Balanced Performance Between Energy-Delay and

Bit Error Rate in UAV Relay Networks

M khalil, Member, IEEE

Abstract—This paper presents a new strategy for nano-satellite communication applications, which are expected
arXiv:2205.00797v2 [eess.SP] 15 May 2022

simultaneously reducing energy consumption, transmission to be an essential part of the 5G networks.

delays, and bit error rate in Unmanned Aerial Vehicle (UAV) The drone’s battery unit, named Battery Eliminator Circuit
networks. A UAV is fitted with a wireless Bidirectional Relay
(BR) to enable coverage network extension and increase (BEC), supplies the necessary energy for the drone, the AF
transmission throughput. The downside of the BR advantages is relay and other components equipped on the drone. Using
the delay in data transmission caused by the UAV’s movement. A AF relay nodes increases the drain of AR’s BEC, particularly
consequence of this delay is increased total energy consumption, in the transmitting mode at which the relay node becomes
causing further degradation in bit error rate performance, active. In the transmission mode, the energy model presents
especially at high SNR levels. In wireless communication, the
trade-off between delay and energy consumption is, fortunately, energy consumption for each transmitted bit. Another factor
possible to improve performance. Therefore, this study aims that increases the drone networks’ energy consumption is
to enhance UAV network performance by reducing energy the delay due to the UAV movement. Further, transmission
consumption, data transmission delay, and bit error rate. A signals in tow-hop significantly impact each node’s energy
multi-objective algorithm is employed to generate an adaptive consumption [4]. Rising energy consumption again degrades
optimal energy allocation strategy based on the balance between
energy consumption and transmission delays. The results of bit error rate, particularly at high levels of signal-to-noise ratio
theoretical analysis are illustrated with several examples. As (SNR) [5]. Moreover, in UAV networks, users and drone nodes
herein demonstrated, the proposed solution effectively balances often rely on a battery with a limited amount of energy. Thus,
delay and energy efficiency in a customised system design and minimizing energy consumption and delay is a fundamental
improves the bit error rate in UAV networks. goal in UAV networks.
Index Terms—UAV, two-way amplify-and-forward, relay, Unfortunately, the delay, energy and bit error rate objectives
energy, delay cannot be minimized simultaneously as these metrics having
conflict with each other. Such a conflicting problem, however,
can be addressed by using a multi-objective solution [6]. One
I. Introduction
of the effective methods in multi-objective is the weight of
Unmanned aerial vehicles relay network is a communication the scalarization method [7]. All objective functions in the
system employing drones fitted with wireless relay devices weight scalarization method are consolidated into a single part
to enhance the scope and flexibility of communication showing a linear function. Thus, we can combine the delay,
systems. UAV is currently one of the essential technologies energy and bit error rate objectives into one function. However,
serving various applications besides telecommunications, solving such a single function for a bidirectional relay network
such as product deliveries, aerial photography, policing is challenging as the communication scenario coincides in
and surveillance, military and infrastructure inspections. An two directions. This paper proposes dividing the problem into
example of UAV’s importance has been illustrated in the two sub-problems to simplify the weight scalarization function
recent outbreak of Coronavirus disease, where UAV was analysis in a bidirectional relay network. In the first problem,
employed to perform surveillance public and persuade them to energy consumption and transmission delay are combined in
follow public health best practices [1]. In a telecommunication one function. The solution to the first problem is employed
application, UAV enables communications between many land to optimize energy allocation parameters. Such parameters
user nodes to take place through UAV nodes. It is recognized reduce delay and energy consumption and enhance bit error
that by fitting Decode-and-Forward (DF) or Amplify-and- rate.
Forward (AF) relay systems, the drone network gives better Many studies have analysed the balance between energy
coverage and better throughput and energy performance [2]. and delay by using information theory; it significantly focuses
An AF relay is considered in this paper as it can theoretically on designing power allocation under various constraints on
be applied with less complexity than DF relay, which requires the information delay, such as an average delay constraint
complete processing, including encoding, re-modulating and for a buffer, queuing delay in [8], per-packet delay constraint
re-transmitting the received signal. Such operations processes [9], in addition, a multipacket transmission [10]. Study [11]
require sophisticated power control, which is unnecessary in uses a minimum departure time as a model of packet delay
an AF relay. The relay node operates as either one-way constraints; this scheme can be applied to model various
(unidirectional), or two-way (bidirectional). This paper focuses quality of service (QoS) constraints. Further, the design of
on the BR AF type as several studies, such as [3] have the system algorithm depends on the availability of Channel
observed that the BR AF system analysis is appropriate for State Information (CSI), which can include fading channels
and time-variation as error rate by selecting the best relay location is impractical.
According to Shannon’s capacity theorem the minimum of Therefore, another work [24] optimised trajectory and energy
energy per bit rate(E) control at the same time. The UAV trajectory is also optimised
noise power density (No) needed to achieve arbitrarily low bit error
 jointly with the device-UAV association and uplink power to
probability as E/No = ln(2). For a given error probability
min minimise the total transmission power according to the number
and code rate with finite bits, the minimum energy has been
of updates in [25].
studied in [12]. Authors [13] sought to maximize the average
throughput, which is the equivalent of minimizing the average All of the aforementioned studies demonstrated effective
delay-per-bit for a given number of bits and input power. Study schemes to improve UAV networks’ performance in terms of
[14] presents a minimal energy solution for transmuting finite UAV placement and energy allocation. However, an essential
bits without delay constraints, and [15] a solution for Energy- factor that is largely ignored in these works is that UAV
delay balance over fading channels has been demonstrated in networks may have slightly larger transmission times, so
[15]. In [16], the model to minimize the energy consumption the data received from the ground user will have various
of the intermediate relay between the source and the receiver SNR levels. Furthermore, works that developed the energy
was adopted for wireless terrestrial relay. The locations of consumption and data transmission delay metrics, jointly or
wireless relays may, however, be random in practical networks. individually, ignored the relationship between these metrics
Hence, in [16] the effects of randomly positioned relays and the bit error rate, which is an essential metric for
were examined, but this study investigated selecting the best evaluating the performance of UAV network applications [26].
relay location based on linear places between the source and Thus, [27] presented a specific system model that can only
destination. However, in these analyses, the CSI is assumed to provide balance energy and delay transmission data for a
be available at the source or destination nodes. Nonetheless, unidirectional UAV AF relay flying in a triangle formation.
the fading channel at mobility relay locations often varies Error-free reception, however, is very limed in the practical
rapidly, which can make it difficult to estimate, especially if environment, particularly in multi-hop networks with varying
the relay is moving in space. channel conditions. Also, the unidirectional relay operates in
one-way communication between the source and destination,
For this reason, Doppler effects are considered by [17],
whereas such communication has limited applicability. To
who assumed the relay was flying at a constant speed
respond to this gap, this paper considers a general trade-
and the destination node able to estimate and compensate.
off energy-delay scenario that enables energy allocation to
The mentioned studies, however, were focused solely on
be an adaptive factor for providing an optimal bit error rate
transmission delay, without taking into consideration network
for UAV bidirectional AF relay networks over mobile fading
energy consumption. In contrast, other researchers [2, 18]
channels. The proposed method is achieved by simultaneously
were mainly focused on network energy consumption. Studies
optimising both energy consumption and transmission delay in
[6, 19] investigated the energy consumption with delay
UAV networks. The optimum energy allocation is distributed
constraints for UAV AF relay network. Based on the results of
between relay and source nodes in UAV networks. Such
these studies, power allocation is essential to maximize UAV
energy allocation solves two problems. First, it enables
throughput. However, in many applications, particularly for
the development of a decision-making platform to achieve
energy-limited appliances like sensor or relay systems, energy
the best trade-off between energy consumption and data
is the main parameter that carries out a specific operation
transmission delay. Second, it enhances the performance of
than its power consumption. Generally, the energy parameter
UAV networks in terms of bit error rate. Hence, the proposed
is subject to time delay, and in the meantime, the UAV network
method simultaneously improves energy consumption, data
has strict delay regarding safety information transmission
transmission delays and bit error rate in UAV networks.
delays. Despite that, [6] recognized that distributing energy
The contribution of this paper is summarised by defining an
value among terrestrial networks nodes reduces delay and
algorithm allowing to compute an optimal energy allocation
overall energy consumption.
to each node (users and relay) of the UAV communication
The location of UAV changes periodically, then the received network, and further, it enhances the bit error rate. Most
SNR has time-varying characteristics; thus, bit rate often notably, the algorithm strikes a balance between transmission
changes [20]. Accordingly, several researchers worked on delays and energy as follows:
determining the effect of channel characteristics on bit error
rate performance. For instance, study [2] adopted variable rate • Define UAV network by three nodes, which are two
protocol to enhance bit error rate performance and achievable terrestrial users S a and S b and a wireless relay (R) fitted
information rate, as the location of relay changes periodically. on a drone.
Another researcher [18] investigated how to place UAVs to • Calculate a flight distance (da ) between S a and R and the
reduce bit error rates optimally. One paper [21] proposed a path (db ) between S b and R.
path loss model that accommodates both Line of Sight (LOS) • Consider the distances da and db to calculate the end-to-
and Non-Line of Sight (NLOS) path loss conditions. Likewise, end SNRs at S a and S b .
the authors of [22] extended their results to include three- • Use expressions of SNRs at S a and S b to calculate energy
dimensional space. In [23], the optimum location of device- allocations for S a as (αa ), relay (αr ) and S b as (αb ).
to-device communications was also considered in UAV to • Define the bit energy consumption (Ea ) for S a and (Eb )
enhance network performance. However, optimising the bit for S b as functions of energy allocations factors, i.e.,
 Ea (αa , αb , αr ) and Eb (αa , αb , αr ), also distance (d) of da and db is varied based on r value. Thus, the
• Define bit transmission times (qa ) at S a and (qb ) at S a Euclidean distance between S a and R is calculated as
2
!1/2
as functions of energy allocations as qa (αa , αb , αr ) and 2 r
qb (αa , αb , αr ). da = d + (3 − cos(2φ)) − ψ , (1)
2
• Use Scalarization optimization method demonstrated
in [28] to optimize Ea (αa , αb , αr ), Eb (αa , αb , αr ), and the distance between R and S b is
!1/2
qa (αa , αb , αr ) and qb (αa , αb , αr ), simultaneously, and this r2
db = d2 + (3 − cos(2φ)) + ψ , (2)
provides an effective method to balance delay-energy 2
performance on customized system design. In other where ψ = 2 r d cosθ.
words, the proposed method offers a decision-making
scheme responsible achieving the best trade-off between TABLE I: Summary of Notations
transmission delay and energy consumption in UAV
networks. Notations Description
• Use optimum αa , αb , αr to maximise end-to-end SNR and
thus lower bit error rates. S a and S b two terrestrial wireless users
The proposed method has been evaluated for the UAV network,
and the analytic outcomes reveal that the proposed approach da and db the distances between
enables energy consumption, transmission delay and bit error source-drone and drone-
rate to be minimized in a well-balanced scheme. receiver, respectively.
Though the energy allocation strategy has been used to
optimise wireless network performance, as in [29], this
pa ,pb and pR allocation powers for S a , S b
paper emphasises the potential of using such a strategy to
and relay, respectively.
simultaneously improve the data rate, energy consumption and
bit error rate performance in UAV networks. According to the
author’s knowledge, a UAV network has not been optimised ℜa , ℜb data rates for S a , S b
using such a schema previously.
Before further discussion, we provide Table I to summarized h and g fading channels for da and
notations used in this manuscript. db , respectively.
The rest of this paper is arranged as follows: Sections II
and III describe the system model and the received signal at Ya and Yb received signal by S b and
destination node analysis, respectively; Section IV introduces S b , respectively.
a new method to calculate energy allocation for UAV users and
relay; Section VI presents the simulation results; and finally, γa and γb SNR at S b and S b .
the conclusion is presented in Section VII.
qa and qb data transmission time by
II. SYSTEM MODEL S b and S b , respectively.
Consider two terrestrial wireless users, S a and S b ,
exchanging information simultaneously via a wireless relay Ea and Eb data transmission energies
(R) node fitted with an drone, i.e., an UAV network. The relay for S b and S b , respectively.
acts in a two-way amplify-and-forward (AF) relay mode. It is
assumed that all the nodes are equipped with a single antenna, αa ,αb and αr energies allocation factors
and all the links among the nodes are half-duplex and use the for S b , S b and the relay,
same carrier frequency. Flying a drone involves flight along respectively.
any path in coordinates of three dimensions (x, y, z), where
(±x, 0, 0) represent the locations of land nodes i.e. S a and S b wa , wb and wr weight coefficients for
and z represents the altitude (H) as shown in Fig (1). S b , S b and the relay,
For analyzing UAV flight path, we suppose that the respectively.
drone flies along a path defined by a series of waypoints, áe bit error rate
which is assumed to be the drone’s initial position, defined
as pi (xi , yi , zi ). Thus, the Cartesian coordinates of pi are
determined from the coordinate transformations as xi =
r cos(θ), yi = r sin(θ) and zi = r sin(φ). The polar coordinates III. Received signal at detestation node
r, θ and φ are measured from the centre O(0, 0, z) of the flight This part derives an expression for the users SNRs of the
path. proposed UAV network. We first assume that the destination
The Euclidean distance among S a , O and Sb are calculated is entirely compensated by the Doppler effect due to the
at r = 0, as ±d which represents the separating distance UAV’s mobility as the UAV follows a trajectory with a fixed
between S a and S b . On the other hand, when r > 0, the total flying speed [30]. Thereby, both flat fading channels Sa − R
 R and variance (σ2 ).
The YR signal is then amplified by the relay amplification
factor (β) given by:
q
       
β n = pR n /(|h|2 da−α pa n + |g|2 db−α pb n + σ2 ), (4)
 
where pR n is the allocated power for R node .
After that, each user receives the amplified signals through
the same flat fading channel, during the second time slot. The
z received signal by S a is then included in the relay-amplified
signal beside the direct signal received from S b node. This
gives   −α   √ −α
nS Ya n = h da2 β n yR + pb χb g db2
R g √ −α
Ðf
+2 pb χb gh |d| 2 + na . (5)
Sa r Node S b also involves the relay-amplified signal and the
o Sb direct S a signal as the following
(-d,0,0) x   −α   √ −α
Ðq (d,0,0)
Yb n = g db2 β n yR + pa χa h da2
√ −α
+2 pa χa gh |d| 2 + nb , (6)
y
where gh is the channel coefficient between user S a and
Fig. 1: System Model user S b with the separating distance 2|d| ,and na and nb are
the users Gaussian noises with a zero mean and variance (σ2 )
for both user S a and user S b , respectively.
and R − Sb have power gains following the free-space path Equations (5) and (6) each have their own transmission
−7 −7
signal mixed with the received signal and this is returned to
attenuation schemas h da2 and g db2 ,respectively, where h is
self-interference (SI). Here, we assume that both users have
the channel coefficient between S a and R, g is the channel
an SI cancellation circuit that allows for their own transmitted
coefficient between S b and R, 7 is the path loss exponent
signal to be cancelled out [32]. Thus, the SNRs at each user
which is commonly estimated in the range of 2 ≤ 7 ≤ 4. It is
are given by
also assumed that the channel’s characteristics are available at
the user nodes. Further, user S a is able to adjust the transmit   pR γb pb γa db−α dh−α
γaι n = + pR pb γab db−α , (7)
powers in S b and relay R. Similarly, S a can regulate S a and γa (pR + pa )da−α + pb γb db−α + 1
relay R powers using the power allocation process.   pR γb pa γa da−α db−α
A further assumption is that the users adopt binary phase γbι n = + pR pa γab da−α , (8)
pa γa da−α + (pR + pb ) γb db−α + 1
shift keying modulation to broadcast their signals χa and
χb , with average power pa and pb , from users S a and where γaι and γbι are the actual line-of-sight SNRs for S a
2 2 2
S b , respectively. Both signals χa and χb are transmitted and S b , respectively, γab = |gh|
σ2
, γa = |h|
σ2
and γb = |g|
σ2
.
during the time interval 0 < t ≤ T ,and each signal follows a In the proposed UAV, the direct link between S a and
circularly symmetric complex Gaussian distribution CN(0, 1), user S b is usually not possible due to long-distance flight
2 2 and degradation in channel quality. Therefore, the destination
i.e. E{ χa } = 1 and E{ χb } = 1, where E{.} denotes an
2 nodes receive only the relayed signals from the R node, i.e.,
expected value and . is the absolute square of a signal.
γab = 0. Equations (7) and (8) then become
The S a and S b nodes transmit their bits according to a
schedule that defines the commencing and the duration of   pR γb pb γa da−α db−α
γae n = , (9)
each bit transmission. Then, χa and χb arrive to their target γa (pR + pa )da−α + pb γb db−α + 1
node in the time interval 0 < t ≤ T . In order to simplify,   pR γb pa γa da−α db−α
T is discretized into n time slots as δt =T /N, where δt γbe n = , (10)
pa γa da−α + (pR + pb ) γb db−α + 1
represents the time slot length. The value of δt is assumed
to be small enough that the UAV’s location can be supposed where γae and γbe are the exact non-line-of-sight SNRs for S a
to be approximately constant within a slot. Thereby, the UAV’s and S b , respectively
trajectory (xi (t), yi (t), zi (t)) over T can be specified by N scope The average power pr , p(a and pb should not exceed the )
as xi [n], yi [n], zi (n) N
 3 P
n ∈ R ,where n = 1, 2, ...N − 1. overall network power (P) P : P ∈ R, P = pi i ∈ (a, r, b) .
The exchange signals between S a and S b occur in two-hops. i

In the first hop, the relay receives signals from both S a and In order to manage P value, the power allocation strategy
S b as is used to allocate a specific power value to the relay and
q
 
q
  −α both user nodes. All nodes have inclusive knowledge of
YR [n] = pa n χa h + pb n χb g db2 + nR, (3) information about channels and maximum transmission power
−α −α for each related node [33], so both S a and S b nodes achieve
where h da2 and g db2 are the fading channels model defined the allocation strategy by employing the reverse channels for
in [31], nR is the Gaussian noise of the relay with zero mean feedback allocated energy factors to the corresponding node.
After that, S a , S b and relay nodes adjust their transmit power Input: P, x(t), αa , αb , and αr
based on these feedback factors [34]. Initialization of parameters
It assumes that the energy allocation factors S a , R and S b , while t , 0 do
respectively, are set as da ← (1)
αa ⇒ {αa :∈ R, → 0 ≤ αa ≤ 1} , (11) db ← (2)
αr ⇒ {αr :∈ R, → 0 ≤ αr ≤ 1} , (12) γa ← EYa [n] ⊲ expectation value of (9)
αb ⇒ {αb :∈ R, → 0 ≤ αb ≤ 1} . (13) γa ← EYb [n] ⊲ expectation value of (10) pa = P αa
; pb = P αb ; pr = P αr
Once received, the corresponding nodes adjust their powers
γa ← (14)
as P αb , P αr and P αa . So, substituting these regulated powers
γb ← (15)
into (9) and (10) and considering high SNR domain, we get end
   
αr n γb αb n γa da−7 db−7 Output: SNR
γa [n] = P       , (14)
γa (αr n + αa n )da−7 + αb n γb db−7 Algorithm 1: Procedure of SNR analysis
   
αr n γb αa n γa da−7 db−7
γb [n] = P       , (15)
αa n γa da−7 + (αr n + αb n ) γb db−7
where γa and γb are the high SNRs domain for (9) and (10), allows total energy consumption and data transmission time
respectively (delay) to be minimised under a balanced approach.
Now, another metric is the channel capacity, which is given In Shannon’s theorem, prolonging the transmission
by 1 time reduces transmitting power. Thus, the total energy
C[n] = log2 (1 + φ[n]). bits/s, (16) consumption in UAV networks is minimised by maximising
2
transmission time. However, increasing transmission delay
where φ[n] = 1 + γa[n] + γb[n] + γa[n]γb [n].
in transmitting the information directly affects the user’s
To adapt reliably transmitted information rate (R), we have
service. Hence, transmission time (q) and power networks
R[n] = h C[n] (0 < h < 1) (17) must be designed with a trade-off scheme. q is the amount
At h = 1,the maximum data rate in (16) becomes of time required by a user to send out a single packet of
1
bits; it depends on the network’s bandwidth and length of
ℜi = ℜa + ℜb = log2 (1 + γa[n]) + log2 (1 + γb[n]) , (18) the packet, as q = (Data size/bandwidth) (sec). By using
2
where ℜi is the total data rate in two way [35]. low-latency algorithms or low-delay transmission protocols,
As mentioned earlier, the data rate R[n] is passed through data transmission amounts can be adjusted, and delays can
two hops, i.e., two cascade channels. The transmission rate be reduced. Such approaches allow the data transmission
through two channels is subjected to an Information Cascade amount to be managed according to the change in delay
(IC) analysis [36], which is defined as a propagation sequence performance due to a change in transmission amount. Each
of data bits transmitting over chaotic channels. Further, the data bit delivered at q can consume energy (E) as E = q P.
mutual-information flow along the cascade of channels cannot Then, the total energy model in UAV is defined from (14),
exceed each channel individually. Along the cascade channels, (15) and (17) as follows
the flow mutual information capacity cannot overpass each        
 2 
Ea n = q[n] (αr n + αa n )H + αb n G 2 q[n] − 1 , (20)
channel capacity. Regarding UAV channels, the first channel
delivers a sequence of bits that transmitted by users nodes to        
 2 
Eb n = q[n] αa n H + (αr n + αb n )G 2 q[n] − 1 , (21)
the R node during a time slot n = 1. The R node requires one-
slot processing to forward the received data to the S a and S b where G = γb db−7 , H = γa da−7 .
node during the second time slots, i.e. n= 2, 3, 4,...N [37]. Concurrent with the information exchanging between S a
Then, the information-causality constrain is obtained as: and S b , the overall energy consumption is typically generated
n n−1      
X X by including (20) and (21), i.e., E n = Ea n + Eb n , as in
R[i] ≤ R[i] . (19)  
[2]. The region E n can also be specified by the union of all
i=2 i=1
| {z } | {z } the possible sets of (Ea , Eb ) , as in [30]. Hence, if the region
 
Sa → R → Sb R → Sa E n is managed by varying αa , αb , αr ,then the union of the
Sb → R → Sa R → Sb two sets of energies in (20) and (21) are subject to definition
1.
S a → R → S b and S b → R → S a represent the transmitted a Definition 1. The regions of two sets Ea and Eb is
signal from S a to S b and from S b to S a , respectively, R → S a the collection of all objects that are in either set. Then,
and R → S b typify the amplified signal forwarded by R to S a the union of Ea and Eb is defined as:
S  
Ei n = Ea ∪
and S b through one channel. Eb {αi : (αi ∈ Ea ) ∨ (αi ∈ Ea )} i ∈ {a, b, r}.
Now, the total energy consumption can be expressed in
IV. Energy Allocation regard to αi as the following
This section will concentrate on an energy allocation    
Ei (αi ) n = Ea n + Eb n
 
i ∈ {a, b, r}. (22)
problem to manage energy consumption in the users and relay
nodes of UAV networks. Energy consumption is related to For each transmitted bit, the value of q can be defined as
 
transmission delay [38], so the proposed energy allocation q n = 1/R[n], so from (17) we have
 ∂(αa ,αb ,αr ,w) T
h ∂(α i
a ,αb ,αr ,w) ∂(αa ,αb ,αr ,w)
∂αa αb ∂αb αr ∂αr αa > 0. Thus, it is
 −1
 αr αb  required to calculate the following equations
qa (αi )[n] = 2 log2 1 + GP G
 , (23)  2 
αr + αa + αb H ∂F(αa , αb , αr , w)
= q[n] H 2 q[n] − 1 (wa + wb )
∂αa
 −1 !
 αr αa  αb H 2 αr GHαr (αb + αr )
qb (αi )[n] = 2 log2 1 + GP  i ∈ {a, b, r} , −wr GP + . (31)
αa + (αr + αb ) G (αbG + Hαr + Hαa )2 (αbG + Gαr + Hαa )2
H  2 
(24) ∂F(αa , αb , αr , w)
−1 = q[n] G 2 q[n] − 1 (wa + wb )
where (.) indicates reciprocal action. ∂αb
!
In two-way relay networks, the powers design of relay H 2 αr (αa + αr ) αaGHαr
and users nodes are defined as (pa [n] + pr [n] + pb[n]) ≤ P n]
 +wr GP − , (32)
(αbG + Hαr + Hαa )2 (αbG + Gαr + Hαa )2
[32]. In this case, the proposed allocated power is expressed ∂F(αa , αb , αr , w)  2

considering (19) as = q[n] H 2 q[n] − 1 (H wa + G wb )
XN ∂r !
 
(pa [n] + pr [n] + pb[n]) ≤ P n . (25) αb H(αa H + αbG) αa H(αa H + αbG)
+wr GP + . (33)
n=2
| {z } (αbG + Hαr + Hαa )2 (αbG + Gαr + Hαa )2
the total transmit power of Also we have
 ∂(αa ,αb ,αr ,w) 
two hops  ∂α α 
 ∂(αa ,αab ,αbr ,w) 
The purpose of this paper is to optimize αi i.e., αa , αb and αr  ∂αb αr  =

∂(αa ,αb ,αr ,w) 
in order to regulate the energy consumption of (20) and (21) ∂αr αa
and the transmission delay of (23) and (24) in a balanced way. 
 −H 2 r(αa H−Gαb +Hαr )GP GHr(αa H−G(αr +αb ))GP 


Hence, the optimisation problem can be formulated as follows  (αa H+Gαb +Hαr )3 (αa H+G(αr +αb )) 3

 GPH 2 α2a H+αa bG+aHα r +2αb Gαr

−αa GH(αa H+αb G−Gαr )GP  .
  (34)
 (H(αa +αr )+αb G)3 (αa H+αb G+Gαr )3 
min Ea (αi )[n], (26)  αb H 2 (αb G+H(αa −αr ))GP 2αa H 2 (αa H+αb G)GP 
(αb G+H(αr +αa ))3 (αb G+Gαr +Hαa )3
min Eb (αi )[n], (27)
Equation (34) illustrates that the second partial derivative
min (qa (αi [n]+qb(αi )[n]), (28) h ∂(αa ,αb ,αr ,w) T
i
test ∂(αa∂α ,αb ,αr ,w)
a αb
∂(αa ,αb ,αr ,w)
∂αb αr ∂αr αa > 0. Thus, the
s.t ∀n αa [n] + αb[n] + αr [n] ≤ 1. (29) minimum local points of equations (31) to (33) are obtained
All objectives (26)-(28) must be minimized at once under after setting the first conditions, i.e., ∂F(αa∂α ,αb ,αr ,w)
a
= 0,
the constraint of (29); however, the issue is that both (26) ∂F(αa ,αb ,αr ,w) ∂F(αa ,αb ,αr ,w)
∂αb = 0 and ∂αr = 0.
and (27) are in contras to (28). Multi-objective optimisation
Before continuing with this analysis, another test should be
techniques, particularly the weight scalarization method [7], 
applied to F[n (αa , αb , αr ). This test aims to find whether the F
are some of the most reliable methods for resolving such an
function is a convex function or not, as the weigh scalarization
issue. In the weight scalarization method, all of the objective
method is inefficient to find a solution with a non-convex
functions are consolidated into a single function that appears as
function [39]. To achieve such a test, the Bordered Hessian
a linear function. Many studies have adopted the scalarization
(Hb ) matrix is employed as follows:
approach to optimise different mathematical functions such h i
as quadratic and logarithmic functions. By employing the z T Hb z = z 1 z 2 z 3 z 4
scalarization approach to minimise (26)-(28) under constraint  ∂F(αa ,αb ,αr ,w) ∂F(αa ,αb ,αr ,w)
 0
(29), the following expression is obtained  ∂F(αa ,αb ,αr ,w) ∂2 F(αa∂α,αab ,αr ,w) ∂F2 (αa∂α b
,αb ,αr ,w)2
N 
X  ∂αa ∂α2a ∂αa ∂αb
F(αa , αb , αr , w)[n] = (wa Ea (αi )[n] + wb Ea (αi )[n]  ∂F(αa ,αb ,αr ,w) ∂2 F(αa ,αb ,αr ,w) ∂2 F(αa ,αb ,αr ,w)
 ∂αb ∂αa ∂αb ∂α2b
n=2 
+wr (qa (αi )[n] + qb(αi )[n])) ∀n, (30)  ∂F(αa ,αb ,αr ,w) ∂2 F(αa ,αb ,αr ,w) ∂2 F(αa ,αb ,αr ,w)
∂αr ∂αa ∂αr ∂αb ∂αr
where wa {wa : ∈ R, 0 < wa ≤ 1}, wb {wb : ∈ R, 0 < wb ≤ 1} ∂F(αa ,αb ,αr ,w) 
  
∂αr
and wr {wr : ∈ R, 0 < wr ≤ 1} . ∂2 F(αa ,αb ,αr ,w)  z1 
 z 
The weight coefficients are limited to the following ∂αa ∂αr   2  . (35)
P ∂2 F(αa ,αb ,αr ,w)
constraint: m i wi ≤ 1,where m is the number of functions ∂αb ∂αr
 z3 
  
and i ∈ {a, b, r}. Equation (30) reveals that both qa (αi )[n] and ∂2 F(αa ,αb ,αr ,w)  z4
qb (αi )[n] express as linear with a single weight coefficient; ∂α2r
this is because the proposed system model assumes that S a Substituting (31)-(34) into (35) leads to the following
and S b , are exchanging information simultaneously with each results: zT Hb z ≥ 0, i.e., Hb is non-negative. Therefore, the
other through R. Hessian matrix is positive semidefinite; hence F(αa , αb , αr , w)
The minimum solution of the objective function is a set of convex.
h iT
F(αa , αb , αr , w) is obtained by satisfying the conditions
h ∂(α ,α ,α ,w) ∂(α ,α ,α ,w) ∂(α ,α ,α ,w) iT Now, we evaluate ∂(αa ,α∂αba,αr ,w) ∂(αa ,α∂αbb,αr ,w) ∂(αa ,α∂αb r,αr ,w) =
a b r a b r a b r
∂αa ∂αb ∂αr = 0 and 0 from (31) to (33). This gives
 q Input: γa , and γb
∗ 1  
Initialization of parameters
αa (w j ) = ψb ± ψ2b + 2ψa ψc ∀n, (36)
ψa while t , 0 do
!
∗ 1  q   H  R ← (18)
αb (w j ) = ψb ± ψ2b + 2ψa ψc −1 +1 Ea (αa , αb , αr ) ← (20)
ψa G
v
u
t  q[n]
u 2
 Eb (αb , αb , αr ) ← (21)
q 2 − 1 [ H − G] qa (αa , αb , αr ) ← (23)
(1 − w3 ) ∀n, (37) qb (αa , αb , αr ) ← (24)
2HGP(w3 ) for wi = 0 : to 1 do
q
∗ 1  
for w j = 0 : to 1 do
αr (w j ) = 1 − ψb ± ψ2b + 2ψa ψc
ψa F(αa , αb , αr , wi, j ) ← (30)
q  ! ∗
1  2
 H
αa (wi, j ) ← (36)
− ψb ± ψb + 2ψa ψc −1 +1 ∗
ψa G αb (wi, j ) ← (37)
v
u ∗
t  q[n]
u 2

αr (wi, j ) ← (38)
q 2 − 1 [ H − G] ∗ ∗ ∗ ∗
(1 − w3 ) ∀n, (38) qi (αa , αb , αr ) ← (39)
∗ ∗ ∗ ∗
2HGP(w3 ) Ei (αa , αb αr ) ← (40)
where (w j ) : j ∈ {a ∨ b, r}, ∨ is ’or’ symbol, ψa = 2 (G − H) end
   2   
  q 2 q[n] − 1 [H − G] (w2 )   end
  1   end
1 −  −  (G − H),
  2HGP(w3 ) (G − H) + H    Output: Ea (αa , αb , αr );Eb (αb , αb , αr );
s ! qa (αb , αb , αr );qb (αb , αb , αr )
2
  q 2 q[n] −1 [ H−G] Algorithm 2: Proposed energy-delay trade-off
H
ψb = G −1 G
2HP(w3 )
(1 − w3 )+2G and
s 2
!
q 2 q[n] −1 [ H−G]
ψc = G (1 − w3) − G. bit error metric defines the number of errors that occur during
2HP(w3 )
The solution of (36)-(38) depend on any variation of two- data transmission on the UAV network. It can be reformatted
weight coefficient, and this gives an expected result because for each transmitted bit (i.e. R = 1) as follows
the proposed system model has two sources, S a and S b ,
and both sources contribute to adjusting the energy allocation 1 1 1
∗
=  +  ,
∗ ∗ ∗ ∗
parameters. Based on several w j , (36)-(38) provide a trade- γ(w j )[n] αr (w j ) + 2αa (w j ) HP αr (w j ) + 2αb (w j ) PG
off between energy consumption and transmission time. This (41)
agrees with the analysis in [40], which demonstrated that ∗
where γ(w j ) the optimal SNR.
the weight scalarization method produce trade-off solution by
repeating the analysis process for several weight coefficients. Equation (41) reveals that the SNR is increasing function 
∗ ∗ ∗
Thus, the solution (28) is obtained by using (36)-(38) as of optimal allocation factor (ϕi ), ϕi ∈ αa (w j ), αb (w j ), αr (w j ) .
follows: ∗   

Further, according to [41], bit error rate decreases when the
q n (w ) = 2/ 1 + log 1 + Φ n .
i j 2 (39)
overall received SNR is maximised. Thus, increasing ϕi in (41)
By adding (26) and (27) together, the total energy allows bit error rate to be minimised as demonstrate in the
consumption solution is calculated in terms of (36)-(38) as following analysis.
follows: In the proposed UAV network, the transmitted signal from a
∗    2   
∗  ∗ ∗
Ei n (w j ) = qi n 2 q[n] − 1 αr (w j ) + 2αa (w j ) H Sa to the Sb node propagates through two cascaded channels,
   as illustrated in (1). Each channel is a Rayleigh fading and, in
∗ ∗
+ αr (w j ) + 2αb (w j ) G j ∈ {a ∨ b, r} , (40) such a link, SNR follows a negative exponential distribution.
∗ ∗
! Then the total SNR obtained from two i.i.d channels follows a
  αr (w j ) αb (w j )
where, Φ n = 1 + log2 1 + GP ∗ ∗ ∗ negative exponential distribution. To find the total probability
αr (w!j )+αa (w j )+αb (w j ) G
∗ ∗
H density function (pdf) of the Sa − R and R − Sb channels,
αr (w j )αb (w j )
+ log2 1 + GP ∗ ∗ ∗ G
distribution of the harmonic mean of two i.i.d. gamma random
αr (w j )+αa (w j )+!αb (w j ) H
∗ ∗ variables demonstrated in [42] is applied as follows: first define
αr (w j ) αa (w j )
1 + GP ∗ ∗ ∗ pdf of the Sa − R as
αa (w j )+(αr (w j )+αb (w j )) G H !
∗ ∗ η
αr (w j ) αa (w j ) −
+ log2 1 + GP ∗ ∗ ∗ . e γh p [n]
h
αa (w j )+(αr (w j )+αb (w j )) HG pd f = , (42)
γh p [n]
The (39) and (40) procedures are listed in Algorithm 2.
and pdf of the R − Sb link as
V. Bit error rate performance
∗ −
η
This section discusses the optimum bit error rate (áe ) e γg p [n]
g
behaviour of the UAV network based on Equation (40). The pd f = , (43)
γg p [n]
 ∗ ∗
∗ αr (w j ) αb (w j ) ∗
where γh p [n] = αr (w j ) HP, γg p [n] = ∗ ∗ GP, and η is Input: ϕi
αr (w j )+αa (w j )
the harmonic mean according to a gamma distribution defined Initialization of parameters
as η = µH (γh p [n], γg p [n]) [43]. while t , 0 do
To joint (42) and (43), the modified harmonic mean for wi = 0 : to 1 do
∗
demonstrated in [44] is applied as follows γ(wi )[n] ← (41)
  
 2  2η  pd f h ← (42)
pd f t (η) =  k0  p 
pd f g ← (43)
γh p [n] γg p [n] γh p [n] γg p [n]
  γh p [n]+γg p [n]
F (η) ← (46)
γh p [n] + γg p [n]  2η  −η ( γ p [n] γ p [n] ) end

k1  p   2η e
 h g ,
γh p [n]γg p [n] 3/2 γh p [n] γg p [n] end
(44) Output: áe [n]
where pd f t (η) is the total pdf, k0 (.) and k1 (.) are the first Algorithm 3: Bit error rate performance
and the second order modified Bessel function of the second
kind.
Equation (44) is simplified by applying the modified Bessel scheme. The results of (20)-(24) are plotted under the name
function properties as k0 (η) → 0 and k1 ( η ) → 1/η. This Sub-optimal Network (SN).
η→0
! Fig. 2 illustrates optimal trade-off curves between energy
  −η 1 1 ∗
γh p [n] + γg p [n] and delay corresponding to various values of αi (wj ). Each
gives pd f t (η) ≈ γ p1[n] + γ p1[n] e . By integrating
h g
t point on the curve corresponds to different energy delay
pd f (η) relative to η, the cumulative distribution function is ∗
levels, and it is calculated by adjusting αi (wj ) within the
obtained ! range between 0 and 1. Based on the trade-off relationship
−η γ p1 [n] + γ p1 [n]
F (η) = 1 − e h g
, (45) between energy and delay, we find that the energy decreases
∗
monotonically with delay. By increasing αi (wj ), the optimum
where F (η) is the cumulative distribution function for pd f t (η). ∗
delay decreases under the same energy, as the αi (wj ) for the
At high SNR domain, the first order expansion of F (η) is delay objective is decreased. In the same manner, increasing
given by ∗
αi (wj ) leads to reduce energy consumption due to the long
! delay in data delivery. As observed, any curve associated
1 1   ∗
F (η) = η + + o x1+ε , 0 < ε < 1 (46) with a specific αi (wj ), follows an exponential decay unit
γh p [n] γg p [n] approaches to a steady when the transmission delay leans
Now, the approximate bit error rate of the UAV at a high to infinity. In this case, the energy converges to a constant
SNR can be roughly estimated by using [45] as ∗
Z ∞ −η as delay approaches αi (wj ) constraint value. The energy, on
n  p o 1 e ∗
Bit error rate = E Q 2γ = √ √ F (η) dη. (47) the other hand, approaches αi (wj ) constraint value when the
2 π 0 η delay tends to infinity. This result is consistent with earlier
Substituting (46) in (47) and evaluating the result lead to researchers results such as [46], which reveals that there is
obtaining the optimal bit error rate of the UAV as always a trade-off between energy and delay that enables the
  performance of the transmission network to be enhanced. As
3 a continuation of these studies, our proposal in PA provides a
∗ Γ 2
á[n] = ∗ √ novel trade-off scheme, adopting an energy allocation strategy
αr (w j )P π to achieve the optimal energy distribution between relay
 
  and user nodes, which, in turn, improves UAV transmission
 1 1 
   +   , (48) network performance by reducing the transmission delay (i.e.,
 ∗ ∗ ∗ ∗  high data rate) or energy consumption (i.e., higher energy
αr (w j ) + 2αa (w j ) H αr (w j ) + 2αb (w j ) G 
efficiency).
where Γ(.) is gamma function.
In Algorithm 3, the (48) procedure is demonstrated. Consider the following practical example to illustrate
how the framework proposed can be implemented for UAV
networks: transmitting signals, by Sa or Sb, over poor channel
VI. Simulation Results conditions consumes more energy than broadcasting over
In this section, numerical simulations are carried out to normal channel conditions. Under such a scenario, using the
assess transmission delay and energy savings for the proposed proposed method enables a user’s node to determine whether
UAV network. The analytical outcomes are validated by to send packets or leave the users idle based on the energy
utilizing Monte Carlo simulations that use 105 samples. The consumption. Then, the proposed approach ensures that energy
rest of the simulation specifications are set as: 100 ≤ d ≤ 700 consumption is conserved in UAV networks by creating a
meters and the maximum powers for users and relay nodes decision-making scheme. In order to clarify the characteristic
are specified by 2 watts. of this decision-making platform, Fig. 3 illustrates the energy
The simulation results, which are plotted simultaneously consumption curves versus total distance d, which represents
using equations (39)-(40) refer to the Proposed Algorithm (PA) the sum of da and db . It is clearly shown that d has
 a significant impact on energy consumption output as it
6 decreases significantly with increasing d. This is because
=0.3 increasing da requires a higher transmit power by Sa node
=0.5 to suppress channel-fading growth, and, hence, higher energy
5 =0.8 consumption. This result agrees with what is observed in [47].
In addition, Fig. 3 shows that the proposed scheme allows
4 energy consumption to be decreased by about 15% more than
the SA scheme.
3 Fig. 4 depicts that with the rise in power, E initially
decreases and then reaches a constant. This is because R
only increases logarithmically with P while the transmission
2 power consumption increases linearly with the transmission
power [48]. Consequently, increasing power divides energy
1 consumption into two stages. In the first stage, the energy is
0 1 2 3 4 5 6 decreased; then, in the second stages, it approaches a constant
rate. This reveals that there is a transition point that allows
Fig. 2: Energy and Delay trade-off for the PA scheme for the improvement of energy consumption. Fig. 4, illustrates
that the domain of the transition point is between 13 and
17 dBm; and a further reduction of energy consumption is
shown by the proposed method of PA, which corresponds to
6.5 previous studies [48, 49]. Further validation is given by Fig.
= 0.8 5, which compares the average data rate and the maximum
Energy consumption (Joule)

i
*
(wj )=0.8 transmit power. It is clear that the average data rate in (39)
6.0 i
tends to become a constant at the high range of the transmit
power (range between 14 and 16 dBm). This is because the
4.5 power allocation strategy enables the transmit power to be
regulated [50]. By using the PA scheme, a higher average
i
= 0.5 system rate is achieved, as the PA minimises both E and
3 *
(wj )=0.5 q (i.e., increases R) simultaneously. Therefore, R increases
i
SN logarithmically, while the transmission power consumption
PA increases linearly with the transmission power [48]. In other
1.5 words, the transition point is improved significantly, and
such enhancement is expected to enhance other performance
0.1 0.2 0.3 0.4 0.5 0.6 parameters in UAV networks. For example, Fig. 6 illustrates
S-D Link distance (km) the relationship between distance and achievable data rate.
Fig. 3: Energy consumption vs link distance for the PA It shows that increasing distance between Sa and Sb nodes
and SN schemes gradually reduces achievable data rates due to high channel
gain degradation. However, the PA is always better than the
SN, which agrees with other results obtained by [29].
4.0 In the PA, the rate of distance change depends on flight
altitude, and this is required to explain the effect of the drone
Energy consumption (Joule)

i
= 0.5 altitude on data rate performance, as illustrated in Fig. 7. As
*
(wj )=0.5 depicted in the figure, when the drone takes off from the Sa
i
3.0
node level at 100 m, the data rate decreases slightly with the
distance travelled by the drone remains within a low path loss
range [51]. By increasing the altitude between Sa node and
2.0 drone, the path loss increases significantly and the data rate,
in turn, starts to decrease gradually. The results of the proposed
algorithm in the PA show a marked improvement because
= 0.8
*
i increasing the altitude extend da and, then, the arrival data
(wj )=0.8
1.0 i
rate at the relay is decreased. Hence, in addition to reducing
transmission delay, both user nodes allocation of a higher
5 10 15 power to each other in order to motivate information growth.
Total power (dBm) The data rate curve is enhanced compared with other studies
that depend on energy allocation only, as in [52]. To explain
Fig. 4: Energy consumption vs total power for the PA ∗
how the proposed αi (w j ) is allocated for the UAV network, Fig.
and SN schemes ∗
8 plots the relationship between αi (w j ) and both P and w j . It
 15 90
Achievable data rate (Mbits/sec)

PA

Achievable data rate (Mbits/sec)

* 80 SN
12 i
(w
j
)=0.8
= 0.8 *
(w )=0.8
i 70 i j
9 = 0.8
i
60
*
6 (w )=0.5 50
i j
= 0.5
i
40
3
*
30 (w )=0.5
i j
= 0.5
0 i
20

0 5 10 15 20 25 0.1 0.2 0.3 0.4 0.5 0.6

Total power (dBm) S-D Link distance (km)
Fig. 5: Achievable data rate vs total power for the PA Fig. 6: Achievable data rate vs link distance for the PA
and SN schemes and SN schemes

∗
can be seen that increasing P allows αi (w j ) to be increased 12
while notably decreasing w j . The user nodes manage such a PA
high P based on many factors, such as channel conditions or Achievable data rate (Mbits/sec) SN
10
distance da and db . Thus, if da is higher than db , the path loss
of da is large and the rate at which data arrives at the relay 8
∗
is low. Therefore, the user nodes allocate higher αi (w j ) to *
i
(wj )=0.8
stimulate an increase in data transmission by each user when 6 = 0.8
i
there is an increase in delay time.
To evaluate UAV performance in (48), Fig. 9. shows results *
4 i
(wj )=0.5
of áe [n] versus ϕi , and for various SNRs values, where = 0.5
∗ ∗ ∗ ∗ i
γ(w j )[n3] > γ(w j )[n2] > γ(w j )[n1]. Each γ(w j )[n] is obtained
2
from a specific value of ϕi (0 < ϕi < 1). It is clearly seen
∗
that the highest SNR, at γ(w j )[n3 ], enhanced the bit error
rate result, and this result confirms the theoretical analysis, 0.1 0.2 0.3 0.4 0.5 0.6
as high SNR calculates from optimal power allocation. This
result agrees with previous studies [53], which indicate that
Fig. 7: Achievable data rate vs altitude flight for the PA
the power allocation ϕi is an effective metric for enhancing
and SN schemes
UAV performance.
Fig. 9. compares between the S N and PA system in
term of ϕi and bit error rate. Using energy allocation in
0.8
(40) rises SNR[n] which in turn increases data rate, and
the result enhancing bit error rate. Fig. 10. also shows that wj =0.1
the
 highest SNR, i.e.  SNR[n3 ] > SNR[n2 ] > SNR[n1 ] SNR[n] ∈ 0.7
∗
γ(w j )[n], γ(w j )[n] , gives the lower bit error rate. This wj =0.3

result is consistent with previous results [54] for terrestrial 0.6

communication systems. Obviously, a higher SNR results in wj =0.5
better performance for the PA as compare in SN, consequently,
∗ 0.5
the proposed á[n] can improve UAV performance. wj =0.7

0.4
VII. Conclusion and Future Work
This paper offers a new method for drawing out the
most effective energy-delay curve for a UAV layout by 0 5 10 15 20 25 30 35
optimising the energy allotments for users and relay nodes. A Total power (dBm)
multi-objective technique for optimising the energy allocation
factors, weight scalarization optimisation, is appraised in Fig. 8: Achievable data rate vs total power for the PA
this paper. The signal transmission distance is taken into scheme
 [2] F. Ono, H. Ochiai, and R. Miura, “A wireless relay network based on
0 unmanned aircraft system with rate optimization,” IEEE Transactions on
10
Wireless Communications, vol. 15, no. 11, pp. 7699–7708, Nov 2016.
[3] W. Zeng, J. Zhang, D. W. K. Ng, B. Ai, and Z. Zhong, “Two-way
hybrid terrestrial-satellite relaying systems: Performance analysis and
-1 relay selection,” IEEE Transactions on Vehicular Technology, vol. 68,
10
no. 7, pp. 7011–7023, 2019.
Bit error rate

[4] R. Zhang and J. Gorce, “Optimal transmission range for minimum

energy consumption in wireless sensor networks,” in 2008 IEEE Wireless
Communications and Networking Conference, 2008, pp. 757–762.
10 -2
* [5] B. Zhou, L. Jiang, S. Zhao, and C. He, “BER analysis of TDD downlink
(wj)[n1 ] multiuser MIMO systems with imperfect channel state information,”
*
(wj)[n2 ] EURASIP Journal on Applied Signal Processing, vol. 2011, no. 1, Dec.
2011.
10 -3 *
(wj)[n3 ] [6] A. Zappone, L. Sanguinetti, and M. Debbah, “Energy-delay
efficient power control in wireless networks,” IEEE Transactions
on Communications, vol. 66, no. 1, pp. 418–431, 2018.
-4 [7] A. Brandt, Optimization Methods for Material Design
10 of Cement-based Composites, ser. Modern Concrete
0.15 0.3 0.45 0.6 0.75 Technology. Taylor & Francis, 1998. [Online]. Available:
https://books.google.com.au/books?id=RZTPi0V1JKEC
[8] P. Nuggehalli, V. Srinivasan, and R. R. Rao, “Delay constrained
Fig. 9: Power allocation vs bit error rate for the PA energy efficient transmission strategies for wireless devices,” in
scheme Proceedings.Twenty-First Annual Joint Conference of the IEEE
Computer and Communications Societies, vol. 3, 2002, pp. 1765–1772
vol.3.
[9] Xiliang Zhong and Cheng-Zhong Xu, “Delay-constrained energy-
40 efficient wireless packet scheduling with qos guarantees,” in
SN [n] GLOBECOM ’05. IEEE Global Telecommunications Conference, 2005.,
SNR[1] vol. 6, 2005, pp. 5 pp.–3340.
PA *(wj)[n]
[10] E. Uysal-Biyikoglu, B. Prabhakar, and A. El Gamal, “Energy-efficient
Data Rate (Mbit/sec)

packet transmission over a wireless link,” IEEE/ACM Transactions on

30 Networking, vol. 10, no. 4, pp. 487–499, 2002.
SNR[2] [11] M. A. Zafer and E. Modiano, “A calculus approach to minimum
energy transmission policies with quality of service guarantees,” in
Proceedings IEEE 24th Annual Joint Conference of the IEEE Computer
20 SNR[3] and Communications Societies., vol. 1, 2005, pp. 548–559 vol. 1.
[12] C. E. Shannon, “Probability of error for optimal codes in a gaussian
channel,” The Bell System Technical Journal, vol. 38, no. 3, pp. 611–
656, 1959.
[13] Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel coding rate in the
10 finite blocklength regime,” IEEE Transactions on Information Theory,
vol. 56, no. 5, pp. 2307–2359, 2010.
[14] ——, “Minimum energy to send k bits through the gaussian channel
-6 -5 -4 -3
10 10 10 10 with and without feedback,” IEEE Transactions on Information Theory,
Bit error rate vol. 57, no. 8, pp. 4880–4902, 2011.
[15] R. A. Berry and R. G. Gallager, “Communication over fading channels
Fig. 10: Achievable data rate vs bit error rate for the PA with delay constraints,” IEEE Transactions on Information Theory,
vol. 48, no. 5, pp. 1135–1149, 2002.
and SN schemes [16] O. Waqar, M. Imran, M. Dianati, and R. Tafazolli, “Energy consumption
analysis and optimization of ber-constrained amplify-and-forward relay
networks,” Vehicular Technology, IEEE Transactions on, vol. 63, no. 3,
consideration to evaluate UAVs. Simulations confirm the pp. 1256–1269, March 2014.
[17] Y. Zeng, R. Zhang, and T. J. Lim, “Throughput maximization
results originating from analytical expressions, and a real- for uav-enabled mobile relaying systems,” IEEE Transactions on
world application scenario demonstrates how the proposed Communications, vol. 64, no. 12, pp. 4983–4996, Dec 2016.
structure will be used. The paper concludes that the proposed [18] Y. Chen, W. Feng, and G. Zheng, “Optimum placement of uav as relays,”
IEEE Communications Letters, vol. 22, no. 2, pp. 248–251, Feb 2018.
technique effectively executes optimal decision-making and [19] X. Jiang, Z. Wu, Z. Yin, and Z. Yang, “Power and trajectory optimization
presents a compromise between energy and delay in UAVs. for uav-enabled amplify-and-forward relay networks,” IEEE Access,
It would be interesting to extend our proposed study vol. 6, pp. 48 688–48 696, 2018.
[20] K. Govindan, K. Zeng, and P. Mohapatra, “Probability density of the
by considering future works’ weighted interval scheduling received power in mobile networks,” IEEE Transactions on Wireless
problem. Also, different system scenarios such as multiple Communications, vol. 10, no. 11, pp. 3613–3619, 2011.
UAVs could be employed instead of a single UAV. Besides [21] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal lap altitude for
maximum coverage,” IEEE Wireless Communications Letters, vol. 3,
that, statistical channel state information estimation can be no. 6, pp. 569–572, 2014.
used instead of instantaneous channel state information. [22] M. Alzenad, A. El-Keyi, F. Lagum, and H. Yanikomeroglu, “3-d
Further, computational complexity is another direction that placement of an unmanned aerial vehicle base station (uav-bs) for
energy-efficient maximal coverage,” IEEE Wireless Communications
would be required to evaluate in future studies. Letters, vol. 6, no. 4, pp. 434–437, 2017.
[23] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Unmanned aerial
vehicle with underlaid device-to-device communications: Performance
References and tradeoffs,” IEEE Transactions on Wireless Communications, vol. 15,
[1] M. I. Khalil, “Energy efficiency maximization of relay aerial no. 6, pp. 3949–3963, June 2016.
robotic networks,” IEEE Transactions on Green Communications and [24] S. Zhang, H. Zhang, Q. He, K. Bian, and L. Song, “Joint trajectory
Networking, vol. 4, no. 4, pp. 1081–1090, 2020. and power optimization for uav relay networks,” IEEE Communications
 Letters, vol. 22, no. 1, pp. 161–164, 2018. Communications, vol. 60, no. 12, pp. 3767–3778, 2012.
[25] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Mobile [49] D. W. K. Ng, E. S. Lo, and R. Schober, “Energy-efficient resource
unmanned aerial vehicles (uavs) for energy-efficient internet of things allocation in ofdma systems with large numbers of base station
communications,” IEEE Transactions on Wireless Communications, antennas,” IEEE Transactions on Wireless Communications, vol. 11,
vol. 16, no. 11, pp. 7574–7589, 2017. no. 9, pp. 3292–3304, 2012.
[26] M. Luo, G. Villemaud, J. M. Gorce, and J. Zhang, “Realistic prediction [50] ——, “Energy-efficient resource allocation in multi-cell ofdma systems
of ber and amc for indoor wireless transmissions,” IEEE Antennas and with limited backhaul capacity,” IEEE Transactions on Wireless
Wireless Propagation Letters, vol. 11, pp. 1084–1087, 2012. Communications, vol. 11, no. 10, pp. 3618–3631, 2012.
[27] M. I. Khalil, “Delay and energy balance for unmanned aerial vehicle [51] W. Khawaja, I. Guvenc, D. W. Matolak, U. Fiebig, and
networks,” in 2021 IEEE 11th Annual Computing and Communication N. Schneckenburger, “A survey of air-to-ground propagation channel
Workshop and Conference (CCWC), 2021, pp. 1453–1458. modeling for unmanned aerial vehicles,” IEEE Communications Surveys
[28] O. Weck, “Multiobjective optimization : History and promise,” in Tutorials, vol. 21, no. 3, pp. 2361–2391, 2019.
MULTIOBJECTIVE OPTIMIZATION : HISTORY AND PROMISE, [52] L. Sboui, H. Ghazzai, Z. Rezki, and M. Alouini, “Achievable rates of
2004. uav-relayed cooperative cognitive radio mimo systems,” IEEE Access,
[29] C. Tian, Z. Qian, X. Wang, and L. Hu, “Analysis of joint relay selection vol. 5, pp. 5190–5204, 2017.
and resource allocation scheme for relay-aided d2d communication [53] M. Mondelli, Q. Zhou, X. Ma, and V. Lottici, “A cooperative approach
networks,” IEEE Access, vol. 7, pp. 142 715–142 725, 2019. for amplify-and-forward differential transmitted reference ir-uwb relay
[30] D. Yang, Q. Wu, Y. Zeng, and R. Zhang, “Energy tradeoff in ground- systems,” in 2012 IEEE International Conference on Acoustics, Speech
to-uav communication via trajectory design,” IEEE Transactions on and Signal Processing (ICASSP), 2012, pp. 2905–2908.
Vehicular Technology, vol. 67, no. 7, pp. 6721–6726, July 2018. [54] W. Al-Hanafy and S. Weiss, “Discrete rate maximisation power
[31] Y. Li, X. Zhang, M. Peng, and W. Wang, “Power provisioning and relay allocation with enhanced bit error ratio,” IET Communications, vol. 6,
positioning for two-way relay channel with analog network coding,” pp. 1019–1024(5), June 2012.
IEEE Signal Processing Letters, vol. 18, no. 9, pp. 517–520, Sept 2011.
[32] M. Khalil, S. Berber, and K. Sowerby, “High snr approximation for
performance analysis of two-way multiple relay networks,” Physical
Communication, vol. 24, pp. 62 – 70, 2017. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1874490716301446
[33] A. S. Y. Poon, D. N. C. Tse, and R. W. Brodersen, “An adaptive
multiantenna transceiver for slowly flat fading channels,” IEEE
Transactions on Communications, vol. 51, no. 11, pp. 1820–1827, 2003.
[34] J. Zander, “Radio resource management in future wireless networks:
requirements and limitations,” IEEE Communications Magazine, vol. 35,
no. 8, pp. 30–36, 1997.
[35] Y. Zhang, Y. Ma, and R. Tafazolli, “Power allocation for bidirectional af
relaying over rayleigh fading channels,” Communications Letters, IEEE,
vol. 14, no. 2, pp. 145–147, 2010, iD: 1.
[36] A. B. Kiely and J. T. Coffey, “On the capacity of a cascade of channels,”
IEEE Transactions on Information Theory, vol. 39, no. 4, pp. 1310–
1321, 1993.
[37] S. Salehkalaibar, M. Wigger, and L. Wang, “Hypothesis testing over
cascade channels,” in 2017 IEEE Information Theory Workshop (ITW),
2017, pp. 369–373.
[38] J. Ren, G. Yu, Y. He, and G. Y. Li, “Collaborative cloud and edge
computing for latency minimization,” IEEE Transactions on Vehicular
Technology, vol. 68, no. 5, pp. 5031–5044, 2019.
[39] J. Koski, “Defectiveness of weighting method in multicriterion
optimization of structures,” Communications in Applied Numerical
Methods, vol. 1, no. 6, pp. 333–337, 1985. [Online]. Available:
https://onlinelibrary.wiley.com/doi/abs/10.1002/cnm.1630010613
[40] T. B. To and U. Korn, “An evolution strategy for the multiobjective
optimization,” The Second International Conference on Genetic
Algorithms, Brno, Czech Republic, 01 1996.
[41] G. Farhadi and N. C. Beaulieu, “On the performance of amplify-
and-forward cooperative systems with fixed gain relays,” Wireless
Communications, IEEE Transactions on, vol. 7, no. 5, pp. 1851–1856,
2008.
[42] M. Hasna and M.-S. Alouini, “Harmonic mean and end-to-end
performance of transmission systems with relays,” Communications,
IEEE Transactions on, vol. 52, no. 1, pp. 130–135, Jan 2004.
[43] ——, “End-to-end performance of transmission systems with relays over
rayleigh-fading channels,” Wireless Communications, IEEE Transactions
on, vol. 2, no. 6, pp. 1126–1131, Nov 2003.
[44] R. H. Louie, Y. Li, H. Suraweera, and B. Vucetic, “Performance analysis
of beamforming in two hop amplify and forward relay networks with
antenna correlation,” Wireless Communications, IEEE Transactions on,
vol. 8, no. 6, pp. 3132–3141, June 2009.
[45] Z. Wang and G. Giannakis, “A simple and general parameterization
quantifying performance in fading channels,” Communications, IEEE
Transactions on, vol. 51, no. 8, pp. 1389–1398, Aug 2003.
[46] B. Gurakan, O. Ozel, and S. Ulukus, “Optimal energy and data routing
in networks with energy cooperation,” IEEE Transactions on Wireless
Communications, vol. 15, no. 2, pp. 857–870, 2016.
[47] K. Yang, S. Martin, C. Xing, J. Wu, and R. Fan, “Energy-efficient power
control for device-to-device communications,” IEEE Journal on Selected
Areas in Communications, vol. 34, no. 12, pp. 3208–3220, 2016.
[48] C. Xiong, G. Y. Li, S. Zhang, Y. Chen, and S. Xu, “Energy-
efficient resource allocation in ofdma networks,” IEEE Transactions on