Ris + Uav

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 20, NO.
9, SEPTEMBER 2021 5769
UAV-Assisted Intelligent Reflecting Surface

Symbiotic Radio System
Meng Hua , Luxi Yang , Senior Member, IEEE, Qingqing Wu , Member, IEEE Cunhua Pan ,
Chunguo Li , Senior Member, IEEE, and A. Lee Swindlehurst , Fellow, IEEE
Abstract— This paper investigates a symbiotic unmanned aer- are provided to evaluate the performance of the proposed designs
ial vehicle (UAV)-assisted intelligent reflecting surface (IRS) radio under different setups, as compared with benchmarks.
system, where the UAV is leveraged to help the IRS reflect its
own signals to the base station, and meanwhile enhance the Index Terms— Intelligent reflecting surface (IRS), unmanned
UAV transmission by passive beamforming at the IRS. First, aerial vehicle (UAV), phase shift optimization, UAV trajectory
we consider the weighted sum bit error rate (BER) minimization optimization.
problem among all IRSs by jointly optimizing the UAV trajectory,
IRS phase shift matrix, and IRS scheduling, subject to the I. I NTRODUCTION
minimum primary rate requirements. To tackle this complicated
problem, a relaxation-based algorithm is proposed. We prove
that the converged relaxation scheduling variables are binary,
which means that no reconstruct strategy is needed, and thus
W ITH the ever-growing sales of mobile devices
and Internet of Things devices, current network
architectures are becoming overwhelmed by growing data
the UAV rate constraints are automatically satisfied. Second, traffic demands [1]. Although numerous technologies such
we consider the fairness BER optimization problem. We find as millimeter wave (mmWave) communications, ultra-dense
that the relaxation-based method cannot solve this fairness BER networks, and massive multiple-input multiple-output
problem since the minimum primary rate requirements may not
be satisfied by the binary reconstruction operation. To address (MIMO) [2]–[4] have been proposed to address this problem,
this issue, we first transform the binary constraints into a series of they are usually realized with very large energy consumption
equivalent equality constraints. Then, a penalty-based algorithm and high hardware cost due to the large number of RF chains
is proposed to obtain a suboptimal solution. Numerical results required at the terminals. Recently, a new technology has
come to the attention of the wireless research community,
Manuscript received August 15, 2020; revised January 20, 2021; accepted namely intelligent reflecting surface (IRS), due to its potential
March 19, 2021. Date of publication April 12, 2021; date of current version
September 10, 2021. This work was supported in part by the National Natural ability to reconfigure the radio propagation environment
Science Foundation of China under Grant U1936201, Grant 61941115, and in a favorable way for transceiver optimization. An IRS is
Grant 61971128 and in part by the National Key Research and Development comprised of a manmade surface of electromagnetic material
Program of China under Grant 2020YFB1804901. The work of Qingqing
Wu was supported in part by FDCT under Grant 0108/2020/A and in part consisting of a large number of square metallic patches,
by the Open Research Fund of National Mobile Communications Research each of which can be digitally controlled to induce different
Laboratory, Southeast University, under Grant 2021D15. The work of A. Lee reflection amplitude, phase, and polarization responses on
Swindlehurst was supported by the U.S. National Science Foundation under
Grant ECCS-2030029. The associate editor coordinating the review of this the incident signals [5], [6]. Since an IRS typically has
article and approving it for publication was N. Michelusi. (Corresponding numerous patch units (such as PIN-diodes), it can provide
author: Luxi Yang.) a significant passive beamforming gain without the need
Meng Hua is with the School of Information Science and Engineering,
Southeast University, Nanjing 210096, China, and also with the State Key for RF chains, thus yielding a cost- and energy-efficient
Laboratory of Internet of Things for Smart City, University of Macau, Macao solution. For example, experiments conducted recently in [7]
999078, China (e-mail: mhua@seu.edu.cn). showed that for a large IRS consisting of 1720 reflecting
Luxi Yang is with the School of Information Science and Engineering,
Southeast University, Nanjing 210096, China, and also with the Purple elements, the total power consumption is only 0.280W.
Mountain Laboratories, Nanjing 211111, China (e-mail: lxyang@seu.edu.cn). In addition, each IRS reflecting element adjusted by the smart
Qingqing Wu is with the State Key Laboratory of Internet of Things for controller is able to induce an independent phase shift on the
Smart City, University of Macau, Macau, 999078, China, and also with the
National Mobile Communications Research Laboratory, Southeast University, incident signal to change the signal propagation such that the
Nanjing 210096, China (e-mail: qingqingwu@um.edu.mo). desired and interfering signals can be added constructively or
Cunhua Pan is with the School of Electronic Engineering and Computer destructively to assist the communication system. Therefore,
Science, Queen Mary University of London, London E1 4NS, U.K. (e-mail:
c.pan@qmul.ac.uk). IRS is a promising solution for improving the spectral and
Chunguo Li is with the School of Information Science and Engineering, energy efficiency of wireless networks, and paving the way
Southeast University, Nanjing 210096, China, and also with the Peng Cheng to the green networks of the future.
Laboratory, Shenzhen 518000, China (e-mail: chunguoli@seu.edu.cn).
A. Lee Swindlehurst is with the Center for Pervasive Communications and The new research paradigm of IRS-aided wireless com-
Computing, University of California at Irvine, Irvine, CA 92697 USA (e-mail: munication has been extensively studied, e.g., see [8]–[10].
swindle@uci.edu). The authors of [8] proposed a software controlled approach,
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TWC.2021.3070014. to enable programmable control over the behavior of
Digital Object Identifier 10.1109/TWC.2021.3070014 IRS-related wireless environments. How the availability of IRS
1536-1276 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: ULAKBIM UASL - DOKUZ EYLUL UNIVERSITESI. Downloaded on July 16,2022 at 17:15:54 UTC from IEEE Xplore. Restrictions apply.
5770 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 20, NO. 9, SEPTEMBER 2021
will allow wireless network operators to redesign common and shift matrix, and IRS scheduling, subject to a minimum data
well-known network communication paradigms was discussed rate requirement for the UAV. We study two optimization
in [9]. The authors of [10] provided an overview of promis- objectives, one based on fairness for the IRS and the other
ing IRS technology for achieving smart and reconfigurable on the weighted sum BER of the IRS. Then, we develop two
environments in future wireless networks, and elaborated the novel algorithms to solve them. The main contributions of this
reflection and channel models, hardware architecture as well paper are summarized as follows:
as various applications. • We first consider the IRS-weighted sum BER optimiza-
Recently, there have been many contributions devoting tion problem, which is a mixed-integer and non-convex
efforts to integrating IRS into current cellular networks. problem. To develop a low complexity algorithm, we pro-
Joint active and passive beamforming design was investigated pose a relaxation-based method. Specifically, we first
to either maximize the system throughput or minimize the relax the binary scheduling variables to continuous
base station (BS) transmit power in [11]–[15]. In particular, variables, and then we develop an alternating optimiza-
the authors in [11] studied the BS transmit power mini- tion (AO) algorithm to solve the relaxed non-convex opti-
mization problem by jointly optimizing the BS beamform- mization problem. We prove that the obtained scheduling
ing matrix and IRS phase shift matrix while satisfying the results are the same as the binary results from the AO
users’ minimum signal-to-interference-plus-noise ratio (SINR) method, which means that no reconstruction strategy is
requirement, and the results showed that for a single-user needed, and thus the primary rate requirements of the
IRS-aided system, the received signal-to-noise ratio (SNR) UAV are always satisfied. Numerical results show the
increases quadratically with the number of reflecting elements. proposed relaxation-based method converges within only
In addition, the applications of IRS are also appealing for a few iterations.
numerous different system setups such as spectrum shar- • We then take into account the fairness among multi-
ing [16], physical layer security [17]–[19], orthogonal multiple ple IRSs, and formulate a fairness BER optimization
access [20], [21], and simultaneous wireless information and problem. The resultant problem is also a mixed integer
power transfer [22]–[24]. and non-convex problem, which is in general difficult
Unlike the above studies in which the IRS is used purely to solve optimally. We show that the commonly used
to assist the transmissions of the existing system, a new relaxation-based method cannot be applied to this prob-
IRS functionality referred to as symbiotic radio transmission lem since the UAV’s rate requirement constraints may
has been proposed (also known as passive beamforming and not be satisfied by the binary reconstruction operation for
information transfer), where the information bits are carried scheduling. To address this issue, a novel penalty-based
by the on/off states of the IRS, while passive beamforming algorithm is proposed. We first transform the binary
is achieved by adjusting the phase shift of each reflecting constraints into a series of equivalent equality constraints,
element [25]–[27]. Specifically, a sensor is integrated into and then propose a two-layer algorithm to solve the
the IRS system, which for example collects environmental problem. Numerical results show the effectiveness of this
information such as temperature, humidity, illuminating light, penalty-based algorithm.
etc., and sends it to a smart controller at the IRS via adjusting • We conduct simulation results for the two proposed
the on/off state of the IRS. Then, the controller transmits the scenarios to illustrate their performance. For the first
collected information to the BS by adjusting the on/off state scenario, we study the impact of weighting factors on
of the IRS. This concept is similar to the spatial modulation the system performance, and find that the optimized
transmission technique, where the choice of active transmit UAV trajectory places it closest to the IRS with a high
antennas is regarded as a source of information to improve weighting factor. For the second scenario, the results
the spectral efficiency [28]. show that the average fairness utility value is highly
In this paper, we study an unmanned aerial vehicle related to the IRS phase shift matrix. In addition, for
(UAV)-assisted IRS symbiotic radio system, where the UAV both scenarios, the system performance is significantly
is leveraged to assist the IRS data transmission. Specifically, improved by the optimized UAV trajectory as well as the
we consider an urban environment, where there are multiple finely tunable IRS phase shift.
IRSs available to sense environmental information. As shown The rest of this paper is organized as follows: Section II
in Fig. 1, the IRS sends its own data to the BS by controlling introduces the system model and problem formulation. In Sec-
its on/off state, and the receiver side (BS) uses the difference in tions III and IV, we study the weighted sum BER and fairness
channel response caused by the on/off state to decode the IRS BER optimization problems, respectively. Numerical results
information. The IRS also simultaneously tunes each reflecting are provided in Section V, and the paper is concluded in
element to align the phase of the signal passing through the Section VI.
UAV-IRS-BS link with that of the UAV-BS link to achieve Notations: Boldface lower-case variables denote vectors.
coherent signal combining at the BS, thereby enhancing the The notation x represents the Euclidean norm of x, the cir-
UAV communication performance. In addition, the UAV’s cularly symmetric complex Gaussian variable x with
mean μ
flexible mobility can be exploited to create favorable channel and variance σ 2 is denoted by x ∼ CN μ, σ 2 , statistical
conditions for the UAV-BS and UAV-IRS links. Our goal in expectation is defined as E {·}, statistical variance is defined
this paper is to minimize the bit error rate (BER) of the as Var {·}, and O (·) denotes the big-O computational com-
IRS by jointly optimizing the UAV trajectory, IRS phase plexity notation.
HUA et al.: UAV-ASSISTED IRS SYMBIOTIC RADIO SYSTEM 5771
mobility constraints are given below:

qu [n] − qu [n − 1] ≤ Vmax δ, ∀n, (1)
qu [0] = qI , qu [N ] = qF , (2)
where qI and qF represent the UAV’s initial and final location,
respectively.
It is assumed that the IRS has M reflecting elements, and the
reflection coefficient matrix of IRS k at time slot n is defined
by the diagonal matrix Θk [n]= diag(ejθk,1 [n] , . . . , ejθk,M [n] ),
where εk,m [n] denotes the phase shift corresponding to the
mth reflecting element of IRS k at time slot n [11], [22],
[29], [30]. Let h1,k [n] ∈ CM×1 , h2,k [n] ∈ CM×1 , and
h3 [n] ∈ C1×1 respectively denote the complex equivalent
baseband channel vector between the UAV and the kth IRS,
between the kth IRS and the BS, and between the UAV and
the BS, at time slot n, ∀k ∈ K. To capture both the large-scale
and small-scale fading, we model all channels as Rician [31].
Fig. 1. UAV assisted IRS Symbiotic Radio System. Specifically, the channel coefficient between the UAV and IRS
k at time slot n is given by [32]
h1,k [n]

K1 1
= β1,k [n] hLoS
[n] + hNLoS
[n] ,
K1 + 1 1,k K1 + 1 1,k
(3)
where β1,k [n] represents the large-scale fading channel coef-
ficient, hLoS
1,k [n] and h1,k
NLoS
[n] denote the deterministic line-
of-sight (LoS) channel component and the small-scale fading
component, respectively, and K1 is the Rician factor. The value
Fig. 2. Transmission frame for the IRS and primary transmission.
of β1,k [n] is related to the communication distance, and is
given by
II. S YSTEM M ODEL AND P ROBLEM F ORMULATION β0
β1,k [n] = , (4)
dα 1
1,k [n]
A. System Model
where β0 denotes the channel power at the reference distance
Consider a UAV-assisted IRS symbiotic radio system con- 2 2
of 1 meter, d1,k [n] = qu [n] − qs,k + (Hu − Hs ) is the
sisting of a single-antenna BS, a single-antenna UAV, and
K IRSs as shown in Fig. 1, where the UAV acts to help distance between the UAV and IRS k, and α1 denotes the path
the IRS transmit its own data to the BS.1 The BS and IRS loss exponent. We assume that the IRS employs a uniform
are in fixed locations, and the UAV can freely adjust its linear array (ULA) of reflecting elements, and thus hLoS
1,k [n]
heading as it moves. The horizontal coordinates of the BS is given by [32]

and the kth IRS are respectively denoted as qb = [qbx , qby ]T hLoS −j
2πd1,k [n]
1,k [n] = e
λ
y T
and qs,k = [qs,k x
, qs,k ] . In addition, the altitude of the BS T
2πd 2π(M −1)d
is denoted as Hb , and the altitudes of all the IRSs are the × 1, e−j λ cos φ1,k [n] , . . . e−j λ cos φ1,k [n]
, (5)
same and denoted as Hs .2 We assume that the UAV flies in a
periodic trajectory at a fixed altitude Hu and with a given where d denotes the IRS element xspacing, λ denotes the carrier
q −qu x
[n]
period T . To make the problem tractable, the period T is wavelength, and cos φ1,k [n] = s,k d1,k [n] is the cosine of the
equally divided into N time slots of duration δ = T /N . As a angle of arrival (AoA) [33], [34]. The elements of hNLoS1,k [n] of
result, the trajectory of the UAV can be approximated by the the non-LoS component are assumed to be independent and
N two-dimensional sequences qu [n] = [qux [n], quy [n]]T . Note identically distributed and follow circularly symmetric com-
that the duration δ should be chosen to be sufficiently small plex Gaussian distribution with zero mean and unit variance.
to satisfy Vmax δ Hu , where Vmax denotes the maximum Similarly, the channel vector between IRS k and the BS at
UAV speed, so that the UAV’s location can be considered time slot n is expressed as
as approximately unchanged within each time slot. The UAV
h2,k [n]

1 For K2 1
ease of exposition, we refer to the UAV and BS as the primary network = β2,k hLoS + hNLoS
[n] ,
to assist the IRS’s information transmission. K2 + 1 2,k K2 + 1 2,k
2 Note that adopting different IRS altitudes does not affect the algorithm
design in this paper. (6)

2 2
where β2,k = dβα02 , d2,k = qb − qs,k + (Hb − Hs ) , where the UAV can only communicate with at most one IRS3
2,k
α2 represents the path loss exponent, and K2 in any time slot n. Define the scheduling variable ak [n], where
is the corresponding Rician factor. hLoS = ak [n] = 1 indicates that IRS k is served by the UAV, and
2πd
2π(M −1)d
2,k
T ak [n] = 0 otherwise. We then have the following scheduling
−j λ2,k −j 2πd cos φ 2,k −j cos φ 2,k
e 1, e λ ,...e λ , where constraints
qx −qx K
cos φ2,k = bd2,ks,k denotes the cosine of the angle of
departure (AoD). The elements of hNLoS [n] are also assumed ak [n] ≤ 1, ∀k, n, (8)
2,k
k=1
to be independent and identically distributed and follow
circularly symmetric complex Gaussian distribution with zero ak [n] ∈ {0, 1} , ∀k, n. (9)
mean and unit variance. If IRS k is communicating with the UAV in time slot n,
Finally, for the UAV-BS link at time slot n we have the signal received by the BS at the lth sub-block of block
n1 within time slot n is given by
h3 [n] yr,k [n, n1 , l]
√
K3 1 = P hH 2,k [n] Θk [n] h1,k [n,] xs,k [n, n1 ] xu [n, n1 , l]
= β3 [n] hLoS [n] + hNLoS [n] ,
K3 + 1 3 K3 + 1 3
IRS-aided link
√
(7) + P h3 [n] xu [n, n1 , l] +w [n, n1 , l] , (10)

direct link
β0
where β3 [n] = α
d3 3 [n]
, d3 [n] = where P denotes the transmit power atthe UAV, xu [n, n1 , l] ∼

qu [n] − qb 2 + (Hb − Hu )2 , hLoS
2πd
[n] = e−j λ ,
3 [n]
CN (0, 1), w[n, n1 , l] ∼ CN 0, σ 2 denotes the noise at
3
path-loss exponent α3 , and Rician factor K3 . In addition, IRS k with power σ 2 . We adopt the simple but widely used
hNLoS [n] ∼ CN (0, 1). on-off keying (OOK) modulation for the IRS’s information
3
Remark 1: Although in this paper we assume a ULA at the transmission, i.e., xs,k [n, n1 ] = {0, 1}.
IRS to facilitate the exposition, all the proposed algorithms Since the IRS’s own signal and UAV’s primary sig-
are applicable to the case of a uniform planar array (UPA) at nal are simultaneously received by the BS, to detect the
the IRS with only slight modifications for optimizing the IRS composite signals correctly, several detectors such as the
phase shifts discussed in (25) in Section III. maximum-likelihood (ML) detector, linear detector, and suc-
Remark 2: Generally, there are two main approaches for the cessive interference cancellation (SIC)-based detector, can be
IRS-involved channel acquisition, depending on whether the applied [39]. Furthermore, the strength of the signal received
IRS elements are equipped with receive radio frequency (RF) from the UAV is generally much larger than that received from
chains or not [5]. For the first approach with active receive the IRS due to the following two reasons. First, the direct link
RF chains, conventional channel estimation methods can be between the UAV and the BS is dominated by the LoS path
applied for the IRS to estimate the channels of the UAV-IRS due to the presence of fewer scatterers. Second, since the IRS
and IRS-BS links, respectively. In contrast, for the second consists of a large number of reflecting elements, it can sig-
approach without receive RF chains at the IRS, the IRS nificantly enhance the UAV’s signal transmission via adjusting
reflection patterns can be designed jointly with uplink pilots to the phase shifters. In addition, the SIC receiver decodes the
estimate the concatenated UAV-IRS-BS channel and UAV-BS stronger signal first, subtracts it from the composite signal, and
channel e.g., [35], [36]. extracts the weaker one from the residue. Therefore, the SIC
Typically, the symbol rate for the IRS transmission is much based detector is practically appealing and applied in this
lower than that for the primary (UAV) transmission due to paper [39], [40].
the limited computational and communication capabilities at Define h [n] = h3 [n] + hH 2,k [n] Θk [n] h1,k [n] xs,k [n, n1 ].
the IRS. To describe it clearly, the frame structure for the IRS It can be observed that h [n] contains xs,k [n, n1 ], which
symbol, primary symbol, and channel coherence time is shown changes relatively fast compared to the channel variation.
in Fig. 2. We assume that the duration of each UAV time slot In other words, the IRS’s reflected signal xs,k [n, n1 ] plays the
equals the channel coherence time, i.e, δ=Tc . In the figure, role of a fast-varying channel component, making the channel
xs,k [n, n1 ] represents the kth IRS symbol transmitted to the h [n] vary over the primary signal xu [n, n1 , l] shown in (10).
BS in the n1 th block of time slot n, and xu [n, n1 , l] is the According to [26], [ [32], Appendix B.7], [41], the achievable
primary symbol transmitted from the UAV to the BS at the rate (bps/Hz) for the primary (UAV) system assisted by IRS
lth sub-block of block n1 within time slot n. Denote by Ts k is given by
and Tu the durations of each IRS symbol and primary symbol, R̄u,k [n, n1 , l]

respectively. Without loss of generality, we assume that each P |h [n]|
2
IRS symbol covers L primary symbols, namely Ts = LTu , = Eh[n] log2 1+
σ2
where L is an integer, and L 1. In addition, we assume
δ = Tc = N1 Ts , where N1 is an integer, and N1 1. 3 Strictly speaking, the UAV directly communicates with the controller at
To facilitate the system design, we consider a widely IRS k rather than IRS itself. In the sequel, we will use the two terminologies
used wake-up communication scheduling approach [37], [38], interchangeably.

2
(a) P |h [n]| ȳr,k [n, n1 ] ∼ N Lσi2 , Lσi4 , under Hi , i = {0, 1}. (16)
= Exs,k [n,n1 ] log2 1+
σ2
⎛ 2 ⎞ Define probability density function ! f (ȳr,k [n, n1 ]|Hi ) =
(ȳr,k+ [n,n1 ]−Lσi2 )
2
⎜ P h3 [n] + h2,k [n] Θk [n] h1,k [n] ⎟

H √ 1
exp − 2Lσ4
, i = {0, 1}. Following
(b) 2πLσ2
= ρlog2 ⎝1 + ⎠ i i
σ2 from [44], the decision criteria is if

f (ȳr,k [n, n1 ]|H1 ) ≥ f (ȳr,k [n, n1 ]|H0 ) , (17)
2
P |h3 [n]|
+ (1 − ρ) log2 1+ , (11) then symbol “1” is chosen, otherwise, symbol “0” is chosen.
σ2
As such, the IRS BER can be derived from
where (a) holds since under a channel coherence time, h3 [n],
h1,k [n], and h2,k [n] are all invariant, while h [n] varies with Pe,k [n, n1 ]

the IRS’s reflected signal xs,k [n, n1 ]. Equality (b) holds since = Pr (H1 ) Pr ȳr,k [n, n1 ] < ȳrth [n, n1 ] |H1

we assume that the probability for sending symbol “1” at IRS + Pr (H0 ) Pr ȳr [n, n1 ] ≥ yrth [n, n1 ] |H0 , (18)
k is ρ (0 ≤ ρ ≤ 1), and that for sending symbol “0” at IRS
k is 1 − ρ, ∀k. It can be seen that the primary rate for each where ȳrth [n, n1 ] denotes the decision threshold. The optimal
sub-block within time slot n is the same. Thus, the achievable decision threshold ȳrth,opt [n, n1 ] can be derived by setting the
rate for the primary system assisted by IRS k at time slot n first derivative of Pe,k [n, n1 ] with respect to (w.r.t.) ȳrth [n, n1 ],
is given by R̄u,k [n] = R̄u,k [n, n1 , l]. leading to
⎡ % ⎤
After correctly decoding the primary signal xu [n, n1 , l], & 2
& 2 (σ 2 + σ 2 ) ln σ1
and subtracting the primary signal from the composite signal, 2 2
Lσ σ ⎢ ' 1 0 σ02 ⎥
ȳrth,opt = 2 1 02 ⎣1 + 1 + ⎦ . (19)
we can obtain the intermediate signal [40] σ1 + σ0 L (σ1 − σ0 )
2 2
ŷr,k [n, n1 , l] = w [n, n1 , l]

√ Therefore, the BER for detecting IRS k’s reflected symbol can
+ P hH 2,k [n] Θk [n] h1,k [n] xs,k [n, n1 ] xu [n, n1 , l] . be derived as
(12)
Pe,k [n, n1 ]

For the different IRS reflected symbols, the signals received Lσ12 − ȳrth,opt ȳrth,opt − Lσ02
at the BS have different amplitude values as = ρQ √ + (1 − ρ)Q √ ,
Lσ12 Lσ02
ỹr,k [n, n⎧
1 , l] !
√ (a) √ σ12 − σ02
⎪ P hH
⎨ 2,k [n] Θk [n] h1,k [n] xu [n, n1 , l]
≈ Q L 2
σ1 + σ02
= +w [n, n1 , l] , for symbol 1, (13) ⎛ 2 ⎞
⎪
⎩ w [n, n , l] , for symbol 0. H
1 ⎜√ P h 2,k [n] Θk [n] h1,k [n] ⎟
= Q⎝ L 2 ⎠, (20)
It is not difficult to check that4 H
P h2,k [n] Θk [n] h1,k [n] +2σ 2

CN 0, σ12 , for symbol 1, + ∞ t2
ỹr,k [n, n1 , l] ∼ (14) with the Q function given by Q (x) = √12π x e− 2 dt. The
CN 0, σ02 , for symbol 0,
result in (a) holds since for a large value L, ȳrth,opt given
2 2Lσ12 σ02
in (19) approaches σ2 +σ 2 . We can see that the BER of IRS
where σ12 = P hH 2,k [n] Θk [n] h1,k [n] + σ and σ0 = σ .
2 2 2
1 0
We adopt a simple joint-energy detector for detect- k within each sub-block of time slot n is the same. Thus,
ing the IRS’s symbols [42], [43]. Define ȳr,k [n, n1 ] = the BER for IRS k at time slot n is given by Pe,k [n] =
L
2
Pe,k [n, n1 ].
|ỹr,k [n, n1 , l]| . It can be readily checked that ȳr,k [n, n1 ] The following theorem establishes the average communica-
l=1
is the sum of L independent identically distributed central tion throughput and average BER. Theorem 1: The average
chi-squared random variables with two degrees of freedom. achievable rate for the primary system, i.e., E{R̄u,k [n]},
Assuming that the symbol “1” hypothesis is H1 and symbol is upper bounded by
!
“0” hypothesis is H0 , we can readily obtain the expectation , - P β3 [n]
2
and variance of |ỹr,k [n, n1 , l]| as E R̄u,k [n] ≤ R̂u,k [n] = (1 − ρ) log2 1 +
σ2
. /
2
E |ỹr,k [n, n1 , l]| = σi2 , under Hi , i = {0, 1}, 2 (K1 +K2 +1)Mβ1,k [n]β2,k β3 [n] !
P |x0,k [n]| + +K
+ρlog2 1+
(K1 +1)(K2 +1) 3 +1
,
2
Var |ỹr,k [n, n1 , l]| = σi4 , under Hi , i = {0, 1}. (15) σ2
(21)
Based on the central limit theorem, when L is large, the dis-
K3 β3 [n] LoS
tribution of ȳr,k [n, n1 ] asymptotically approaches a Gaussian where x0,k [n] = K3 +1 h3 [n] +
distribution: . /H
K1 K2 β1,k [n]β2,k
(K1 +1)(K2 +1) h2,k [n] Φk [n] hLoS
LoS
1,k [n].
4 Note that although x [n, n , l] is known at the BS after decoding,
u 1 Proof : Please refer to Appendix A.
xu [n, n1 , l] can still be regarded as a circularly symmetric complex Gaussian
random variable each IRS symbol since each IRS symbol covers L primary In Theorem 1, we can see that R̂u,k [n] is deter-
symbols. mined by the deterministic LoS channel components

h3 LoS [n] , hLoS
1,k [n] , hLoS
2,k [n] , the large-scale fading coef- the same reasons as those discussed above for the weighted
ficients {β1,k [n] , β2,k , β3 [n]}, and the IRS phase shift matrix sum BER problem.
Φk [n]. It is worth pointing out that the above approximation Remark 3: It can be seen that (20) is complicated and
will be tight if the SNR is sufficiently high [45]. challenging to analyze directly. It can also be readily verified
|hH [n]Θk [n]h1,k [n]| that the Q function is monotonically decreasing with SNR
Define SNR γ̄k [n] = 2,k σ2 . Similarly, we can
γ̄k [n]. Instead of maximizing the SNR directly, we introduce
obtain the average SNR for IRS k at time slot n as
. / a utility function F (γ̄k [n]), which is a differential, concave
2 (K +K +1)Mβ [n]β2,k

|x̄0,k [n]| + 1 (K21 +1)(K21,k +1)
and monotonically increasing function w.r.t. γ̄k [n], to replace
γk [n] = , (22) the BER Pe,k [n]. In addition, to obtain the average BER
σ2
. /H E {Pe,k [n]}, we set the upper bound of E {F (γ̄k [n])},
where x̄0,k [n] =
K1 K2 β1,k [n]β2,k
hLoS
[n] i.e., F (E {γ̄k [n]}), as our design metric for facilitating the
(K1 +1)(K2 +1) 2,k
algorithm design. In the subsequent sections, instead of mini-
Φk [n] hLoS
1,k [n]. mizing the BER directly, we aim to maximize the correspond-
ing utility functions.
B. Problem Formulation
For the first scenario, our goal is to minimize the weighted III. R ELAXATION -BASED A LGORITHM FOR W EIGHTED
sum BER among all IRSs over all the time slots by jointly S UM BER O PTIMIZATION P ROBLEM
optimizing the UAV trajectory, IRS phase shift matrix, and In this section, we propose a relaxation-based algorithm
IRS scheduling. Accordingly, the problem can be formulated to solve problem (23). Specifically, we first relax the binary
as scheduling variables into continuous variables, divide the
K N relaxed non-convex problem into two sub-problems, and then
min wk ak [n] E {Pe,k [n]} (23a) solve these two sub-problems. However, even with this decom-
θk,m [n],qu [n],ak [n]
k=1 n=1 position, the problem is still difficult to handle due to the
K non-convex cosine in both the objective function and con-
s.t. ak [n]R̂u,k [n] ≥ Rth , ∀n, (23b) straints. To address this issue, we first obtain a closed-form
k=1 solution for the IRS phase shift matrix for a given UAV
0 ≤ εk,m [n] ≤ 2π, ∀m, k, n, (23c) trajectory and communication scheduling, and then substitute
(1), (2), (8), (9), (23d) this expression into the original problem resulting in a joint
IRS scheduling and UAV trajectory optimization problem.
where wk denotes the weighting factor for IRS k, with a higher
We first develop the following theorem:
value representing a higher priority over other IRSs, and Rth
Theorem 2: For any given UAV trajectory and IRS schedul-
is the minimum rate requirement for the primary transmission
ing, the optimal IRS phase shift matrix that maximizes the
system in any time slot n. Problem (23) is challenging to solve
primary rate and SNR is given by
mainly due to the following three reasons. First, the optimiza-
tion variables ak [n] for communication scheduling are binary opt 2πd ((cos φ2,k − cos φ1,k [n]) (m − 1))
εk,m [n] = −
and thus (9), (23a), and (23b) involve integer constraints. λ
Second, the IRS phase shift matrix, UAV trajectory, and 2πd (− (d1,k [n] − d2,k [n]) +d3 [n])
− , ∀k, m, n. (25)
IRS scheduling are intricately coupled in (23a) and (23b), λ
which makes the problem non-convex. Third, the expression Proof : Please refer to Appendix B.
of E{Pe,k [n]} in the objective function is implicit w.r.t. the From Appendix B, we can see that the maximizer of the
optimization variables. In general, there is no efficient method terms |x0,k [n]|2 and |x̄0,k [n]|2 are respectively given by
to optimally solve problem (23). ∗ 2
For the second scenario, our goal is to minimize the x0,k [n]2 = K3 β3 [n] + K1 K2 M β1,k [n] β2,k
K3 + 1 (K + 1) (K2 + 1)
maximum BER among all IRS over all the time slots by 0 1
jointly optimizing the UAV trajectory, the IRS phase shift K1 K2 K3 β1,k [n] β2,k β3 [n]
matrix, and the IRS scheduling. Accordingly, the problem can + 2M , (26)
(K1 + 1) (K2 + 1) (K3 + 1)
be formulated as
and
min R (24a) ∗ 2
θk,m [n],qu [n],ak [n],R x̄ [n]2 = K1 K2 M β1,k [n] β2,k . (27)
0,k
N (K1 + 1) (K2 + 1)
1
s.t. ak [n] E {Pe,k [n]} ≤ R, ∀k, (24b) Substituting (26) and (27) in (21) and (22), respectively,
N n=1 we have
K
ak [n]R̂u,k [n] ≥ Rth , ∀n, Ru,k [n]
(24c) !
k=1 P β3 [n] P β3 [n]
= (1 − ρ) log2 1 + + ρlog2 1 +
(1), (2), (8), (9), (23c). (24d) σ2 σ2
!
The left hand side of (24b) denotes the average BER of IRS k ck,2 β1,k [n] β3 [n]+ (ck,1 +ck,3 ) β1,k [n]
× (28)
over all N time slots. Problem (24) is also difficult to solve for σ2
and linear optimization problem, it has very low computational

(ck,1 +ck,3 ) β1,k [n] complexity [48].
γk [n] = , (29)
σ2
K K M2β
1 2 2,k B. UAV Trajectory Optimization
where ck,1 = (K1 +1)(K2 +1) , ck,2 =
For any given IRS scheduling ak [n], the UAV trajectory
K1 K2 K3 β2,k (1+K1 +K2 )Mβ2,k
2M (K1 +1)(K2 +1)(K3 +1) , and ck,3 = (K1 +1)(K2 +1) . optimization problem of (30) becomes
As a result, the weighted sum BER optimization problem !
K N
(23) can be simplified as (ck,1 +ck,3 ) β1,k [n]
max wk ak [n] F (33a)
K N ! qu [n]
n=1
σ2
(ck,1 +ck,3 ) β1,k [n] k=1
max wk ak [n] F s.t. (1), (2), (30b). (33b)
qu [n],ak [n]
n=1
σ2
k=1
(30a) Note that (33) is neither concave nor quasi-concave due to
K
the non-convex constraints (30b) and non-convex objective
P (ck,1 +ck,3 ) β1,k [n]
s.t. ak [n] ρlog2 1 + function (33a). In general, there is no efficient method to
σ2 obtain the optimal solution. In the following, we adopt the
k=1
!
ck,2 β1,k [n] β3 [n]+β3 [n] successive convex optimization technique to solve (33). To this
σ2
+ end, we introduce additional slack variables {z1,k [n]} and
! {z3 [n]}, and recast (33) as
P β3 [n] !
(1 − ρ) log2 1 + ≥ Rth , ∀n, (30b) K N
(ck,1 +ck,3 ) z1,k [n]
σ2 max wk ak [n] F
(1), (2), (8), (9). (30c) qu [n],z1,k [n],z3,k [n]
n=1
σ2
k=1
(34a)
It can be seen that (30) only involves two variables, qu [n] and K

ak [n], and the cosine function no longer appears, which thus P (ck,1 +ck,3 ) z1,k [n]
s.t. ak [n] ρlog2 1 +
make the problem easier to solve. In the following, a low com- σ2
k=1
plexity algorithm based on the relaxation method is proposed. !
ck,2 z1,k [n] z3 [n]+z3 [n]
Note that the proposed approach is unlike [38], [46], [47], +
σ2
where the resulting continuous scheduling variables need to be
!
converted into binary. However, for problem (30), we prove P z3 [n]
that the converged relaxation scheduling variables are binary, (1 − ρ) log2 1 + ≥ Rth , ∀n, (34b)
σ2
which means that no reconstruction strategy is needed, and
β1,k [n] ≥ z1,k [n] , ∀k, n, (34c)
thus the UAV rate constraints (30b) are automatically satisfied.
Specifically, we first relax the binary variable ak [n] to a β3 [n] ≥ z3 [n] , ∀n, (34d)
continuous variable, and rewrite constraint (9) as follows [38], (1), (2). (34e)
[46], [47]:
It can be shown that at the optimal solution to (34), we must
0 ≤ ak [n] ≤ 1, ∀k, n. (31) have β1,k [n] = z1,k [n] and β3 [n] = z3 [n] , ∀k, n, since
otherwise we can always increase z1,k [n] (or z3 [n]) without
We then decompose the relaxed problem into two separate
decreasing the value of the objective. Therefore, problem (34)
subproblems, IRS scheduling and UAV trajectory optimiza-
is equivalent to problem (33). With this reformulation, objec-
tion, and then alternately optimize each one.
tive function (34a) is now concave w.r.t. z1,k [n], but with
the new non-convex constraints (34c) and (34d). The key
A. IRS Scheduling Optimization observation is that in (34c), although β1,k [n], defined in (4),
For any given UAV trajectory qu [n], the IRS scheduling 2
is not convex w.r.t. qu [n], it is convex w.r.t. qu [n] − qs,k .
problem of (30) becomes Recall that any convex function is globally lower-bounded
K N ! by its first-order Taylor expansion at any feasible point [49].
(ck,1 +ck,3 ) β1,k [n] 2
max wk ak [n] F (32a) Therefore, for any local point qru [n] − qs,k obtained at the
ak [n]
n=1
σ2 rth iteration, we have
k=1
s.t. (8), (30b), (31). (32b)
β1,k [n]
Since both the objective function and constraints are linear β0
w.r.t. ak [n], problem (32) is thus a linear optimization problem. ≥ . /α1 /2
2 2
Theorem 3: The optimal solution aopt qru [n] − qs,k + (Hu − Hs )
k [n] to problem (32)
is binary, i.e., aopt
k [n] ∈ {0, 1}. α1 β0
Proof : Please refer to Appendix C. − . / α1 +1
2 2 2
Theorem 3 shows that even though the binary constraint 2 qru [n] − qs,k + (Hu − Hs )
in the IRS scheduling problem of (32) has been relaxed, . /
the obtained solution is still a binary result. As such, no recon- × qu [n] − qs,k 2 −qru [n] − qs,k 2 = ϕlb (β1,k [n]) .
struction operation is needed. In addition, since (32) is a (35)
Define the new constraint as Algorithm 1 Proposed Relaxation-Based Algorithm for Solv-
ing Problem (30)
ϕlb (β1,k [n]) ≥ z1,k [n] , ∀k, n, (36)
2
lb
1: Initialize qru [n] − qs,k , r = 0, rmax .
which is convex since ϕ (β1,k [n]) is a quadratic function 2: Relax binary scheduling variables as continuous variables,
w.r.t. qu [n]. Similarly, for any local point qru [n] − qb 2 and set ark [n]=1/K.
obtained at the rth iteration, with β3 [n] = dαβ30[n] and 3: Repeat
3
2 2 4: Solve problem (32) for given {qru [n]}, and denote the
d3 [n] = qu [n] − qb + (Hb − Hu ) , β3 [n] which is
optimal solution as {ar+1
k [n]}.
defined in (34d) can be replaced by Solve problem (40) for given {ar+1
5: k [n]}, and denote
β0 the optimal solution as {qr+1u [n]}.
. / α23 6: Update r ← r + 1.
2 2
qru [n] −qb + (Hu − Hb ) 7: Until the fractional increase in the objective value of (30)
α3 β0 is below a threshold or the
− . / α3 +1 maximum number of iterations rmax is reached.
2 2 2
2 qu [n] − qb + (Hu − Hb )
r
. /
2 2
× qu [n] − qb −qru [n] − qb ≥ z3 [n] , ∀
In step 5, the complexity for solving (40) by the interior point
n, (37)
method is O(2KN + 3N )3.5 [51], where 2KN + 3N denotes
which is also a convex constraint. the number of variables.
. .Therefore, the total complexity
// of
3.5
In addition, to tackle the non-convexity of constraint (34b), Algorithm 1 is O Liter KN +(2KN + 3N ) , where
we introduce variable z2,k [n], and reformulate (34b) as Liter stands for the number of iterations required to reach
K
convergence.
P (ck,1 +ck,3 ) z1,k [n]
ak [n] ρlog2 1 +
σ2
!
k=1
! IV. P ENALTY-BASED A LGORITHM FOR FAIRNESS BER
ck,2 z2,k [n]+z3 [n] P z3 [n] O PTIMIZATION P ROBLEM
+ (1 − ρ) log2 1 +
σ2 σ2
In this section, we aim to solve problem (24). Based on
≥ Rth , ∀n, (38) Theorem 2, (28), and (29) in Section III, problem (24) is
with the additional constraint simplified as
2
z2,k [n] max R (41a)
z1,k [n] ≥ , ∀k, n. (39) qu [n],ak [n],R
z3 [n] !
N
1 (ck,1 +ck,3 ) β1,k [n]
Both constraints (38) and (39) are convex since we can see s.t. ak [n] F ≥ R, ∀k,
that the left hand side of (38) is a log function, which is N n=1
σ2
concave, and the right hand side of (39) is a quadratic- (41b)

over-linear fractional function, which is convex. Define z [n] = K
P (ck,1 +ck,3 ) β1,k [n]
{z1,k [n] , z2,k [n] , z3 [n]}. As a result, for any given local ak [n] ρlog2 1 +
2 2 σ2
points qru [n] − qs,k and qru [n] − qb , we have the k=1
!
following optimization problem ck,2 β1,k [n] β3 [n]+β3 [n]
+
K N ! σ2
(ck,1 +ck,3 ) z1,k [n] !
max wk ak [n] F P β3 [n]
qu [n],z[n]
n=1
σ2 (1 − ρ) log2 1 + ≥ Rth , ∀n, (41c)
k=1
σ2
(40a)
s.t. (1), (2), (36), (37), (38), (39). (40b) (1), (2), (8), (9). (41d)
Unfortunately, the low-complexity algorithm based on the

C. Convergence Analysis and Computational Complexity relaxation-based method cannot be applied to problem (41)
In the proposed AO algorithm, we solve the relaxed prob- due to the primary rate requirement (41c). More specifically,
lem (30) by iteratively solving problems (32) and (40), where when converting the continuous-valued solutions for the ak [n]
the solution obtained for one subproblem in each iteration is obtained by the relaxed problem to binary, e.g., using the
used as the initial point for the other. The detailed procedure rounding function [46], constraint (41c) will in general no
for solving (30) is summarized in Algorithm 1. The conver- longer be satisfied. In this section, we propose a two-layer
gence of Algorithm 1 has been well studied in [50], and is penalty-based algorithm to solve (41). The inner layer solves
omitted here for brevity. a penalized optimization problem by applying the AO method,
We now analyze the complexity of Algorithm 1. In step 4, while the outer layer updates the penalty coefficient, until con-
(32) is a linear optimization problem whose complexity is vergence is achieved. Specifically, in the inner layer, the orig-
O (KN ) [48], where KN denotes the number of variables. inal problem (41) is decomposed into three subproblems: IRS
phase shift matrix optimization, IRS scheduling optimization, 2) Optimizing ak [n] for given āk [n] and qu [n]. This sub-
and UAV trajectory optimization. problem is written as
We first introduce slack variables {āk [n]} to transform
min −R
the binary constraints into a series of equivalent equality ak [n],R
constraints. Specifically, (9) can be rewritten as
1
K N . /
2 2
+ |ak [n] (1−āk [n])| +|ak [n] −āk [n]|
ak [n] (1−āk [n]) = 0, ∀k, n, (42) 2η
k=1 n=1
ak [n] = āk [n] , ∀k, n. (43) (47a)
s.t. (8), (41b), (41c). (47b)
From (42) and (43), we can readily derive that the ak [n] that
satisfies the above two constraints must be either 1 or 0, which It can be seen that (47) is convex with a quadratic objective
confirms the equivalence of the transformation of (9) into the function and linear inequality constraints, which can be numer-
two constraints. We then use (42) and (43) in a penalty term ically solved by standard convex optimization techniques, such
that is added to the objective function of (41), yielding the as the interior-point method [49].
following optimization problem 3) Optimizing qu [n] for given ak [n] and āk [n]. Ignoring the
constant terms that are irrelevant to the UAV trajectory, this
min −R subproblem is formulated as:
qu [n],ak [n],R,āk [n]
1
K N . / max R (48a)
2 2 qu [n],R
+ |ak [n] (1−āk [n])| +|ak [n] −āk [n]|
2η s.t. (1), (2), (41b), (41c). (48b)
k=1 n=1
(44a) Note that (48) is neither concave nor quasi-concave due to the
s.t. (1), (2), (8), (41b), (41c), (44b) non-convex constraints (41b) and (41c). In general, there is no
efficient method to obtain the optimal solution. In the follow-
where η > 0 is the penalty coefficient used to penalize the ing, we adopt the successive convex optimization technique to
violation of the equality constraints (42) and (43) [50]. While solve (48). Using the previous analysis of the UAV trajectory
these equality constraints become satisfied as η → 0, it is not optimization for problem (33) in Section III-B, by introducing
effective to initially set η to be a very small value since in this the same slack variables {z1,k [n], z2,k [n], z3 [n]} and local
2 2
case the objective will be dominated by the penalty terms, and points qru [n] − qs,k and qru [n] − qb , we can directly
the term −R will be diminished. In contrast, initializing η with derive the following equivalent convex optimization problem
a larger value allows us to obtain a good starting point for the
proposed algorithm. Then, by gradually decreasing the value max R (49a)
qu [n],z1,k [n],z2,k [n],z3 [n],R
of η, we can finally obtain a solution that satisfies (42) and N !
(43) within a predefined accuracy. Note that, for any given 1 (ck,1 +ck,3 ) z1,k [n]
s.t. ak [n] F
penalty coefficient η, problem (44) is still non-convex due N n=1
σ2
to the non-convex constraints (41b) and (41c). We then apply ≥ R, ∀k,
the AO method to iteratively optimize the primary variables in (49b)
different blocks [50]. Specifically, in the inner layer, problem
(44) is divided into three subproblems in which {āk [n]}, (1), (2), (36), (37), (38), (39). (49c)
{ak [n]}, and {qu [n]} are optimized iteratively as follows: Based on the previous discussions, the objective function and
all of the constraints are convex. Thus, (49) is a convex
optimization problem that can be efficiently solved by, for
A. Inner Layer Iteration
example, the interior point method [49].
1) Optimizing āk [n] for given ak [n] and qu [n]. This sub-
problem can be expressed as B. Outer Layer Iteration
In the outer layer, we gradually decrease the value of the
min −R
R,āk [n] penalty coefficient η as follows:
1
K N . / η = cη, (50)
2 2
+ |ak [n] (1−āk [n])| +|ak [n] −āk [n]|
2η where c (0 < c < 1) is a scaling factor, where a larger value
k=1 n=1
(45a) of c can achieve better performance but at the cost of more
s.t. (41b). (45b) iterations in the outer layer.
We can see that only the auxiliary variable āk [n] is involved C. Convergence Analysis and Computational Complexity
in the objective function. Therefore, setting the derivative of To show the converged solutions of the proposed
(45) w.r.t. āk [n] to zero, the solution can be obtained as penalty-based algorithm, the terminal criteria for the outer
ak [n] + a2k [n] layer is given as follows;
āopt
k [n] = , ∀k, n. (46)
1 + a2k [n] ξ= max {|ak [n] (1−āk [n])| , |ak [n] −āk [n]| , ∀k, n} , (51)
Algorithm 2 Proposed Penalty-Based Algorithm for Solving

Problem (41)
1: Initialize ark1 [n], qru1 [n] − qs,k 2 , η, r1 = 0, r2 = 0 ε1 ,
ε2 , rmax .
2: Repeat: outer layer
3: Repeat: inner layer
4: Update ārk1 [n] based on (46).
5: Update ark1 [n] by solving problem (47).
6: Update qru1 [n] by solving problem (49).
7: r1 ← r1 + 1.
8: Until the fractional decrease of the objective value of
(44) is below a threshold ε1 or the
maximum number of iterations rmax is reached.
9: Update penalty coefficient η r2 based on (50). Fig. 3. Convergence behaviour of the proposed Algorithm 1 for different
periods T .
10: r2 ← r2 + 1, and r1 ← 0.
11: Until the constraint violation ξ is below a threshold ε2
where ξ sets a predefined level of accuracy. The detailed

penalty-based algorithm is summarized in Algorithm 2. In the
inner layer, with the given penalty coefficient, the objective
function of (44) is non-increasing over each iteration after
applying the AO method, and the objective of (44) is bounded
due to the limited flying time T and transmit power P .
As such, a stationary point can be achieved in the inner
layer. In the outer layer, we gradually decrease the penalty
coefficient so that the equality constraints (42) and (43) are
ultimately satisfied. Based on the results in [52, Appendix B],
this penalty-based framework is guaranteed to converge.
The complexity of Algorithm 2 can be quantified as follows. Fig. 4. Optimized UAV trajectory for different weighting factors.
In the inner layer, the main complexity of Algorithm 2
comes from steps 5 and 6. In step 5, the complexity
of computing ak [n] is O(KN + 2N + 1)3.5 [51], where δ = 0.1 s. The path loss exponents for the UAV-IRS link,
KN + 2N + 1 stands for the number of variables [53]. IRS-BS link, and UAV-BS link all assumed to be 2.4, and
Similarly, in step 6, the complexity required to compute the Rician factors for the above links are set to be 10 dB.
the UAV trajectory is O(2KN + 3N + 1)3.5 [51], Without loss of generality, we set the utility function F (·) as
where 2KN + 3N + 1 denotes the number of variables. a logarithm function with base 2, which naturally achieves a
Therefore, the total complexity of Algorithm 2 is certain degree of fairness among the information transmission
3.5 3.5
O(Louter Linner((KN + 2N + 1) +(2KN + 3N + 1) )), of multiple IRSs, and has been widely adopted in the literature,
where Linner and Louter respectively denote the number of such as [56]. Unless otherwise specified, we set rmax = 300,
iterations required for reaching convergence in the inner layer ρ = 0.5, ε1 = 10−3 , ε2 = 10−10 , η = 500, c = 0.7.
and outer layer.
A. Weighted Sum BER Optimization
V. N UMERICAL R ESULTS This subsection evaluates the performance of Algorithm 1
In this section, we provide numerical results to verify the for the weighted sum BER problem (23). We consider 5 IRS,
performance of the proposed algorithm for the UAV assisted which are located at qs,1 = [30 m, 30 m]T , qs,2 =
IRS symbiotic radio transmission system. In the simulation, [−30 m, 30 m]T , qs,3 = [−40 m, 0]T , qs,4 =
we consider a system that operates on a carrier frequency of [−30 m, −30 m]T , qs,5 = [30 m, −30 m]T in a horizontal
755 MHz with a system bandwidth of 1 MHz and effective plane. Unless otherwise specified, the weighting factors
noise power density −120dBm/Hz. As such, the noise power are set as w = [1, 1, 1, 1, 1]T . To show the efficiency of
at the BS and the channel gain are set to σ 2 = −60 dBm Algorithm 1, its convergence behaviour for the two different
and β0 = −30 dB, respectively [54]. In addition, we set periods T is plotted in Fig. 3. It is observed that the average
d/λ = 1/2 [55]. The UAV altitude is fixed at Hu = 30 m with weighted sum utility value increases quickly with the number
transmit power P = 20 dBm and maximum speed Vmax = of iterations, and in both cases converges within only
10 m/s. The UAV’s initial and final location are set to qI = 3 iterations.
qF = [15m 0]T . The altitudes of the BS and IRS are both In Fig. 4, the optimized UAV trajectories obtained by Algo-
set to Hs = Hb = 10 m. The duration of each time slot is rithm 1 when T = 40 s are studied for two different weighting
Fig. 5. UAV speed under T = 40 s. Fig. 7. Average weighted sum utility value versus period T .
Fig. 8. Average weighted sum utility value versus the number of IRS
Fig. 6. IRS scheduling under T = 40 s.
reflecting elements.
T T
factors, i.e., w1 = [1, 1, 1, 1, 1] and w2 = [1, 1, 0.5, 1, 1] . fixed phase shift examples, the UAV trajectory is set to be the
We see that the UAV sequentially visits all IRS for the result obtained by the proposed scheme. The upper bound for
weighting factor w1 , since the path loss between the UAV and the weighted sum BER problem is given by the solution to
IRS is significantly reduced when the UAV is nearby, thereby
1 !2
improving the utility value. However, for weighting factor w2 , LP (ck,1 + ck,3 ) β0
the UAV only does a close fly-by of IRS 3 rather than hovering max log2 1 + α . (52)
∀k σ 2 (Hu − Hs ) 1
above it, since w2 places a lower weight on IRS 3 and hence
reduces its priority relative to the others. To see this more It is observed from Fig. 7 that our proposed algorithm sub-
clearly, in Fig. 5 the UAV speed profile for the two weighting stantially outperforms the other methods in terms of average
factors is plotted. Compared with w1 , for w2 the UAV spends weighted sum utility value. This is expected since an optimized
less time hovering above IRS 3. UAV trajectory can establish better channel conditions for the
In Fig. 6, the IRS scheduling for T = 40 s is plotted. IRS, which significantly increases the IRS SNR. In addition,
We see that for optimizing the weighted sum utility, the IRSs by adjusting the IRS phase shifts to align the cascaded AoA
are scheduled for different lengths of time as shown in Fig. 6. and AoD with the UAV-BS link, i.e., as shown in Theo-
As before, the IRS scheduling results are indeed binary, which rem 2, the SNR of the UAV-IRS-BS link will be significantly
verifies the effectiveness of Algorithm 1. increased.
In Fig. 7, we compare the average weighted sum utility In Fig. 8, we study the average weighted sum utility
value versus T achieved by the following schemes: 1) Pro- value versus the number of IRS reflecting elements M . The
posed scheme in Algorithm 1; 2) Circular trajectory, where the performance gain of the proposed approach and the circular
UAV flies with a circular path of radius 15 m (corresponding trajectory increases with M , since more reflecting elements
to the distance from the BS to the UAV’s initial/final location) help achieve higher passive beamforming gain. In addition,
and center [0, −0]T ; 3) Fixed phase shifts, where the IRS phase our proposed approach outperforms the circular trajectory by
shifts for all the elements are fixed at either π or π/2. For the leveraging the UAV mobility. Clearly, the IRS has a significant
Fig. 9. Convergence behaviour of Algorithm 2.
Fig. 10. Optimized UAV trajectories for different T . Fig. 11. Optimized UAV speed for T = 40 s.
impact on the system performance, and the IRS phase shifts increases, the UAV exploits its mobility to adaptively enlarge
must be finely tuned in the system design. and adjust its trajectory to move closer to each IRS. When
T becomes sufficiently large, i.e., T = 40 s, the UAV is
able to sequentially visit all the IRS and stay stationary above
B. Fairness BER Optimization each of them for a certain amount of time. This is expected
This subsection evaluates the performance of Algorithm 2 since when the distance between the UAV and IRS is small,
for the BER fairness problem (24). The initial setup for the the length of the double channel fading propagation, i.e., the
BER fairness simulations are the same as those used for evalu- UAV-IRS-BS link, will be reduced, thus improving the IRS
ating the weighted sum BER approach discussed above. Fig. 9 transmission SNR. To see this more clearly, Fig. 11 plots the
shows the penalty violation ξ in (51) and the convergence UAV speed for the case when T = 40 s. We see that the
behavior of Algorithm 2 under different periods T . It can UAV flies either with maximum or zero speed, indicating that
be seen from Fig. 9(a) that ξ converges very fast with the the UAV flies with maximum UAV speed to move closer to
value decreasing to 10−10 after 34 iterations for T = 20 s. the IRS, and then remains stationary above it. Additionally,
Even when T = 40 s, the constraint is eventually satisfied we observe in Fig. 12 that the IRS sequentially communicates
within the predefined accuracy (i.e., 10−10 ) by 34 iterations, with each UAV to experience better channel conditions, and
which indicates that the proposed penalty-based algorithm can the scheduling results are indeed binary, which demonstrates
effectively tackle the binary scheduling constraints. In addi- that the constraints in (42) and (43) are satisfied by the
tion, in Fig. 9(b), we plot the fairness utility value versus proposed Algorithm 2.
the number of outer layer iterations. We see that the utility In Fig. 13, we study the average fairness utility value versus
increases quickly with the number of outer layer iterations for period T for our proposed scheme compared with the same
both the T = 20 s and T = 40 s cases, and convergence to a benchmarks as those considered for the weighted sum BER
fraction of the final value is achieved only 4 iterations. problem in Fig. 7. Fig. 13 shows that the fairness utility for
In Fig. 10, the optimized trajectories obtained by Algo- the circular trajectory is constant regardless of the period T
rithm 2 for T = 20 s and T = 40 s are plotted. As T due to the time-invariant air-to-ground channels. In contrast,
Fig. 12. Optimized IRS scheduling for T = 40s.

Fig. 14. Average fairness utility value versus the number of IRS reflecting
elements.
the number of IRS reflecting elements increases, since more

reflecting elements help achieve higher passive beamforming
gain. In addition, the performance of the fixed IRS phase shift
scheme is very poor, and the average fairness utility value
nearly approaches zero due to the unaligned angles of the
UAB-IRS-BS and UAV-BS links, which implies that the IRS
phase shift must be carefully tuned.
VI. C ONCLUSION
In this paper, we studied a UAV-assisted IRS symbiotic
radio system. We exploited the UAV mobility to maximize
Fig. 13. Average fairness utility value versus period T . the data information transferred from several IRSs to a given
BS. We first studied a weighted sum BER minimization
problem by jointly optimizing the UAV trajectory, IRS phase
the utility achieved by the proposed scheme increases with T , shifts, and IRS scheduling, and proposed a low-complexity
which further demonstrates the benefits of leveraging the UAV relaxation-based method to solve it. We proved that the solu-
mobility. The calculation of the upper bound for the BER tion to the relaxed problem provides binary scheduling results,
fairness problem is different from that for the weighted-sum and hence no additional operation is needed to enforce this
BER problem. When T is sufficiently large, it can be assumed constraint. We then considered fairness among the IRSs, and
that the amount of time each IRS served is equal. As for the developed a BER fairness optimization problem. To handle the
case when the UAV hovers above the IRS, an upper bound resulting mixed integer non-convex problem, we transformed
for the BER fairness problem can be obtained by solving the the binary constraints into an equivalent set of equality con-
following problem straints, and proposed a penalty-based method to address the
constraints. The effectiveness of this approach was justified
max Rupper (53a) by the numerical simulations. Simulation results demonstrated
xk ≥0,Rupper
! that the system performance can be significantly improved by
LP (ck,1 + ck,3 ) β0
s.t. xk log2 1 + α ≥ Rupper, ∀k, optimizing the UAV trajectory as well as the IRS phase shifts.
σ 2 (Hu − Hs ) 1
(53b)
K A PPENDIX A
xk = 1, (53c) P ROOF OF T HEOREM 1
k=1 To show Theorem 1, we first define the function f (z) =
. / log2 (1 + z) , z ≥ 0. It can be readily checked that f (z) is
LP (c 0+c )β
where the term log2 1 + σ2 (Hk,1 k,3
u −Hs ) 1
α represents the concave with respect to z. Thus, based on Jensen’s inequal-
achievable rate for the IRS when the UAV is directly above ity [49], we have E {f (z)} ≤ log2 (1 + E {z}). Therefore,
IRS k, and xk denotes the travel time ratio for IRS k. the following inequality holds
Problem (53) is a linear optimization problem, and thus can , -
be easily solved by the interior point method. E R̄u,k [n]
. 2 2 /
In Fig. 14, the average fairness utility value versus the ≤ ρlog2 1 + P E h3 [n] + hH 2,k [n] Θ k [n] h1,k [n] /σ
number of IRS reflecting elements is studied. We see that . /
2
the performance gain of the proposed scheme increases as + (1 − ρ) log2 1 + P E |h3 [n]| /σ 2 . (54)
0
!
Since the small-scale fading channel coefficients hNLoS [n], K3 β3 [n] 2πd3 [n]
=
3
hNLoS [n], and h NLoS
[n] are independent of each other, exp −j
1,k 2,k K3 + 1 λ
we can obtain 0
2 !
K1 K2 β1,k [n] β2,k 2π (d1,k [n] − d2,k )
E h3 [n] + hH 2,k [n] Θ k [n] h1,k [n] + exp −j
(K1 + 1) (K2 + 1) λ
2 2
= |x0,k [n]| + E |x1,k [n]| M
2πd (cos φ2,k − cos φ1,k [n]) (m − 1)
× exp j
+E |x2,k [n]|2 + E |x3,k [n]|2 + E |x4,k [n]|2 , m=1
λ
(55) +εk,m [n]))|
0
!
K3 β3 [n] LoS K1 K2 β1,k [n]β2,k (a) K β [n] 2πd [n]
≤ +
where x0,k [n] = 3 3 3
K3 +1 h3 [n] + (K1 +1)(K2 +1) exp −j
β3 [n] NLoS K3 + 1 λ
(h2,k [n]) Φk [n]h1,k [n], x1,k [n]= K3 +1 h3
LoS H LoS
[n], 0
. /H K K β [n] β !
K β1,k [n]β2,k
x2,k [n] = (K11 +1)(K hNLoS [n] Φk [n] hLoS 1 2 1,k 2,k 2π (d1,k [n] − d2,k )
2 +1) 2,k 1,k [n], × exp −j
. /H (K1 + 1) (K2 + 1) λ
K β1,k [n]β2,k
x3,k [n] = (K21 +1)(K 2 +1)
hLoS
2,k [n] Φk [n] hNLoS
1,k [n], and M
. /H 2πd (cos φ2,k − cos φ1,k [n]) (m − 1)
β [n]β2,k × exp j
+1)(K2 +1) h2,k
x4,k [n] = (K11,k [n] Φk [n, l] hNLoS
NLoS
1,k [n]. λ
m=1
We first calculate
+ εk,m [n]))| , (58)
2
E |x2,k [n]| where (a) is due to the triangle inequality, which holds with
K1 β1,k [n] β2,k LoS H
2π(d1,k [n]−d2,k )
H equality if and only if −j 2πdλ3 [n] = −j +
= h [n] (Φk [n]) 2πd(cos φ2,k −cos φ1,k [n])(m−1)
λ
(K1 + 1) (K2 + 1) 1,k j + εk,m [n], ∀m. This indicates
NLoS H
λ
×E hNLoS [n] h2,k [n] Φk [n] hLoS that the mth phase shift at IRS k should be tuned such that
2,k 1,k [n]
the phase of the signal that passes through the UAV-IRS
(a) K1 M β1,k [n] β2,k and IRS-BS links is aligned with that of the signal over the
= , (56)
(K1 + 1) (K2 + 1) UAV-BS direct link to achieve coherent signal combining at
1 . /H 2 the BS. Thus, we can obtain the closed-form IRS phase shift
where (a) holds since E hNLoS 2,k [n] hNLoS
2,k [n] = IM , expression in (25). In addition, it can be easily checked that
opt
. /H εk,m [n] in (25) is also the optimal solution that maximizes
(Φk [n])H Φk [n] = IM , and hLoS 1,k [n] hLoS
1,k [n] = M . We the IRS reflecting rate in (22). This completes the proof of
can obtain the remaining terms as follows: Theorem 2.
β3 [n]
2
E |x1,k [n]| = , A PPENDIX C
K3 + 1
K2 M β1,k [n] β2,k
P ROOF OF T HEOREM 3
2
E |x3,k [n]| = , It can be readily verified that problem (32) satisfies Slater’s
(K1 + 1) (K2 + 1)
M β1,k [n] β2,k
condition, and thus strong duality holds and its optimal
2
E |x4,k [n]| = . (57) solution can be obtained by solving its dual problem [49].
(K1 + 1) (K2 + 1)
Specifically, we first introduce the dual variables {λ[n] ≥ 0}

2 associated with the primary rate constraints (30b), and derive
In addition, we have E |h3 [n]| = β3 [n]. Combining all of
the partial Lagrangian of problem (32) as follows
the above results, we can directly arrive at (21).
L (ak [n] , λ [n])
K N
A PPENDIX B
P ROOF OF T HEOREM 2 = wk ak [n] F (γk [n])
k=1 n=1
Here we derive a closed-form solution for the IRS phase N
K

shifts that maximize the primary rate expression R̂u,k [n] + λ [n] ak [n] Ru,k [n] − Rth
in (21). We have the following inequality n=1 k=1
N
K
= (wk F (γk [n]) + λ [n] Ru,k [n]) ak [n]
|x0,k [n]|
0 n=1 k=1
−λ [n] Rth ) . (59)
K3 β3 [n] LoS
= h3 [n]
K3 + 1 The Lagrange dual function of (32) is defined as
0
K1 K2 β1,k [n] β2,k LoS H g (λ [n]) = max L (ak [n] , λ [n]) (60a)

+ h2,k [n] Φk [n] h1,k [n]
LoS ak [n]
(K1 + 1) (K2 + 1) s.t. (8), (31). (60b)
It can be seen that the dual function (60) can be divided [16] X. Guan, Q. Wu, and R. Zhang, “Joint power control and passive
into N subproblems that can be solved in parallel. The beamforming in IRS-assisted spectrum sharing,” IEEE Commun. Lett.,
vol. 24, no. 7, pp. 1553–1557, Jul. 2020.
n -th subproblem of (60) can be written as [17] H. Yang, Z. Xiong, J. Zhao, D. Niyato, L. Xiao, and Q. Wu, “Deep
K reinforcement learning-based intelligent reflecting surface for secure
wireless communications,” IEEE Trans. Wireless Commun., vol. 20,
max (wk F (γk [n ]) + λ [n ] Ru,k [n ]) ak [n ] no. 1, pp. 375–388, Jan. 2021, doi: 10.1109/TWC.2020.3024860.
ak [n ]
k=1 [18] G. Zhou, C. Pan, H. Ren, K. Wang, and A. Nallanathan, “A framework
−λ [n ] Rth (61a) of robust transmission design for IRS-aided MISO communications with
imperfect cascaded channels,” IEEE Trans. Signal Process., vol. 68,

s.t. 0 ≤ ak [n ] ≤ 1, ∀k, (61b) pp. 5092–5106, 2020.
K [19] X. Lu, W. Yang, X. Guan, Q. Wu, and Y. Cai, “Robust and secure
beamforming for intelligent reflecting surface aided mmWave MISO
ak [n ] ≤ 1. (61c) systems,” IEEE Wireless Commun. Lett., vol. 9, no. 12, pp. 2068–2072,
k=1 Dec. 2020.
[20] Y. Yang, B. Zheng, S. Zhang, and R. Zhang, “Intelligent reflecting
It can be easily derived that the optimal solution aopt k [n ] surface meets OFDM: Protocol design and rate maximization,” IEEE
opt opt
that maximizes (61) is either ak [n ] = 1 or ak [n ] = 0 Trans. Commun., vol. 68, no. 7, pp. 4522–4535, Jul. 2020.
for k = k, where subscript k corresponds to the index [21] Y. Yang, S. Zhang, and R. Zhang, “IRS-enhanced OFDMA: Joint
resource allocation and passive beamforming optimization,” IEEE Wire-
that maximizes wk F (γk [n ]) + λ [n ] Ru,k [n ] among all less Commun. Lett., vol. 9, no. 6, pp. 760–764, Jun. 2020.
k ∈ {1, . . . , K}. This also holds for the case that there are [22] Q. Wu and R. Zhang, “Weighted sum power maximization for intelligent
more than two IRS that have the same maximum value of reflecting surface aided SWIPT,” IEEE Wireless Commun. Lett., vol. 9,
no. 5, pp. 586–590, May 2020.
wk F (γk [n ]) + λ [n ] Ru,k [n ] among all k ∈ {1, . . . , K}. [23] C. Pan et al., “Intelligent reflecting surface aided MIMO broadcasting
This thus completes the proof of Theorem 3. for simultaneous wireless information and power transfer,” IEEE J. Sel.
Areas Commun., vol. 38, no. 8, pp. 1719–1734, Aug. 2020.
[24] Z. Li, W. Chen, and Q. Wu, “Joint beamforming design and power
R EFERENCES splitting optimization in IRS-assisted SWIPT NOMA networks,” 2020,
arXiv:2011.14778. [Online]. Available: http://arxiv.org/abs/2011.14778
[1] Q. Wu, G. Y. Li, W. Chen, D. W. K. Ng, and R. Schober, “An overview
[25] W. Yan, X. Yuan, and X. Kuai, “Passive beamforming and information
of sustainable green 5G networks,” IEEE Wireless Commun., vol. 24,
transfer via large intelligent surface,” IEEE Wireless Commun. Lett.,
no. 4, pp. 72–80, Aug. 2017.
vol. 9, no. 4, pp. 533–537, Apr. 2020.
[2] A. L. Swindlehurst, E. Ayanoglu, P. Heydari, and F. Capolino,
[26] Q. Zhang, Y.-C. Liang, and H. Vincent Poor, “Large intelligent
“Millimeter-wave massive MIMO: The next wireless revolution?” IEEE
Surface/Antennas (LISA) assisted symbiotic radio for IoT
Commun. Mag., vol. 52, no. 9, pp. 56–62, Sep. 2014.
communications,” 2020, arXiv:2002.00340. [Online]. Available:
[3] M. Kamel, W. Hamouda, and A. Youssef, “Ultra-dense networks:
http://arxiv.org/abs/2002.00340
A survey,” IEEE Commun. Surveys Tuts., vol. 18, no. 4, pp. 2522–2545,
4th Quart., 2016. [27] J. Hu, Y.-C. Liang, and Y. Pei, “Reconfigurable intelligent surface
[4] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, enhanced multi-user MISO symbiotic radio system,” IEEE Trans. Com-
“An overview of massive MIMO: Benefits and challenges,” IEEE J. Sel. mun., early access, Dec. 25, 2020, doi: 10.1109/TCOMM.2020.3047444.
Topics Signal Process., vol. 8, no. 5, pp. 742–758, Oct. 2014. [28] R. Y. Mesleh, H. Haas, S. Sinanovic, C. W. Ahn, and S. Yun, “Spatial
[5] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment: modulation,” IEEE Trans. Veh. Technol., vol. 57, no. 4, pp. 2228–2241,
Intelligent reflecting surface aided wireless network,” IEEE Commun. Jul. 2008.
Mag., vol. 58, no. 1, pp. 106–112, Jan. 2020. [29] G. Zhou, C. Pan, H. Ren, K. Wang, and A. Nallanathan, “Intelligent
[6] T. J. Cui, S. Liu, and L. Zhang, “Information metamaterials and reflecting surface aided multigroup multicast MISO communication
metasurfaces,” J. Mater. Chem. C, vol. 5, no. 15, pp. 3644–3668, 2017. systems,” IEEE Trans. Signal Process., vol. 68, pp. 3236–3251, 2020.
[7] W. Tang et al., “Wireless communications with reconfigurable intelligent [30] S. Zhang and R. Zhang, “Capacity characterization for intelligent
surface: Path loss modeling and experimental measurement,” IEEE reflecting surface aided MIMO communication,” IEEE J. Sel. Areas
Trans. Wireless Commun., vol. 20, no. 1, pp. 421–439, Jan. 2021. Commun., vol. 38, no. 8, pp. 1823–1838, Aug. 2020.
[8] C. Liaskos, S. Nie, A. Tsioliaridou, A. Pitsillides, S. Ioannidis, [31] C. Zhan and Y. Zeng, “Aerial–ground cost tradeoff for multi-UAV-
and I. Akyildiz, “A new wireless communication paradigm through enabled data collection in wireless sensor networks,” IEEE Trans.
software-controlled metasurfaces,” IEEE Commun. Mag., vol. 56, no. 9, Commun., vol. 68, no. 3, pp. 1937–1950, Mar. 2020.
pp. 162–169, Sep. 2018. [32] D. Tse and P. Viswanath, Fundamentals Wireless Communications.
[9] M. D. Renzo et al., “Smart radio environments empowered by recon- Cambridge, U.K.: Cambridge Univ. Press, 2005.
figurable AI meta-surfaces: An idea whose time has come,” EURASIP [33] S. Li, B. Duo, X. Yuan, Y.-C. Liang, and M. Di Renzo, “Reconfigurable
J. Wireless Commun. Netw., vol. 2019, no. 1, pp. 1–20, May 2019. intelligent surface assisted UAV communication: Joint trajectory design
[10] Q. Wu, S. Zhang, B. Zheng, C. You, and R. Zhang, “Intelligent reflecting and passive beamforming,” IEEE Wireless Commun. Lett., vol. 9, no. 5,
surface aided wireless communications: A tutorial,” IEEE Trans. Com- pp. 716–720, May 2020.
mun., early access, Jan. 18, 2021, doi: 10.1109/TCOMM.2021.3051897. [34] H. Long et al., “Reflections in the sky: Joint trajectory and pas-
[11] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless sive beamforming design for secure UAV networks with reconfig-
network via joint active and passive beamforming,” IEEE Trans. Wireless urable intelligent surface,” 2020, arXiv:2005.10559. [Online]. Available:
Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019. http://arxiv.org/abs/2005.10559
[12] C. Huang, R. Mo, and C. Yuen, “Reconfigurable intelligent surface [35] S. Lin, B. Zheng, G. C. Alexandropoulos, M. Wen, F. Chen, and
assisted multiuser MISO systems exploiting deep reinforcement learn- S. sMumtaz, “Adaptive transmission for reconfigurable intelligent
ing,” IEEE J. Sel. Areas Commun., vol. 38, no. 8, pp. 1839–1850, surface-assisted OFDM wireless communications,” IEEE J. Sel. Areas
Aug. 2020. Commun., vol. 38, no. 11, pp. 2653–2665, Nov. 2020.
[13] C. Pan et al., “Multicell MIMO communications relying on intelligent [36] B. Zheng and R. Zhang, “Intelligent reflecting surface-enhanced OFDM:
reflecting surfaces,” IEEE Trans. Wireless Commun., vol. 19, no. 8, Channel estimation and reflection optimization,” IEEE Wireless Com-
pp. 5218–5233, Aug. 2020. mun. Lett., vol. 9, no. 4, pp. 518–522, Apr. 2020.
[14] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and [37] M. Hua, Y. Wang, Q. Wu, H. Dai, Y. Huang, and L. Yang, “Energy-
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in efficient cooperative secure transmission in Multi-UAV-Enabled wireless
wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8, networks,” IEEE Trans. Veh. Technol., vol. 68, no. 8, pp. 7761–7775,
pp. 4157–4170, Aug. 2019. Aug. 2019.
[15] S. Zhang and R. Zhang, “Intelligent reflecting surface aided mul- [38] Q. Wu, Y. Zeng, and R. Zhang, “Joint trajectory and communication
tiple access: Capacity region and deployment strategy,” 2020, design for multi-UAV enabled wireless networks,” IEEE Trans. Wireless
arXiv:2002.07091. [Online]. Available: http://arxiv.org/abs/2002.07091 Commun., vol. 17, no. 3, pp. 2109–2121, Mar. 2018.
[39] G. Yang, Q. Zhang, and Y.-C. Liang, “Cooperative ambient backscatter Luxi Yang (Senior Member, IEEE) received the
communications for green Internet-of-Things,” IEEE Internet Things J., M.S. and Ph.D. degrees in electrical engineer-
vol. 5, no. 2, pp. 1116–1130, Apr. 2018. ing from Southeast University, Nanjing, China,
[40] S. Verdu et al., Multiuser Detection. Cambridge, U.K.: Cambridge Univ. in 1990 and 1993, respectively. Since 1993, he has
Press, 1998. been with the Department of Radio Engineering,
[41] R. Long, Y.-C. Liang, H. Guo, G. Yang, and R. Zhang, “Symbiotic radio: Southeast University, where he is currently a Full
A new communication paradigm for passive Internet of Things,” IEEE Professor of information systems and communica-
Internet Things J., vol. 7, no. 2, pp. 1350–1363, Feb. 2020. tions, and the Director of the Digital Signal Process-
[42] J. Qian, F. Gao, G. Wang, S. Jin, and H. Zhu, “Noncoherent detec- ing Division. He has authored or coauthored of two
tions for ambient backscatter system,” IEEE Trans. Wireless Commun., published books and more than 200 journal articles,
vol. 16, no. 3, pp. 1412–1422, Mar. 2017. and holds 50 patents. His current research interests
[43] M. Hua, L. Yang, C. Li, Z. Zhu, and I. Lee, “Bistatic backscatter include signal processing for wireless communications, MIMO communica-
communication: Shunt network design,” IEEE Internet Things J., early tions, intelligent wireless communications, and statistical signal processing.
access, Nov. 30, 2020, doi: 10.1109/JIOT.2020.3040963. He is also a member of the Signal Processing Committee, Chinese Institute
[44] S. M. Kay, Fundamentals of Statistical Signal Processing. of Electronics. He received the first and second class prizes of science and
Upper Saddle River, NJ, USA: Prentice-Hall, 1993. technology progress awards from the State Education Ministry of China
[45] Y. Han, W. Tang, S. Jin, C.-K. Wen, and X. Ma, “Large intelli- in 1998, 2002, and 2014.
gent surface-assisted wireless communication exploiting statistical CSI,”
Qingqing Wu (Member, IEEE) received the B.Eng.
IEEE Trans. Veh. Technol., vol. 68, no. 8, pp. 8238–8242, Aug. 2019.
degree in electronic engineering from the South
[46] M. Hua, L. Yang, C. Pan, and A. Nallanathan, “Throughput maxi-
China University of Technology in 2012, and the
mization for full-duplex UAV aided small cell wireless systems,” IEEE
Ph.D. degree in electronic engineering from Shang-
Wireless Commun. Lett., vol. 9, no. 4, pp. 475–479, Apr. 2020.
hai Jiao Tong University (SJTU) in 2016.
[47] M. Hua, Y. Wang, Z. Zhang, C. Li, Y. Huang, and L. Yang, “Power-
From 2016 to 2020, he was a Research Fellow
efficient communication in UAV-aided wireless sensor networks,” IEEE
with the Department of Electrical and Computer
Commun. Lett., vol. 22, no. 6, pp. 1264–1267, Jun. 2018.
Engineering, National University of Singapore. He is
[48] J. Gondzio and T. Terlaky, “A computational view of interior point
currently an Assistant Professor with the State Key
methods,” in Advances in Linear and Integer Programming (Oxford
Laboratory of Internet of Things for Smart City,
Lecture Series in Mathematics and its Applications), vol. 4. 1996,
University of Macau. His current research interests
pp. 103–144.
include intelligent reflecting surface (IRS), unmanned aerial vehicle (UAV)
[49] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
communications, and MIMO transceiver design. He was listed as a World’s
Cambridge Univ. Press, 2004.
Top 2% Scientist by Stanford University in 2020. He was the recipient of
[50] D. P. Bertsekas, Nonlinear Programming. Belmont, MA, USA:
the IEEE WCSP Best Paper Award in 2015, the Outstanding Ph.D. Thesis
Athena Scientific, 1999.
Funding in SJTU in 2016, and the Outstanding Ph.D. Thesis Award of China
[51] G. Zhang, Q. Wu, M. Cui, and R. Zhang, “Securing UAV communi-
Institute of Communications in 2017. He was the Exemplary Editor of IEEE
cations via joint trajectory and power control,” IEEE Trans. Wireless
C OMMUNICATIONS L ETTERS in 2019 and the Exemplary Reviewer of several
Commun., vol. 18, no. 2, pp. 1376–1389, Feb. 2019.
IEEE journals. He serves as an Associate Editor for IEEE C OMMUNICATIONS
[52] Y. Cai, Q. Shi, B. Champagne, and G. Y. Li, “Joint transceiver design for
L ETTERS , IEEE W IRELESS C OMMUNICATIONS L ETTERS , IEEE O PEN
secure downlink communications over an amplify-and-forward MIMO
J OURNAL OF C OMMUNICATIONS S OCIETY (OJ-COMS), and IEEE O PEN
relay,” IEEE Trans. Commun., vol. 65, no. 9, pp. 3691–3704, Sep. 2017.
J OURNAL OF V EHICULAR T ECHNOLOGY (OJVT). He is also the Lead Guest
[53] S. Boyd. EE364b Convex Optimization. Accessed: Apr. 6, 2021.
Editor for IEEE J OURNAL ON S ELECTED A REAS IN C OMMUNICATIONS on
[Online]. Available: https://web.stanford.edu/class/ee364b/lectures.html
UAV Communications in 5G and Beyond Networks’ and the Guest Editor
[54] J. Xu, Y. Zeng, and R. Zhang, “UAV-enabled wireless power transfer:
for IEEE OJVT on 6G Intelligent Communications’ and IEEE OJ-COMS on
Trajectory design and energy optimization,” IEEE Trans. Wireless Com-
Reconfigurable Intelligent Surface-Based Communications for 6G Wireless
mun., vol. 17, no. 8, pp. 5092–5106, Aug. 2018.
Networks. He is also the Workshop Co-Chair of IEEE ICC 2019–2021
[55] Q. Wu and R. Zhang, “Joint active and passive beamforming opti-
workshop on Integrating UAVs into 5G and Beyond, IEEE GLOBECOM
mization for intelligent reflecting surface assisted SWIPT under QoS
2020, and ICC 2021 workshop on Reconfigurable Intelligent Surfaces for
constraints,” IEEE J. Sel. Areas Commun., vol. 38, no. 8, pp. 1735–1748,
Wireless Communication for Beyond 5G. He serves as the Workshops and
Aug. 2020.
Symposia Officer of Reconfigurable Intelligent Surfaces Emerging Technology
[56] Q. Ye, B. Rong, Y. Chen, M. Al-Shalash, C. Caramanis, and
Initiative and Research Blog Officer of Aerial Communications Emerging
J. G. Andrews, “User association for load balancing in heterogeneous
Technology Initiative. He is also the IEEE ComSoc Young Professional Chair
cellular networks,” IEEE Trans. Wireless Commun., vol. 12, no. 6,
in APB.
pp. 2706–2716, Jun. 2013.
Cunhua Pan received the B.S. and Ph.D. degrees
from the School of Information Science and
Engineering, Southeast University, Nanjing, China,
in 2010 and 2015, respectively.
From 2015 to 2016, he was a Research Asso-
ciate with the University of Kent, U.K. He held a
post-doctoral position at the Queen Mary University
of London, U.K., from 2016 and 2019, where he is
currently a Lecturer. His research interests include
reconfigurable intelligent surfaces (RIS), intelligent
reflection surface (IRS), ultra-reliable low latency
communication (URLLC), machine learning, UAV, the Internet of Things,
Meng Hua received the M.S. degree in electrical and and mobile edge computing. He serves as a TPC Member for numerous con-
information engineering from the Nanjing University ferences, such as ICC and GLOBECOM, and the Student Travel Grant Chair
of Science and Technology, Nanjing, China, in 2016, for ICC 2019. He is currently an Editor of IEEE W IRELESS C OMMUNI -
and the Ph.D. degree from the School of Informa- CATION L ETTERS , IEEE C OMMUNICATIONS L ETTERS , and IEEE A CCESS .
tion Science and Engineering, Southeast University, He also serves as a Leading Guest Editor for IEEE J OURNAL OF S ELECTED
Nanjing, in 2021. He is currently a Post-Doctoral T OPICS IN S IGNAL P ROCESSING (JSTSP) Special Issue on Advanced Signal
Researcher with the State Key Laboratory of Internet Processing for Reconfigurable Intelligent Surface-aided 6G Networks, IEEE
of Things for Smart City, University of Macau. His O PEN J OURNAL OF V EHICULAR T ECHNOLOGY on the Special Issue on
current research interests include intelligent reflect- Reconfigurable Intelligent Surface Empowered Wireless Communications
ing surface, UAV assisted communication, backscat- in 6G and Beyond, and IEEE A CCESS Special Issue on Reconfigurable
ter communication, energy-efficient wireless com- Intelligent Surface Aided Communications for 6G and Beyond. He is also
munication, X-connectivity, cognitive radio networks, secure transmission, and the Workshops and Symposia officer for Reconfigurable Intelligent Surfaces
optimization theory. Emerging Technology Initiative.
Chunguo Li (Senior Member, IEEE) received the A. Lee Swindlehurst (Fellow, IEEE) received the
bachelor’s degree in wireless communications from B.S. and M.S. degrees from Brigham Young Univer-
Shandong University in 2005, and the Ph.D. degree sity (BYU), in 1985 and 1986, respectively, and the
in wireless communications from Southeast Univer- Ph.D. degree from Stanford University, in 1991, all
sity in 2010. In July 2010, he joined the Faculty in electrical engineering. He was with the Depart-
of Southeast University, Nanjing, China, where he ment of Electrical and Computer Engineering with
is currently an Advisor of Ph.D. candidates and BYU from 1990 to 2007, where he served as the
a Full Professor. From June 2012 to June 2013, Department Chair from 2003 to 2006. From 1996 to
he was the Post-Doctoral Researcher with Concordia 1997, he held a joint appointment as a Visiting
University, Montreal, Canada. From July 2013 to Scholar with Uppsala University and the Royal Insti-
August 2014, he was with the DSL Laboratory of tute of Technology, Sweden. From 2006 to 2007,
Stanford University as a Visiting Associate Professor. From August 2017 to he was on leave working as the Vice President of Research for ArrayComm
July 2019, he was an Adjunct Professor of Xizang Minzu University under LLC, San Jose, CA. Since 2007, he has been a Professor with the Electrical
the supporting Tibet Program organized by the China National Human Engineering and Computer Science Department, University of California
Resources Ministry. His research interests include cell-free distributed MIMO Irvine, where he served as the Associate Dean for Research and Graduate
wireless communications and cyberspace security, and machine learning-based Studies with the Samueli School of Engineering from 2013 to 2016. From
image/video signal processing. He is also an IET fellow and the IEEE CIS 2014 to 2017, he was also a Hans Fischer Senior Fellow with the Institute
Nanjing Chapter Chair. for Advanced Studies, Technical University of Munich. In 2016, he was
elected as a Foreign Member of the Royal Swedish Academy of Engineering
Sciences (IVA). His research interests include array signal processing for
radar, wireless communications, and biomedical applications, and he has
over 300 publications in these areas. He was the inaugural Editor-in-Chief
of the IEEE J OURNAL OF S ELECTED T OPICS IN S IGNAL P ROCESSING.
He received the 2000 IEEE W. R. G. Baker Prize Paper Award, the 2006 IEEE
Communications Society Stephen O. Rice Prize in the field of communication
theory, the 2006 and 2010 IEEE Signal Processing Society’s Best Paper
Awards, and the 2017 IEEE Signal Processing Society Donald G. Fink
Overview Paper Award.

Ris + Uav

Uploaded by

Copyright:

Available Formats

Ris + Uav

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ris + Uav

Uploaded by

Copyright:

Available Formats

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 20, NO.

9, SEPTEMBER 2021 5769

UAV-Assisted Intelligent Reflecting Surface

mobility constraints are given below:

heading as it moves. The horizontal coordinates of the BS is given by [32]

⎜ P h3 [n] + h2,k [n] Θk [n] h1,k [n] ⎟

σ2 from [44], the decision criteria is if

ŷr,k [n, n1 , l] = w [n, n1 , l]

and linear optimization problem, it has very low computational

Unfortunately, the low-complexity algorithm based on the

Algorithm 2 Proposed Penalty-Based Algorithm for Solving

where ξ sets a predefined level of accuracy. The detailed

Fig. 9. Convergence behaviour of Algorithm 2.

Fig. 12. Optimized IRS scheduling for T = 40s.

the number of IRS reflecting elements increases, since more

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Ris + Uav

Uploaded by

Copyright:

Available Formats

Ris + Uav

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ris + Uav

Uploaded by

Copyright:

Available Formats

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 20, NO.

9, SEPTEMBER 2021 5769

UAV-Assisted Intelligent Reflecting Surface

mobility constraints are given below:

heading as it moves. The horizontal coordinates of the BS is given by [32]

⎜ P h3 [n] + h2,k [n] Θk [n] h1,k [n] ⎟

σ2 from [44], the decision criteria is if

ŷr,k [n, n1 , l] = w [n, n1 , l]

and linear optimization problem, it has very low computational

Unfortunately, the low-complexity algorithm based on the

Algorithm 2 Proposed Penalty-Based Algorithm for Solving

where ξ sets a predefined level of accuracy. The detailed

Fig. 9. Convergence behaviour of Algorithm 2.

Fig. 12. Optimized IRS scheduling for T = 40s.

the number of IRS reflecting elements increases, since more

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

⎜ P h3 [n] + h2,k [n] Θk [n] h1,k [n] ⎟