You are on page 1of 24

1

A Comprehensive Performance Evaluation of a


DF-Based Multi-Hop System Over α − κ − µ
and α − κ − µ-Extreme Fading Channels
Tau Raphael Rasethuntsa, Sandeep Kumar† , Manpreet Kaur?
arXiv:1903.09353v1 [cs.IT] 22 Mar 2019

Abstract

In this work, an integrated performance evaluation of a decode-and-forward (DF) multi-hop wireless


communication system is undertaken over the non-linear generalized α − κ − µ and α − κ − µ-Extreme
fading models. Analytical formulas for the probability density function (PDF) and the cumulative
distribution function (CDF) of the received signal-to-noise ratio (SNR) as well as its generalized moments
and moment generating function (MGF) are derived. Based on the derived PDFs, novel closed-form
expressions for traditional performance metrics such as amount of fading (AF), outage probability (OP),
bit error rate (BER) under coherent and non-coherent modulation schemes as well as channel capacity
under various adaptive transmission techniques are derived. Additionally, asymptotic analyses of BER
based on Poincare series expansions of SNR PDFs are carried out and results show good approximations
for low SNR regimes. The correctness of the proposed solutions has been corroborated by comparing
them with Monte Carlo simulation results.

Index Terms

Multi-hop system, decode-and-forward relay, bivariate Fox H-function, non-linear generalized fad-
ing, Marcum Q-function, Nuttall Q-function

Tau Raphael Rasethuntsa is based in, Lenasia Extension 9, Gauteng Province, Johannesburg, South Africa and Ha
Abia/Mahlabatheng, Moreneng, Maseru, Lesotho, e-mail: tauzand@gmail.com
Sandeep Kumar and Manpreet Kaur are with the Central Research Laboratory, Bharat Electronics Limited, Ghaziadbad, Uttar
Pradesh, India, e-mail: sann.kaushik@gmail.com† , manpreettiet@gmail.com?
2

I. I NTRODUCTION

The use of multi-hop communication systems as a way of increasing network coverage area
and energy-efficiency with high data rates is gaining importance in next-generation wireless
networks. In a multi-hop wireless communication system, information is transmitted from a
source device to a destination device through several intermediate nodes, referred to as relays.
Based on the nature and complexity of the relaying system, multi-hop communication systems
can be classified into two categories, namely (i) decode-and-forward (DF) and (ii) amplify-and-
forward [1]. For a DF-based multi-hop system, the relaying node receives the encoded signal,
decodes it to regenerate the original symbol and forwards the newly encoded signal to the next
node [2]. DF-based multi-hop systems have several applications in ad-hoc networks, sensor
networks and microwave links among many others [3].
The performance analysis of multi-hop communication systems over numerous fading distribut-
ions has been well documented in the literature [3], [4], [5], [6], [7]. The ergodic capacity of a
multi-hop relaying system over Rayleigh fading channels has been studied in [4]. The end-to-
end performance of a DF relaying system based on Ricean and Nakagami-m fading channels
was explored in [5]. Yang et al. [8] studied the unified performance analysis of a DF-based
multi-hop system based on second order statistics in average outage duration (AOD) and level
crossing rate (LCR) over Nakagami, Rayleigh and Rician fading channels. The performance
study of a multi-hop communication system with regenerative relays has been investigated over
generalized η − µ, κ − µ and α − µ fading channels in [6], [7] and [9], respectively. Using
traditional performance metrics as well as LCR and amount of fade duration (AFD), Cao et
al. [3] presented the performance analysis of end-to-end wireless links over the Generalized-K
fading channel.
The modeling of mobile radio channel is a challenging task as the wireless signal is subjected
to various hurdles/corruptions in its propagation path from the transmitter to the receiver [10].
Accurate characterization of the wireless channel is therefore of paramount importance for
the realistic assessment of wireless systems performances. Several statistical distributions were
developed in the literature to model small-scale fluctuations of the fading channel envelope [11].
For instance, motivated by the pioneering works in [12] and [13], the non-linear generalized
α − κ − µ and α − κ − µ-Extreme fading models were proposed in [14], [15] and [16]. The
α − κ − µ distribution is a very flexible model that does not assume homogeneous scattering and
3

can account well for non-linearity of multi-path fading [17]. The model includes as special cases
several other well-known distributions. In particular, when α = 2, the α − κ − µ distribution
reduces to the κ − µ distribution [12] and for κ → 0, the α − µ is obtained [7]. The α − κ − µ-
Extreme fading model is derived from the α − κ − µ distribution by allowing µ → 0 and κ → ∞
in order to better characterize small-scale variations of mobile radio propagation under non-linear
severe multi-path fading effects usually encountered in enclosed environments [15]. For α = 2,
the α − κ − µ-Extreme distribution reduces to a linear severe fading model, the κ − µ-Extreme
distribution introduced by Rabelo et al. [13]. Although there are many works related to the study
of DF-based multi-hop relaying systems over different fading channels, none of the previous
works have investigated the end-to-end performance over α−κ−µ and α−κ−µ-Extreme fading
channels. Motivated by this void, we have derived the closed-form mathematical expressions for
amount of fading (AF), outage probability (OP), bit error rate (BER) (coherent and non-coherent)
and channel capacity for ene-to-end DF relaying system under various adaptive transmissions
over α−κ−µ and α−κ−µ-Extreme fading channels. The expressions derived here are presented
in a simple manner to optimize clarity and readability. Furthermore, the results produced here
are generalized expressions and valid for other fading distributions as special cases. The rest of
the paper is organized as follows : The system and channel model are described in Section II. In
Section III, we provide the closed-form expressions for the probability density function (PDF),
cumulative distribution function (CDF) generalized moments and moment generating function
of the end-to-end signal-to-noise ratio (SNR). Different performance metrics for the DF-based
multi-hop wireless system over α − κ − µ and α − κ − µ-Extreme fading channels are presented
in Section IV. Asymptotic analysis of BER based on Poincare expansions of fading model
(PDFs) is carried out in Section V. Performance evaluation, simulation results and discussions
are provided in Section VI. Finally, we round up the paper in Section VII with conclusions and
possible future considerations.

II. M ODEL F ORMULATION AND D ESCRIPTION

A. System and Channel Models

Consider an n-hop system shown in Figure 1 transmitting information from the source S to
the destination D via n − 1 intermediate relays R1 , R2 , . . . , Rn−1 in time-sharing principles. If
the SNR of the i-th hop is denoted by γi , with PDF fγi , i = 1, . . . n, the statistical properties
4

1 2  n 1 n

S R1 R2 Rn-1 D

Relays

Figure 1. A n-link DF-based multi-hop system.

of the end-to-end system SNR are then determined by the SNR of the last hop γn . According
to [18], the PDF of the end-to-end system SNR γ can be written as

fγ (γ) = Aδ(γ) + Āfγn (γ), (1)

where A denotes the probability of the first outage occurring for the first n − 1 hops, Ā = 1 − A

and δ(·) is the Dirac delta function. The probability A is given by [18]

A = 1 − P r (γ1 > γth , γ2 > γth , · · · , γm−1 > γth )


m−1
Y m−1
Y (2)
=1− P r (γn > γth ) = 1 − [1 − Fγn (γth )] ,
k=1 k=1

where Fγn is the CDF of γn .

1) The α − κ − µ and α − κ − µ-Extreme fading models: The PDF of the instantaneous SNR
of the n-th hop γn > 0 following the αn − κn − µn distribution is given by
αn
q 
( α2n )ωn −1 −σn γn2 αn
fγn (γn ) = Kn γn e Iµn −1 θn γn2 , (3)

where, in order to improve clarity in presentation, we have set

αn σnωn e−µn κn µn + 1 µn (κn + 1)


Kn = , ω n = , σ n = αn ,
2(µn κn )ωn −1 2 γ̄n2 (4)
θn = 4µn κn σn
with αn , µn , κn > 0. The average SNR per symbol γ̄n is given by

2
eµn κn [µn (κn + 1)] αn E
γ̄n =     b,
Γ 2
αn
+ µn Φ̃ α2n + µn ; µn ; µn κn N0

where Eb denotes the energy per symbol and N0 is the single-sided power spectral density. Using

the method of [19], one can easily show that the CDF of γn is written as
5

 q 
p αn
Fγn (γn ) = 1 − Qµn 2κn µn , 2σn γn2 , γn > 0, (5)

where Q(·, ·) is the generalized Marcum Q-function. The PDF of the instantaneous SNR of the

n-th hop γn > 0 following the αn − κn − µn -Extreme distribution is given by


α /4−1
an γn n e−2mn
q  r 
αn γn
gγn (γ) = αn I1 cn γn2
+ √ δ , (6)
ebn γn
2 2 γn γ̄n γ̄
where
αn mn e−2mn 2mn (16m2n )
an = α /4
,b = αn , cn = αn , (7)
γ̄n n γ̄n2 γ̄n2
with αn , µn , κn > 0 and mn = µn (κn + 1)2 /(2κn + 1) as the Nakagami parameter which is

inversely proportional to the fading severity. The average SNR per symbol is given by
2
e2mn (2mn ) αn −1 E
γ̄n =     b.
Γ 2
αn
+ µn Φ α2n + µn ; µn ; µn κn N0

The CDF of the α − κ − µ-Extreme fading model can also be derived using the method of [19]

resulting in
 

q
α
Gγ (γ) = 1 − Q0 2 mn , 2 mn (γ/γ̄) 2 . (8)

The symbols in (4) and (7) will be employed throughout the paper unless the original notation

produces shorter expressions.

Remark 1. It is also worth noting that the proposed PDFs representations renders, purely on
a symbolic level, the α − κ − µ-Extreme fading model PDF without the extra δ(·) term as a
special case of the PDF of the α − κ − µ model in a sense that Kn = an , ωn = 1/2, σn = bn ,
θn = cn and µn = 2. This observation will be particularly useful for reducing redundancy in the
forthcoming derivations.

III. S TATISTICAL P ROPERTIES

A. PDF and CDF of End-to-End SNR

Case I : Assume that the SNRs of the n hops are independent and not necessarily identically
distributed (INID) random variables following the α − κ − µ distribution. Then we have from
(3) and (5) that the PDF and CDF of the end-to-end SNR γ must be given respectively by
6

αn
)ωn −1
γ(
q 
2 αn
pγ (γ) = Aδ(γ) + Kn Ā αn Iµn −1 θn γ 2 (9)
eσn γ 2

 q αk

Qn−1
where A = 1 − k=1 Qµk 2κk µk , 2σk γth2 and

 q 
p αn
Pγ (γ) = 1 − ĀQµn 2κn µn , 2σn γ 2 . (10)

Case II : On the other hand if the SNRs of the n hops are INID random variables following the

α − κ − µ-Extreme distribution, then we have from (6) and (8) that the PDF of the end-to-end
SNR γ must be given by
αn
−1
 q  
an γ 4 αn

peγ (γ) e
= A δ(γ) + Ā e
αn I1 cn γ 2 + δ (γ) (11)
ebn γ 2

e−2mn
r 
∗ γn
whereδ (γ) = √ δ
2 γγ̄n γ̄
 

q
Qn−1 αk
e
and A = 1 − k=1 Q0 2 mk , 2 mk (γk /γ̄k ) 2 . The CDF of γ can be written as
 

q
αn
Pγe (γ) = 1 − Ā Q0 e
2 mn , 2 mn (γ/γ̄) 2 . (12)

B. Moments of End-to-End SNR

Generalized moments of the end-to-end SNR are useful for obtaining r-th-order performance
metrics such as the mean, variance as well as kurtosis and skewness of the distribution of γ.
The r-th moment of a positive random variable X with density fX is defined by
Z ∞
r
EhX i , xr fX (x)dx (13)
0

where Eh·i denotes the expectation operation. For Case I, substituting (9) into (13) and setting

y = γ α2 /2 , the r-th moment of γ can be written with the aid of [20, Eq. 6.643(2)] and [20, Eq.
6.220(2)] as
   
2r 2r
Āγ̄nr Γ αn
+ µn Φ̃ αn
+ µn ; µn ; µn κn
Ehγ r i = 2r (14)
eµn κn [µn (κn + 1)] αn
where Φ̃ is the regularized confluent hypergeometric function [20, Eq. 9.210(1)]. Similarly, for

Case II, we have the r-th moment of γ as


   
Āe γ̄nr Γ α2rn + 1 Φ α2rn + 1; 2; 2mn
Ee hγ r i = 2r (15)
e2mn (2mn ) αn −1
7

It is also worth noting that as expected, upon substituting r = 1 in (14) and (15), one obtains

γ̄ = Āγ̄n and γ̄ = Āe γ̄n for each case, respectively.

C. Moment Generating Function

Another useful statistical characteristic of random variables is the moment generating function
(MGF) used to obtain n-th order moments or BER under non-coherent modulation, see Section
IV. The MGF of γ, denoted by Mγ , is defined as
Z ∞
Mγ (s) , e−sγ fγ (γ)dγ. (16)
0
For Case I, we substitute (9) into (16) to get the integral
I
z }|1   {
Z ∞ ( α2n )ωn −1
γ
q
αn
Mγ (s) = A + Kn Ā αn Iµn −1 θn γ 2 dγ (17)
0 esγ eσn γ 2
αn /2
In order to evaluate I1 , we set y = γ and employ [21, Eq. 1.25] and [22, Eq. 8.4.22(1)] as

well as the relationship between the Meijer-G and the Fox-H functions as
 
Z ∞


1,0  α2n
y ωn −1 e−σn y H0,1

I1 =π s y  
αn
0 0, 2
  (18)
µn

1,0  θn
2
,1
× H1,3 y   dy
4 (ωn − 1, 1), (1 − ωn , 1), µn , 1
2

Now, expanding the two Fox H-functions in terms of their definition as given in [23, Eq. 1.1.1]

and reversing the order of integration, we have that


G
Z Z zZ ∞
{ }|
1 ωn +r+t−1 −σn y
I1 =π y e dy
(2πi)2 Lr Lt 0 (19)
αn r t
αn θ
  
Γ − 2 r Γ(ωn − 1 − t) s 2 n
4 
drdt
× µn
 µn
Γ(ωn + t)Γ 1 − 2 + t Γ 2 − t
where Lr and Lt are two suitable contours in the complex r and t planes respectively. The

integral G can be evaluated with the aid of [20, Eq. 3.326(2)] to yield
Γ − α2n r Γ(ωn − 1 − t)
Z Z   
π
I 1 = ωn
σn (2πi)2 Lr Lt Γ(ωn + t)
αn r  θn t (20)
Γ (ωn + r + t) s 2 /σn 4σn
drdt
×
Γ 1 − µ2n + t Γ µ2n − t
 
8

Table I
MOMENT GENERATING FUNCTOINS

αn
 
s 2
 σ ∆ (αn , µn , ωn ) 
π 0,1:1,0;1,0  n 
Mγ (s) = A + ĀKn ωn H1,0:1,0;1,3   (21)
σn  
 θn ∇ (αn , µn , ωn ) 

4σn
 αn 
s 2
1

 bn ∆ αn , 2, 2 
e e e an π 0,1:1,0;1,0 
Mγ (s) = A + Ā √ H1,0:1,0;1,3  . (22)


bn  c
n ∇ αn , 2, 1
 
2
4bn

µ 
n
∆ ≡ (1 − ωn ; 1, 1) : − ; ,1
2
 α  µ 
n n
∇ ≡ − : 0, ; (ωn − 1, 1), (1 − ωn , 1), ,1
2 2

where <(ωn +r +t) > 0. Now comparing (20) with the definition of the bivariate Fox H-function

given in [21, Eq. 2.57] yields a closed-form expression for the MGF as shown in (21) of Table I.
Remark 1 suggests that the MGF for Case II can be obtained from (21) by inspection resulting
in the MGF for Case II being as shown in (22) of Table I.

IV. P ERFORMANCE A NALYSIS OF M ULTI -H OP DF-BASED R ELAY S YSTEMS

A. Amount of Fading

The second-order AF is a useful performance metric that can be used to account for the
severity of fading. AF is defined as the ratio of the variance to the square average SNR per
(2)
symbol, AFγ , Var(γ)/γ̄ 2 . That is,

Ehγ 2 i − Ehγi2
AFγ(2) , . (23)
Ehγi2
(n)
This can be generalized to higher-order fading as AFγ , Ehγ n i/(Ehγin ) − 1. Substitution of

Ehγi and Ehγ 2 i from (14) into (23) yields after some basic simplifications the second-order AF
of γ under Case I as
9

   
µn κn 4 4
e Γ αn
+ µn Φ̃ µn + αn
; µn ; κn µn
AFγ =  2  2 − 1 (24)
2
ĀΓ αn
+ µn Φ̃ µn + α2n ; µn ; κn µn
Likewise, for Case II, the second-order amount of fading can be derived using (15) in (23)

resulting in
   
2mn 4 4
e Γ
+1 Φ + 1; 2; 2mn
αn αn
AFγe = 2  2 − 1 (25)
e 2 2
Ā 2mn Γ αn + 1 Φ αn + 1; 2; 2mn

B. Outage probability

The OP is defined as the probability that the SNR per symbol falls below a certain threshold
γth . By employing (10) and (12), this probability can be written for Case I as

Pout = P (γ ≤ γth )
 q  (26)
αn /2
p
= 1 − ĀQµn 2κn µn , 2σn γth .

Similarly, for Case II, we have that OP can be written as


 

q
e e α /2
Pout = 1 − Ā Q0 2 mn , 2 mn (γth /γ̄) n . (27)

C. Bit Error Rate

1) Coherent Modulation Schemes: For a given SNR γ, the BER for a variety of coherent
modulation schemes can be obtained by averaging the PDF fγ over the conditional additive
white Gaussian noise (AWGN) BER given as [24, Eq. 17]
 p 
Pb (e|γ) , (φ/2)erfc ρ γ/2 (28)

where erfc(·) is the complementary error function [20, Eq. 8.250(4)]. The constants φ and ρ vary

depending on the type of modulation scheme. For BFSK, φ = 1 and ρ = 1. For BPSK, φ = 1

and ρ = 2. For 4-QAM and QPSK, φ = 2 and ρ = 1. Finally, for M-PAM, φ = 2(1 − 1/M )
p
and ρ = 6/(M 2 − 1). Thus, for Case I we average (9) over (28) to obtain the integral

φ ∞
Z  p 
Pb = pγ (γ)erfc ρ γ/2 dγ. (29)
2 0
Substituting (9) into (29), one has that
φA ∞ φKn Ā
Z  p 
Pb = δ(γ)erfc ρ γ/2 dγ + I2 (30)
2 0 2
10

Table II
BER UNDER COHERENT MODULATION

√ √ !ωn
φ φ π ( 2)αn
Pb = A + ĀKn
2 2 ραn

( 2)αn
 
 σn ραn Φ(αn , µn , ωn )  (32)
0,2:1,0;1,0 
 
× H2,1:0,1;1,3  √ α


 ( 2) n Ψ(αn , µn , ωn ) 
θn
4ραn

√ √ α ! 12
φ φ π ( 2) n
Pbe = Ae + an Āe
2 2 ρ αn
√   
( 2)αn

1
 bn ραn Φ αn , 2, 2  (33)
0,2:1,0;1,0 
 
× H2,1:0,1;1,3  √ α

  .

 ( 2) n 1 
cn Ψ α n , 2,
4ραn
2
 
 αn αn αn  1 αn αn αn µ 
n
Φ≡ 1− ωn ; , , − ωn ; , : −; ,1
2 2 2 2 2 2 2 2
 α αn αn  µ 
n n
Ψ ≡ − ωn ; , : (0, 1) ; (ωn − 1, 1), (1 − ωn , 1), ,1
2 2 2 2

where

γ (αn /2)ωn −1
Z q   r 
αn γ
I2 = Iµn −1 θn γ 2 erfc ρ dγ. (31)
0 eσn γ αn /2 2
Following the same method as in Section III, using [22, Eq. 8.4.14(2)] to express the comple-

mentary error function in terms of its Mellin-Barnes representation, we obtain after some simple
mathematical manipulations the BER under coherent modulation schemes for Case I and Case
II as shown in (33) and (33) of Table II, respectively.
2) Non-Coherent Modulation Schemes: Under a range of different non-coherent modulation
schemes, the BER of the end-to-end SNR γ can be written as [24, Eq. 27]
Z ∞
Pbn , φ e−ργ fγ (γ)dγ. (34)
0

The values of φ and ρ also vary depending on the modulation scheme. Specifically, for BFSK,
11

1
φ= 2
and ρ = 12 . For DBPSK, φ = 1
2
and ρ = 1 and for M -FSK φ = (M − 1)/2 and ρ = 12 .
Upon substituting (9) into (34), it is clear that the resulting integral is of the form of the MGF
integral given in (17). It is therefore straightforward to show that

Pbn = φMγ (ρ). (35)

Similarly, the BER for non-coherent modulation schemes under Case II, can be derived using

(22) resulting in
e
Pbn = Pbn = φMeγ (ρ). (36)

D. Channel Capacity

In this section, we investigate the capacity of α−κ−µ and α−κ−µ-Extreme fading channels
under various adaptive transmission schemes.
1) Optimal Rate Adaptation: The channel capacity of the received ene-to-end SNR γ with
density function fγ (γ) under ORA (also known as ergodic capacity) is obtained by averaging
the capacity of an additive AWGN C O = B log2 (1 + γ) over fγ (γ), where B is the bandwidth
of the channel. For a n-hop DF-based multi-hop system, we have
Z ∞
B
CO = ln(1 + γ)fγ (γ)dγ. (37)
n ln 2 0
For Case I, we substitute (9) into (37) and set y = γ αn /2 to obtain

I
z  }|3
 {
2

nC O Āαn Kn
Z ∞ ln 1 + y αn p 
= Iµn −1 θn y dy (38)
B ln 4 0 y 1−ωn eσn y
Proceeding as before, making routine use of the definition of the Fox H-function of a single

variable from [23, Eq. 1.1.1] yields the double Mellin-Barnes integral
h   i
Z Z Γ ωn + 2 r + t Γ(ωn − 1 − t)
1 αn
I3 =
Γ 1 − µ2n + t Γ µ2n − t
2
 
(2πi) Lr Lt
  α2r  t (39)
n
πΓ(1 − r)Γ(r)Γ(r) σ1n θn
4σn
drdt
×
σnωn Γ(1 + r)Γ(ωn + t)Γ µ2n − t


where Re (ωn + 2r/αn + t) > 0. Comparing (39) with [21, Eq. 2.57] results i the closed-form

expression for C O under Case I as shown in (40) of Table III. Using Remark 1, we obtain by
12

Table III
ORA CHANNEL CAPACITY

   α2 
1 n


σn
Ω(αn , µn , ωn ) 
B Āπαn Kn 0,1:1,2;1,0  
CO = H   (40)
2nσnωn ln 2 1,0:2,2;1,3  
 θn
Π(αn , µn , ωn ) 
4σn

   α2   
1 n 1
Ω αn , 2,
B Āe παn an 0,1:1,2;1,0 

bn

2  
C eO = √ H1,0:2,2;1,3   (41)
2n bn ln 2 
 cn


  
1 
Π αn , 2,
4bn 2
 
2 µ 
n
Ω ≡ 1 − ωn ; , 1 : (1, 1), (1, 1); ,1
αn 2
µ 
n
Π ≡ − : (1, 1), (0, 1); (ωn − 1, 1), (1 − ωn , 1), ,1
2

inspection from (40) the ORA capacity for Case II as in (41) of Table III. Using integration by
parts, [25] showed that (37) has the alternative representation
Z ∞
B [1 − Fγ (γ)]
CO = dγ. (42)
n ln 2 0 1+γ
Therefore, substituting (10) and (12) into (42) yields expressions for channel capacity with ORA

under Case I and Case II respectively as


Z ∞  
B Ā
q
p αn
CO = Qµn 2κn µn , 2σn γ 2 /(1 + γ)dγ, (43)
n ln 2 0


B Āe
 

Z q
αn
C eO = Q0 2 mn , 2 mn (γ/γ̄) 2 /(1 + γ)dγ. (44)
n ln 2 0

2) Optimal Power and Rate Adaptation: Under OPRA, the channel capacity of the end-to-end
SNR γ of a n-link DF-based multi-hop system with density fγ is defined as [25, Eq. 10]

B ∞
Z  
γ
CP , log2 fγ (γ)dγ
n γo γo
Z ∞ (45)
γo B
= ln(y)fγ (γo y)dy, upon setting γ = γo y.
n ln 2 1
13

The optimum cutoff γo must satisfy the following relation


Z ∞
(1/γo − 1/γ) fγ (γ)dγ = 1. (46)
γo

In order to evaluate (45) for Case I, we express the PDF (9) in terms of the Fox H-function as

was done in Section III to obtain after some simple manipulations the integral
H
z}|  {
 2
Z ∞ ln y αn
Z Z
1
C P =K∗ dy
(2πi)2 Lr Lt 1 y 1−ωn −r−t (47)
 αn  r  αn  t
θn
Γ(−r)Γ(ωn − 1 − t) σn γo 2 γ 2 drdt
4 o
×
Γ(ωn + t)Γ 1 − µ2n + t Γ µ2n − t
 
 αn 
∗ ωn
where K = 2γo 2
πBKn /(αn n ln 2Ā−1 ). The integral H can be evaluated using integration

by parts as H = 2/ [αn (ωn + r + t)2 ] resulting in the double Mellin-Barnes contour integral

2K∗ 1 Γ(−r)Γ(ωn + r + t)2


Z Z
CP =
αn (2πi)2 Lr Lt Γ(1 + ωn + r + t)2
 αn  r 
θn
αn  t (48)
Γ(ωn − 1 − t) σn γo 2 4
γo
2
drdt
×
Γ(ωn + t)Γ 1 − µ2n + t Γ µ2n − t
 

where <(ωn + r + t) < 0. Finally, comparing (48) with [21, Eq. 2.57] gives a closed-form

expression for the OPRA channel capacity under Case I and Case II as shown in (49) and
(50) of of Table IV, respectively. Applying integration by parts in (46) results in the alternative
representation for the cutoff SNR relation as [25, Eq. 13]
Z ∞
[1 − Fγ (γo )] /γo − fγ (γ)/γdγ = 1
γo
(51)
1 − Fγ (γo )
=⇒ γo = R∞ = ξ(γo ).
1 + γo fγ (γ)/γdγ
It is clear that the solution to (51) is a fixed point of the non-linear equation γo = ξ(γo ). According

to [25], γo will always lie in the interval [0, 1] regardless of the fading model and number of
relays used for the wireless system. The explanation for the latter relies on the properties of the
CDF of γ and is fully documented in Section 3 of [25]. Thus, choosing a starting point in [0, 1],
γo can be computed via iterative schemes such as the Newton-Raphson method as

γoi = ξ(γoi )
14

Table IV
OPRA CHANNEL CAPACITY

 αn 
αn
ωn σn γo 2
♠(µn , ωn )
4γo 2
πBKn 0,2:1,0;1,0  
CP = G   (49)
αn2 n ln 2Ā−1 2,2:0,1;1,3  θn α2n
♣(µ , ω )

γo n n
4

   
αn

♠ 2, 1
e 4
αn  bn γo 2
2 
e 4π Ā γo Ban 0,2:1,0;1,0  
CP = G2,2:0,1;1,3   (50)
αn2 n ln 2  c n αn 
1
 
 γo 2
♣ 2,

4 2
µ 
n
♠ ≡ (1 − ωn ; 1, 1), (1 − ωn ; 1, 1) : −; ,1
2 µ 
n
♣ ≡ (−ωn ; 1, 1); (−ωn ; 1, 1) : (0, 1); (ωn − 1, 1), (1 − ωn , 1), ,1
2

where γoi is the i-th iterate of γo . The iterative procedure will be stopped when |γoi+1 − γoi | is

sufficiently small. We also note again that using integration by parts on (45) yields [25, Eq. 11]

Z ∞
B
CP = [1 − Fγ (γ)] /γdγ. (52)
n ln 2 γo

Therefore, we can use (10) and (12) in (52) to obtain the OPRA channel capacities for Case I

and Case II respectively as


∞  
B Ā
Z q
p αn
CP = Qµn 2κn µn , 2σn γ 2 )/γdγ, (53)
n ln 2 γo


B Āe
 

Z q
αn
C eP = Q0 2 mn , 2 mn (γ/γ̄) 2 /γdγ (54)
n ln 2 γo

Numerical integration routines from standard packages can be employed to evaluate (53) and

(54).
3) Channel Inversion With Fixed Rate: Under this technique, the channel capacity of the
end-to-end SNR γ of a n-link DF-based multi-hop system with density fγ is given by
 
B 1
C c , log2 1 + R ∞ (55)
n 0
(fγ (γ)/γ) dγ
15

Since the expression obtained using (14) and (15) are highly restrictive, we can obtain more

general ones by employing a similar procedure to the one used in deriving the MGF function
of Section III. We only express the bessel function in each SNR PDF in terms of the Meijer
G-function and then reverse the order of integration to obtain after some simplifications the
single Mellin-Barnes integral
 
2
Z ∞
pγ (γ) 2πKn Ā 1
I Γ ωn − αn
+r
dγ = ωn − 2
0 γ αn σn αn 2πi Lr Γ(ωn + r)
 r (56)
θn
Γ(ωn − 1 − r) 4σ n
dr
×
Γ 1 − µ2n + r Γ µ2n − r
 

 
2
where Re ωn − αn
+ r > 0. Comparing the contour integral in (56) with [21, Eq. 1.1] yields

the CIFR channel capacity for Case I as


   −1 

nC c  2πKn Ā 1,1  θn N  
= log2 1 +  ωn − α2
G2,3 , (57)
B 
αn σ n n 4σ n H

where N ≡ 1 + − ωn , µ2n and H ≡ ωn − 1, 1 − ωn , µ2n . In an analogous fashion, the CIFR


2
αn
 
channel capacity for Case II can be derived with the contour condition Re 1/2 − α2n + r > 0
as
  −1 

nC ec e
2πan Ā 1,1  cn
 N 
= log2 1 +  1
− 2
G2,3   (58)
B 
αn bn2 αn 4bn H ∗ 

where N ∗ ≡ 1
2
+ 2
αn
,1 and H ∗ ≡ − 12 , 12 , 1 The Meijer G-function has been implemented in

MATHEMATICAr and MATLABr .


4) Truncated Channel Inversion With Fixed Rate: The channel inversion technique with fixed
rate may suffer large capacity penalties as compared to other techniques. To circumvent this
shortcoming, a modified inversion which inverts the channel fading above a predetermined
truncated fade γo is often used and is defined for n-link DF-based multi-hop systems as
!
B 1
C T , log2 1 + R ∞ P r (γ > γo ) . (59)
n f (γ)/γdγ
γo γ

Note that C T → C c from above as γo → 0. That is, the gap between C T and C c depends on

P r (γ > γo ) in (59). Making the substitution y = γ/γo , the integral in (59) can be re-written as
16

Table V
TIFR CHANNEL CAPACITY INTEGRAL J

αn
 
αn
σn γo 2 ♠∗ (µn , ωn )
 
ωn
−2π Ā γo Kn2

G0,1:1,0;1,0
 
J = 
1,2:0,1;1,3 
 (61)
αn γo θn α2n ♣∗ (µ , ω ) 
γo n n
4
   
α n
∗ 1
 bn γo 2 ♠ 2, 2 
 αn 
−2π Āe γo 4 an
G0,1:1,0;1,0
 
J = 
1,1:0,1;1,3 
 (62)
αn c α
 n γo 2 n 
1
 

♣∗ 2,
4 2
 
∗ 2 µ 
n
♠ ≡ 1+ − ωn ; 1, 1 : −; ,1
αn 2
   
∗ 2 2
♣ ≡ − ωn ; 1, 1 , − ωn ; 1, 1 : (0, 1); (ωn − 1, 1),
αn αn
µ 
n
(1 − ωn , 1), ,1
2

Z ∞ Z ∞
1
J = fγ (γ)/γdγ = fy (γo y)dy (60)
γo 1 y

Substituting (9) into (60) and following the simple procedure undertaken in deriving OPRA chan-
 
2
nel capacity yields The double Mellin-Barnes contour integral must satisfy Re ωn − αn + r + t <
0. Therefore, putting (61) and (62) into (59) gives the TIRF channel capacity for Case I and
Case II, respectively. Alternatively, for Case I, another closed-form expression for the integral
in (59) can be obtained by making use of the consecutive substitutions x = γ αn /2 followed by
√ √
x = u/ 2σn which yield
2  q 
Ā(2σn ) αn p αn
J = Qµ∗ ,µn −1 2µn κn , 2σn γo 2 (63)
2(2µn κn )ωn −1

where µ∗ = µn − 4
αn
, αn µn > 4 and QM,N (·, ·) is the Nuttall Q-function defined as [26, Eq.

2] with b, M, N ≥ 0 and a > 0. A similar procedure can be followed for Case II. However,
the resulting integral does not satisfy the M ≥ 0 condition and hence fails to attain a Nuttall
Q-function representation. The Nuttall Q-function is a generalization of the Marcum Q-function.
17

Specifically, when M = N + 1, QN +1,N (a, b) = aN QN +1 (a, b). The latter is clearly evident from
(63) in the case α = 2.

V. BER AND C APACITY A PPROXIMATIONS FOR L OW SNR R EGIMES

In this section, we explore the asymptotic behaviour of the BER by employing asymptotic
Poincare series expansions for both the α − κ − µ and α − κ − µ-Extreme fading models in
(9) and (11). Following the idea from [27] of using [20, Eq. 8.445] and [20, Eq. 0.316], we
have that the PDFs of the SNRs for the n-th hop of Case I and Case II fading channels have
Poincare asymptotic series expansion near the origin given respectively by
 ωn −1 N −1
θn X
dk γ tk + O γ t+N −1

˙ Aδ(γ) + ĀKn
pγ (γ) ∼ (64)
4 k=0

where O denotes the order of a term in the asymptotic series and

k k−j
X (−1)j σnj θ4n
dk = , tk = αn (k + µn )/2 − 1. (65)
j=0
j!(k − j)!Γ(k + µn − j)

√ N −1
e an cn X e te
peγ (γ) e
dk γ k + O γ t+N −1

˙ A δ(γ) + Ā
∼ (66)
2 k=0

where
k k−j
X (−1)j bjn c4n
dek = , te = αn (k + 1)/2 − 1. (67)
j=0
j!(k − j)!Γ(k + 2 − j) k

A. Asymptotic BER at low SNR levels

We will now attempt to derive asymptotic BER expressions based on the series expansions in
(64) and (66).
1) Coherent Modulation Schemes: Substituting (64) and (66) into (28) and employing [22,
Eq. 8.4.15(2)], we have that the asymptotic BER under coherent modulation for Case I and
Case II are given respectively by
ωn −1 N −1
2tk Γ 23 + tk
 
φ θn φ X
Pb ∼
˙ A + ĀKn √ dk , (68)
2 4 π k=0 (tk + 1)ρ2(tk +1)

√ N −1 tek 3 e

φ e e an c n e 2 Γ 2 + tk
X
Pbe ˙ A + φĀ √
∼ d . (69)
2 π k=0 k (tek + 1)ρ2(tk +1)
e
2
18

2) Non-coherent Modulation Schemes: Similarly, by substituting (64) and (66) into (34) and
employing [22, Eq. 8.4.3(1)], we have that the asymptotic BER under non-coherent modulation
for Case I and Case II are given respectively by
 ωn −1 N −1
θn X Γ(tk + 1)
Pbn ∼
˙ φA + φĀKn dk (70)
4 k=0
ρtk +1

√ N −1
e e e an cn X e Γ(tek + 1)
Pbn ˙ φA + φĀ
∼ dk tek +1
. (71)
2 k=0
ρ

VI. N UMERICAL E XPERIMENTS AND S IMULATIONS

In this section we have presented the graphical results of the performance metrics derived in the
preceding sections. All computations are carried out in MATLABr (version 2014a).Nonetheless,
based on linear optimization, [28] recently proposed versatile GPU-enabled MATLABr and
C/M EX codes with automated contour computation for the general multivariate Fox H-function
of the type in [21] and we use a modified version of it in the present work. Figure 2 and Figure
3 present the AF vs average SNR under severe and moderate fading conditions for both Case
I and Case II, respectively. It is evident that AF decreases with increments in values for any
of the fading parameters of both models. This makes sense since the parameter m is inversely
proportional to the fading severity. We also note that both models perform almost identically
for moderate fading with the extreme distribution showing a slightly lower AF. Furthermore,
AF under Case II stabilizes to a constant as SNR decreases indefinitely. While increasing the
number of hops appears to lower the AF for moderate fading conditions, for severe fading, there
is a steep increase in AF at low SNR with increases in signal power, the number of multi-clusters
µ or m having little to no positive effect. Figure 4 illustrates the OP vs average SNR curves
for Case I under moderate fading conditions. From the figure, we can deduce that increasing
any of the fading parameters values results in notable system performance enhancement with
steeper increases noted for increments in µ, the multi-cluster parameter. Increasing the number of
hops also improves system performance, especially at lower SNR levels which is a particularly
desirable trait. The increase in performance is albeit of a diminishing type. This implies that
there is an optimal number/range of hops that maximizes system performance. That is, relaying
for the sole purpose of improving outage is only fruitful up to a certain number of relays
upon which any additional relaying is futile. Figure 5 depicts the behaviour of BER under non-
19

8 90
,=2.0 5=3.5; 7=2.5 ,=2.0; 5=0.5; 7=0.3
,=2.5 5=3.5; 7=2.5 ,=2.5; 5=0.5; 7=0.3
7 ,=3.0 5=3.5; 7=2.5 80 ,=3.0; 5=0.5; 7=0.3
,=3.0 5=4.0; 7=2.5 n = 5 hops ,=3.0; 5=0.8; 7=0.3
,=3.0 5=4.0; 7=3.0 ,=3.0; 5=0.8; 7=0.5
,=3.0 5=4.0; 7=3. 70 ,=3.0; 5=0.8; 7=0.5
6
Simulation Simulation
60

Amount of Fading (AoF)


5

50
4
n = 3 hops
n = 3 hops 40

3
30
n = 5 hops
2
20

1 10

0 0
10 12 14 16 18 20 10 12 14 16 18 20
SNR (dB) SNR (dB)

Figure 2. AF curves for the α − κ − µ case showing the effects of variation in model parameter values and hop numbers under
moderate and severe fading conditions.

25 8
, = 2:0; m = 0:34 , = 2:0; m = 6:33
, = 2:5; m = 0:34 , = 2:5; m = 6:33
, = 3:0; m = 0:34 , = 3:0; m = 6:33
7
, = 3:0; m = 0:37 , = 3:0; m = 6:94
20 , = 3:0; m = 0:62 , = 3:0; m = 8:33
, = 3:0; m = 0:62 , = 3:0; m = 8:33
6
Amount of Fading (AoF e )

Simulation Simulation

5 n = 5 hops
15
n = 5 hops

4 n = 3 hops

10
3

n = 3 hops 2
5

0 0
-10 0 10 20 30 10 12 14 16 18 20
SNR (dB) SNR (dB)

Figure 3. AF curves for the α − κ − µ-Extreme case showing the effects of variation in model parameter values and hop
numbers under moderate and severe fading conditions similar to Case I.

coherent DBPSK modulation scheme for Case I in moderate fading conditions. Consistent with
observations from AF and OP analyses, there is notable system improvement with increments
in fading model parameter values and a diminishing system performance improvement with hop
number increase at low SNR levels. An additional system behavior is observed for high SNR
level which shows that there is a non-null BER even as signal power increases indefinitely. This
characteristic was observed by Rabelo et al. [13] for the linear κ−µ-Extreme fading model where
20

100
, = 2:5; 5 = 2:5; 7 = 2:5
, = 2:7; 5 = 2:5; 7 = 2:5
10 -1 n = 2 hops , = 3:0; 5 = 2:5; 7 = 2:5
, = 3:0; 5 = 3:0; 7 = 2:5
, = 3:0; 5 = 3:0; 7 = 3:0
10-2 , = 3:0; 5 = 3:0; 7 = 3:0; n = 3 hops

Outage Probability (OP)


, = 3:0; 5 = 3:0; 7 = 3:0; n = 4 hops
, = 3:0; 5 = 3:0; 7 = 3:0; n = 5 hops
10-3 Simulation

10-4

10-5

10-6

10-7

10-8
10 15 20 25 30 35 40 45 50
SNR (dB)

Figure 4. OP curves for the α − κ − µ case under moderate fading showing the effects of increments in hop number and fading
model parameter values.

they showed that as SNR increases, BER under non-coherent DPSK modulation approaches the
constant (1/2) exp(−2m).

100
, = 2:5; 5 = 2:5; 7 = 2:5
, = 2:7; 5 = 2:5; 7 = 2:5
, = 3:0; 5 = 2:5; 7 = 2:5
, = 3:0; 5 = 3:0; 7 = 2:5
10-1 , = 3:0; 5 = 3:0; 7 = 2:8
n = 2 hops , = 3:0; 5 = 3:0; 7 = 2:8; n = 3 hops
, = 3:0; 5 = 3:0; 7 = 2:8; n = 4 hops
, = 3:0; 5 = 3:0; 7 = 2:8; n = 5 hops
Bit Error Rate (BER) Pbn

10-2 Simulation

10-3

10-4

10-5

-6
10
5 10 15 20 25 30 35 40 45 50
SNR (dB)

Figure 5. BER curves for the α − κ − µ case under non-coherent DBPSK modulation showing the effects of increasing model
parameters and the number of hops and N in Poincare approximations for moderate fading conditions

Under similar fading conditions of Case I, Figure 6 presents BER curves under non-coherent
DBPSK modulation for Case II. We once more observe a consistent behaviour in terms of
parameter and hop number variation. Furthermore, BER levels are considerably lower as com-
21

100
, = 2:5; m = 6:33
n = 2 hops , = 3:5; m = 6:33
, = 4:0; m = 6:33
10-1 , = 4:0; m = 6:94
Poincare Approx. N = 1,2,3,4,5,6
, = 4:0; m = 9:72
, = 4:0; m = 9:72; n = 3 hops
-2
10 , = 4:0; m = 9:72; n = 4 hops
, = 4:0; m = 9:72; n = 5 hops

Bit Error Rate Pbn


Simulation

e
10-3 5

3
10-4

10-5

1
10-6

10-7

10-8
5 10 15 20 25 30 35 40 45 50
SNR (dB)

Figure 6. BER curves for the α − κ − µ-Extreme case under non-coherent DBPSK modulation showing the effects of increasing
model parameters, the number of hops and N in Poincare approximations for moderate fading conditions

pared to Case I even under moderate fading severity. Figure 6 also shows that Poincare series-
based asymptotic BER approximations perform quite well at low SNR regimes for the choice in
parameter values, but that an increase in N does not seem to improve convergence or guarantee
improvement in accuracy of approximations. Moreover, an increase in hop number or µ results
in a higher non-null BER level under Case I than for increases in m under Case II. Figure 7

3
, = 2:0; 5 = 1:5; 7 = 2:0
, = 3:0; 5 = 1:5; 7 = 2:0
, = 5:0; 5 = 1:5; 7 = 2:0
2.5 , = 5:0; 5 = 3:0; 7 = 2:0
CORA Capacity (bits per sec per Hz)

, = 5:0; 5 = 3:0; 7 = 3:0


, = 2:0; 5 = 1:5; 7 = 2:0; n = 3 hops
, = 2:0; 5 = 1:5; 7 = 2:0; n = 4 hops
2 , = 2:0; 5 = 1:5; 7 = 2:0; n = 5 hops
Simulation

n = 2 hops

1.5

0.5

0
12 14 16 18 20 22 24 26 28 30
SNR (dB)

Figure 7. C O curves for the α − κ − µ case showing the effect of increasing number relays and fading model parameter values
under moderate fading conditions
22

4.5

Channel Capacity (bits/s/Hz)


3.5
Moderate Fading
3

2.5 COP RA ; CORA ; CT IF R ; CCIF R

2
Severe Fading

1.5

0.5

15 20 25 30 35 40 45 50
SNR (dB)

Figure 8. Channel capacity curves for the α − κ − µ case under various adaptive transmission schemes for a 3-hop DF-based
system

presents the channel capacity under ORA adaptive transmission for Case I. There is improvement
in channel capacity for increments in any of the fading model parameter values in Figure 7
in moderate fading. However, the improvements are very thin and even larger increments in
parameter values do not cause a bigger gap between the curves. The gap between OPRA and
ORA capacities gets smaller as SNR increases. Based on the analyses for AF, OP and BER, a
similar behavior is expected for the other three transmission protocols. However, there appears
to be an a diminishing reduction in channel capacity as more relays are added to the DF-
based system. According to [7] this is due to the half-duplex capability of the relays which
employs time-sharing multiple-access mechanism. Consequently, increasing the number of hops
will require more time to deliver information from S to D, which hampers system capacity.
Finally, Figure 8 shows the channel capacity under all the adaptive transmission schemes for
moderate fading conditions. The results are in agreement with the findings of [29] and prove that
as expected C P ≥ C O ≥ C T ≥ C c . We expect similar results to the trend set in the preceding
analyses for Case II and we therefore avoid displaying the figure for the sake of brevity.

VII. C ONCLUSIONS

In this paper, various performance measures of a DF-based multi-hop system over generalized
α−κ−µ and α−κ−µ-Extreme fading channels have been examined in detail. We have provided
the closed-form expressions for the OP, amount of fading, BER and channel capacity under
23

various adaptive schemes for an end-to-end DF-based multi-hop relaying system. The effects
of the variation of the fading parameters and the number of hops on the system performance
is demonstrated. System performance was shown to improve in a diminishing manner as the
number of hop links increased under moderate fading. However, for severe fading conditions,
there is little gain or actual performance degradation. From the analyses carried out in the present
work, a few points which could also be of interest for further research are : (i) Determining the
number of hops at which a negligible gain in system performance kicks in so as to avoid futile
additional relaying if the purpose of relaying is to improve system performance. (ii) Determining
the exact limit of BER under non-coherent modulation schemes as in [13] for the present fading
models.

R EFERENCES

[1] G. Kramer, M. Gastpar, and P. Gupta, “Cooperative strategies and capacity theorems for relay networks,” IEEE Transactions
on Information Theory, vol. 51, no. 9, pp. 3037–3063, 2005.
[2] S. N. Islam, P. Sadeghi, and S. Durrani, “Error performance analysis of decode-and-forward and amplify-and-forward
multi-way relay networks with binary phase shift keying modulation,” IET Communications, vol. 7, no. 15, pp. 1605–
1616, 2013.
[3] J. Cao, L.-L. Yang, and Z. Zhong, “Performance analysis of multihop wireless links over generalized-k fading channels,”
IEEE Transactions on Vehicular Technology, vol. 61, no. 4, pp. 1590–1598, 2012.
[4] G. Farhadi and N. Beaulieu, “On the ergodic capacity of wireless relaying systems over rayleigh fading channels,” IEEE
Transactions on Wireless Communications, vol. 11, no. 7, pp. 4462–4467, 2008.
[5] M. R. Bhatnagar, “On the capacity of decode-and-forward relaying over rician fading channels,” IEEE Communications
Letters, vol. 17, no. 6, pp. 1100–1103, 2013.
[6] D. Dixit and P. Sahu, “Performance of multihop communication systems with regenerative relays in η-µ fading channels,”
in Vehicular Technology Conference (VTC Spring), 2014 IEEE 79th. IEEE, 2014, pp. 1–5.
[7] F. H. Tha’er, H. B. Salameh, and T. Aldalgamouni, “Performance study of multi-hop communication systems with decode-
and-forward relays over α- µ fading channels,” IET Communications, vol. 11, no. 10, pp. 1641–1648, 2017.
[8] L. Yang, M. O. Hasna, and M.-S. Alouini, “Average outage duration of multihop communication systems with regenerative
relays,” IEEE Transactions on Wireless Communications, vol. 4, no. 4, pp. 1366–1371, 2005.
[9] D. Dixit and P. Sahu, “Exact closed-form aber for multi-hop regenerative relay systems over κ − µ fading,” IEEE Wireless
Communications Letters, vol. 6, no. 2, pp. 246–249, 2017.
[10] I. B. G. Pôrto and M. D. Yacoub, “On the phase statistics of the κ-µ process.” IEEE Trans. Wireless Communications,
vol. 15, no. 7, pp. 4732–4744, 2016.
[11] S. Kumar, “Performance of ED based spectrum sensing over α–η–µ fading channel,” Wireless Personal Communications,
vol. 100, no. 4, pp. 1845–1857, 2018.
[12] M. D. Yacoub, “The κ-µ distribution and the η-µ distribution,” IEEE Antennas and Propagation Magazine, vol. 49, no. 1,
pp. 68–81, 2007.
24

[13] G. S. Rabelo, U. S. Dias, and M. D. Yacoub, “The κ-µ extreme distribution: Characterizing severe fading conditions,” in
Microwave and Optoelectronics Conference (IMOC), 2009 SBMO/IEEE MTT-S International. IEEE, 2009, pp. 244–248.
[14] G. Fraidenraich and M. D. Yacoub, “The α-η-µ and α-κ-µ fading distributions,” in Spread Spectrum Techniques and
Applications, 2006 IEEE Ninth International Symposium on. IEEE, 2006, pp. 16–20.
[15] P. C. Sofotasios and S. Freear, “The α-κ-µ extreme distribution: Characterizing non-linear severe fading conditions,” in
Telecommunication Networks and Applications Conference (ATNAC), 2011 Australasian. IEEE, 2011, pp. 1–4.
[16] T. R. Rasethuntsa and S. Kumar, “An integrated performance evaluation of ed-based spectrum sensing over α-κ-µ and
α-κ-µ-extreme fading channels,” Transactions on Emerging Telecommunications Technologies, 2019.
[17] S. Kumar, M. Kaur, N. K. Singh, K. Singh, and P. S. Chauhan, “Energy detection based spectrum sensing for gamma
shadowed α-η-µ and α-κ-µ fading channels,” AEU-International Journal of Electronics and Communications, vol. 93, pp.
26–31, 2018.
[18] N. C. Beaulieu and J. Hu, “A closed-form expression for the outage probability of decode-and-forward relaying in dissimilar
rayleigh fading channels,” IEEE Communications Letters, vol. 10, no. 12, 2006.
[19] B. Kumbhani and R. S. Kshetrimayum, MIMO Wireless Communications over Generalized Fading Channels. CRC Press,
2017.
[20] I. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Products. Elsevier Science, 2014.
[21] A. M. Mathai, R. K. Saxena, and H. J. Haubold, The H-function: theory and applications. Springer Science & Business
Media, 2009.
[22] Y. A. Brychkov, O. Marichev, and A. Prudnikov, Integrals and Series, vol 3: more special functions. Gordon and Breach
science publishers, 1990.
[23] A. M. Mathai and R. K. Saxena, The H function with applications in statistics and other disciplines. Wiley, 1978.
[24] O. S. Badarneh and M. S. Aloqlah, “Performance analysis of digital communication systems over α-η-µ fading channels.”
IEEE Trans. Vehicular Technology, vol. 65, no. 10, pp. 7972–7981, 2016.
[25] A. Annamalai, R. Palat, and J. Matyjas, “Estimating ergodic capacity of cooperative analog relaying under different adaptive
source transmission techniques,” in 2010 IEEE Sarnoff Symposium. IEEE, 2010, pp. 1–5.
[26] Y. Sun, Á. Baricz, and S. Zhou, “On the monotonicity, log-concavity, and tight bounds of the generalized marcum and
nuttall Q-functions,” IEEE Transactions on Information Theory, vol. 56, no. 3, pp. 1166–1186, 2010.
[27] A. Annamalai and E. Adebola, “Asymptotic analysis of digital modulations in κ − µ, η − µ and α − µ fading channels,”
IET Communications, vol. 8, no. 17, pp. 3081–3094, 2014.
[28] H. Chergui, M. Benjillali, and M.-S. Alouini, “Rician k-factor-based analysis of xlos service probability in 5g outdoor
ultra-dense networks,” arXiv preprint arXiv:1804.08101, 2018.
[29] J. Zhang, M. Matthaiou, Z. Tan, and H. Wang, “Performance analysis of digital communication systems over composite
η − µ/gamma fading channels,” IEEE Transactions on Vehicular Technology, vol. 61, no. 7, pp. 3114–3124, 2012.