IRS-Aided Energy Efficient UAV Communication - arXiv
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
1
IRS-Aided Energy Efficient UAV Communication
Hyesang Cho, Student Member, IEEE and Junil Choi, Senior Member, IEEE
Abstract— Unmanned aerial vehicles (UAVs) have steadily Even with their potential, UAV communication systems
gained attention to overcome the harsh propagation loss and have their limitations such as energy consumption [21]. While
blockage issue of millimeter-wave communication. However, UAV traditional devices replenish energy from a dedicated source,
communication systems suffer from energy consumption, which
limits the flying time of UAVs. In this paper, we propose several it is improbable for the UAV to replenish energy due to
UAV energy consumption minimization techniques through the its wireless feature. Also, due to its hovering characteristic,
the UAV has additional power consumption, which is usually
arXiv:2108.02406v1 [cs.IT] 5 Aug 2021
aid of multiple intelligent reflecting surfaces (IRSs). In specific,
we introduce a tractable model to effectively capture the char- orders of magnitude larger than communication power [22].
acteristics of multiple IRSs and multiple user equipments (UEs). To address this issue, there have been works to maximize the
Then, we derive a closed form expression for the UE achievable
rate, resulting in tractable optimization problems. Accordingly, energy efficiency of UAV communication systems. Especially,
we effectively solve the optimization problems by adopting the [23] minimized the energy consumption of a rotary-wing
successive convex approximation technique. To compensate for UAV communication system with a specific data constraint
the high complexity of the optimization problems, we propose a through jointly optimizing the UAV trajectory, velocity, and
low complexity algorithm that has marginal performance loss. communication time.
In the numerical results, we show that the proposed algorithms
can save UAV energy consumption significantly compared to the Recently, there have been attempts to combine UAV com-
benchmark with no IRSs, justifying that exploiting the IRSs is munication systems with IRSs, which can be divided into
indeed favorable to UAV energy consumption minimization. two large categories [24]–[27]. In the first category, the UAV
Index Terms— UAV communication, intelligent reflecting sur- itself is equipped with the IRS. In particular, [28] analyzed
face, energy consumption minimization, successive convex ap- the outage probability, ergodic capacity, and energy efficiency
proximation of wireless communication systems supported by the IRS
equipped UAVs. In the second category, the UAV commu-
I. I NTRODUCTION nication system is assisted through terrestrial IRSs. In [29], it
Current wireless communication systems are setting foot has been shown that exploiting a terrestrial IRS is beneficial
on millimeter-wave (mmWave) communication via fifth- for UAV energy consumption minimization. However, [29]
generation technology [1]–[3]. By exploiting the abundant assumed the LoS channel model and performed the task with
and vacant frequency bands, mmWave communication has the only a single IRS. Also, [30] maximized the average reception
potential to serve the increasing demand of data. Nonetheless, power in a given time constraint using multiple IRSs by
the high frequency region has its disadvantages, e.g., due to optimizing the IRS phase shifts, transmit beamformer, and
the characteristics of high frequency signals, the signals suffer UAV trajectory. While [30] discussed a multiple IRS scenario,
from extreme propagation loss and are vulnerable to blockage it did not reveal the tradeoff between the reception power and
[4], [5]. UAV flight power, e.g., the UAV may consume excess flight
To overcome the blockage issue, intelligent reflecting sur- power to maximize the reception power. Accordingly, this
faces (IRSs) have been proposed as a favorable candidate [6]– tempts us to explore UAV energy consumption minimization
[8]. An IRS is a planar surface consisting of multiple passive in a multiple terrestrial IRS scenario.
elements, where each element has the ability to independently In this paper, we propose several UAV energy consumption
shift the phase of the electromagnetic waves impinging on minimization algorithms with a specific data constraint for
itself [9]–[12]. By thoroughly adjusting the phase controller, various scenarios including the most general multiple IRS
the IRS can enhance the signal gain for a desired location, multiple UE (MIMU) case. The main insight is that by
resulting in a virtual line-of-sight (LoS) path. Therefore, the increasing the received data rate at the UEs through the IRSs,
IRS can function as a controllable relay with minimal power we can reduce the flight time, thus, reduce the UAV energy
consumption. consumption. In specific, we first expand the probabilistic LoS
Another effective approach to overcome the blockage issue model to capture the advantage of deploying IRSs on tall
is to exploit unmanned aerial vehicles (UAVs) [13]–[16]. A buildings [31]. To the best of our knowledge, this is the first
UAV is a flying platform that freely hovers in the sky, which work to consider the popular probabilistic LoS channel model
can be used for communication by using it as a mobile base for UAV scenarios in IRS assisted environments. By utilizing
station (BS) or a relay [17]–[20]. Unlike conventional ground the new system model, we derive the optimal IRS phase shifts
nodes, the UAV can enjoy favorable channel conditions and and develop a closed form of the UE achievable rate. We
avoid blockage via their high altitude. then formulate the optimization problems that jointly optimize
the trajectory, flight speed, and communication time of the
The authors are with the School of Electrical Engineering, Korea Advanced UAV. Afterwards, we transform the optimization problems into
Institute of Science and Technology, Daejeon 34141, South Korea (e-mail:
{nanjohn96, junil}@kaist.ac.kr). tractable forms exploiting slack variables and effectively solve
them through the well known successive convex approximation2
location and height of the k-th UE denoted as qUk ∈ R2×1
and HUk ∈ R, respectively.
A. Channel Model
The channel between any two nodes in the system, i.e.,
UAV, IRSs, and UEs, are denoted as
s
β0
h̃Iw (t) = α g̃I (t), (1)
dIwIw (t) w
s
β0
h̃wk (t) = αwk g̃wk (t), (2)
dwk
s
β0
h̃Uk (t) = αUk g̃Uk (t), (3)
Fig. 1: Single UAV downlink system with K UEs and W IRSs. dUk (t)
where h̃Iw (t) ∈ CMw ×1 , h̃wk (t) ∈ CMw ×1 , and h̃Uk (t) ∈ C
(SCA) technique [26]. Finally, we propose a low complexity denote the UAV to the w-th IRS channel, the w-th IRS to
the k-th UE channel, and the UAV to the k-th UE channel,
algorithm that tries to follow the behavior of a specific
respectively. We split the channel into the large scale fading
optimization problem solution, resulting in comparable per-
formance with respect to the jointly optimized results. factors and small scale fading channel, where for the large
scale fading factors, the distances between the UAV to the
The rest of paper is organized as follows. Section II de-
w-th IRS, the w-th IRS to the k-th UE, and the UAV to the
scribes the channel model of the MIMU scenario and the
k-th UE are denoted as dIw (t), dwk , and dUk (t), respectively,
power consumption model of the UAV. In Section III, we
where we observe that dwk is constant since the locations of
derive and formulate the UAV energy consumption minimiza-
UEs and IRSs are fixed. The pathloss exponent is denoted as
tion problem for the single IRS case. In Section IV, we
αi , i ∈ {Iw , wk, Uk }, and β0 is the pathloss at a reference
generalize the system to the multiple IRS case, and introduce
distance. The small scale fading channels are denoted as
several algorithms to effectively solve the joint optimization
r
problems. Section V shows the simulation results of the
r
κi 1
proposed techniques with a benchmark scheme without any g̃i (t) = ei (t) + ḡi (t), i ∈ {Iw , wk} ,
κi + 1 κi + 1
IRSs. Finally, Section VI concludes the paper. (4)
Notation: Lower and upper boldface letters represent col- s
umn vectors and matrices. A∗ and AH denotes the conjugate, κ Uk 1
r
g̃Uk (t) = exp(jψUk )(t) + ḡU (t), (5)
and conjugate transpose of the matrix A. diag(a) returns the κ Uk + 1 κUk + 1 k
b
diagonal matrix with a on its diagonal. {a} represents the
set of length b vectors with each element from a. Cm×n and where g̃Iw (t), g̃wk (t), and g̃Uk (t) denote the small scale fading
Rm×n represent the set of all m×n complex and real matrices. channels between the UAV to the w-th IRS, the w-th IRS
|·| denotes the amplitude of the scalar, and k·k represents the to the k-th UE, and the UAV to the k-th UE, respectively.
T
ℓ2 -norm of the vector. O denotes the Big-O notation. 0m is The vectors eIw = exp(jψIw,1 ), ..., exp(jψIw,Mw ) and
T
used for the m × 1 all zero vector, and Im denotes the m × m ewk = [exp(jψw,1k ), ..., exp(jψw,Mw k )] denote the LoS path
identity matrix. CN (m, Σ ) denotes the circularly symmetric phases with ψIw,m and ψw,mk representing the phases of the
complex Gaussian distribution with mean m and variance Σ . channels between the UAV to the mw -th IRS element of the w-
th IRS and the mw -th IRS element of the w-th IRS to the k-th
II. S YSTEM M ODEL UE, respectively. The LoS path phase of the channel between
the UAV to the k-th UE is denoted as ψUk .The small scale
In this paper, we consider a wireless communication system channels follow the independent Rician distribution with unit
where a rotary-wing UAV supports multiple UEs with the aid power and the K-factor as κi , i ∈ {Iw , wk, Uk }, and with
of multiple IRSs as in Fig. 1. We assume a single antenna ḡIw (t) ∼ CN (00Mw , IMw ) , ḡwk (t) ∼ CN (00Mw , IMw ), and
UAV, K single antenna UEs, and W IRSs, with the w-th ḡUk (t) ∼ CN (0, 1).
IRS having Mw IRS elements. The UAV freely hovers around We assume that the LoS paths between the IRSs and UEs
the horizontal plane, where the horizontal location at time always exist. For the channels associated with the UAV, we
t is denoted as q(t) ∈ R2×1 , and the altitude is fixed as adopt the well-known probabilistic LoS model [31], which
HA ∈ R. Since the IRSs are typically located to gain a better models the LoS path existence probability as a variable
LoS condition, we assume that the IRSs are installed on the dependent on the elevation angle between a node and a UAV.
outer walls of tall buildings, thus, having more height than the The LoS path existence probability is given as [31]
UEs. We assume that the w-th IRS is fixed with the horizontal
location qIw ∈ R2×1 and height HIw ∈ R. Also, we assume 1
pi (t) = , i ∈ {Iw , Uk } , (6)
that the UEs are located without mobility with the horizontal 1 + ai exp (−bi (θi (t) − ai ))3
where ai and bi are the design parameters dependent on the TABLE I: UAV flight power consumption variables.
environment, and θi (t) is the elevation
angle,
which can be Variable Description Value
expressed as θi (t) = 180 sin−1 HA −Hi
. Note that, due P0 Blade profile power 79.86
π di (t) Pi Induced power 88.63
to the height of the installed IRSs, we consider the design Utip Rotor blade tip speed 120
parameters ai and bi as variables. Defined in [31], the design v0 Rotor induced velocity 4.03
d0 Fuselage drage ratio 0.6
parameters are dependent on the average building height and ρ Air density 1.225
building density with respect to the ground. Therefore, by s Motor solidity 0.05
taking the height of the installed IRSs on buildings as the Ap Rotor disc area 0.503
new effective ground, the average building height and density
which the IRSs see will decrease, resulting in a relatively rural
environment with respect to the original environment. This where V is the UAV speed and the other variables are listed
new consideration can effectively capture the fact that highly in Table I. The total energy consumption of the UAV can be
located IRSs are favorable in obtaining the LoS path. Note expressed as
that, the design parameters ai and bi can adapt to each IRS Z T
or UE, thus, this model can effectively consider the different Etot = τ Pc + P (V (t)) dt, (12)
heights of the IRSs or even consider the advantages of highly 0
located UEs. Accordingly, the effective channel between the where τ is the communication time and T is the total flight
nodes can be expressed as time of the UAV. We observe that the power consumption
is dependent on the UAV speed, and that the model itself is
(
h̃Iw (t), Probability pIw (t),
hIw (t) = (7) quite complicated. This leads us to believe that the UAV speed
νIw h̃Iw (t), Probability 1 − pIw (t),
( should also be jointly optimized to effectively minimize the
h̃Uk (t), Probability pUk (t), UAV power consumption. Also, we can confirm that while
hUk (t) = (8) the power consumption for communication usually has the
νUk h̃Uk (t), Probability 1 − pUk (t),
magnitude of 1 W or smaller, the power consumption from
where νi < 1, i ∈ {Iw , Uk } , is the additional attenuation UAV hovering has the magnitude of 100 W. Thus, it is obvious
factor due to the loss of the LoS path. From hereafter, we take that we must consider the UAV flight power to successively
νi = 0 for equational conciseness, which will be explained achieve energy efficient UAV communication systems.
in Section III. Note that, the proposed methods can still be
applied when νi 6= 0. In conclusion, the overall channel for III. S INGLE IRS O PTIMIZATION
the k-th UE is given as
In this section, we define and solve the optimization prob-
W
X lems to minimize the UAV energy consumption in a simple
hH
h̄k (t) = Φ w (t)hIw (t) + hUk (t),
wk (t)Φ (9)
single IRS case. Specifically, we first assume the most ele-
w=1
mentary single IRS single UE (SISU) case, where we derive
where Φ w is the phase control matrix of the w-th IRS. The the closed form expression for the IRS phase shifts and the
phase control matrix can be represented as Φ w = diag (φ φw ), achievable rate. Thereafter, we define the joint optimization
T
with φ w = [exp(jφw,1 ), ..., exp(jφw,Mw )] , thus, the m-th problem and develop it into a tractable form. Next, we expand
IRS element of the w-th IRS will shift the phase of the the scenario to the single IRS multiple UE (SIMU) case, which
impinging signal with the factor φw,m . can be easily extended from the SISU case. The last expansion
We adopt the time-division multiple access (TDMA) tech- to the MIMU case will be held on Section IV.
nique at the UAV to serve the K UEs to efficiently prevent
inter-user interference. The achievable rate for the k-th UE
can be denoted as A. SISU Scenario Achievable Rate
Pc |h̄k (t)|2 In the SISU case, we omit the indexes w and k for brevity.
R̄k (t) = B log2 1 + , (10) To minimize the UAV energy consumption while sufficing
Bσ 2
a specific data constraint, it is trivial to operate the IRS
where B is the bandwidth, Pc is the transmit power from the
phase shifts to maximize the achievable rate. Since the UAV
UAV, and σ 2 denotes the noise spectral density.
and UE are both single antenna devices, the optimal phase
shifts for the IRS are given in a simple closed form. Note
B. UAV Power Consumption Model
that, we need the instantaneous channel state information to
Since we assumed to use the rotary-wing UAV, we adopt derive the optimal phase shifts. To focus on the achievable
the rotary-wing UAV flight power consumption model from theoretical performance, we assume to have perfect channel
[23], which is given as state information (CSI) throughout this paper. The reflection
! s !1/2 coefficient of the m-th IRS element is then given as
3V 2 V4 V2
P (V ) = P0 1 + 2 + Pi 1+ 4 − 2 φm (t) = ∠ (hU (t)) − ∠ (hI,m (t)) + ∠ (hm (t)) , (13)
Utip 4v0 2v0
1 where hI,m (t) and hm (t) denote the m-th element of the chan-
+ d0 ρsAp V 3 , (11)
2 nel vectors hI (t) and h(t), respectively. This will coherently4
combine the UAV-UE channel with the UAV-IRS-UE channel,
optimally maximizing the achievable rate. 24
Analytical
To gain tractability, we adopt two frequently used assump- Simulate
tions as in [23]. First, to capture the randomness of the 23
channel, we use the expected achievable rate instead of the
Rate (Mbps)
instantaneous achievable rate. Note that, this might seem as if 22
perfect CSI is unnecessary. However, to obtain the expected
achievable rate, we need the IRS to constantly adapt to the
instantaneous channels for successful coherent combination, 21
thus, the assumption of perfect CSI is necessary. Second,
due to the involved form of (6), we assume that the LoS 20
path existence probabilities are constant with respect to time,
e.g., fix the probabilities with the average elevation angles. 19
Nevertheless, the simplified model still captures the advantage 0 100 200 300 400 500
of the highly located IRSs. Distance (m)
To effectively express the probabilistic LoS model, we Fig. 2: Achievable rate comparison between the approximated
define a state variable as analytical values and the simulated results for the SISU case.
s = [sU , sI ]T , si ∈ {0, 1} , (14) The UE is located at 250 m and the IRS is located at 245 m.
for i ∈ {U, I}, where sU and sI denote the states of the
UAV-UE path and UAV-IRS path, respectively, and si = 1 number of IRS elements, i.e., M , is predicted to be installed
represents the existence of the selected path. With this variable, in large quantities, we employ the central limit theorem and
we can upper bound the expected achievable rate using the assume that C(t) follows the Gaussian distribution. Overall,
Jensen’s inequality as (16) can be shown as
E R̄(t) X Y 1−s
R̂(t) = psi i (1 − pi ) i log2 (1 + γ̂s (t)) , (19)
Pc |h̄s (t)|2
1−s
X Y
= psi i (1 − pi ) i E B log2 1 + s∈{0,1}2 i
2 i
Bσ 2
s∈{0,1}
with the overall signal-to-noise ratio (SNR) in the closed form
(15)
! expression as
2
X Y 1−si Pc E |h̄s (t)|
≤ psi i (1 − pi ) B log2 1+ γ̂s (t) = sU γU2 d−αU
(t) + 2sU sI γU′ γI′ dU
−αU /2 −αI /2
(t)dI (t)
2 i
Bσ 2 U
s∈{0,1}
(16) + sI γI2 d−α
I
I
(t), (20)
= R̂(t), i ∈ {U, I} , where s2i is denoted as si since s2i = si , and the time
where h̄s (t) is the overall channel between the UAV and the independent variables γi and γi′ are defined as
UE. Due to the channel states, the overall channel is now r
a variable dependent on s. We observe that β0 Pc
the closed form γU = 2
, (21)
of(16) can be derived by computing E |h̄s (t)|2 . However, r Bσ
E |h̄s (t)|2 is difficult to derive due to the joint consideration β0 Pc
γU′ = 2
µ (κU ) , (22)
of the UAV-UE channel and the UAV-IRS-UE channel. r Bσ
To compute the expected channel gain, we first assume that β02 Pc
q
the IRS phases are always coherently combining the UAV-UE γI = α 2
µ2I + 1, (23)
channel and the UAV-IRS-UE channel. In result, the overall r d Bσ
β02 Pc
channel norm value is given as γI′ = µI , (24)
dα Bσ 2
M
X
|h̄s (t)| = sU |hU (t)| + sI |hI,m (t)||hm (t)|. (17) where µ (κi ) denotes the mean of the Rician distribution with
m=1 the K-factor as κi , and µI = M µ (κI ) µ (κ). Note that in
PM (17), the channel terms without LoS paths are zero due to the
By defining a new variable C(t) = m=1 |hI,m (t)||hm (t)|, assumption νi = 0, where the case νi 6= 0 gives only constant
we can expand the expected channel gain as differences in (21)-(24).
E |h̄s (t)|2 = s2U E |hU (t)|2 In Fig. 2, we plot the approximated achievable rate and the
simulated achievable rate for the SISU case with the same
+ 2sU sI E [|hU (t)|] E [C(t)] + s2I E C 2 (t) ,
parameters in Section V. We observe that the approximated
(18)
achievable rate is tightly bound with the simulated achievable
where |hU (t)|C(t) can be separated due to independence. Note rate. Thus, hereinafter, we consider R̂(t) as the achievable rate
that C(t) is the sum of M i.i.d. variables. Considering that the of the UE.5
B. SISU Scenario Problem Formulation upper bound the UAV-IRS distance and the UAV-UE distance,
respectively, which are given as
In this subsection, we formulate the optimization problem
for the SISU case. Since optimizing with respect to continuous
dI,n = k[qT [n], HA ] − [qTI , HI ]k, (36)
time will result in an infinite number of variables, we adopt the
T
path discretization technique as in [23], where the trajectory is dU,n = k[q [n], HA ] − [qTU , HU ]k. (37)
discretized into small segments. By segmenting the trajectory
into N pieces, the UAV energy consumption minimization Also, Rn is the achievable rate with the slack variables un
problem can be formulated as and vn instead of dI,n and dU,n , respectively, which can be
expressed as
N
X
(P1): min Tn P (Vn ) + τn Pc Rn =
X Y 1−si
psi i (1 − pi ) log2 (1 + γs,n ) , (38)
T,τ ,q
n=1
s∈{0,1} 2 i
s.t. q[1] = q0 , q[N + 1] = qF , (25)
γs,n = sU γU2 vn−αU + 2sU sI γU′ γI′ vn−αU /2 u−α
n
I /2
τn ≤ Tn , τn ≥ 0, (26)
+ sI γI2 un−αI . (39)
∆n ≤ min{∆max , Vmax Tn }, (27)
N
X
τn R̂n ≥ Q, (28) We observe that by considering the IRS, the achievable rate
n=1
has an involved form compared to the achievable rate with
no IRSs, which makes the analysis difficult. To the best of
where Tn , τn , Vn , q [n] and R̂n are the UAV flight time, UAV our knowledge, this is the first paper to acknowledge the
transmission time, UAV speed, UAV horizontal location, and achievable rate for a general number of IRSs.
achievable rate at the n-th segment, respectively. We assume While the objective function is a convex function, the
the UAV has fixed initial and final locations q0 and qF , respec- constraints (33)-(35) do not follow the standard convex op-
tively, as in (25). The transmission time is trivially bounded timization problem form. To resolve this issue, we adopt the
by the flight time as in (26), and we assume that the UE has well known SCA technique. By deriving a global bound of a
a fixed data constraint Q as in (28). One characteristic of the constraint that meets specific conditions, the SCA technique
path discretization technique is it assumes that the channels iteratively solves the optimization problem and guarantees to
are constant within a segment. To uphold this assumption, converge to a KKT condition point. One way to obtain the
for the segment length ∆n = kq[n + 1] − q[n]k, we bound global lower bound satisfying the specific conditions is to
the length with min{∆max , Vmax Tn }, where ∆max is the apply the first-order Taylor expansion to a convex function
maximum length of a segment to conserve this assumption, on a local point. The right hand side of the constraints (33)
and Vmax is the maximum speed of the UAV. and (35) have been shown to be convex in [23], while Rn
Similar to (P4) in [23], we transform (P1) into an equivalent from (34), which is defined in (38), needs to be analyzed.
problem as Therefore, we propose the following lemma.
N
Lemma 1. For given non-negative constants ǫz , ζz , αz , z ∈
δb ∆2n ∆3
{1, 2, ..., Z}, the function given as
X
(P2): min τn Pc + P0 Tn + + δp n2
A,y,τ ,T,q,u,v
n=1
Tn Tn
+ Pi yn f (u1 , ..., uZ )
s.t. q[1] = q0 , q[N + 1] = qF , (29) Z Z X
Z
−αi /2
X X
= log2 1 + ǫz u−α
z
z
+ ζz ζi uz−αz /2 ui ,
∆n ≤ min{∆max , Vmax Tn }, (30)
z=1 z=1 i6=z
τn ≤ Tn , τn ≥ 0, (31) (40)
dI,n ≤ un , dU,n ≤ vn , (32)
N
X is convex with respect to uz > 0, z ∈ {1, 2, ..., Z} for any
Q≤ A2n , (33) non-negative integer Z.
n=1
A2n Proof. The proof is given in Appendix A.
≤ Rn , (34)
τn
Tn4 ∆2 Since pi and (1 − pi ) are non-negative constants, Rn is
≤ yn2 + 2n , (35)
2
yn v0 a weighted linear sum of the function f (u1 , ..., uZ ) with
different values of Z, thus, through Lemma 1, we confirm
−2
where δb = 3Utip , and δp = 12 d0 ρsAp . The transformation that Rn is indeed a convex function. Due to the convexity of
with regard to slack variables An and yn , which are the n- Rn , we can successfully use the first-order Taylor expansion
th elements of A and y, are equivalent as in [23], thus, we of Rn as the lower bound.
omit the transformation details. The slack variables un and Adopting the lower bound of the constraints (33)-(35),
vn are the n-th elements of the vectors u and v, where they the overall optimization problem in the ℓ-th iteration can be6
Algorithm 1 Pseudo-code for the SISU scenario UAV energy to the UE constraint Q, e.g., Q + χ for some positive constant
consumption minimization algorithm χ. Thus, by restraining the constraint, it will have a larger
1:
(0) (0) (0)
Initialization: Set An , yn , ∆n , q(0) [n], un , vn , and
(0) (0) probability to satisfy the constraint in practice.
(0)
Rn . Let ℓ = 0.
2: repeat
3: Set ℓ = ℓ + 1. C. SIMU Scenario Problem Formulation
4: Solve the optimization problem (P3). In this subsection, we formulate the optimization problem to
(ℓ) (ℓ) (ℓ)
5: Update the solution of (P3) as An , yn , ∆n , minimize the UAV energy consumption in a SIMU scenario.
(ℓ) (ℓ) (ℓ) (ℓ)
q [n], un , vn , and Rn . However, this can be performed very easily with the similar
6: until The UAV energy consumption Etot decreases by a expression as in [23]. The resulting optimization problem is
fraction below a predefined threshold. given as
N K
( )
X X
expressed as (P4): min Tn P (Vn ) + τnk Pc (50)
T,τ ,q
n=1 k=1
N
δb ∆2n ∆3 s.t. q[1] = q0 , q[N + 1] = qF , (51)
X
(P3): min τn Pc + P0 Tn + + δp n2 K
A,y,τ ,T,q,u,v
n=1
Tn Tn X
τnk ≤ Tn , τnk ≥ 0, (52)
+ Pi yn k=1
s.t. q[1] = q0 , q[N + 1] = qF , (41) XN
∆n ≤ min{∆max , Vmax Tn }, (42) τnk R̂nk ≥ Qk , (53)
n=1
τn ≤ Tn , τn ≥ 0, (43)
∆n ≤ min{∆max , Vmax Tn }, (54)
dI,n ≤ un , dU,n ≤ vn , (44)
N
X A2n where τnk and R̂nk are the transmit time with the k-th UE at
Q≤ Ã(ℓ)
n , ≤ R̃n(ℓ) , (45) the n-th segment and the achievable rate for the k-th UE at the
n=1
τn
n-th segment, respectively, and Qk is the data constraint for the
Tn4 k-th UE. Since we assume that the channels are approximately
≤ Ỹn(ℓ) , (46)
yn2 constant within a segment, the TDMA assumption is legitimate
where the lower bound functions are defined as with the constraint (52). With the multiple UE expansion, we
observe that the transmission for all UEs must be considered
Ã(ℓ)
n = An
(ℓ)2
+ 2A(ℓ)
n An − A(ℓ)
n , (47) in each flight time as in (52), and the data constraint for each
(ℓ)2 UE must be sufficed independently as in (53). We neglect the
∆n transformation process of (P4) due to redundancy. However,
Ỹn(ℓ) = yn(ℓ)2 + 2yn(ℓ) yn − yn(ℓ) −
v02 (P4) will be a subset within the general MIMU case, thus, can
2 (ℓ) be effectively solved with the latter solutions in Section IV.
+ (q [n + 1] − q(ℓ) [n])T (q[n + 1] − q[n]) , (48)
v02
" #!
(ℓ)
un − un
R̃n(ℓ) (ℓ)
= Rn + ∇Rn (ℓ)T
(ℓ) . (49) IV. M ULTIPLE IRS O PTIMIZATION
vn − vn
(ℓ) (ℓ) (ℓ) (ℓ) (ℓ) (ℓ) In this section, we consider the most general scenario of
The values An , yn , ∆n , q(ℓ) [n], un , vn , and Rn are the MIMU case. To minimize the UAV energy consumption, we
local points to obtain the lower bound, which are the results propose two optimization problems to effectively minimize
of the (ℓ − 1)-th iteration, and the gradient ∇Rn can be the UAV energy consumption with similar performance. In
calculated through the result of Appendix B. Note that, for addition, we propose an algorithm to mimic the behavior of
the first iteration, we must define an initial case to compute a specific optimization problem to achieve low UAV energy
the problem (P3). We will specify the initial case in Section V. consumption with minimal complexity.
Since the lower bounded constraints are all affine functions,
the problem (P3) is now in the standard convex optimization
problem form, thus, can be solved effectively via problem A. General IRS Usage Case
solving tools such as CVX [32]. By iteratively solving the
optimization problem and updating the local points, the so- In the MIMU scenario, the most straightforward approach to
lutions of (P3) will be in the feasible region of the original exploit the multiple IRSs is to use all of them simultaneously.
problem (P2), which converges to a KKT condition point. The We denote this approach as the general IRS usage case. In
proposed optimization technique is summarized in Algorithm order to realize this concept, we must define the UE achievable
1. Note that, since the optimization problem is from the upper rate with regard to multiple IRSs. Fortunately, due to TDMA,
bound in (16), the resulting solution may not satisfy the data only one UE is considered in a time instance, thus, the phase
constraint in reality. To settle this issue, we can give a margin shifts of the IRSs can be easily obtained with the same7
approach as (13). Assuming that the IRSs concentrate on the IRS matching case. The formulated optimization problem will
k-th UE, the achievable rate is obtained as effectively evaluate and decide which IRS to use per instance.
X Y 1−s The overall problem can be expressed as
R̂k (t) = psi i (1 − pi ) i log2 (1 + γ̂s (t)) ,
N K
( )
s∈{0,1}W +1 i X X
(55) (P6): min Tn P (Vn ) + τnk Pc (64)
T,τ ,q
n=1 k=1
i ∈ {Uk , I1 , ..., IW },
s.t. q[1] = q0 , q[N + 1] = qF , (65)
with the state vector as s = [sUk , sI1 , ..., sIW ]T , and the SNR K
X
as τnk ≤ Tn , τnk ≥ 0, (66)
k=1
X
γ̂s (t) = si γi2 d−α
i
i
(t) N
i
X
XX −αi /2 −αj /2
τnk max R̂nwk ≥ Qk , (67)
+ si sj γi′ γj′ di (t)dj (t), (56) n=1
w
i j6=i ∆n ≤ min{∆max , Vmax Tn }, (68)
where γi and γi′ follow the same definitions as (21)-(24).
In result, the optimization problem will have the same where R̂nwk represents the achievable rate of the k-th UE
expression as (P4), and the problem can be transformed in at the n-th segment when the UE is matched with the w-
the equivalent form as th IRS. Hence, with the maximum function in (67), we
pick the best IRS per instance for communication. Since we
N XK
δb ∆2n exploit only a single IRS, R̂nwk is equivalent to (19) by
X
(P5): min τnk Pc + P0 Tn + considering the signals from only the w-th IRS, where we
A,y,τ ,T,q,u,v
n=1
Tn
k=1
assume the signals from other IRSs are negligible. While we
∆3
+ δp n2 + Pi yn can still adopt the SCA technique due to the convexity of
Tn maxw R̂nwk , the gradient is quite difficult to compute. Instead,
s.t. q[1] = q0 , q[N + 1] = qF , (57) we introduce a matching variable ηnwk , where it represents the
∆n ≤ min{∆max , Vmax Tn }, (58) communication time between the k-th UE in the n-th segment
K using the w-th IRS. The overall problem is given as
X
τnk ≤ Tn , τnk ≥ 0, (59) N
( )
X X
k=1
(P7): min Tn P (Vn ) + τnk Pc (69)
dIw ,n ≤ unw , dUk ,n ≤ vnk , (60) T,τ ,q,η
n=1 k
N
X s.t. q[1] = q0 , q[N + 1] = qF , (70)
Qk ≤ A2nk , (61) K
X
n=1
τnk ≤ Tn , τnk ≥ 0, (71)
A2nk k=1
≤ Rnk , (62)
τnk XW
Tn4 2 ∆2n ηnwk ≤ τnk , ηnwk ≥ 0, (72)
≤ y n + , (63)
yn2 v02 w=1
X
where we observe that the slack variables vnk , Ank , and unw ηnwk R̂nwk ≥ Qk , (73)
n,w
have an additional index due to the MIMU generalization.
Using Lemma 1, we have shown that Rnk is a convex function ∆n ≤ min{∆max , Vmax Tn }, (74)
with respect to unw and vnk for a general number of IRSs.
Thus, we can effectively solve (P5) with the same method as where with the additional constraints (72) and (73), the solu-
(P2). tion will effectively force only one ηnwk , which is associated
with the largest R̂nwk , to τnk , and the other ηnw′ k values to
zero. By adopting the matching variable ηnwk , we successfully
B. IRS Matching Case removed the maximum operation in (67), and the gradient of
While exploiting all the IRSs simultaneously might be the R̂nwk can be calculated through the results from the previous
most straightforward method, it may not be the most probable section since R̂nwk has the same expression as (19). As we
solution. Considering that the IRS is a communication assist- can observe, (P7) is the same as (P4) with the additional linear
ing object, the IRSs are likely to be located far apart. Thus, due constraint (72), thus, we can solve the problem equivalently
to the double propagation characteristic of the IRS, the power through the SCA technique. Note that, by choosing a single
from the IRSs that are far apart from a UE is likely to be IRS for each instance, the IRS matching case can also com-
negligible. Also, the assumption that every IRS is controlled pensate for the rather strong assumption that the LoS path
simultaneously for a single UE may be implausible. Thus, the between the IRSs and UEs always exist, since IRSs closely
resulting solutions may be less practical. located to a UE will have a high probability to have an LoS
Accordingly, we consider a more practical case that each path. Thus, the IRS matching case is more practical than the
UE is supported by only one IRS, where we denote it as the general IRS usage case.8
Algorithm 2 Pseudo-code for the low complexity UAV energy the internal division point between q̄k and q̂k using f . With
consumption minimization algorithm the transmit locations {q⋆k }, we solve the traveling salesman
1: Pair each UE to its closest IRS problem (TSP) to find the shortest trajectory [23]. Finally,
2: Find q̂k and q̄k . the UAV moves with speed Ve and the transmission takes
3: Through the toy trajectory, find f . place while the UAV is moving, and while the UAV is at the
4: if fk ≥ 1, ∀k, then hovering locations. In specific, we first calculate the amount of
5: Set the toy trajectory as the solution. transmitted data while the UAV is moving, and the remaining
6: else data is transmitted while hovering at the fixed locations {q⋆k }.
7: Compute (76). In conclusion, the UAV will fly straight to qF for small data
8: Solve the TSP of {q⋆k } to find the shortest trajectory. constraints, and will fly to all the designated transmit locations
9: Set the shortest trajectory as the solution. {q⋆k } in straight lines for large data constraints. In this paper,
10: end if we define gk (f ) = min(fk , 1). While determining the optimal
gk (f ) is itself another area of study, we consider it outside the
scope of this paper. Nevertheless, the simple gk (f ) is shown
C. Low Complexity Algorithm to have considerable performance in Section V. The overall
In this subsection, we propose a low complexity algorithm algorithm is summarized in Algorithm 2.
that attempts to follow the behavior of the optimized solutions
described in the previous subsection. Through the simulation V. S IMULATION R ESULTS
results in Section V, we observe several factors. First, the IRS
matching case has marginal performance loss than the general In this section, we verify and compare the performance
IRS usage case. Second, when the data constraint is small, the of the proposed algorithms. For simulations, the altitude and
UAV directly goes to the final location with speed Ve , where Ve heights are fixed as HA = 100 m, HIw = 20 m, and HUk = 0,
is the speed with the minimum energy consumption per meter, and the number of IRS elements is fixed as Mw = 500.
i.e., Ve = minV P (V )/V . Finally, when the data constraint For the probabilistic LoS model, the parameters are fixed as
is large, the UAV tends to communicate at fixed locations. aU = 30, bU = 0.15, aI = 15, bI = 0.18, θU = 60◦ , and
Accordingly, to actively adapt to these characteristics, we θI = 54.2◦ . For the channel model, the coefficients are fixed
introduce the following low complexity algorithm. as β0 = 0.01, αI = 2.2, αU = 2.5, α = 3, κI = 30,
For this algorithm, we pair each UE with the closest IRS κU = 10, κ = 5, σ 2 = −174 dBm, and B = 1 MHz.
to mimic the IRS matching case, given the fact that the IRS For the communication scenario, the parameters are fixed as
matching case can suffice considerable performance. Before q0 = [0, 0]T , qF = [100, 100]T, ∆max = 1 m, Ve = 18.3 m/s,
we specify the trajectory, we define several auxiliary variables. Vmax = 30 m/s, and we assume that Qk is the same for all
Through a one dimensional search between each UE and its UEs for simplicity. We construct a simple solution, i.e., hover
paired IRS, we find the largest achievable rate location for each and transmit at the UE locations, and use it for the initial case
UE, denoted as q̂k . Afterwards, we draw a straight line from for the algorithms adopting the SCA technique. In the figures,
q0 to qF , denoted as a toy trajectory. Next, we find the closest the UEs are denoted as black circles and the IRSs are denoted
point from q̂k to the toy trajectory as q̄k by landing a vertical as black diamonds.
line on the toy trajectory from q̂k . Using these variables, we Additionally, we simulate the no IRS case as a benchmark
will acquire the trajectory for the UAV. by setting pIw = 0, which will have the same result as [23].
First, we simulate a toy scenario which follows the toy While other UAV trajectory optimization techniques exist for a
trajectory, moves with speed Ve , and performs equal time multiple IRS scenario, there were no results considering UAV
transmission for each UE at every segment, i.e., τnk = Tn /K. energy consumption. Thus, we only adopt the no IRS case as
From this toy scenario, we can compute the transmitted data a benchmark.
as In Fig. 3, we observe the UAV trajectory for the IRS
T matching case with different data constraints. The figure shows
r = [f1 Q1 , ..., fK QK ] , (75)
that the UAV trajectory follows a straight line for a small
where fk is the fraction of data the UAV has transmitted with data constraint, and steadily shifts towards the UE locations
respect to the k-th UE with data constraint Qk . We can see as the data constraint increases. These results follow our
that the fraction fk indirectly implies the magnitude of the data intuition. When the data constraint is small, the UAV can
constraint and distance between the toy trajectory and the UE, easily satisfy the data constraint from a far distance, thus, it
e.g., fk will be small if Qk is too large or the achievable rate uses the minimum flight power by simply flying on a straight
is too small due to the large UE distance. Note that, if fk ≥ line. As the data constraint increases, the UAV must find a
1, ∀k ∈ {1, 2, ..., K}, we can simply select the toy scenario as balance between the energy consumption due to flying and
the algorithm solution. Otherwise, we fix the transmit locations the increasing achievable rate from shorter distance between
as the UAV and UEs. Thus, as the rate constraint increases, there
is an increasing need for high achievable rate, resulting in
q⋆k = q̂k + gk (f ) (q̄k − q̂k ) , (76)
a trajectory shift towards the UEs. Note that, we observe
where q⋆k denotes the transmit location for the k-th UE, f = some unnatural behavior in some of the trajectories such as
T
[f1 , ..., fK ] , and gk (f ) is a well defined function to choose loops. This phenomenon seems to occur due to the UAV power9
100 15
IRS matching, Q: 10 Mbit IRS matching, Q: 10 Mbit
IRS matching, Q: 100 Mbit IRS matching, Q: 100 Mbit
80 IRS matching, Q: 200 Mbit IRS matching, Q: 200 Mbit
Flight time (min)
IRS matching, Q: 500 Mbit IRS matching, Q: 500 Mbit
10
y-axis (m)
60
40
5
20
0 0
0 20 40 60 80 100 0 50 100 150
x-axis (m) n
Fig. 3: UAV trajectories for the IRS matching case for different Fig. 5: UAV flight time with respect to the path segment n for
data constraints. the IRS matching case with different data constraints.
20 100
No IRS, Q: 50 Mbit
Low complexity, Q: 50 Mbit
IRS matching, Q: 50 Mbit
80 General IRS, Q: 50 Mbit
UAV speed (m/s)
15 IRS matching, Q: 10 Mbit No IRS, Q: 1000 Mbit
Low complexity, Q: 1000 Mbit
IRS matching, Q: 100 Mbit IRS matching, Q: 1000 Mbit
60 General IRS, Q: 1000 Mbit
y-axis (m)
IRS matching, Q: 200 Mbit
10 IRS matching, Q: 500 Mbit
40
5 20
0
0 0 10 20 30 40 50 60 70 80 90 100
0 50 100 150 x-axis (m)
n Fig. 6: UAV trajectories in extreme data constraints.
Fig. 4: UAV flight speed with respect to the path segment n
for the IRS matching case with different data constraints.
on a fixed location, which is the basis of our proposed low
complexity algorithm.
consumption model, i.e., the UAV power consumption is not In Fig. 6, we plot the trajectory of the UAV for the various
minimized at V = 0, therefore; the UAV may need to keep proposed techniques in two extreme data constraints. For
moving near the UEs to reduce its power consumption unless the low data constraint, we observe that all the techniques
the data constraint is too high. Note also that the solution of converge to the same trajectory, confirming that the algorithms
the optimization problem only converges to a local optimum work in a consistent manner for low data constraints. Also, it
since the problem is not in the standard convex optimization confirms that the proposed low complexity algorithm adapts
form. well to the low data constraint. For the high data constraint, we
In Figs. 4 and 5, we plot the UAV speed and UAV flight time observe that the trajectories behave in a more involved manner,
with respect to the path discretization segment n for the IRS confirming that proper adjustment is needed for different
matching case with the scenario equivalent to Fig. 3. Note that, communication requirements.
the flight time for the small data constraint cases have minimal In Fig. 7, we plot the UAV energy consumption for different
flight time, thus, are all shown at the bottom of Fig. 5. We techniques with respect to the data constraint for the scenario
first observe that the UAV speed for the lowest data constraint equivalent to Fig. 6. Note that, since one UE has two IRSs
case is a constant with value Ve . This is expected, since, to in close proximity, the results will be beneficial to the general
reduce UAV energy consumption, the UAV must choose not IRS usage case. As expected, the most prominent result is that
only the shortest trajectory, but also the optimal speed for the proposed optimization problem solutions have outstanding
energy efficiency. Also, by jointly observing the speed and performance with respect to the no IRS case, confirming that
flight time, we find out that as the data constraint increases, exploiting the IRS for UAV energy consumption minimization
the UAV has the tendency to slow down and communicate is a legitimate approach. Also, the general IRS usage case10
follows its performance. By comparison with the no IRS case,
25
No IRS
we confirmed that exploiting the IRSs in UAV communication
Low complexity systems is indeed favorable to minimize the UAV energy
20 IRS matching
General IRS
consumption.
Energy (kJ)
15
A PPENDIX A
10 P ROOF OF L EMMA 1
5 Note that the functions
log2 ǫz u−α
0 z
z
= −αz log2 (uz ) + log2 (ǫz ) ,
100 200 300 400 500 600 700 800 900 1000
Q (Mbits)
(77)
−α /2
α z
Fig. 7: UAV energy consumption with respect to the data log2 ζz ζi uz−αz /2 ui i = − log2 (uz )
2
constraint. αi
− log2 (ui ) + log2 (ζz )
2
+ log2 (ζi ) , (78)
works as a lower bound for any data constraint. This result
is expected since the general IRS usage case exploits all the are convex with respect to uz and ui , since the logarithmic
IRSs in the system, thus, will have the best performance. We function is concave and αz ≥ 0. Thus, the functions ǫz u−α z
z
also observe that the performance of the IRS matching case −αz /2 −αi /2
and ζz ζi uz ui are log-convex functions. With the
declines faster than the general IRS usage case. This is due
same approach, we can also find out that a constant value
to the composition of the UAV energy consumption model.
is also a log-convex function.
While for small data constraints, the UAV energy consumption
By exploiting the fact that the sum of log-convex functions
is dominated by the flying power to the final location, which
is log-convex, we see that the function f (u1 , ..., uZ ) is the
is inevitable. However, as the data constraint increases, the
sum of log-convex functions, which finishes the proof.
UAV hovers in a fixed location for communication, thus,
the energy consumption from hovering for communication
becomes dominant. Since the general IRS usage case will have
A PPENDIX B
the largest achievable rate, the energy consumption of hovering
for communication will be generally smaller, resulting in better
For the function
performance than the IRS matching case. Finally, we observe
the performance of the proposed low complexity algorithm. In
f (u1 , ..., uZ )
the low data constraint region, the low complexity algorithm
has some performance loss due to the constant speed of the Z Z X
Z
−α /2
X X
UAV. However, the low complexity algorithm has marginal = log2 1 + ǫz u−α
z
z
+ ζz ζi uz−αz /2 ui i
performance loss with respect to the IRS matching case in z=1 z=1 i6=z
the high data constraint region. Indeed, we have confirmed = log2 (g(u1 , ..., uZ )) , (79)
that with extremely high data constraints, the performance of
the low complexity case converges to the IRS matching case. the derivative with respect to uz is given as
This confirms that the proposed low complexity algorithm
PZ −α /2−1 −αi /2
works sufficiently well in the high data constraint region with df −αz ǫz u−α
z
z −1
− i6=z −αz ζz ζi uz z ui
a simply defined gk (f ), while it can be further improved = .
duz g(u1 , ..., uZ ) ln 2
by optimizing the speed of UAV for the low data constraint (80)
region, which is an interesting future research topic.
In result, the derivative of f (u1 , ..., uZ ) is expressed as
VI. C ONCLUSION
T
In this paper, we proposed several techniques to minimize
df df
the UAV energy consumption in a general MIMU scenario. We ∇f = , ..., , (81)
du1 duZ
successfully developed the probabilistic LoS channel model
to ensure the advantages of highly located IRSs. Through which finishes the derivation.
some approximations, we transformed and derived the new
system into a tractable form, which could be solved by the
SCA technique. Employing various approaches, we derived ACKNOWLEDGEMENT
an algorithm that exploits all the IRSs simultaneously and an
algorithm that actively selects the best IRS for transmission. The authors thank S. Kang and B. Ko at KAIST for helping
Using the results, we proposed a low complexity algorithm that with the derivation of Appendix A and designing Fig. 1.11
R EFERENCES [22] Y. Zeng and R. Zhang, “Energy-Efficient UAV Communication With
Trajectory Optimization,” IEEE Transactions on Wireless Communica-
tions, vol. 16, no. 6, pp. 3747–3760, Mar. 2017.
[1] F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski,
[23] Y. Zeng, J. Xu, and R. Zhang, “Energy Minimization for Wireless Com-
“Five Disruptive Technology Directions for 5G,” IEEE Communications
munication With Rotary-Wing UAV,” IEEE Transactions on Wireless
Magazine, vol. 52, no. 2, pp. 74–80, Feb. 2014.
Communications, vol. 18, no. 4, pp. 2329–2345, Mar. 2019.
[2] T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N. [24] S. Li, B. Duo, X. Yuan, Y. Liang, and M. D. Renzo, “Reconfigurable
Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter Wave
Intelligent Surface Assisted UAV Communication: Joint Trajectory De-
Mobile Communications for 5G Cellular: It Will Work!” IEEE Access,
sign and Passive Beamforming,” IEEE Wireless Communications Letters,
vol. 1, pp. 335–349, May 2013. vol. 9, no. 5, pp. 716–720, Jan. 2020.
[3] W. Roh, J. Seol, J. Park, B. Lee, J. Lee, Y. Kim, J. Cho, K. Cheun, and [25] D. Ma, M. Ding, and M. Hassan, “Enhancing Cellular Communications
F. Aryanfar, “Millimeter-wave Beamforming as an Enabling Technology for UAVs via Intelligent Reflective Surface,” 2020 IEEE Wireless
for 5G Cellular Communications: Theoretical Feasibility and Prototype Communications and Networking Conference (WCNC), pp. 1–6, May
Results,” IEEE Communications Magazine, vol. 52, no. 2, pp. 106–113, 2020.
Feb. 2014. [26] Z. Wei, Y. Cai, Z. Sun, D. W. K. Ng, J. Yuan, M. Zhou, and L. Sun,
[4] Y. Niu, Y. Li, D. Jin, L. Su, and A. V. Vasilakos, “A Survey of “Sum-Rate Maximization for IRS-Assisted UAV OFDMA Communica-
Millimeter Wave Communications (mmWave) for 5G: Opportunities and tion Systems,” IEEE Transactions on Wireless Communications, vol. 20,
Challenges,” Wireless Networks, vol. 21, pp. 2657–2676, Apr. 2015. no. 4, pp. 2530–2550, Dec. 2021.
[5] S. Rangan, T. S. Rappaport, and E. Erkip, “Millimeter-Wave Cellular [27] S. Fang, G. Chen, and Y. Li, “Joint Optimization for Secure Intelligent
Wireless Networks: Potentials and Challenges,” Proceedings of the Reflecting Surface Assisted UAV Networks,” IEEE Wireless Communi-
IEEE, vol. 102, no. 3, pp. 366–385, Feb. 2014. cations Letters, vol. 10, no. 2, pp. 276–280, Sep. 2021.
[6] W. Saad, M. Bennis, and M. Chen, “A Vision of 6G Wireless Systems: [28] T. Shafique, H. Tabassum, and E. Hossain, “Optimization of Wireless
Applications, Trends, Technologies, and Open Research Problems,” Relaying With Flexible UAV-Borne Reflecting Surfaces,” IEEE Trans-
IEEE Network, vol. 34, no. 3, pp. 134–142, Oct. 2020. actions on Communications, vol. 69, no. 1, pp. 309–325, Oct. 2021.
[7] E. Björnson, L. Sanguinetti, H. Wymeersch, J. Hoydis, and T. L. [29] Y. Cai, Z. Wei, S. Hu, D. W. K. Ng, and J. Yuan, “Resource Allocation
Marzetta, “Massive MIMO is a Reality—What is Next?: Five Promising for Power-Efficient IRS-Assisted UAV Communications,” 2020 IEEE
Research Directions for Antenna Arrays,” Digital Signal Processing, International Conference on Communications Workshops (ICC Work-
vol. 94, pp. 3 – 20, Nov. 2019. shops), pp. 1–7, Jun. 2020.
[8] H. Cho and J. Choi, “Alternating Beamforming With Intelligent Re- [30] L. Ge, P. Dong, H. Zhang, J. Wang, and X. You, “Joint Beamforming
flecting Surface Element Allocation,” IEEE Wireless Communications and Trajectory Optimization for Intelligent Reflecting Surfaces-Assisted
Letters, vol. 10, no. 6, pp. 1232–1236, Mar. 2021. UAV Communications,” IEEE Access, vol. 8, pp. 78 702–78 712, Apr.
[9] M. Renzo, M. Debbah, and D. PhanHuy, “Smart Radio Environments 2020.
Empowered by Reconfigurable AI Meta-Surfaces: An idea whose time [31] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP Altitude
has come.” EURASIP Journal on Wireless Com Network, no. 129, May for Maximum Coverage,” IEEE Wireless Communications Letters, vol. 3,
2019. no. 6, pp. 569–572, Jul. 2014.
[10] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, and [32] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex
I. Akyildiz, “A New Wireless Communication Paradigm Through programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014.
Software-Controlled Metasurfaces,” IEEE Communications Magazine,
vol. 56, no. 9, pp. 162–169, Sep. 2018.
[11] E. Basar, M. Renzo, J. Rosny, M. Debbah, M. Alouini, and R. Zhang,
“Wireless Communications Through Reconfigurable Intelligent Sur-
faces,” IEEE Access, vol. 7, pp. 116 753–116 773, Aug. 2019.
[12] H. Guo, Y. Liang, J. Chen, and E. G. Larsson, “Weighted Sum-Rate
Maximization for Reconfigurable Intelligent Surface Aided Wireless
Networks,” IEEE Transactions on Wireless Communications, vol. 19,
no. 5, pp. 3064–3076, Feb. 2020.
[13] Z. Xiao, P. Xia, and X. Xia, “Enabling UAV Cellular With Millimeter-
Wave Communication: Potentials and Approaches,” IEEE Communica-
tions Magazine, vol. 54, no. 5, pp. 66–73, May 2016.
[14] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless Communications With
Unmanned Aerial Vehicles: Opportunities and Challenges,” IEEE Com-
munications Magazine, vol. 54, no. 5, pp. 36–42, May 2016.
[15] E. P. de Freitas, T. Heimfarth, I. F. Netto, C. E. Lino, C. E. Pereira,
A. M. Ferreira, F. R. Wagner, and T. Larsson, “UAV Relay Network to
Support WSN Connectivity,” International Congress on Ultra Modern
Telecommunications and Control Systems, pp. 309–314, Dec. 2010.
[16] B. Ji, Y. Li, B. Zhou, C. Li, K. Song, and H. Wen, “Performance Analysis
of UAV Relay Assisted IoT Communication Network Enhanced With
Energy Harvesting,” IEEE Access, vol. 7, pp. 38 738–38 747, Mar. 2019.
[17] M. Alzenad, A. El-Keyi, F. Lagum, and H. Yanikomeroglu, “3-D
Placement of an Unmanned Aerial Vehicle Base Station (UAV-BS) for
Energy-Efficient Maximal Coverage,” IEEE Wireless Communications
Letters, vol. 6, no. 4, pp. 434–437, May 2017.
[18] V. Saxena, J. Jaldén, and H. Klessig, “Optimal UAV Base Station Tra-
jectories Using Flow-Level Models for Reinforcement Learning,” IEEE
Transactions on Cognitive Communications and Networking, vol. 5,
no. 4, pp. 1101–1112, Oct. 2019.
[19] S. Zhang, H. Zhang, Q. He, K. Bian, and L. Song, “Joint Trajectory and
Power Optimization for UAV Relay Networks,” IEEE Communications
Letters, vol. 22, no. 1, pp. 161–164, Oct. 2018.
[20] X. Chen, X. Hu, Q. Zhu, W. Zhong, and B. Chen, “Channel Modeling
and Performance Analysis for UAV Relay Systems,” China Communi-
cations, vol. 15, no. 12, pp. 89–97, Dec. 2018.
[21] Y. Zeng, Q. Wu, and R. Zhang, “Accessing From the Sky: A Tutorial on
UAV Communications for 5G and Beyond,” Proceedings of the IEEE,
vol. 107, no. 12, pp. 2327–2375, Dec. 2019.You can also read
