Matlab Codes

Available online at www.sciencedirect.
com
Journal of the Franklin Institute 358 (2021) 8012–8044

www.elsevier.com/locate/jfranklin
Multihop Weibull-fading communications:

Performance analysis framework and applications
Abdelaziz Soulimani a, Mustapha Benjillali a,∗, Hatim Chergui b,
Daniel B. da Costa c,d
a Communications Systems Department, INPT, Rabat, Morocco
b CTTC,
Barcelona, Spain
c Future Technology Research Center, National Yunlin University of Science and Technology, Taiwan, R.O.C.
d Department of Computer Engineering, Federal University of Ceará, Sobral, CE, Brazil
Received 13 May 2020; received in revised form 28 February 2021; accepted 10 August 2021
Available online 20 August 2021
Abstract
The paper presents a comprehensive closed-form performance analysis framework for multihop com-
munications over Weibull fading channels. This framework may be of interest in different applications
in the contexts of beyond-5G (B5G) and Internet of Things (IoT) use cases. The analyzed scheme
consists generally of multiple regenerative relays, and we also consider generalized high-order quadra-
ture amplitude modulation (M-QAM) transmissions. To take into consideration the channel fading, we
adopt the Weibull model for its largely flexible ability to cover different channel conditions in different
application contexts. The end-to-end performance is evaluated in terms of outage probability, bit error
probability (BER), symbol error probability (SER), block error rate (BLER), ergodic capacity, and en-
ergy efficiency (EE). For all the metrics, we present exact closed-form expressions—along with their
asymptotic behavior—capitalizing on the powerful generalized hypergeometric functions. To illustrate
the utility of the obtained analytical results, we derive two BER- and EE-optimal transmit power al-
location strategies, and we discuss the resulting performance gains. The exactness of our analysis is
illustrated by numerical examples, and assessed via Monte-Carlo simulations for different system and
channel parameters. Finally, as a secondary contribution, and noting the increasing popularity of single
∗ Corresponding author.
E-mail address: benjillali@ieee.org (M. Benjillali).
https://doi.org/10.1016/j.jfranklin.2021.08.004
0016-0032/© 2021 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
A. Soulimani, M. Benjillali, H. Chergui et al. Journal of the Franklin Institute 358 (2021) 8012–8044
and bivariate Fox’s H-function, we provide generalized M ATLAB codes for computing these functions
which are of practical utility in many fields.
© 2021 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.
1. Introduction
Cooperative and multihop communications [1] had emerged as a practical solution to in-
crease the coverage, while preserving high throughput and reliability, with low transmission
powers. It is a simple approach to overcome the severe channel conditions that usually im-
pact the transmissions (more particularly the millimeter wave (mmWave) signals in the B5G
systems). On the other hand, the drawbacks of multihop relaying (in terms of increased chan-
nel use, coordination overhead, and delay) may be dealt with using optimized transmission
parameters and network protocols in a comprehensive and cross-layer framework.
The double-parameterized Weibull distribution has been shown to accurately model both
narrow- and large-band small scale fading channels [2–4]. As an example, the authors
of [2] showed that this distribution models very well the studied environment, with the best-fit
distribution to the measurements compared to several other distributions. The authors provided
an estimation of the shape and the scale parameters of the distribution given the first and the
second sample central moments of the received field strength. Moreover, the scale parame-
ter of the Weibull distribution can be modeled as a log-normal distribution. As a result, the
Weibull distribution can be used to model mmWave communications, both in cellular B5G
and IoT applications.
Millimeter wave communications [5] (and references therein) have gained great interest as
a key enabler technology for the fifth generation (5G) of mobile communication systems. The
mmWave band offers large unlicensed bandwidths to answer the huge demand for increased
capacity, and higher spectral-efficiency. In addition, it presents interesting anti-interference
abilities, together with the new spatial processing techniques allowed by the short wavelengths.
While this band was under-utilized in the previous wireless systems—mainly due to practical
implementation limitations, cost, and stability—it is nowadays very attractive thanks to cost-
effective hardware technologies, and novel directional high-gain antennas. Given their short
range and weak penetration into different materials, mmWaves also offer efficient spectrum
utilization and secure transmissions.
Besides the heavy path loss in the mmWave band, large scale blockage is a real challenge
too. Even though, these drawbacks can be exploited to increase the efficiency of heterogeneous
networks. The main model of the path loss in the mmWave band is the well-known log-normal
distribution, whose parameters were computed using collected data from the measurement
campaigns in [6]. On the other hand, several models are proposed for the blockage effect in
the mmWave band. Some of them are summarized in [7, Sec. III].
There is a significant amount of work on the performance analysis of multihop relaying
communications, such as [8–11]. We only list two here that are in line with the perspective
of this work. For instance, in [8], the authors have analyzed the performance of multihop
relaying systems over Weibull fading channels in terms of bit error rate (BER) and outage
probability. However, the authors considered the amplify-and-forward (AF) strategy, and based
their analysis on an approximation of the end-to-end signal-to-noise ratio (SNR) resulting in a
8013
lower-bound discussion. A similar analysis of the BER and outage probability over multihop
Weibull fading channels was also conducted in [9], but in the context of free-space optical
communications with only binary pulse position modulations.
To the best of our knowledge, the performance of regenerative1 multihop schemes, over
the mmWave Weibull-modeled fading channels, remains an open problem, especially in terms
of error probabilities with high-order M-ary quadrature amplitude modulation (QAM) that is
of interest in modern communication systems. The present paper extends and generalizes our
results from [15,16] by proposing a considerably more comprehensive analysis framework,
with a more general perspective, and based on a more realistic system model. For instance,
path loss and recent blockage models are taken into consideration, and the analysis is extended
for real valued Weibull fading shape parameters.
The contributions in this work can be summarized in the following points:
• We generalize the performance analysis framework in [15] by, mainly, extending the shape
parameter to real values, resulting in generalized compact expressions in Fox’s H-function
rather than Meijer’s G-function. We also adapt the results of [16] to the current system
model chiefly by adding the realistic path-loss and blockage models. This new overview
provides a compact and handy survey of the common metrics, all in one place for interested
readers from the performance analysis community. All expressions are derived in terms of
single and bivariate Fox’s H-functions and bivariate Meijer’s G-function.
• For all the analyzed performance metrics, we derive asymptotic expressions that are simple
and offer tractable tools for engineers and designers to solve resource allocation and system
optimization problems.
• We highlight the effect of the path loss and the blockage facing mmWave signals.
• To illustrate the usability of the results, especially the asymptotic expressions, we derive
optimal transmit power allocation strategies. We show that considerable performance gains
can be obtained with these allocation schemes, when compared to a uniform allocation for
example. Note that a similar method can be adopted to optimize the other performance
metrics.
• We provide and discuss a rich set of numerical results reflecting a wide selection of
applications in B5G and IoT eco-systems.
• Finally, and as a secondary yet important contribution, we implement the single and bivari-
ate Fox’s H-functions in M ATLAB. Since they are gaining in popularity both in software
computation tools (like M ATHEMATICA and M ATLAB) and among the performance anal-
ysis community. The proposed MATLAB implementation code of the functions—unlike a
few existing versions2 —is general and independent of specific numerical examples.
The remainder of this paper is organized as follows. Section 2 introduces the adopted
system model and notations. Next, the expressions of the analyzed performance metrics are
1 In general, two main relaying classes are available for cooperative communications, namely, the non-regenerative
and the regenerative strategies [12]. In this work, we adopt the regenerative “detect-and-forward” (DetF) relaying
scheme [13], where the relays re-transmit (without decoding and re-encoding) the demodulated binary sequences.
The adoption of DetF is motivated by its simplicity compared to several relaying techniques, reduced delay, and
interesting performance with lower processing complexity and channel state information constraints [14].
2 For example, in [17], the author is gives the MATLAB code to implement only one specific equation in the
paper; and in [18], the authors provide a Mathematica code, where the integration is done only over one of the three
contours that must be checked for general cases.
8014
Fig. 1. Adopted system model with (N −1) regenerative multihop relays.
derived in Section 3. The obtained results are used in Section 4 to derive optimal power
allocation strategies for multihop schemes. Numerical examples along with simulation re-
sults, using our implementation of the hypergeometric functions, are presented in Section 5.
Section 6 summarizes a few possible extensions of this paper, and Section 7 concludes the
discussion. Finally, in the appendices, we include the new M ATLAB implementation codes
for single and bivariate Fox’s H-functions.
2. System model
In this work, we consider a cooperative transmission from a source node (S) to a destination
(D) through (N − 1) regenerative relay nodes Ri , i = 1, . . . , N − 1, as shown in Fig. 1. We
assume that all nodes are operating in the half-duplex mode, with the same modulation order
M, and that all transmissions are orthogonal3 , e.g., over different time or frequency resources.
Each receiving node considers only the previous adjacent transmitter, and the direct link
between (S) and (D) is not taken into consideration (i.e., no signal is received at the destination
directly from the source because of considerable path loss).
The transmitted signal from the source node is denoted by x, while yi and x˜i are, respec-
tively, the received and transmitted signals at the i-th node, and y is the received signal at the
destination. Thus, we can express the communication over the i-th link, i = 1, . . . , N , under
the form

t B
Pi−1 i
yi = gi x˜i−1 + ni = hi x˜i−1 + ni , (1)
Li
3 Although this is not the optimal serial transmission protocol, it is adopted to simplify the analysis and avoid all
the considerations that are out of the scope of the contribution of this paper.
8015
where x˜0 = x, yN = y, and ni denotes the zero-mean additive white Gaussian noise (AWGN)
at the i-th node, with the same variance N0 over all links. The distance between the source (S)
and the destination (D) is d meters, and the distance between adjacent nodes is di meters,
for i = 1, . . . , N . The communication between nodes i − 1 and i is subject to a path-loss Li
given by [19]
Li [dB] = a + 10 · b · log10 (di ), (2)
where a and b are given by [19, Table 1] for the 28GHz and 73GHz mmWave bands. Bi
represents the blockage probability from the 3GPP model [20] given by

18 −di −di
Bi = min , 1 1 − e 63 + e 63 , (3)
di
i
for urban areas, and Bi = exp −d 200
for suburban areas.
Channel coefficients gi are assumed to be perfectly known at the receiver, since channel
estimation is out of scope here. The coefficients are also assumed to follow independent but
not necessarily identically distributed (i.n.i.d.) flat Weibull fading profiles with parameters
(βi , i ).
2.1. Perfect channel state information
We first suppose that the receiver estimates the fading channel perfectly i.e. the instan-
taneous signal-to-noise ratio (SNR) over the i-th hop at the receiver may be written as
γi = |g|2i γ i , where γ i = Pi−1
t
Bi /N0 /Li denotes the average SNR at the i-th node, where Pit is
the
transmitting power of the i-th node. γi is also Weibull distributed [21] with parameters
αi = βi /2, φi = γ i 2i . Its PDF and CDF are respectively

αi αi −1 −γ αi
pγi (γ ) = γ exp , (4)
φi φi
αi
−γ
and Fγi (γ ) = 1 − exp . (5)
φi
2.2. Outdated channel state information
Due to the outdated channel state information at the receiver, the fading effect can not be
compensated. The resulting signal at the output of the equalizer is
gi
yi = x˜i−1 + ni , (6)
g˜ i
where gi is the channel coefficient, g˜ i is the estimated fading which, due to the fast fading,
is correlated with the true realization gi . ni is the noise after the equalizer, its variance is
Ni = 1/γ i . Since |gi |2 and |g˜ i |2 are also Weibull RVs correlated with a factor of ρ, we
begging by using the result in [22], to get the PDF of Gi = |gi |2 /|g˜ i |2 defined as
+∞
pGi (z) = x p|gi |2 ,|g˜i |2 (zx , x ) dx (7)
0
8016
∞
αi2 αi −1 2αi −1 −(zαi + 1)x αi
= z x exp
(1 − ρ)2i 0 (1 − ρ)i

2 zαi x 2αi
× I0 ρ dx (8)
1−ρ 2i
αi (1 − ρ)zαi −1 (zαi + 1)
= 3/2 , (9)
z2αi + (2 − 4ρ)zαi + 1
= π (1 − ρ)αi zαi −1 (10)

⎛ ⎞
(−1;1,2) ( 21 ,1)
0;1,0;1,0 ⎝ αi αi ⎠.
× H11,,0;0, 1;1,3 z , ρz
(0, 10),(0,1),( 1 ,1)
(0, 1) 2
Hence the PDF of the instantaneous SNR of the ith hop γ˜i = Gi /Ni = γ i Gi is
π (1 − ρ)αi αi −1
pγ˜i (γ ) = γ (11)
γ αi i
⎛ ⎞
ργ αi (−1;1,2) ( ,
αi 1
1,0;1,0;1,0 ⎝γ 1)
× H1,0;0,1;1,3 αi , αi
2
⎠.
γi γi
(0,1)
(0,1),(0,1),( 1 ,1)
2
Integrating (9), we get the CDF of γ˜i as

1 1 γ αi − γ αi i
Fγ˜i (γ ) = + . (12)
2 2 γ 2αi + (2 − 4ρ)γ αi γ αi + γ 2αi
i i
3. Comprehensive performance analysis
In this section, we investigate the end-to-end performance of the proposed scheme in terms
of outage probability, ergodic capacity, BER, SER, EE, and BLER.
3.1. Outage probability
3.1.1. Perfect CSI

For the proposed system, an end-to-end outage event occurs when at least one hop go
into an outage, namely the transmission rate of the i-th hop is higher than the mutual
information over the equivalent channel between the transmitting and receiving nodes. The
end-to-end outage probability can be written, for an outage SNR threshold γth = 2 − 1, as

Pout = Pr min (γ1 , . . . , γN ) ≤ γth , (13)

= 1 − Pr min (γ1 , . . . , γN ) ≥ γth (14)
8017

N

=1− 1 − Fγi (γth ) , (15)
i=1
Substituting (5) in (15) yields the following compact closed-form expression for the end-to-
end outage probability

N
γ αi
Pout = 1 − exp − th
. (16)
i=1
φi
For low γth values, (16) can be approximated by

N
γ αi
Pout th
. (17)
i=1
φi
Assuming identical channel parameters (φi = φ, αi = α for all i = 1, . . . , N ), the outage prob-
ability can be further simplified to
N γthα
Pout . (18)
φ
One direct application of this expression, for a target outage probability Π , is the minimum
number of hops needed between the source and the destination is obtained as

φΠ
Nmin = , (19)
γthα
where . denotes the ceiling function.
3.1.2. Outdated CSI

When the estimated fading channel is outdated the end-to-end outage probability becomes

N

Pout = 1 − 1 − Fγ˜i (γth ) , (20)
i=1
⎛ ⎞

N
1 1 γthαi − γ αi i
=1− ⎝ − ⎠. (21)
i=1
2 2 γ 2αi + (2 − 4ρ)γ αi γ αi + γ 2αi
th i th i
3.2. Bit error rate
For regenerative relays, it was shown in [23] that BER can be expressed in terms of the
average BER values of each hop as

N
N

BER = BERi 1 − 2BER j , (22)
i=1 j=i+1
where BERi is the average BER of the individual ith hop.
8018
3.2.1. Closed-form exact analysis

Over fading channels, BERi is
+∞
BERi = Pb (γ ) pγi (γ )dγ , (23)
0
where, in our case, pγi (·) is given by (4), and Pb (·) denotes the exact instantaneous BER of
an M-QAM transmission over a Gaussian channel, which was derived in [24] for an arbitrary
order M under the form
√
1
log2 M
Pb (γ ) = √ Pb (γ , m), (24)
log2 M m=1
with
νm
1 √
Pb (γ , m) = √ Φ m,n erfc ωn γ , (25)
M n=1
and
√ 3(2n + 1)2 log2 M
νm = 1 − 2−m M − 1, ωn = ,
2M − 2
and

m−1
n2√ n2m−1 1
Φ m,n = (−1) M 2m−1 − √ + .
M 2
Hence, the ith hop BER can be written as
√
νm
M
log2
1
BERi = √ √ Φ m,n ζn,i , (26)
M log2 M m=1 n=0
where
+∞ √
ζn,i = erfc ωn γ pγi (γ )dγ (27)
0

αi +∞ αi −1 √ −γ αi
= γ erfc ωn γ exp dγ . (28)
φi 0 φi
To evaluate this integral, we express the exp(·) and erfc(·) functions, under its generalized
Fox H-function representation,

—
1,0
exp(−x) = H0,1 x
(0, 1)

√ 1 2,0 (1, 1)
erfc x = √ H1,2 x , (29)
π (0, 1), (1/2, 1)
then, using [25, Theorem 2.9], we get

αi ωn−αi 1,2 (1 − αi ; αi ), (1/2 − αi ; αi )
−αi
ζn,i = √ H φi ω n . (30)
π φi 2,2 (0, 1), (−αi , αi )
By substituting (30) in (26), and then in (22), we obtain a closed-form expression for BER.
A general Matlab code to implement Fox H-function is provided in Appendix A.
8019
3.2.2. Asymptotic analysis

The asymptotic behavior of the resulting hypergeometric functions may be obtained using
a direct expansion [25, Th. 1.7 and 1.11]. However, this method requires the satisfaction
of restricted conditions. Consequently, we adopt a general method, based on the fact that
the functions are defined by Mellin-Barnes integrals, using the residue theorem, as explained
in [26].
For high γ i values, (30) becomes
−αi
ζn,i = √ωn 21 + αi . (31)
γ →∞ πφii
By replacing (31) in (26), we get

√
log2 M νm
21 + αi
BERi = √ √ Φ m,n ωn−αi . (32)
γ i →∞ φi π M log2 M m=1 n=0
One direct application of this result is the diversity order of each hop
log BERi (γ i )
δi = − lim = αi , (33)
γ i →∞ log γ i
and, using (22) and (26), we get the diversity order of the end-to-end system
N
δe2e = min (αi ). (34)
i=1
N
Since j=i+1 1 − 2BER j → 1 for high values of γ i (i = 1, . . . , N ), (22) can be simpli-
fied to
√
log2 M
μm
N0 Li αi
N
( 21 + αi ) m,n
BER = √ √ , (35)
γ m→∞
i=1
π M log2 M Pi Bi i
t 2
m=1 n=1
ωnαi

were γ m = minNi=1 γ i .
3.2.3. Outdated CSI

With the same notations as (26), the BER of the i-th hop in an outdated CSI environment
becomes
√
νm
M
log2
1
BERi = √ √ Φ m,n ζñ,i , (36)
M log2 M m=1 n=0
+∞ √
ζñ,i = erfc ωn γ pγ˜i (γ ) dγ . (37)
0
Using [27, Eq. 2.1], we get this integral in closed form as

√
π (1 − ρ)αi
ζñ,i = α α
γ i i ωn i
⎛ ⎞
−α −αi (1 − α ; α , α ), ( 1 − α ; α , α ), (−1; 1, 2) ( 21 , 1)
ωn i ρωn i i i 2 i i i
H32,,10;1 ,0;1,0 ⎝
α ,
⎠. (38)
;0,1;1,3
γi i
α
γi i
(−αi ; αi , αi ) (0, 1) (0, 1), (0, 1), ( 21 , 1)
8020
3.2.4. Beamforming
Although not the focus of this work, we here discuss the impact of beamforming enhanced
directivity, since it is a key enabler for mmWave communications. In this case the channel
model should be revisited.
Each node is equipped with a high number of antennas, and it is using t transmitting
antennas and r < t receiving antennas. At the i-th hop, we suppose that the channel ma-
trix Hi , Weibull randomly distributed, is perfectly known at the transmitter and the receiver.
The singular-value decomposition of the channel matrix is Hi = rUi .Si .ViH , where Ui and
Vi are respectively r × r and t × r matrices, .H is the Hermitian transpose symbol, and
Si = diag(σi,1 , σi,2 , . . . , σi,r ) where σi,1 ≥ σi,2 ≥ . . . ≥ σi,r > 0 are the ordered singular val-
ues of Hi /r.
The streamed signal Xi = [xi,1 , xi,2 , . . . , xi,r ]T is precoded using V , where xi,1 , xi,2 , . . . , xi,r
are the modulated symbols, and .T is the transpose symbol. Hence the stream at the i-th hop
is X i = Vi Xi .
The receiver gets the signal

Y i = Pi−1t B /L H .X + W =
i i i i i
t B /L U S X + W ,
Pi−1 i i i i i i

where Wi is the noise vector at the receiver. Multiplying by Li /Pi−1
t /B U H we get as a
i i
result Y˜i = Si Xi + W˜ i , namely, the k-th signal at the receiver is

Y˜i,k = σi,k xi,k + wi,k , k = 1, . . . , r. (39)
Thus, the SNR at each receiving antenna can be written as
γi,k = σi,k
2
γ i,k , k = 1, . . . , r, (40)
where γ i,k is the average SNR for the ith hop at the kth receiving antenna. Assuming a large
number of antennas, according to the Marčenko Pastur law [28], the PDF of γi,k converges
almost surely to
+
+ 1 x + x
fi,k (x) = (1 − c) δ(x) + b− −a , (41)
2π csi2 x γ i,k γ i,k
+
√ 2
√ 2 δ(. ) is the Dirac delta function, (x ) = max (0, x ), c = r/t, a = (1 − c ) , b = (1 +
where
c ) , and si is the standard deviation of Hi .
Finally, the average BER of the ith hop is
√
νm
M
log2
2
BERi = √ √ Φ m,n
M log2 M m=1 n=0
∞ √
× fi,k (x)erfc ωn x dx (42)
0
√
νm
M
log2
2 Φ m,n
=√ √ Ja,b,n , (43)
M log2 M m=1 n=0
2π csi2 γ i,k
8021
where a = aγ i,k , b = bγ i,k , and

b √
(x − a )(b − x) √
Ja,b,n = erfc ωn x dx. (44)
a x
Using [29, eq. 2.75] this integral can be evaluated in terms of bivariate Meijer G function as
a (b√

−a )2

Ja,b,n = π
23

2 1, 1/2−1/2 (45)
1;0,2;1,1
× G10,,0;2, 1;1,2
1
ωn a
, b −a
a .
—0 0, −2
3.3. Symbol error rate
To alleviate the presentation, we refer the reader to [30] where the authors have derived
the SER of M-QAM communications using regenerative relays over Nakagami-m channels.
Here, we use the same approach.

The calculations of the SER reduce to the following integral
+∞
√ √
ISERi = Q A γ Q B γ pγi (γ )dγ , (46)
0
where A and B are two positive constant coefficients (note that they cannot be both null), and
pγi (. ) is given by (4). To evaluate (46), two cases need to be differentiated:
• A × B = 0: we denote C = max (A, B). This case is similar to (28), and hence
αi
αi 2
ISERi (C ) = √
4φi π
C
(1 − αi , αi ), ( 21 − αi , αi )
1,2 1 2 αi
(47)
× H2,2 φi C .
(0, 1), (−αi , αi )
• A × B > 0: In this case, we replace the Q-function and the exponential with their Fox H-
Function representations. Using [27, eq. 2.3], we get the closed-form expression of (46) in
terms of the bivariate Fox H-Function as
αi
αi 2
ISERi (A, B ) =
4φi π A

αi
(1 ,

(—)

2,0;2,0;1,0 B 1 2 (1 − αi ; 1 , αi ) , ( 1
2
− αi ; 1 , αi ) 1)
× H2, 1;1,2;0,1 , .
A φi A (−αi ; 1, αi ) (0, 1), ( 1 , 1) (0, 1)
2
(48)
A general M ATLAB code to implement the bivariate Fox H-Function in Appendix B.
3.3.2. Asymptotic analysis

To derive the asymptotic expressions of (47) and (48), we again reuse the residue method
to get
αi
1 2
ISERi (C ) = √ (1/2 + αi ) , (49)
γ i →∞ 4φi π C
8022
and

−αi 2 αi 2,2 B αi , 1/2 + αi , 1
ISERi(A, B ) = G3,3 , (50)
γ i →∞ 8φi π 2 A A 0, 1 , 1 + αi
2
where the Meijer G-Function [31] is used.

Apart from their simplicity, we emphasize the usefulness of these asymptotic expressions
especially in terms of computation time for high-order modulations compared to [30, eq. 17]
and (48).
3.4. Block error rate
In this case we are assuming that the fading channel remains invariant for the duration of
a block; TB = lTs , where TB is the block time, Ts is the symbol time, and l is the number of
symbols per block called also channel uses (c.u.).
Based on [32], the average end-to-end block error rate (BLER), noted E , can be defined
recursively as
E = E (D) = E (RN−1 ) + (1 − E (RN−1 ))EN ,
(51)
E (R1 ) = E1 ,
where E (Rk ) is the cumulated BLER at the kth node, and Ei is the average individual BLER
of the ith hop.
Given this definition, we can express the average end-to-end BLER in function of the
average individual BLER of each hop only. To do so, we can observe that

N
N
E = Ei − E (Ri−1 )Ei . (52)
i=1 i=2
By recurrence, it can be shown that

N
N
N
N
N
N
E = Ei − Ei1 Ei2 + Ei1 Ei2 Ei3
i1 =1 i1 =1 i2 =i1 +1 i1 =1 i2 =i1 +1 i3 =i2 +1

N
N
N
k
+ . . . + (−1)k−1 ··· Eik
i1 =1 i2 =i1 +1 ik =ik−1 +1 j=1

N
+ . . . + (−1)N−1 Eik , (53)
j=1
which can be made more compact under the form

N (Ni )

E = (−1) i−1
Ek , (54)
i=1 j=1 k∈Si, j
N Si, j = {k1 , k2 , . . . , ki } j for 1 ≤ k1 < k2 < · · · < ki ≤ N , representing the j-th set of the
with
i
sets with i elements, chosen from {1, 2, . . . , N }.
8023
All that is left is to determine the expression of the average individual BLER of each hop.
It is defined [33] for the ith hop as
+∞
(C(γ ) − )
Ei = Q √ pγi (γ ) dγ , (55)
0 V (γ )/l
where is the transmission rate and
C(γ ) = log2 (1 + γ ),
+2) 2 (56)
V (γ ) = γ(γ(γ+1) 2 log2 e .
It is clear that the integral in (55) is very challenging to compute in closed-form. The
linear approximation of the term with the Q-function seems to be very helpful;
⎧
⎪ ⎪1 γ ≤ γ-
(C(γ ) − ) ⎨ √
Q √ ≈ 2 − λ l (γ − γth ) γ- < γ < γ+ ,
1
(57)
V (γ )/l ⎪
⎪
⎩
0 γ+ ≥ γ
where λ = 2π √212 −1 , γth = 2 − 1, and γ± = γth ± 1√
2λ l
.
Hence the integral (55) becomes
√ γ+
Ei = λ l Fγ (γ ) dγ (58)
γ-
√
λ l α1i 1 γ-αi 1 γ+αi
=1− φ , − , . (59)
αi i αi φi αi φi
3.5. Ergodic capacity
The ergodic capacity corresponds to the maximum long-term achievable rate averaged over
all states of the time-varying channel. In the present context of multihop communication with
regenerative relays, the average ergodic capacity can be expressed as
C = min Ci , (60)
i=1,...,N
where C i is the bandwidth-normalized average ergodic capacity of the ith hop, given by
+∞
Ci = log2 (1 + γ ) pγi (γ ) dγ . (61)
0

In order to evaluate the integral in (61), we express the exp(·) (29) and the log(·) functions
through the Fox H-function representation

(1, 1), (1, 1)
1,2
log (1 + x ) = H2,2 x , (62)
(1, 1), (0, 1)
and using [25, Theorem 2.9], we get (61) in closed-form as

αi (−αi , αi ), (1 − αi , αi )
3,1 1
Ci = H . (63)
φi log 2 2,3 φi (0, 1), (−αi , αi ), (−αi , αi )
8024
3.5.2. Asymptotic expression α

For high values of γ i , recalling that φi = γ i 2i i , the asymptotic behavior of (63) can be
performed by using again [26, sec. IV]
0 (1 )
Ci = + log2 (φi ), (64)
γ i →∞ αi log 2
where 0 (. ) denotes the digamma function [34].
3.6. Energy efficiency

Without loss of generality, we adopt the definition of energy efficiency (EE) based on the
consumption factor metric [35]. For the proposed system model, the end-to-end bandwidth
normalized EE can be expressed as
1
EE e2e = log (1 + γe2e ), (65)
PT
where γe2e = mini=1,...,N γi denotes the equivalent instantaneous end-to-end SNR. While
PT is
the total consumed power in the system (circuit and transmission powers) PT = Pc + N−1 t
i=0 Pi ,
t
where Pc is referring to the circuit power consumed during the transmission (Pc ), reception
(Pcr ), modulation (Pcm ), demodulation (Pcd ), and in the idle mode (Pci ). Thus

N−1
t
N−1 N
Pc = Pc,i + Pcm,i + Pcr ,i + Pcd,i + Pci ,i
i=0 i=0 i=0

= N Pct ,1 + Pcr ,1 + Pcm,1 + Pcd,1 + (N + 1 )Pci ,1 , (66)
where it is implicitly assumed in (66) that all nodes have a similar power consumption profile.
The average EE may be obtained directly from (65) as
∞
1
EE e2e = E[EE e2e ] = 1 − Fγe2e (ex − 1 ) dx, (67)
PT 0
where E[·] stands for the mathematical expectation and

Fγe2e(γ ) = Pr min γi ≤ γ
i=1,...,N

N
N
γ αi
=1− 1 − Fγi (γ ) = 1 − exp − . (68)
i=1 i=1
φi
To proceed further with the derivation of EE e2e , we suppose that αi = α, i = 1, . . . , N .

Then, taking the
simplified form of (68) into consideration in (67), and using the short hand
notation ψ = 1/ Ni=1 (1/φi ), we get
∞
1 (ex − 1 )α
EE e2e = exp − dx (69)
PT 0 ψ
∞ α
1 1 v
= exp − dv. (70)
PT 0 v+1 ψ
8025
By writing the exponential function and 1/(v + 1) in terms of the Fox H function, we get
∞

(0, 1) α —
1 1,1 1,0 v
EE e2e = H1,1 v H0,1 dv (71)
PT 0 (0, 1) ψ (0, 1)

1 2,1 1 (0, α)
= H . (72)
PT 1,2 ψ (0, 1), (0, α)
3.6.2. Asymptotic expression

Thanks to the residue approach once again, we can easily obtain an asymptotic expression
for the end-to-end EE as

N
1 1
EE e2e ≈ 0 (1 ) + ln . (73)
αPT φ
i=1 i
4. Power allocation optimization
In this section, we derive the optimal power allocation strategy for the analyzed system.
Note that the exact expressions can be used to compute a numerical solution to the problem.
On the other hand, we can adopt the asymptotic expressions of the BER and EE to derive
closed-form analytical approximations. Both approaches yield similar solutions, and it is the
second approach that we use in the subsequent investigations.
Note that the power allocation strategies derived in this section are based on BER and
EE results only, a similar method can be adopted to optimize the other performance metrics.
Since the analysis is quite straightforward, we do not detail all the other cases in the current
paper.
4.1. BER-Optimal Power allocation
As a reference for comparison, we first derive the optimal power allocation strategy mini-
mizing the BER given a total transmit power budget, i.e., the following problem needs to be
solved
min BER
P0t ,P1t ,...,PN−1
t
⎧
⎨ Pt = P
⎪N−1
i max
s.t. i=0 (74)
⎪
⎩P > 0, i = 0, . . . , N − 1
i
where Pmax is the maximal transmission √power budget √ of the system. Using the result
in (35) and the notation ϕ = (1/2 + α)/( π M log2 M ), we get the Lagrangian cost func-
tion as
α log
√
M μm

N
N
N0 Li 2 m,n
Jber = ϕ + λber Pi−1 − PT ,
t
(75)
i=1
t 2
Bi Pi−1 i m=1 n=1
ωnα i=1
8026
where λber is the Lagrange multiplier. The N + 1 Karush-Kuhn-Tucker (KKT) conditions can
be expressed as
⎧ t −α−1 √
⎨ ∂Jber = −ϕ N0 Li α (Pi−1 ) × m=1
log2 M μm m,n
∂Pi−1
t
Bi 2i α n=1 ωnα + λber = 0
(76)
⎩ ∂Jber = N Pt − P = 0;
∂λber i=1 i−1 T
yielding a system of equations whose solutions, i.e., the optimal transmit powers, are easily
obtained as
Ai
t
Pi−1 = N PT , i = 1, . . . , N, (77)
n=0 An
where, for n = 0, . . . , N ,
⎛ √ ⎞1/(α+1)
α log M μm

N L 2

An = ⎝
0 n m,n ⎠
ϕ α
. (78)
Bn n m=1 n=1
ω n
4.2. EE-Optimal Power allocation
In this subsection we derive the power allocation maximizing the energy efficiency of our
system, i.e.,
min − EE e2e
P0t ,P1t ,...,PN−1
t
N−1
i=0 Pit ≤ Pmax
s.t. (79)
Pit > 0, i = 0, . . . , N − 1
The Lagrangian of this problem is

N−1
t t
Jee = −EE e2e P0 , P1 , . . . , PN−1
t
− λee Pit − Pmax , (80)
i=0
where λee is the Lagrange multiplier in this case. The N + 1 KKT conditions are then
⎧ −α−1
⎪ t
⎨ a j P j−1
N = λee /E0 , j = 1, . . . , N
a P t −α (81)
⎪
⎩N ti=1 i i−1
i=1 Pi−1 = Pmax
where E0 = 1/PT , and ai = (N0 Li /Bi /2i )α for i = 1, . . . , N . The solutions of (81) can be
easily obtained as
E0
λee = Pmax
,
(82)
t
Pk−1 = N
Pmax
1/ (α+1) , k = 1, . . . , N,
i=1 (ai /ak )
yielding the optimal transmit powers maximizing the end-to-end EE.
5. Numerical results
To assess the accuracy of the theoretical analysis and illustrate the performance of multi-hop
relaying systems in the adopted context of mmWave, we present in this section a few numerical
scenarios of interest and we compare our analytical results to Monte-Carlo simulations.
8027
Fig. 2. The end-to-end outage probability Pout versus the EIRP per hop for multi-hop Weibull fading channels with
similar shape parameters βi = 2. In the legend, the highlighted parameter is the one that has been changed in each
scenario.
5.1. Setup
In order to simplify the readability of the figures, we follow this same convention in
all the figures when necessary: solid lines represent the exact analytical results, simulations
are represented with markers (only, no lines), and asymptotic expressions correspond to the
dashed lines. Hence, when the markers are on a line, this should be interpreted as a perfect
match between simulation and analytical results.
We took into consideration data parameters from ITU, 3GPP, and FCC specifications to
reflect realistic mmWave systems. For instance, we consider the following figures:
• noise power: −174 dBm/Hz,

• single antenna element gain: 5 dB,
• receiver front end loss: 4 dB,
• noise figure: 5 dB.
All the studied metrics are plotted in terms of the equivalent isotropically radiated
power (EIRP) since its values are directly related to the SNR; user devices’ maximal EIRP
8028
Fig. 3. End-to-end bandwidth-normalized average ergodic capacity C /Bw versus the EIRP per node for a 3 hops
communication system.
is fixed at 23 dBm, base station’s maximal peak EIRP value can reach up to 85 dBm for high
antenna gains (51 dBi), but is in general limited to 43 dBm.
5.2. Outage probability
Fig. 2 shows the end-to-end outage probability for a three-hop communication. Several
scenarios have been studied to highlight some of the system parameters:
• The bandwidth has a horizontal shifting effect on the outage probability of the system as
it appears only in the noise power: N0 [dBm] = −174 + 10 log10 (Bw ). The same applies
to the BER, SER, BLER, and ergodic capacity metrics. We consider in all the remaining
figures, when not specified, that Bw = 200 MHz. For other values of Bw (in MHz), the
curves can be simply obtained by a horizontal shift of 10 log10 (Bw /200) [dB].
• The distance has a double horizontal offset effect on the outage probability, since it appears
in the blockage (3) and the path-loss (2) models.
• Increasing the number of hops seems to decrease the outage probability. This is not always
true as it will be discussed later for the end-to-end BER.
8029
Fig. 4. Ratio of the theoretical end-to-end average ergodic capacities: R(W1 , W2 ) = C (28GHz )/C (73GHz ) versus the
end-to-end distance (d) for Bw = 200 MHz.
• In narrow band communications (Bw ∝ 100kHz), for IoT applications, the outage proba-
bility decreases significantly even with low transmitting power.
From the figure, it is clear that the exact closed-form expression (16) matches with sim-
ulation results, confirming the exactness of the analysis. We also note the tightness of the
asymptotic bound for the region of interest Pout < 0.1.
5.3. Capacity
Fig. 3 depicts the bandwidth-normalized end-to-end ergodic capacity C as a function of

the EIRP per hop for several scenarios. In this example, we are assuming an environment of
3 hops with similar Weibull fading parameters βi = 2 and i = 1, i = 1, . . . , 3. The figure
shows, as expected, that the increasing modulation order increases the end-to-end capacity
with a factor related to the difference between the compared modulation orders. Similar to
the outage probability, the distance between the end nodes (S and D) has a discernible impact
on the ergodic capacity.
In order to compare the performance of the system in the two mmWave frequency bands
W1 = 28 GHz and W2 = 73 GHz, the ratio R(W1 , W2 ) = C (28GHz )/C (73GHz ) versus the end
to end distance d is drawn in Fig. 4. Here, the communication is established through three
8030
Fig. 5. End-to-end average bit error rate versus average EIRP per hop.
hops with the same distance di = d/3 and similar Weibull parameters in two scenarios; one
with an EIRP of 30 dBm, and the other with 50 dBm. It is clear that C (W1 ) is much
larger than C (W2 ) since R(W1 , W2 ) > 4. We see also that curves slopes are similar within
the same scenario, however, R(W1 , W2 ) changes significantly even with small variations of
the shape parameter (notice that the difference between the values of βi is less than 0.08 yet
the ratio deviates with around 0.5). The steepness of the curves depends mainly on the SNR
represented here with the EIRP. Nonetheless, it is worth noting that this comparison is a bit
unfair since we use the same bandwidth Bw = 200 MHz value for both capacities, yet, the
73 GHz range of frequencies offers more bandwidth and is suitable for very small distance
ranges. Therefore, to decrease R(W1 , W2 ), optimal bandwidth values and distances could be
chosen.
5.4. BER
In Fig. 5, several scenarios of the end-to-end BER are plotted for a three-hop mmWave
communication system based on DetF relaying over Weibull fading channels. The figure
shows that the modulation order, the end-to-end distance, and the frequency band are key
players in designing the system based on the BER metric. For uplink communications or
IoT applications, it is better to use low order modulation to get better performance given the
low power used by these devices (maximal EIRP of 23 dBm). For low power transmissions,
the performance can also be improved by using low bandwidths, as an example, a 180kHz
8031
Fig. 6. End-to-end average BER versus the number of hops N, with βi = 2 and i = 1.
bandwidth (the adopted bandwidth value in UL NB-IoT [36]) would enhance the system’s
performance by more than 30 dB.
Fig. 6 depicts the effect of the number of hops on the end-to-end BER. In general, in-
creasing the number of hops tends to decrease the BER. For low power communications
(EIRP < 23 dBm) the BER increases while increasing the number of hops, for instance, for
an EIRP = 20 dBm, the increase has a low impact on the end-to-end SNR.
In Fig. 7, the end-to-end BER has increased significantly when the estimated channel is
outdated even for a high correlation factor ρ = 0.95. The curves of the BER, in the two
frequency bands, tend to have the same value in higher SNRs, since the effect of additive
noise becomes insignificant compared to the outdated CSI. On the obverse side, beamforming
has increased the system’s performance. Noting that, the total power given to all antennas
is equal to the power allocated to the single antenna transmission. Hence, the effectiveness
of beamforming against multipath degradation, not to mention the increased gains at the
transmission and reception.
5.5. SER
In Fig. 8, we present in several scenarios, the average end-to-end SER for a dual-hop
communication (β1 = 1, β2 = 2) versus the EIRP per hop of the proposed system:
8032
Fig. 7. End-to-end average bit error rate BER versus average EIRP per hop for a dual-hop communication, using
16-QAM modulation highlighting the context of outdated CSI and beamforming.
• The results confirm, again, the vulnerability of the mmWave signals to the distance, How-
ever for small distances, ideal environment of IoT applications, it is very promising.
• The frequency band has also a perceptible effect, similarly to the other metrics.
Again, the perfect match between analytical and simulated results, and the tight correspon-
dence between exact and asymptotic analysis, can be appreciated from the figure.
5.6. BLER
Fig. 9 displays the evolution of the BLER in terms of the EIRP. In the figure, we highlight
some system parameters effect on this metric. First, we must note that the approximation of the
Q-function (57) gives good results with an accuracy depending on and the number of hops.
Like the other metrics, the BLER tends to deteriorate when increasing the modulation order.
However, high modulation orders have promising results with low distance communications
or sufficient number of hops.
Large bandwidth communications tend to worsen the BLER due to the accumulated noise.
On the other hand, nodes communicating over the 1 MHz bandwidth show lower BLER
values, hence better anti-error performance.
8033
Fig. 8. End-to-end average symbol error rate SER versus EIRP per hop for a dual-hop communication.
Finally, we used relatively small values of the number of c.u. l to imply the ultra-reliable
low latency communications (URLLC) as a key enabler of the next cellular networks gener-
ation. It seems to have minimal impact on the end-to-end BLER.
5.7. EE
The power analysis to get the total consumed power by the system has been done with
the help of realistic results from recent works [37–43] on mmWave transceivers. Circuit
and transmit powers range from a few tens of milliwatts to a few watts depending on the
application. The power consumed in the idle mode drops to less than 1% of the consumed
power in the connected mode. In the remainder of this subsection, with the exception of the
last figure, we are assuming that the total circuit power consumed in each node is equal to
0.5 Watts.
In Fig. 10, the end-to-end bandwidth-normalized energy efficiency (BwNEE) is represented
in eight scenarios as a function of the average EIRP per hop. The figure highlights the effect
8034
Fig. 9. End-to-end average block error rate BLER versus average EIRP.
of the end-to-end-distance (d), adjacent node distances (di ), the number of hops (N ), the
frequency band (W1 and W2 ), and the shape parameter (α). The figure shows, from S3 and
S4, that distinct distributions of the nodes along this distance affect significantly the BwNEE
of the system. Using the proposed power allocation optimization, as illustrated by S4 and S5,
the BwNEE has considerably increased. We may also notice, from comparing S0 and S1, that
the system becomes less energy efficient when the channel conditions improve (increasing α)
for low values of the EIRPP (less than 25dBm) which may be beneficial for the low power
applications. By studying S1, S2, and S3, we perceive that the increased number of hops
decreases the system’s BwNEE.
The previous observations about the shape parameter and the number of hops may be
misleading. In fact, as shown in Fig. 11, where we plot the BwNEE versus the number
of hops and the EIRP per hop for two values of the shape parameter (α = βi /2 = 1 and
α = βi /2 = π /2). The figure shows that the behavior of the system’s BwNEE depends on all
the parameters. This behavior of the BwNEE is due to the dual role of the transmit power;
it acts in the advantage of the BwNEE inside the SNR and acts harmfully within the total
consumed power of the system.
Finally, in Fig. 12, we studied the energy efficiency of low-power (LP) narrow-band (NB)
communications, emulating an IoT environment. We notice that increasing the bandwidth
results in a significant gain in the energy efficiency of the system. However, in long distance
(d = 400m) communications with at most one relay, the very low power narrow-band system
8035
Fig. 10. End-to-end energy efficiency versus the total power in the bands W1 = 28GHz and W2 = 73GHz.
[Pc = 100mW, Bw = 180 kHz] (equivalent to NB-IoT [36]) is much energy efficient than
low power system [Pc = 200mW, Bw = 1.4 MHz] (equivalent to CAT-M1 [36]). For short
distances, adding relaying nodes seems to be without interest in terms of EE, it becomes
more interesting when the end nodes (S and D) get more distant.
6. Extensions discussion
Several techniques and aspects of the next generation of mobile networks are not taken
into consideration or are not investigated in details in this work. A non exhaustive list of
these aspects can be summarized in the following.
6.1. Directivity
Beamforming is one of the main key enablers of the fifth generation systems. Some aspects
that can be investigated to extend this work are:
• a more sophisticated analytical analysis of the impact of beamforming,

• optimizing the pilot allocation to reduce the effect of pilot contamination,
• highlighting the effect of angle of arrival/departure (AoA/AoD) estimation on the system,
• multi-user and access techniques in respect to multiple antenna communications.
8036
Fig. 11. End-to-end EE EE e2e versus the number of relays and the EIRP per hop. The end-to-end distance of the
system is d = 300 m.
6.2. Channel estimation effect
The channel estimation in a mmWave context represents a real challenge especially when
coupled with massive MIMO schemes. Besides outdated CSI, the effect of CSI can be dis-
cussed from several other points of view:
• limited feedback communication,
• erroneous channel estimation,
• pilot contamination effect.
6.3. Interference
As a performance limiting factor, this work can be extended by including the interference
aspects into the analysis. This can be done assuming:
• full-duplex relays, and analyzing the effect of the residual interference,

• inter node interference, as a result of scheduling schemes,
• different recently proposed waveforms.
7. Conclusion
In this paper, we analyzed and discussed the performance of multihop regenerative relaying
in the context of mmWave communications as a key enabler for the next generation of mobile
8037
Fig. 12. End-to-end EE EE e2e versus the number of relays (N-1). Nodes are communicating in the 28GHz band
using an EIRP= 8 dBm with βi = 2.
communication systems. Considering a general M-QAM modulation order, exact closed-form

and asymptotic physical-layer level end-to-end performance metrics (outage probability, BER,
SER, BLER, ergodic capacity, and energy efficiency) were derived for Weibull fading links,
based on the generalized Fox’s H- and Meijer’s G-functions expressions. Based on the ob-
tained results, we computed error and energy efficiency optimal transmit power allocation
strategies, and we showed that they offer considerable gains.
Simulation results confirmed the accuracy of our analysis for a large selection of chan-
nel and system parameters. As a secondary contribution, we proposed new and generalized
implementations of Fox’s single and bivariate H-functions in M ATLAB.
To complete the analysis, more investigations are necessary, and it will be very interesting
to take other aspects into consideration (for example power constraints, delay, and transmission
scheduling) to get a wider and cross-layer insight into the design and optimization of multihop
communication schemes.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.
8038
Appendix A. Fox’s H-Function M ATLAB Code
8039
Appendix B. Bivariate Fox’s H-Function M ATLAB Code
8040
8041
References
[1] H. Li, M. Lott, M. Weckerle, W. Zirwas, E. Schulz, Multihop communications in future mobile radio networks,
in: The 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC),
Lisbon, Portugal, 2002.
[2] J. Reig, M. Martínez-Inglés, L. Rubio, V. Rodrigo-Peñarrocha, J. Molina-García-Pardo, Fading evaluation in the
60 GHz band in line-of-sight conditions, Int. J. Antennas Propag. 2014 (2014) 1–12.
[3] G.R. MacCartney, T.S. Rappaport, S. Rangan, Rapid fading due to human blockage in pedestrian crowds at 5G
millimeter-wave frequencies, in: IEEE Global Communications Conference, Singapore, 2017, pp. 1–7.
[4] C. Lim, M. Lee, R. Burkholder, J. Volakis, R. Marhefka, 60 GHz indoor propagation studies for wireless
communications based on a ray-tracing method, EURASIP J. Wirel. Commun. Netw. 2007 (2007) 49–54.
[5] Y. Niu, Y. Li, D. Jin, L. Su, A.V. Vasilakos, A survey of millimeter wave (mmWave) communications for 5G:
opportunities and challenges, Wirel. Netw. 21 (8) (2015) 2657–2676.
[6] T.S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G.N. Wong, J.K. Schulz, M. Samimi, F. Gutier-
rez Jr, Millimeter wave mobile communications for 5G cellular: it will work!, IEEE Access 1 (1) (2013)
335–349, doi:10.1109/ACCESS.2013.2260813.
[7] J.G. Andrews, T. Bai, M.N. Kulkarni, A. Alkhateeb, A.K. Gupta, R.W. Heath, Modeling and analyzing millimeter
wave cellular systems, IEEE Trans. Commun. 65 (1) (2017) 403–430.
[8] S.S. Ikki, M.H. Ahmed, Performance of multi-hop relaying systems over Weibull fading channels, in: New
Technologies, Mobility and Security, Springer Netherlands, 2007, pp. 31–38, doi:10.1007/978- 1- 4020- 6270- 4_3.
[9] P. Wang, J. Zhang, L. Guo, T. Shang, T. Cao, R. Wang, Y. Yang, Performance analysis for relay-aided multihop
BPPM FSO communication system over exponentiated Weibull fading channels with pointing error impairments,
IEEE Photon. J. 7 (4) (2015) 1–20.
[10] H. Hashemi, J. Haghighat, M. Eslami, Performance analysis of relay-aided millimeter-wave communications
with optimal and suboptimal combining at destination, Phys. Commun. 39 (2020) 100991.
[11] N. Hajri, R. Khedhiri, N. Youssef, On selection combining diversity in dual-hop relaying systems over double
Rice channels: Fade statistics and performance analysis, IEEE Access 8 (2020) 72188–72203.
[12] M. Dohler, Y. Li, Cooperative Communications: Hardware, Channel and PHY, John Wiley & Sons, 2010.
[13] D. Chen, J.N. Laneman, Modulation and demodulation for cooperative diversity in wireless systems, IEEE
Trans. Wirel. Commun. 5 (7) (2006) 1785–1794.
[14] M. Benjillali, L. Szczecinski, A simple detect-and-forward scheme for fading channels, IEEE Commun. Lett.
13 (5) (2009) 309–311.
8042
[15] A. Soulimani, M. Benjillali, Closed-form performance analysis of generalized M-QAM over multihop Weibull
fading channels, in: Proc. International Conference on Wireless Networks and Mobile Communications (WIN-
COM), Marrakech, Morocco, 2015, pp. 1–5.
[16] A. Soulimani, M. Benjillali, H. Chergui, Energy-efficient multihop schemes over Weibull-fading channels: Per-
formance analysis and optimization, in: 13th International Wireless Communications and Mobile Computing
Conference (IWCMC), Valencia, Spain, 2017, pp. 2009–2014.
[17] K.P. Peppas, A new formula for the average bit error probability of dual-hop amplify-and-forward relaying
systems over generalized shadowed fading channels, IEEE Wirel. Commun. Lett. 1 (2) (2012) 85–88.
[18] F. Yilmaz, M.S. Alouini, Product of the powers of generalized Nakagami-m variates and performance of cascaded
fading channels, in: IEEE Global Telecommunications Conference (GLOBECOM), IEEE, Hawaii, US, 2009,
pp. 1–8.
[19] M.R. Akdeniz, Y. Liu, M.K. Samimi, S. Sun, S. Rangan, T.S. Rappaport, E. Erkip, Millimeter wave channel
modeling and cellular capacity evaluation, IEEE J. Sel. Areas Commun. 32 (6) (2014) 1164–1179.
[20] 3GPP, Study on 3d channel model for LTE, 2017. Technical report 36.873 v12.5.1.
[21] M.K. Simon, M.S. Alouini, Digital Communication Over Fading Channels, John Wiley & Sons, 2005.
[22] M.D. Yacoub, D.B.d. Costa, U.S. Dias, G. Fraidenraich, Joint statistics for two correlated Weibull variates, IEEE
Antennas Wirel. Propag. Lett. 4 (2005) 129–132.
[23] E. Morgado, I. Mora-Jiménez, J.J. Vinagre, J. Ramos, A.J. Caamaño, End-to-end average BER in multihop
wireless networks over fading channels, IEEE Trans. Wirel. Commun. 9 (8) (2010) 2478–2487.
[24] K. Cho, D. Yoon, On the general BER expression of one-and two-dimensional amplitude modulations, IEEE
Trans. Commun. 50 (7) (2002) 1074–1080.
[25] A.A. Kilbas, H-Transforms: Theory and Applications, CRC Press, 2004.
[26] H. Chergui, M. Benjillali, S. Saoudi, Performance analysis of project-and-forward relaying in mixed MIMO-
pinhole and Rayleigh dual-hop channel, IEEE Commun. Lett. 20 (3) (2016) 610–613, doi:10.1109/LCOMM.
2016.2514348.
[27] P.K. Mittal, K.C. Gupta, An integral involving generalized function of two variables, in: Proceedings of the
Indian Academy of Sciences-Section A, 75, 1972, pp. 117–123.
[28] V.A. Marčenko, L.A. Pastur, Distribution of eigenvalues in certain sets of random matrices, Mat. Sb. (N.S.) 72
(114) (1967) 507–536.
[29] A.M. Mathai, R.K. Saxena, H.J. Haubold, The H-function: Theory and Applications, Springer Science & Busi-
ness Media, 2009.
[30] A. Muller, J. Speidel, Symbol error probability of M-QAM in multihop communication systems with re-
generative relays, in: IEEE Vehicular Technology Conference (VTC-Spring), Singapore, 2008. 1004–1008.
doi:10.1109/VETECS.2008.215
[31] Meijer G-function: primary definition, http:// functions.wolfram.com/ HypergeometricFunctions/MeijerG/02/
[32] Y. Gu, H. Chen, Y. Li, B. Vucetic, Ultra-reliable short-packet communications: half-duplex or full-duplex re-
laying? IEEE Wirel. Commun. Lett. 7 (3) (2018) 348–351.
[33] P. Mary, J.M. Gorce, A. Unsal, H.V. Poor, Finite blocklength information theory: what is the practical impact on
wireless communications? in: IEEE Globecom Workshops (GC Wkshps), Washington, DC, USA, 2016, pp. 1–6.
[34] J.M. Bernardo, Algorithm AS 103: PSI (digamma) function, J. R. Stat. Soc. Ser. C 25 (3) (1976) 315–317.
[35] J.N. Murdock, T.S. Rappaport, Consumption factor and power-efficiency factor: atheory for evaluating the energy
efficiency of cascaded communication systems, IEEE J. Sel. Areas Commun. 32 (2) (2014) 221–236.
[36] GSMA, 3GPP low power wide area technologies - GSMA white paper.
[37] N. Estes, K. Gao, B. Hochwald, J.N. Laneman, J. Chisum, Efficient modeling of low-resolution millimeter-wave
transceivers for massive MIMO wireless communications systems, Microwave and Optical Technology Letters.
[38] B. Sadhu, Y. Tousi, J. Hallin, S. Sahl, S. Reynolds, O. Renström, K. Sjögren, O. Haapalahti, N. Mazor,
B. Bokinge, et al., A 28GHz 32-element phased-array transceiver IC with concurrent dual polarized beams
and 1.4 degree beam-steering resolution for 5G communication, in: IEEE International Solid-State Circuits
Conference (ISSCC), San Francisco, CA, USA, 2017, pp. 128–129.
[39] M. Kraemer, D. Dragomirescu, R. Plana, Design of a very low-power, low-cost 60 GHz receiver front-end
implemented in 65 nm CMOS technology, Int. J. Microwave Wirel. Technol. 3 (2) (2011) 131–138.
[40] W. Deng, Z. Song, R. Ma, J. Lin, Y. Li, J. Ye, S. Kong, S. Hu, H. Jia, B. Chi, An energy-efficient 10-Gb/s
CMOS millimeter-wave transceiver with direct-modulation digital transmitter and I/Q phase-coupled frequency
synthesizer, IEEE J. Solid-State Circt. 55 (8) (2020) 2027–2042.
[41] V. Vidojkovic, G. Mangraviti, K. Khalaf, V. Szortyka, K. Vaesen, W.V. Thillo, B. Parvais, M. Libois, S. Thijs,
J.R. Long, et al., A low-power 57-to-66GHz transceiver in 40nm LP CMOS with- 17DB EVM at 7Gb/s, in:
8043
IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC), San Francisco, CA,
USA, 2012, pp. 268–270.
[42] R. Wu, R. Minami, Y. Tsukui, S. Kawai, Y. Seo, S. Sato, K. Kimura, S. Kondo, T. Ueno, N. Fajri, S. Maki,
N. Nagashima, Y. Takeuchi, T. Yamaguchi, A. Musa, K.K. Tokgoz, T. Siriburanon, B. Liu, Y. Wang, J. Pang,
N. Li, M. Miyahara, K. Okada, A. Matsuzawa, 64-QAM 60-GHz CMOS transceivers for IEEE 802.11ad/ay,
IEEE J. Solid-State Circt. 52 (11) (2017) 2871–2891.
[43] N. Fourty, A.V.D. Bossche, T. Val, An advanced study of energy consumption in an IEEE 802.15.4 based
network: everything but the truth on 802.15.4 node lifetime, Comput. Commun. 35 (14) (2012) 1759–1767.
8044

Matlab Codes

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Matlab Codes

Uploaded by

Copyright:

Available Formats

Available online at www.sciencedirect.

Journal of the Franklin Institute 358 (2021) 8012–8044

Multihop Weibull-fading communications:

Fig. 1. Adopted system model with (N −1) regenerative multihop relays.

2.1. Perfect channel state information

2.2. Outdated channel state information

= π (1 − ρ)αi zαi −1 (10)

Integrating (9), we get the CDF of γ˜i as

3. Comprehensive performance analysis

3.1. Outage probability

3.1.1. Perfect CSI

For low γth values, (16) can be approximated by

3.1.2. Outdated CSI

3.2. Bit error rate

where BERi is the average BER of the individual ith hop.

3.2.1. Closed-form exact analysis

3.2.2. Asymptotic analysis

By replacing (31) in (26), we get

3.2.3. Outdated CSI

Using [27, Eq. 2.1], we get this integral in closed form as

result Y˜i = Si Xi + W˜ i , namely, the k-th signal at the receiver is

where a = aγ i,k , b = bγ i,k , and

3.3. Symbol error rate

3.3.1. Closed-form exact analysis

3.3.2. Asymptotic analysis

where the Meijer G-Function [31] is used.

3.4. Block error rate

By recurrence, it can be shown that

which can be made more compact under the form

3.5. Ergodic capacity

3.5.1. Closed-form exact analysis

3.5.2. Asymptotic expression  α

3.6. Energy efficiency

3.6.1. Closed-form exact analysis

To proceed further with the derivation of EE e2e , we suppose that αi = α, i = 1, . . . , N .

3.6.2. Asymptotic expression

4. Power allocation optimization

4.1. BER-Optimal Power allocation

4.2. EE-Optimal Power allocation

i=1 Pi−1 = Pmax

yielding the optimal transmit powers maximizing the end-to-end EE.

• noise power: −174 dBm/Hz,

5.2. Outage probability

Fig. 3 depicts the bandwidth-normalized end-to-end ergodic capacity C as a function of

• a more sophisticated analytical analysis of the impact of beamforming,

• full-duplex relays, and analyzing the effect of the residual interference,

communication systems. Considering a general M-QAM modulation order, exact closed-form

Declaration of Competing Interest

Appendix A. Fox’s H-Function M ATLAB Code

Appendix B. Bivariate Fox’s H-Function M ATLAB Code

You might also like

3.5.2. Asymptotic expression α