometric function is a standard built-in function available
in Mathematica. Specically, the command lines given in Eq. (23) can be utilized as an
efcient and straightforward manner to evaluate the MGF:
z1 2nsnb1nb2=b1b2 snb1nb2
Sqrtb1b2^ 2 snb1nb2n2nb1 2nb2 snb1nb2
z2 2nsnb1nb2=b1b2 snb1nb2
Sqrtb1b2^ 2 snb1nb2n2nb1 2nb2 snb1nb2
MGF b1=b2^ a2nAppellF1a2,a1 a2,a1 a2,a1 a2,z1,z2 23
In Maple, a fast and computationally efcient method to evaluate F
1
can be obtained using its
integral representation and symbolic integration. To prove that our newly derived MGF
expression is a valid one, we evaluate it at s0. In this case, the value of the MGF should be
equal to the area under the PDF curve, namely equal to one. Without loss of generality, we
assume b
1
4b
2
. One can easily observe that z
1
0 0. The value of z
2
s at s0 can be
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 850
Author's personal copy
obtained using the LHospital rule and after some straightforward algebraic manipulations as
z
2
0 lim
s-0
2sb
1
b
2
b
1
b
2
sb
1
b
2

b
1
b
2

2
sb
1
b
2
2b
1
2b
2
sb
1
b
2

2b
1
b
2
b
1
b
2

b
1
b
2
b
1
b
2

b
1
b
2
1
b
1
b
2
24
Then, we make use of the identity [22, Eq. (7.2.4.58)] F
1
a,b
1
,b
2
,c; 0,y
2

2
F
1
a,b
2
; c; y
2

where
2
F
1
is the Gauss hypergeometric function [22, Eq. (7.2.1.1)] to write M
H
X
1
,X
2
0 as
M
H
X
1
,X
2
0
b
1
b
2
_ _
a
2
2
F
1
a
2
,a
1
a
2
; a
1
a
2
; 1
b
1
b
2
_ _
25
Finally using the identity [22, Eq. (7.3.1.27)]
2
F
1
a,b; b; z 1z
a
, we obtain the desired
result M
H
X
1
,X
2
0 1. For b
1
rb
2
the same result can be obtained by observing that in this
case z
2
0 0 and following a similar process to evaluate z
1
0.
3. Special cases
In this section, we consider some special cases for the evaluation of the MGF. Two cases
are considered: The rst case of interest is the i.i.d. case, namely a

a and
b

b, 8 1; 2. The second case emerges when a

are restricted to positive integers.


3.1. The i.i.d. case
In the i.i.d. case, we show that our MGF expression is reduced to a previously known
result [14, Eq. (40)].
1
Specically, we prove the following theorem:
Theorem 2. For a
1
a
2
a and b
1
b
2
b, Eq. (9) is reduced to
M
H
X
1
,X
2
s
2
F
1
a,2a; a
1
2
;
sb
4
_ _
26
Proof. Our starting point is the integral representation of the MGF given by Eq. (16).
For a
1
a
2
a and b
1
b
2
b, Eq. (16) can be written as
M
H
X
1
,X
2
s
G2a
Ga
2
_
1
0
t
a1
1t
a1
sbt1t 1
2a
dt 27
Again, this integral is not included in tables of classical reference books such as
Gradshteyn and Ryzhik [10] and Prudnikov et al. [22]. However, performing the change of
variables t sin
2
x, M
H
X
1
,X
2
s can be expressed as
M
H
X
1
,X
2
s 2
22a
G2a
Ga
2
_
p=2
0
sin
2a1
2x
sb
4
sin
2
2x 1
_
2a
dx 28
1
In Hasna and Alouini [14] the MGF of a random variable X is dened as M
X
s E/expsXS. Moreover,
Hasna and Alouini [14] deals with the harmonic mean of two random variables, not the half harmonic mean. For
consistency, we note that the MGF of the RV Y 2X, M
Y
s can be expressed in terms of the MGF of X as
M
X
s M
Y
s=2.
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 851
Author's personal copy
Performing a second change of variables, y 2x, M
H
X
1
,X
2
s is expressed as
M
H
X
1
,X
2
s 2
12a
G2a
Ga
2
_
p
0
sin
2a1
y
sb
4
sin
2
y 1
_
2a
dy
2
12a
G2a
Ga
2
_
p=2
0
sin
2a1
y
sb
4
sin
2
y 1
_
2a
dy
2
12a
G2a
Ga
2
_
p
p=2
sin
2a1
y
sb
4
sin
2
y 1
_
2a
dy
9J
1
J
2
29
Regarding the integral J
2
, by substituting y pw one can easily show that J
1
J
2
.
Therefore
M
H
X
1
,X
2
s 2
22a
G2a
Ga
2
_
p=2
0
sin
2a1
w
sb
4
sin
2
w 1
_
2a
dw 30
After performing a nal change of variables, sin
2
w z, M
H
X
1
,X
2
s is obtained as
M
H
X
1
,X
2
s 2
12a
G2a
Ga
2
_
1
0
z
a1
sb
4
z 1
_ _
2a

1z
p
dz 31
Using the integral representation of the Gauss hypergeometric function, namely
2
F
1
a,b; c; y
Gc
GaGca
_
1
0
t
a1
1t
ca1
1yt
b
dt, Rc4Ra40 32
as well as the Legendre duplication formula for the gamma function [10, Eq. (6.1.18)],
G2a
1

2p
p 2
2a1=2
GaG a
1
2
_ _
33
the MGF of H
X
1
,X
2
is obtained in the compact form given by Eq. (26). &
Corollary 1. The following identity is valid 8A,a40
F
1
a,2a,2a,2a;
2
1

1 A
p ,
2
1

1 A
p
_ _

2
F
1
a,2a; a
1
2
;
1
A
_ _
34
This identity, which is regarded as an extra piece of information emerged from our
original study, is readily obtained by setting a
1
a
2
a, b
1
b
2
b and sb=4 1=A to
Eq. (9) and with the help of Eq. (26).
3.2. The case of positive integer a

For positive integer values of a

we prove the following theorem:


Theorem 3. The MGF of the half harmonic mean of X
1
Ga
1
,b
1
and X
2
Ga
2
,b
2
,
M
H
X
1
,X
2
s when a
1
and a
2
are positive integers, is given in closed-form as
M
H
X
1
,X
2
s
b
1
b
2
_ _
a
2
Ga
1
a
2

Ga
1
Ga
2

As
a
1
a
2

a
1
1
i 0

a
1
a
2
1
j 0
1
i
a
1
1
i
_ _
_
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 852
Author's personal copy

j a
1
a
2
1
a
1
a
2
1
_ _
Bs
j
Ki a
2
1,a
1
a
2
j1,z
2
s
Bs
a
1
a
2

a
1
1
i 0

a
1
a
2
1
k 0
1
i
a
1
1
i
_ _
k a
1
a
2
1
a
1
a
2
1
_ _
As
k
Ki a
2
1,a
1
a
2
k1,z
1
s
_
35
where
As
z
2
s
z
2
sz
1
s
, Bs
z
1
s
z
1
sz
2
s
36a
Kp,q,x x
p1

p
l 0
laq
p
l
1
l
11x
lq
lq

p
q
_ _
1
q
ln1x
_
_
_
_
36b
and z
1;2
s are given by Eq. (10).
Proof. Using Cuyt et al. [8, Lemma (3.1)], for integer values of a
1
and a
2
the integrand in
Eq. (19) can be written as
t
a
2
1
1t
a
1
1
1tz
1
s
a
1
a
2
1tz
2
s
a
1
a
2
As
a
1
a
2

a
1
1
i 0

a
1
a
2
1
j 0
1
i
_

a
1
1
i
_ _
j a
1
a
2
1
a
1
a
2
1
_ _
Bs
j
t
ia
2
1
1tz
2
s
a
1
a
2
j
z
2
s
ia
2
Bs
a
1
a
2

a
1
1
i 0

a
1
a
2
1
k 0
1
i
a
1
1
i
_ _
k a
1
a
2
1
a
1
a
2
1
_ _
As
k
z
1
s
ia
2
t
ia
2
1
1tz
1
s
a
1
a
2
k
_
37
where A(s) and B(s) are given by Eq. (36a). To evaluate the integral (19), we dene
Kp,q,x9
_
1
0
t
p
1xt
q1
dt 38
where p, q are positive integers. This integral can be readily evaluated using Cuyt et al.
[8, Lemma (3.2)] as in Eq. (36b). Thus, after some straightforward algebraic manipulations
Eq. (35) is obtained. &
4. A useful integral in wireless communication theory
In this section, we use our previously derived results to evaluate integrals of the form
Ia
1
,a
2
,w
1
,w
2
,y
_
1
0
t
a
1
a
2
J
1
ytK
a
1
w
1
tK
a
2
w
2
t dt, a
1
,a
2
,w
1
,w
2
,y40 39
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 853
Author's personal copy
where J
n
(x) is the n-th order Bessel function of the rst kind [10, Eq. (8.402)] and K
n
(x) is
the n-th order modied Bessel function of the second kind [10, Eq. (8.407)]. This integral,
which can be considered as a special case of the integrals presented in Trigui et al. [31,
Eq. (3)] and Prudnikov et al. [23, Eq. (2.16.43.1)], appears in a number of wireless
applications including the evaluation of the error probabilities of non-coherent binary
modulation schemes of AF wireless systems. The analytical expressions of the integrals in
Trigui et al. [31, Eq. (3)] and Prudnikov et al. [23, Eq. (2.16.43.1)] make use of the
Lauricella F
C
[22, Eq. (7.2.4.14)] and the Appell F
4
[22, Eq. (7.2.4.4)] functions,
respectively. However, there are not so many computer programs for evaluating the
Lauricella F
C
and the Appell F
4
functions. In fact, we are not aware of any built-in routine,
available in standard mathematical software packages such as Maple or Mathematica that
can evaluate these functions. In the following, we give expressions for Eq. (39) that
involves the Appell F
1
hypergeometric function. We prove the following theorem:
Theorem 4. The following identity is valid:
Ia
1
,a
2
,w
1
,w
2
,y C 1
w
2
w
1

2a
2
F
1
a
2
,a
1
a
2
,a
1
a
2
; a
1
a
2
; r
1
,r
2

_ _
40
where
C
Ga
1
Ga
2
2
a
1
a
2
2
w
a
1
1
w
a
2
2
y
41a
r
1;2

2y
2
w
2
2
w
2
1
y
2
7

w
2
2
w
2
1

2
y
2
2w
2
1
2w
2
2
y
2

_ 41b
Proof. Our starting point is the MGF inversion formula presented in Renzo et al.
[24, Theorem 1] along with a change of variables, leading to the following result:
M
H
X
1
,X
2
s 12

s
p
_
1
0
J
1
2

s
p
tM
X
1
1
t
2
M
X
1
2
t
2
dt 42
where M
X
1

s, 8 1; 2, are the MGFs of X


1

. Assuming that X

Ga

,b

, the MGFs of
X
1

can be evaluated with the help of Gradshteyn and Ryzhik [10, Eq. (3.471.9)] as
M
X
1

s9
_
1
0
e
s=x

x
a

e
x

=b

b
a

Ga

dx


2
Ga

s
b

_ _
a

=2
K
a

s
b

_ _ _
43
Substituting Eq. (42) to Eq. (43) and with the help of Eq. (9), we obtain the desired result. &
To verify the correctness of Eq. (40), Eq. (39) is evaluated numerically for a wide range
of values of the involved parameters. Specically, by performing the change of variables
t tanx in Eq. (39), and applying the GaussChebyshev quadrature (GCQ) technique
[1], Eq. (39) can be expressed as
Ia
1
,a
2
,w
1
,w
2
,y

Q
j 0
p
j
u
a
1
a
2
j
J
1
yu
j
K
a
1
w
1
u
j
K
a
2
w
2
u
j
44
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 854
Author's personal copy
where Q is the number of integration points, u
j
are the abscissas and p
j
are the integration
weights. Using Yilmaz and Alouini [33, Eq. (15, 16)], u
j
and p
j
can be expressed as
u
j
tan
p
4
cos
2j1
2Q
p
_ _

p
4
_ _
45a
p
j

p
2
sin
2j1
2Q
p
_ _
4Q cos
2
p
4
cos
2j1
2Q
p
_ _

p
4
_ _
45b
As expected, our conducted numerical experiments have shown that the values obtained
from both Eqs. (44) and (39) were identical for all considered test cases.
5. Applications in the error performance analysis of AF systems over Nakagami-m
fading channels
In this section, we make use of our previously derived results to provide an extended
error-rate analysis of relaying systems over Nakagami-m fading channels. We consider a
dual-hop system with a source node, S, communicating with a destination node, D,
through a relay node R. Let h and g be the channel coefcients between the source S and R
and between R and D, respectively. These channel coefcients are modeled as Nakagami-m
RVs with parameters m
1
and m
2
respectively.
At a given time slot, S broadcasts the information symbol to the R and to D. The relay
receives and multiplies it by a gain factor G before retransmitting it to D. Without any loss
of generality we assume that all additive white Gaussian noise (AWGN) terms have zero
mean and equal variance N
0
. Assuming that the relay gain G is selected according to the
instantaneous Channel State Information assisted relay gain [12, Eqs. (4), (7)] the end-to-
end signal-to-noise ratio (SNR) can be written as [2]
g
eq

g
1
g
2
g
1
g
2
c
46
where g
1
h
2
E
s
=N
0
is the instantaneous SNR between S and R, g
2
g
2
E
s
=N
0
is the
instantaneous SNR between R and D and E
s
is the signal transmitted energy.
In addition, exact g
eq
is given by substituting c1, and well approximated at high SNR
by substituting c0. Note that according to Hasna and Alouini [12], the case c0
corresponds to an ideal relay that just amplies the incoming signal with the inverse of the
channel of the previous hop, regardless of the noise of that hop. Such a relay serves as a
benchmark for all practical multi-hop systems employing AF relays. The case c1 is more
difcult to analyze; therefore, it will only be investigated using Monte-Carlo simulations.
The instantaneous SNR of the S-R link is g
1
Gm
1
,g
1
=m
1
where m
1
Z1=2 is the
severity of fading parameter and g
1
E/h
2
SE
s
=N
0
. Similarly, for the R-D link,
g
2
Gm
2
,g
2
=m
2
where g
2
E/g
2
SE
s
=N
0
. The MGF of g
eq
can be expressed in closed-
form for c0 using Theorem 1. To apply Theorem 1, the following modications should
be made: a
1
and a
2
should be replaced with m
1
and m
2
, respectively, while b
1
and b
2
should
be replaced with g
1
=m
1
and g
2
=m
2
, respectively.
Following the MGF-based approach [30], the error rate performance of AF systems for
various modulation formats is analytically obtained. Based on Eq. (9), the ABEP for non-
coherent binary frequency shift keying (NBFSK) and binary differential phase shift keying
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 855
Author's personal copy
(BDPSK) modulation signalings can be directly calculated. Specically, the ABEP P
be
of
such schemes can be expressed as
P
be
AM
g
eq
B 47
where A 1=2 and B1 for BDPSK, A B 1=2 for NBFSK. Closed-form tight bounds
for binary coherent modulation schemes can be obtained using the fact that the ABEP of
digital receivers operating over fading channels can be evaluated by averaging the
probability of error conditioned on the instantaneous SNR, P
e
ejg, over the PDF of g
eq
,
that is
P
be

_
1
0
P
e
ejgf
g
e
q
g dg 48
For binary coherent modulations P
e
ejg is given by
P
e
ejg Q

2cg
_
49
where Qx 1=

2p
p

_
1
x
expx
2
=2 dx is the Gaussian Q-function. Moreover, c 1 for
coherent binary phase shift keying (BPSK), c 1=2 for coherent binary frequency shift
keying (BFSK) and c 0:715 for coherent BFSK with minimum correlation. Note that
the Q-function can be expressed in terms of the complementary error function erfc as
Qx
1
2
erfcx=

2
p
. Using the tight exponential bound for the Gaussian Q-function
proposed in Chiani et al. [7], i.e.
QxC
1
12
expx
2
=2
1
4
exp2x
2
=3 50
as well as the denition of the MGF, an accurate closed-form approximation for the ABEP
of binary modulation schemes can be obtained as
P
be
C
1
12
M
g
eq
c
1
4
M
g
eq
4c
3
_ _
51
Note that our analysis can be easily extended to cooperative diversity systems with
maximal-ratio combining reception. In this case, assuming that signals from each incoming
branch are modeled as independent RVs, the MGF of the end-to-end SNR is readily
obtained as the product of the MGFs of each individual branch, based on Eq. (9).
6. Numerical and computer simulation results
In this section, numerical and computer simulation results of the ABEP for BDPSK and
BPSK modulations are presented. The following pairs of values of the fading parameters m
h
and m
g
are assumed: m
g
,m
h
1:5,0:8, m
g
,m
h
2:25,1:25 and m
g
,m
h
2:2,1:8. For
all test scenarios, we consider E/g
2
S0:75 and E/h
2
S 1. To evaluate the Appell
hypergeometric functions we used both Mathematica and Maple. As mentioned in previous
sections, in the latter case, the computation is based on the integral representation of the
Appell F
1
function, given by Eq. (20). Note that this method was proved to be accurate and
highly efcient in terms of computational complexity. To double-check the correctness of
our proposed mathematical analysis, equivalent ABEP results obtained using Monte-Carlo
simulations are also presented. Specically, for given values of m
g
, m
h
, E
s
=N
0
, E/g
2
S and
E/h
2
S three million independent samples of g
eq
are generated, based on Eq. (46). Note that
independent, gamma distributed samples can be easily generated using the Matlab build-in
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 856
Author's personal copy
function gamrnd. A sample program written in Matlab for the evaluation of the ABEP of
BDPSK and BPSK by means of Monte-Carlo simulations is given as
function BDPSK, BPSK simulatemg, mh, Eg, Eh, dB, c, SIZE
%mg : the Nakagami-m parameter of R4D link
%mh : the Nakagami-m parameter of S4R link
%Eg : Eog^ 24
%Eh : Eoh^ 24
%dB : Es=No in dB
%c f0; 1g
%SIZE : Number of samples
Esn0 10^ 0:1 n dB;
gg Eg n Esn0;
gh Eh n Esn0;
%Generation of independent gamma samples for each branch
datag gamrndmg, gg=mg, 1, SIZE;
datah gamrndmh, gh=mh, 1, SIZE;
data datah: n datag:=datah datag c;
%ABEP of BDPSK
BDPSK sum0:5 n expdata=SIZE;
%ABEP of BPSK
BPSK sum0:5 n erfcsqrtdata=SIZE; 52
Fig. 1. ABEP of BDPSK of dual-hop systems operating over Nakagami-m fading channels versus E
s
=N
0
for
different values of m
h
and m
g
.
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 857
Author's personal copy
In Fig. 1, the ABEP of BDPSK and BPSK of dual-hop AF systems as a function of
E
s
=N
0
for c0 and c1. For c1, ABEP results were obtained by means of Monte-Carlo
simulations only. As it is evident, the difference of the ABEP results for c0 and c1 is
indeed small, especially at medium and high values of E
s
=N
0
. Moreover, as expected, we
observe that ABEP performance improves as E
s
=N
0
increases or m
h
, m
g
increase. For the
case c0, additional ABEP results obtained using Monte-Carlo simulations (square
patterns) have been included and as it is obvious, an excellent match with the analytically
obtained results is observed, thus verifying the correctness of our analysis.
In Fig. 2, the ABEP of BPSK of dual-hop AF systems is plotted as a function of E
s
=N
0
for c0 and c1. As it can be observed, Eq. (51) provides a very tight bound for the
evaluation of ABEP for binary constellation schemes. Regarding the impact of the various
parameters on the error performance, one can verify similar ndings with the ones
presented in the previous test case.
7. Conclusions
In this paper, we presented a new result on the MGF of the harmonic mean of two non-
identical gamma variates, and applied these results to evaluate the ABEP performance of
wireless cooperative communication systems with amplify-and-forward relays. The MGF
is expressed in terms of the Appell hypergeometric functions which can be efciently
evaluated using well known mathematical software packages. The proposed mathematical
analysis has been illustrated by extensive numerical and computer simulation results.
Fig. 2. ABEP of BPSK of dual-hop systems operating over Nakagami-m fading channels versus E
s
=N
0
for
different values of m
h
and m
g
.
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 858
Author's personal copy
References
[1] M. Abramovitz, I. Stegun, in: Handbook of Mathematical Functions with Formulas, Graphs, and
Mathematical Tables, Dover, New York, 1964. ISBN 0-486-61272-4.
[2] P. Anghel, M. Kaveh, Exact symbol error probability of a cooperative network in a Rayleigh-fading
environment, IEEE Transactions on Wireless Communications 3 (2004) 14161421.
[3] V. Bhaskara, Distribution and density functions of probability of error over slowly fading channels with
diversity combining, Journal of the Franklin Institute 348 (2011) 11531159.
[4] A. Bletsas, A. Khisti, D.P. Reed, A. Lippman, A simple cooperative diversity method based on network path
selection, IEEE Journal on Selected Areas on Communications 24 (2006) 659672.
[5] J. Boyer, D.D. Falconer, H. Yanikomeroglu, Multihop diversity in wireless relaying channels, IEEE
Transactions on Communications 52 (2004) 18201830.
[6] J. Chen, A unied approach to average LCR and AFD performance of SC diversity in generalized fading
environments, Journal of the Franklin Institute 344 (2007) 889911.
[7] M. Chiani, D. Dardari, M.K. Simon, New exponential bounds and approximations for the computation of
error probability in fading channels, IEEE Transactions on Wireless Communications 2 (2003) 840845.
[8] A. Cuyt, K. Driver, J. Tan, B. Verdonk, A nite sum representation of the Appell series F
1
a,b,b
0
; c; x,y,
Journal of Computational and Applied Mathematics 105 (1999) 213219.
[9] George P. Efthymoglou, Nikolaos Bissias, Valentine A. Aalo, On the error rate analysis of dual-hop amplify-
and-forward relaying in generalized-K fading channels, Journal of Electrical and Computer Engineering
(2010) 5 pp. Article ID 584594.
[10] I. Gradshteyn, I.M. Ryzhik, Tables of Integrals, Series, and Products, 6th ed., Academic Press, New York, 2000.
[11] Z. Hadzi-Velkov, N. Zlatanov, G.K. Karagiannidis, On the second order statistics of the multihop Rayleigh
fading channel, IEEE Transactions on Communications 57 (2009) 18151823.
[12] M.O. Hasna, M.-S. Alouini, End-to-end performance of transmission systems with relays over Rayleigh
fading channels, IEEE Transactions on Wireless Communications 2 (2003) 11261131.
[13] M.O. Hasna, M.-S. Alouini, Outage probability of multihop transmission over Nakagami fading channels,
IEEE Communications Letters 7 (2003) 216218.
[14] M.O. Hasna, M.-S. Alouini, Harmonic mean and end-to-end performance of transmission systems with
relays, IEEE Transactions on Communications 52 (2004) 130135.
[15] G.K. Karagiannidis, Performance bounds of multihop wireless communications with blind relays over
generalized fading channels, IEEE Transactions on Wireless Communications 5 (2006) 498503.
[16] J.N. Laneman, D.N.C. Tse, G.W. Wornell, Cooperative diversity in wireless networks efcient protocols and
outage behaviour, IEEE Transactions on Information Theory 50 (2004) 30623080.
[17] J.N. Laneman, G.W. Wornell, Distributed spacetime-coded protocols for exploiting cooperative diversity in
wireless networks, IEEE Transactions on Communications 49 (2003) 24152425.
[18] B. Maham, A. Hjorungnes, Asymptotic performance analysis of amplify-and-forward cooperative networks
in a Nakagami-m fading environment, IEEE Communications Letters 13 (2009) 300302.
[19] D.S. Michalopoulos, G. Karagiannidis, G. Tombras, Symbol error probability of decode and forward
cooperative diversity in Nakagami-m fading channels, Journal of the Franklin Institute 345 (2008) 723728.
[20] A. Panajotovic, M. Stefanovic, D. Draca, Cochannel interference effect on BEP performance of SSC receiver
in correlated Rician fading, Journal of the Franklin Institute 347 (2010) 12421252.
[21] K.P. Peppas, C.K. Datsikas, H.E. Nistazakis, G.S. Tombras, Dual-hop relaying communications over
generalized-K (KG) fading channels, Journal of the Franklin Institute 347 (2010) 16431653.
[22] A.P. Prudnikov, Y.A. Brychkov, O.I. Marichev, Integrals and Series Volume 3: More Special Functions,
1st ed., Gordon and Breach Science Publishers, 1986.
[23] A.P. Prudnikov, Y.A. Brychkov, O.I. Marichev, Integrals and Series Volume 2: Special Functions, 1st ed.,
Taylor and Francis Ltd., 1998.
[24] M.D. Renzo, F. Graziosi, F. Santucci, A unied framework for performance analysis of CSIassisted
cooperative communications over fading channels, IEEE Transactions on Communications 57 (2009)
25512557.
[25] A. Ribeiro, X. Cai, G.B. Giannakis, Symbol error probabilities for general cooperative links, IEEE
Transactions on Wireless Communications 4 (2005) 12641273.
[26] N.C. Sagias, G. Tombras, On the cascaded Weibull fading channel, Journal of the Franklin Institute
344 (2007) 111.
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 859
Author's personal copy
[27] D. Senaratne, C. Tellambura, Unied exact performance analysis of two-hop amplify-and-forward relaying
in Nakagami fading, IEEE Transactions on Vehicular Technology 59 (2010) 15291534.
[28] A. Sendonaris, E. Erkip, B. Aazhang, User cooperation diversityPart I: system description, IEEE
Transactions on Communications 51 (2003) 19271939.
[29] A. Sendonaris, E. Erkip, B. Aazhang, User cooperation diversityPart II: implementation aspects and
performance analysis, IEEE Transactions on Communications 51 (2003) 19391948.
[30] M.K. Simon, M.S. Alouini, Digital Communication over Fading Channels, 2nd ed., Wiley, New York, 2005.
[31] I. Trigui, S. Affes, A. Stephenne, A useful integral for wireless communication theory and its application in
amplify-and-forward multihop relaying, in: IEEE Global Telecommunications Conference, GLOBECOM2010.
[32] L.-L. Yang, H.-H. Chen, Error probability of digital communications using relay diversity over Nakagami-m
fading channels, IEEE Transactions on Wireless Communications 7 (2008) 18061811.
[33] F. Yilmaz, M.-S. Alouini, A new simple model for composite fading channels: second order statistics and
channel capacity, in: IEEE Proceedings ISWCS2010.
[34] J. Zhang, T. Zhang, J. Huang, R. Yuan, ABEP of amplify and forward cooperation in Nakagami-m fading
channels with arbitrary m, IEEE Transactions on Wireless Communications 8 (2009) 44454449.
[35] Y. Zhao, R. Adve, T.J. Lim, Outage probability at arbitrary SNR with cooperative diversity, IEEE
Communications Letters 9 (2005) 700702.
K. Peppas / Journal of the Franklin Institute 349 (2012) 845860 860

Vous aimerez peut-être aussi