Sum of Weibull Variates
Sum of Weibull Variates
Sum of Weibull Variates
Systems
*
Ferkan Yilmaz and Mohamed-Slim Alouini
Dept. of Electrical & Computer Eng.,Texas A&M University at Qatar, Education City, 23874, Doha, Qatar
E-mail(s): {ferkan.yilmaz, alouini}@qatar.tamu.edu
ABSTRACT
Sum of Weibull random variables (RVs) is naturally of prime
importance in wireless communications and related areas.
Through the medium of the selection of poles as orthogonal
Laguerre polynomials in Cauchy residue theorem, the moment-
generation function (MGF), the probability density function
(PDF) and the cumulative distribution function (CDF) of the sum
of I 2 mutually independent any random variables (RVs) are
represented in terms of fast convergent series, and the obtained
results are applied to the sum of Weibull RVs in order to find
symbol error rate (SER) and outage probability (OP)
performance.
Categories and Subject Descriptors
B.4.1 [Input/Output and Data Communications Devices]: Data
Communications Devices - Receivers, B.4.4 [Input/Output and
Data Communications Devices]: Performance Analysis and
Design Aids - Formal models, Simulation and Verification, C.2.1
[Network Architecture and Design]: Wireless Communication.
General Terms
Performance, Reliability, Design.
Keywords
Sum of random variables, Weibull distribution, probability
density function, cumulative distribution function, outage
probability, symbol error rate.
1. INTRODUCTION
A unified moment generating function (MGF)-based approach
was adopted over the last years for the exact average error-rate
analysis of wireless applications not only with linear receivers
using maximal ratio combining (MRC), but also for other kinds of
diversity combining techniques [1]. In addition to the average
error rate, outage probability (OP) is another standard
performance criterion of communication systems operating over
fading channels [2]. It is defined as the probability that the
instantaneous error rate exceeds a specified value or equivalently
that the (instantaneous) combined signal-to-noise ratio (SNR), y
falls below a certain specified threshold y
th
, i.e.,
P
out
= Pr|u y < y
th
] = _
y
(r)Jr
y
th
0
(1)
where
y
(r) is the probability density function (PDF) of the
random variable (RV), y. In other words, P
out
is the cumulative
distribution function (CDF) of the RV, y evaluated at y
th
. In this
paper, we focus on the statistics of the sum of L (I 2) mutually
independent any RVs in terms of convergent series constructed by
the selection of poles in Cauchy residue theorem as orthogonal
Laguerre polynomials. The obtained results are applied to the sum
of mutually independent Weibull RVs with the application of
symbol error rate (SER) and OP calculation of MRC diversity
systems over Weibull fading channels. Indeed, only very recently,
the topic of digital communications over Weibull fading channels
has begun to receive some interest and as such the sum statistics is
of prime importance since the Weibull distribution seems to
exhibit good fit to experimental fading channel measurements, for
both indoor [3], [4] and outdoor [5][7] environments, with a
reasonable physical justification given in [8]. To the best of our
knowledge, and as indicated in [9], no results are available about
PDF and CDF for the sum of mutually independent Weibull RVs
in part due to the exponents involved in Weibull distributed RVs
[10].
The remainder of this paper is organized as follows. In Section II,
we reintroduce and discuss Weibull RVs and their applications in
wireless communication channels. In Section III, we propose a
statistical analysis of the sum of L (I 2) mutually independent
any RVs in terms of convergent series based on the selection of
poles as orthogonal Laguerre polynomials, and we then apply the
obtained results to the sum of Weibull RVs in order to find PDF,
CDF and MGF. In Section IV, SER and OP performance results
are presented. Finally, Section V summarizes the main results of
the paper.
2. WEIBULL CHANNEL FADING
Weibull fading occurs when the signal recovered from the
wireless channel is composed of clusters of a multipath wave,
each of which propagates in non-homogenous environment such
that they possess similar delay times and with the delay-time
spreads of different clusters and their phases are independent [3],
[4], [11]. In this case, the amplitude envelope of the received
signal, Z
and quadrature
Gaussian elements as
Z
= (X
+]
)
1 {
(2)
with E(X
) = E(
) = u and E(X
2
) = E(
2
) =
2 , where
] = -1 is the imaginary operator and
= |X
+]
|
2
as follows
P
= (X
+]
)
1 {
(X
-]
)
1 {
= R
1 {
(3)
Then, the PDF of P
(r) =
r
{
-1
exp _-
r
{
_, r > u (4)
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
IWCMC09, June 2124, 2009, Leipzig, Germany. Copyright 2009
ACM 978-1-60558-569-7/09/06...$5.00.
*
This work supported by Qatar National Research Fund (A
member of Qatar Foundation).
247
P
(r) =
1
1 {
B
0,1
1,0
_
r
1 {
__1 -
1
,
1
]_ , r > u (5)
where B
p,q
m,n
|] is the Foxs H function defined in [12]. Then,
exploiting the Mellin transform of the Foxs H function [13,
eq.(3.3)], the mean of P
), can be readily
obtained as
E(P
) = _ r
(r)Jr = _1 +
1
1 {
(6)
where () is Gamma function defined in [15, eq.(18.39)].
From (6), we can see that the power of the received signal
does not change linearly with the average power
so this PDF
does not make sense for analytical accuracy for simulations and
computations of wireless communications. In other words, if
represents the average power of fading, then
P
(r)Jr =
for any
and
< 1 then this PDF does not give meaningful results for
wireless communications since
P
1.
Adding to that the normalization,
= (
(1 +1
) )
{
, the
PDF given in (4) can be transformed into the following
representation:
Definition 1 (Channel Power Weibull Fading RV): A random
variable P
and
, denoted as P
~W(
), if its PDF is
given by
P
(r|
) =
_
[
]
{
r
{
-1
e
-_
[
]
(
(7)
=
[
B
0,1
1,0
_
[
r_ (1 -1 2
, 1 2
)_ (8)
for r u, where
> u and
= (1 +1 2
) is
the power exponent coefficient.
As mentioned above, if (7) represents the PDF of the power of
the receive signal, then (7) undoubtedly maintains the following
conditions for different
and
values
P
(r|
) u, (9)
_
P
(r|
)Jr
0
= 1, (10)
_ r
P
(r|
)Jr
0
=
, (11)
where
and
= 1 and
= 2 and
(r) = 2r exp(-r
2
2
), r > u (13)
In addition, the CDF of the RV, P
~W(
) is given by
F
P
(r|
) = _
P
(u|
)Ju
0
(14)
Substituting (8) into (14) and using [18, eq.(3.7)] and [12,
eq.(2.7.9)], we get
F
P
(r|
) = -B
1,2
2,0
_
[
r _
(1,1)
(u,1), (1, 1
)
_
(15)
= 1 -exp(-([
)
{
), r > u (16)
Using the Mellin transform of the Foxs H function [13, eq.(3.3)],
the nth order moments can be obtained as
E(P
n
) = _ r
n
(r|
)Jr
0
= _1 +
n
] _
]
n
(17)
where n = u,1,2, . Consequently, the MGF of Weibull channel-
power fading RV, M
P
(s) E(exp(-sP
)) (18)
= B
1,1
1,1
_
[
s
_
(1,1)
(1, 1
)
_ (19)
When
= 1 and
(s) =
1
1 +s
(20)
which is the MGF of the exponential RV, as expected.
3. STATISTICAL MEASURES OF THE
SUM OF WEIBULL RVS
It is well known that, for the I-branch linear diversity receivers
with MRC in an additive white Gaussian noise (AWGN) channel,
the signal-to-noise ratio (SNR) at the output is the sum of SNRs
of branches (diversity paths) as
P = P
L
=1
(21)
In a wireless Weibull fading channels, the closed-form PDF,
CDF, the nth order moments and MGF expressions of the
instantaneous SNR, P
~W(
)]
1
L
RVs are themselves independent. In such
instances, the MGF M
P
(s) can be expressed as the product of the
MGFs associated with each branch [i.e. M
P
(s) = M
P
(s)
L
=1
],
then, upon using the previously obtained result given by (19), the
MGF can be easily expressed as
M
P
(s) = _B
1,1
1,1
_
[
s
_
(1,1)
(1, 1
)
_
L
=1
(22)
Using [12, eq.(1.1.1)], (22) can be expressed in terms of Mellin
Barnes integrals as
M
P
(s) = _
1
2n]
]
L
_ _
C
L
C
1
__ _1 +
p
] (-p
) _
[
]
p
_
L
=1
s
L
k=1
p
k
Jp
1
Jp
L
(23)
248
where C
-] and terminating
at the point c
+], where -
< c
0
(27)
= (n + +1) _ u
1,2
1,0
_r _
n + 1
u, -
_
X
(r)Jr (28)
=
(-1)
k
(k +1)
_
n +
n -k
]
k
n
k=0
E(X
k
)
(29)
where E(X
k
) is the kth order moment of the RV, X and I
n
u
() and
u
p,q
m,n
|] denote the generalized Laguerre function [23, Section
10.12] and the Meijers G function [22, Sections 5.3-5.6],
respectively.
Laguerre moment of a RV is a very useful tool to obtain the
statistical measures such as moments, PDF, CDF and MGF of the
sum of the mutually independent any type of RVs. Exploiting
Laguerre moment and using the orthogonal pole separation in
complex domain gives three key theorems, [i.e., Theorems 1-3].
Theorem 1 (MGF Based on the Selection of Poles as Orthogonal
Laguerre Polynomials): Let X
1
, X
2
, , X
L
be any mutually
independent RVs with the corresponding PDFs,
X
1
(r),
X
2
(r), ,
X
L
(r), each of which is defined for r > u, and
the corresponding MGFs, H
X
1
(s), H
X
2
(s), , H
X
L
(s), each of
which is defined for +{s] u, then the MGF of X = X
1
+X
2
+
+X
L
,[i.e., M
X
(s) H
X
(s)
L
=1
], is given by (24) in which
K
1
(s) and K
2
(s) are arbitrary two functions that support
K
1
(s)K
2
(s) = s and +{K
1
(s)] A +{K
2
(s)] > u for all +{s] e
R
+
, z e R
+
is a arbitrary number.
Theorem 2 (PDF Based on the Selection of Poles as Orthogonal
Laguerre Polynomials): Let X
1
, X
2
, , X
L
be any mutually
independent RVs with the corresponding PDFs,
X
1
(r),
X
2
(r), ,
X
L
(r), each of which is defined for r > u, then the
PDF of X = X
1
+X
2
++X
L
, [i.e.,
X
(r)], is given by (25) in
which N
1
(r) and N
2
(r) are arbitrary two functions that support
N
1
(r) N
2
(r) = r and +{N
1
(r)] A +{N
2
(r)] > u for all r e R
+
,
z e R
+
is a arbitrary number.
Theorem 3 (CDF Based on the Selection of Poles as Orthogonal
Laguerre Polynomials): Let X
1
, X
2
, , X
L
be mutually
independent any random variables with the corresponding PDFs,
X
1
(r),
X
2
(r), ,
X
L
(r), each of which is defined for r > u, then
the CDF of X = X
1
+X
2
++X
L
, [i.e., F
X
(r)], is given by (26)
in which
1
(r) and
2
(r) are arbitrary two functions that support
1
(r)
2
(r) = r and +{
1
(r)] A +{
2
(r)] > u for all r e R
+
,
z e R
+
is a arbitrary number.
What we obtain from Theorems 1-3 is that we can give the MGF,
PDF and CDF expressions of the sum of any L (I 2) mutually
independent RVs by convergent Laguerre polynomial series based
on the selection of poles in Cauchy residue theorem as orthogonal
Laguerre polynomials. The MGF, PDF and CDF expressions
given in Theorems 1-3 absolutely converge according to their
asymptotic behaviors. Furthermore, the selection of the arbitrary
functions, K
1
(s) and K
2
(s) in Theorem 1, N
1
(r) and N
2
(r) in
Theorem 2 and
1
(r) and
2
(r) in Theorem 3, determine the
complexity and the speed of convergence.
The advantage of these expressions one can readily observe from
(24), (25) and (26) is that, in order to find the MGF, PDF and
CDF of the sum of mutually independent any random variables,
only the computation of the coefficients E
k
(1+x-L) L
(X
|) is
needed. Regarding to the analytical simplicity and accuracy of
(24), (25) and (26) for the sum of L (I 2) mutually independent
channel power Weibull RVs, the coefficients E
k
(1 L -1)
(X
|) for
= 1,2, , I can be shown to be expressible as
E
k
1+x-L
L
(X
|) = _
Ik +z +1
I
]
B
2,2
1,1
l
l
l
l
_
_
_u,
1
] , (k +1,1)
(u,1), _
I -z -1
I
, 1]
1
1
1
1
(30)
Note that the Foxs H function in (30) can be considered to be
difficult to evaluate since it is typically not a built-in function in
standard mathematical software packages such as Matlab and
Mathematica. However, the Foxs H function can be expressed in
terms of the more familiar Meijers G function. In addition, in our
special case, by the help of (27), we can fortunately write
E
k
1+x-L
L
(X
|) = _ L
k
1+x-L
L
(r)
(r|
)Jr (31)
M
X
(s) lim
N-
K
1
(s)
n
(1 +K
1
(s))
n+x+1
N
n=0
_E
k
1+x-L
L
(X
|K
2
(s))
L
=1 k
1
+k
2
++k
L
=n
(24)
X
(r)
N
1
(r)
x
N
2
(r)
(z +1)
lim
N-
F
1 1
(1 +z +n; 1 +z; -N
1
(r))
N
n=0
_E
k
1+x-L
L
(X
|N
2
(r))
L
=1 k
1
+k
2
++k
L
=n
(25)
F
X
(r)
1
(z + 2)
1
(r)
x+1
lim
N-
F
1 1
(1 +z +n; 2 +z; -
1
(r))
N
n=0
_E
k
1+x-L
L
(X
|
2
(r))
L
=1 k
1
+k
2
++k
L
=n
(26)
249
E
k
1+x-L
L
(X
|) =
(-1)
k
k!
n
k=0
_
In +z +1
I
-1
n -k
_
k
E(P
k
) (32)
where
P
(r|
n
) is given in (17).
In Theorem 1, if the arbitrary parameter and functions[i.e, z,
K
1
(s) and K
2
(s)] are selected as,
z = I -1 (33)
K
1
(s) =
2s
=1
L
E(X
)
I
(34)
K
2
(s) =
I
2
=1
L
E(X
)
(35)
where K
1
(s)K
2
(s) = s, then the MGF of the sum of mutually
independent Weibull RVs is given as a simplification of (24) in
the following Corollary I.
Corollary I (MGF for the Sum of Mutually Independent Weibull
RVs): Let P
1
, P
2
, , P
L
be the mutually independent and non-
identically distributed Weibull channel power fading random
variables, {P
~W(
)]
1
L
, then the MGF, H
P
(s) of P = P
1
+
P
2
++P
L
is given by
M
P
(s) lim
N-
1
n
s
n
(1 +
1
s)
n+L
N
n=0
_E
k
(X
| 1
1
)
L
=1 k
1
+k
2
++k
L
=n
(36)
where
1
= 2 (
L
=1
) I .
As a double check for the analytical accuracy of (36), the
corresponding MGF, M
P
(s) is depicted in Figure 1 in which the
numerical and simulations results show perfect agreement for
N = Su. Furthermore, as seen in Figure 1, increasing the number
of Weibull RVs, I > 1, increases the gradient of the MGF, which
means in communications perspective that the performance of the
transmission in wireless channels increases. When I - , the
MGF approximates into
M
P
(s) ]
1, s = u
u, s = u
, I - (37)
which means that there occurs no error at the receiver.
In addition, in Theorem 2, if the arbitrary parameter and
functions are selected as follows,
z = I - 1 (38)
N
1
(r) =
I
2 E(X
)
L
=1
r (39)
N
2
(r) =
I
2 E(X
)
L
=1
(40)
where N
1
(r) N
2
(r) = r, then the PDF of the sum of mutually
independent Weibull RVs is given as a simplification of (25) in
the following Corollary II.
Corollary II (PDF for the Sum of Mutually Independent Weibull
RVs): Let P
1
, P
2
, , P
L
be the mutually independent and non-
identically distributed Weibull channel power fading random
variables, {P
~W(
)]
1
L
, then the PDF,
P
(r) of P = P
1
+
P
2
++P
L
is given by
P
(r)
[
r
1
(I)
lim
N-
F
1 1
_n +I; I; -
r
1
]
N
n=0
_E
k
[X
_
1
L
=1 k
1
++k
L
=n
(41)
where
1
= 2 (
L
=1
) I .
The PDF given above is depicted in Figure 2 for the analytical
accuracy. Furthermore, the summation limit, N in (41) is chosen
50 (N = Su). Although N = Su is small, our analysis and
simulations are in agreements as seen in Figure 2. Furthermore,
one can readily see that increasing the number of RVs shapes the
PDF around
L
=1
, which means in communications
perspective that the effects of fading has been alleviated.
As following the same procedure in Corollary II, if the
arbitrary parameter and functions in Theorem 3 are
z = I -1 (42)
1
(r) =
Ir
2 E(X
)
L
=1
(43)
Figure 1. MGF of the sum of the Weibull RVs (N = 5).
Figure 2. PDF of the sum of the Weibull RVs (N = 5).
250
2
(r) =
I
2 E(X
)
L
=1
(44)
where
1
(r)
2
(r) = r, then we can simplify (26) as in the
following Corollary III.
Corollary III (CDF for the sum of mutually independent Weibull
RVs): Let P
1
, P
2
, , P
L
be the mutually independent and non-
identically distributed Weibull channel power fading random
variables, {P
~W(
)]
1
L
, then the CDF, F
P
(r) of P = P
1
+
P
2
++P
L
is given by
F
P
(r) =
[
r
L
I!
lim
N-
F
1 1
_n + I; I +1; -
r
1
]
N
n=0
_E
k
(X
| 1
1
)
L
=1 k
1
+k
2
++k
L
=n
(48)
where
1
= 2 (
L
=1
) I .
As seen in Figure 3, the CDF given in (48) is depicted for the sum
of several numbers of Weibull RVs, Furthermore, as seen in
Figure 3, our analysis and simulations are in agreement.
4. PERFORMANCE MEASURES FOR
MRC OVER WEIBULL CHANNEL
FADING
Having the MGF and CDF of the sum of channel power Weibull
RVs, we can compute the SER and OP performance of MRC
diversity systems over Weibull fading channels. More
specifically, using (36) based-on the MGF-based approach for the
performance evaluation of digital modulations over fading
channels [1], the SER computation of MRC diversity systems can
be found for various M-ary modulations [such as M-ary phase
shift keying (MPSK), M-ary differential phase shift keying
(MDPSK), and M-ary quadrature amplitude modulation
(MQAM)] as in (45), in which , E
d
, 0
d
, v
d
,
d
and
d
are the
performance parameters for M-ary modulation which are given in
Table I. Furthermore, the performance expression given by (45)
reduces [1] to (46) for coherent binary DPSK (BDPSK)
modulations.
Table 1. Parameters for specific modulations
M-ary
Modulation
Schemes
E
d
v
d
d
d
0
d
MPSK 1
1
n
0 -
1
2
sin
2
[
n
H
nH-n
H
MDPSK 1
2
n
1
+cos [
n
H
cos [
n
H
sin
2
[
n
H
nH-n
H
MQAM 2
4
n
_1 -
1
H
] 0 -
1
2
S 2
H- 1
n
2
-
4
n
_1 -
1
H
]
2
0 -
1
2
S 2
H- 1
n
4
MPAM 1
2H-2
nH
0 -
1
2
S
H
2
-1
n
2
In addition, if the power of available channels of MRC diversity
systems are exponentially decaying with p, (i.e.
0
exp(-p), = 1,2, , I -1), then the parameter
1
in (45)
and (46) can be given by
1
=
2
0
I
1 -exp(-pI)
1 - exp(-p)
(49)
If exponentially decaying rate, p is zero, then
1
is given as
1
= 2
0
I (50)
Based on (46), Figure 4 compares the bit error rate (BER)
performances of MRC diversity systems over Weibull channels as
a function of the number of available diversity paths. As seen in
Figure 4, increasing the number of available channels decreases
the probability of an error in communications. In that case, using
(48), we can expresses the OP of MRC diversity systems as in
(47). In Figure 5, the OP performance for MRC diversity system
over Weibull fading channels is depicted. Similar to the results
obtained from Figure 4, it is clear from Figure 5 that the
performance improves as the number of the diversity paths
increase, which is as expected.
P
SLR
= lim
N-
`
1
1
E
d
_
_
1
d
(v
d
-2
d
sin
2
0)
]
n
_1 +
1
d
(v
d
-2
d
sin
2
0)
]
n+L
J0
0
d
0
d=1
1
1
1
N
n=0
_E
k
[X
_
1
L
=1 k
1
+k
2
++k
L
=n
(45)
P
BLR
= lim
N-
(
1
)
n
(1 +
1
)
n+L
N
n=0
_E
k
[X
_
1
L
=1 k
1
+k
2
++k
L
=n
(46)
P
0P
=
1
I!
_
r
1
]
L
lim
N-
F
1 1
_n +I; I + 1; -
r
1
] _E
k
[X
_
1
L
=1 k
1
++k
L
=n
N
n=0
(47)