warwick.ac.uk/lib-publications
Manuscript version: Author’s Accepted Manuscript
The version presented in WRAP is the author’s accepted manuscript and may differ from the
published version or Version of Record.
Persistent WRAP URL:
http://wrap.warwick.ac.uk/109017
How to cite:
Please refer to published version for the most recent bibliographic citation information.
If a published version is known of, the repository item page linked to above, will contain
details on accessing it.
Copyright and reuse:
The Warwick Research Archive Portal (WRAP) makes this work by researchers of the
University of Warwick available open access under the following conditions.
Copyright © and all moral rights to the version of the paper presented here belong to the
individual author(s) and/or other copyright owners. To the extent reasonable and
practicable the material made available in WRAP has been checked for eligibility before
being made available.
Copies of full items can be used for personal research or study, educational, or not-for-profit
purposes without prior permission or charge. Provided that the authors, title and full
bibliographic details are credited, a hyperlink and/or URL is given for the original metadata
page and the content is not changed in any way.
Publisher’s statement:
Please refer to the repository item page, publisher’s statement section, for further
information.
For more information, please contact the WRAP Team at: wrap@warwick.ac.uk.
1
UAV-Aided MIMO Communications for
5G Internet of Things
Wei Feng, Member, IEEE, Jingchao Wang, Yunfei Chen, Senior Member, IEEE, Xuanxuan Wang,
Ning Ge, Member, IEEE, Jianhua Lu, Fellow, IEEE
Abstract—The unmanned aerial vehicle (UAV) is a promising
enabler of the Internet of Things (IoT) vision, due to its agile
maneuverability. In this paper, we explore the potential gain
of UAV-aided data collection in a generalized IoT scenario.
Particularly, a composite channel model including both the large-
scale and the small-scale channel fading is used to depict typical
propagation environments. Besides, not only the peak transmit
power constraint but also the budget power and the total energy
constraints are considered to characterize practical IoT devices. A
multi-antenna UAV is employed, which follows a circular trajec-
tory and visits the IoT devices in a sequential manner. For each
transmission, the UAV communicates with a cluster of single-
antenna IoT devices to form a virtual MIMO link. On these basis,
we formulate a whole-trajectory-oriented optimization problem,
where the transmission duration time and the transmit power
of all devices are jointly designed for the whole flight. Different
from previous studies, only the slowly-varying large-scale channel
state information (CSI) is assumed available, to coincide with the
fact that practically it is quite difficult to predictively acquire
the random small-scale channel fading prior to the UAV flight.
We propose an iterative scheme by leveraging the maxmin
optimization and convex optimization tools, to overcome the non-
convexity of the problem. The presented scheme adapts well to
the rigorous energy constraints as well as the large-scale CSI
condition, thus, it can provide a significant performance gain
over traditional schemes and converges quickly.
Index Terms—Internet of Things (IoT), unmanned aerial ve-
hicle (UAV), energy constraint, MIMO, large-scale channel state
information (CSI).
I. INTRODUCTION
Nowadays, the unmanned aerial vehicle (UAV) has been
widely investigated in conjunction with wireless communi-
cation networks [1], [2], [3], [4], as it can provide an on-
demand flexible platform for deploying aerial base stations,
or mounting mobile access points (APs) [5]. Particularly, the
UAV has been recognized as one of the main key enablers
of the Internet of Things (IoT) vision, thanks to its agile
usability [6].
Different from the other applications of the ongoing Fifth
Generation (5G) system, IoT devices are usually energy-
limited [7], and they may be randomly scattered over a wide
W. Feng, X. Wang, N. Ge, and J. Lu are with the Beijing National
Research Center for Information Science and Technology, Department of
Electronic Engineering, Tsinghua University, Beijing 100084, China (e-
mail: fengwei@tsinghua.edu.cn, wangxuanxuan@mail.tsinghua.edu.cn, gen-
ing@tsinghua.edu.cn, lhh-dee@tsinghua.edu.cn).
J. Wang is with the China Electronic Equipment System Engineering
Company, Beijing 100141, China (e-mail: wangjc61s@163.com).
Y. Chen is with the School of Engineering, University of Warwick,
Coventry, United Kingdom (e-mail: Yunfei.Chen@warwick.ac.uk).
area for environmental monitoring or other targets. These
issues will inevitably aggravate the coverage problem for IoT
communications. By leveraging the UAV’s agile mobility and
maneuverability [8], it is a promising way to sequentially visit
the IoT devices and move sufficiently close to the devices to
collect the sensing data from them. Thus, the UAV-aided IoT
communications may solve the coverage problem and signifi-
cantly reduce the overhead of IoT communication networks. In
addition, UAVs are cost-effective, which makes them suitable
for emergency on-demand missions in IoT applications [9].
A. Related Work
Basically, the UAV can be either static at a fix location
or mobile along a predefined/dynamically adjusted trajectory,
providing static and mobile aerial APs, respectively. In the
literature, many studies focused on the placement optimization
problem of UAVs, which is an important issue for IoT applica-
tions. Particularly in [10], the authors presented an efficient an-
alytical approach to optimize the altitude of UAV, which may
improve the wireless coverage of aerial APs. Extending to the
three-dimensional scenario, the authors of [11] formulated a
UAV placement problem to maximize the revenue of the whole
network, and proposed a computationally efficient numerical
solution. In [12], the optimum placement of a relaying UAV for
maximum reliability was studied, where the total power loss,
the overall outage and the overall bit error rate were derived,
and the optimum altitude was investigated for both static and
mobile UAVs. The authors of [13] considered the problem
of minimizing the number of UAVs while guaranteeing the
wireless coverage performance, so as to further reduce the
system cost.
These results have shown insightful results with respect to
UAV’s placement. However, the transmission strategies they
assumed are all simple, which can not be directly applied to
some severe IoT scenarios. Generally for IoT applications, the
transmission strategy should be rigorously designed, so as to
efficiently exploit the potential ability of energy-limited and
randomly-scattered IoT devices.
Some researchers have also identified the difference in
terms of transmission strategy optimization between UAV
communications and traditional terrestrial communications.
In [14], the authors studied a mobile relaying technique with
high-mobility UAV relays. The end-to-end throughput was
maximized via optimizing both the relay’s trajectory as well as
the source/relay’s power allocation. Furthermore, the authors
of [15] investigated the energy-efficient UAV communications
2
via trajectory optimization, where the propulsion energy con-
sumption of UAVs was particularly considered. Therein, a
theoretical model on the UAV’s propulsion energy consump-
tion was derived, based on which the energy efficiency of
UAV communications was defined and then optimized. In [16],
the authors maximized the approximate ergodic sum rate via
dynamically adjusting the UAV heading. In [17], the minimum
throughput over all ground users in the downlink UAV com-
munication was maximized to achieve fair performance among
users. These studies have uncovered useful researching points
for the optimization of UAV communication strategies.
B. Main Contribution
Despite of the aforementioned fruitful results, utilizing UAV
in an IoT scenario still faces some open challenges with
respect to energy constraints of IoT devices as well as the
channel model and the priori knowledge of channel state.
For IoT devices, not only the peak transmit power con-
straint, as considered in the previous studies, but also the total
energy constraint, should be taken into account when opti-
mizing the communication strategy, because for most cases,
it is difficult to recharge the battery equipped at IoT devices.
Besides, in IoT applications, the sensing data volume may
dynamically change. Accordingly, the communication strategy
should adapt to the data sensing status. In practice, a budget
power constraint (more sensing data more budget power)
can be introduced to simply link the communication strategy
optimization and data sensing condition, which however is still
untouched in the literature.
In order to depict a typical propagation environment for
IoT applications, a practical channel model should be con-
sidered. In general, both the large-scale and the small-scale
channel fading should be taken into account [18], [19]. As a
contrast, quite a number of previous studies have assumed the
free-space path-loss model to simplify mathematical analysis.
Under the composite channel model, the priori knowledge
for optimizing communication strategy should be carefully
considered. Different from the free-space model, it becomes
not feasible to assume perfect channel state information (CSI)
in this case, because practically it is quite difficult to pre-
dictively acquire the random small-scale channel fading prior
to the UAV flight. Towards this end, we can optimize the
communication strategy in a whole-trajectory-oriented manner
on the basis of only the slowly-varying large-scale CSI.
However, to the best of the authors’ knowledge, it is still an
open issue to consider the composite channel model for UAV
communications and perform optimization using the large-
scale CSI only.
In this paper, we optimize the data collection efficiency of a
UAV-aided IoT communication system, subject to a series of
practical energy constraints as well as practical channel model
and the large-scale CSI condition. The main contribution of
this paper is summarized as follows.
• We consider a multi-antenna UAV, which follows a cir-
cular trajectory and visits the IoT devices in a sequential
manner. For each transmission, the UAV communicates
with a cluster of IoT devices. Thus, a virtual MIMO trans-
mission link is formed, to adapt to the low transmit power
of IoT devices and enable multiple access within a time-
frequency resource block. Both analysis and simulation
results show that this operation is promising to improve
the efficiency of sensing data collection.
• We consider a practical channel model consisting of
both the large-scale and the small-scale channel fading,
to characterize various propagation environments of IoT
applications. In this case, the full CSI assumption, as
used in most previous studies, becomes not applicable,
as the fast-varying small-scale channel fading is difficult
to obtain perfectly prior to the UAV flight. To this end,
we assume that only the large-scale CSI is known. This
is viable because the large-scale channel fading highly
depends on the relative position of UAV and IoT devices,
which can be calculated according to the historical chan-
nel sounding data together with the positional information
of both UAV and IoT devices.
• We consider the peak transmit power, the budget power
and the total energy constraints for IoT devices. Fur-
ther based on the large-scale CSI condition, a whole-
trajectory-oriented optimization problem is formulated to
maximize the data collection efficiency by optimizing
the transmission duration time of all devices, as well
as their transmit power. Notably, in this work all the
communication parameters are designed in a predefined
manner, i.e., prior to the UAV flight. By adopting the
random matrix theory, the maxmin optimization theory,
as well as the convex optimization tools, we propose
an efficient algorithm to solve the problem. Based on
that, we also discuss the impact of UAV’s deploying
parameters on system performance.
C. Organization and Notation
The rest of the paper is organized as follows. Section II
introduces the system model. The data collection efficiency of
UAV-aided MIMO communications is analyzed and optimized
in Section III. Section IV shows the simulation results and
discussions, and finally, concluding remarks are given in
Section V.
Throughout this paper, lower case and upper case boldface
symbols denote vectors and matrices, respectively. I
N
repre-
sents an N × N identity matrix, and C
M×N
represents the
collection of all M × N complex matrices, and CN (0, σ
2
)
denotes the complex Gaussian distribution with zero mean
and σ
2
variance. (·)
H
and (·)
T
represent the transpose conju-
gate and the transpose, respectively. E
x
(·) is the expectation
operator with respect to x. det(·) represents the determinant
operator.
II. SYSTEM MODEL
As shown in Fig. 1, we consider a UAV-aided IoT communi-
cation system consisting of K IoT devices and an on-demand
dispatched UAV. Due to the limited size of sensing devices,
at each only a single antenna is equipped. The UAV, acting
as an on-demand aerial AP, is equipped with M antennas, so
as to promote the efficiency of data collection. Without loss
of generality, we consider a circular coverage area centered at
3
x
z
Cloud
y
IoTdevice
Clusteringofdevices
toenableMIMO
communications
MultiͲantennaUAV
in
r
in
T
in
d
U
h
U
r
Fig. 1. Illustration of a UAV-aided IoT communication scenario, where the
multi-antenna UAV servers as an on-demand aerial AP to communicate in
a MIMO fashion with a cluster of IoT devices, and all the sensing data is
collected, and forward to the cloud for further use.
(0, 0, 0), with a radius of r
c
. All the devices are randomly
deployed. The UAV flies right above the coverage area, at an
altitude of h
U
, following a circular trajectory of radius r
U
centered at (0, 0, h
U
), with a duration time of T . In practice
these parameters, i.e., h
U
, r
U
, T, are coupled with each other,
influencing the total system performance, which should be
carefully designed.
During the UAV flight, all the IoT devices will connect to
the aerial AP when scheduled for data reporting
1
. We divide
all the devices into orthogonal N clusters, each containing M
adjacent devices, as shown in Fig. 1. In practice, these device
clusters could be predefined for ease of management. At the
cost of extra overhead, they can also be adjusted adaptively to
fit with the UAV’s dynamic position. For simplicity, we assume
K = MN. Denote the polar coordinates of device i in cluster
n as (r
in
, θ
in
). The corresponding rectangular coordinates
would be (r
in
cos(θ
in
), r
in
sin(θ
in
), 0). In this work, these N
device clusters are scheduling in a round robin way under a
time-division multiple access (TDMA) regime for fair access
2
.
Particularly, if cluster n is scheduled, the corresponding M
devices will simultaneously transmit data to the UAV, forming
an M × M MIMO communication link. To be agile, the
transmission duration time τ
n
for cluster n is adjustable
3
,
while satisfying
N
i=n
τ
n
≤ T. (1)
As the total energy for data transmission for each device
is quite limited, we assume both the peak transmit power
1
The scheduling issue is also quite important, which however is out of
the scope of this paper.
2
In practice, when the number of IoT devices is much larger than UAV,
non-orthogonal multiple access can be used [20], [21], which however would
increase the processing complexity at the UAV.
3
This is feasible in practice, where the velocity of the UAV can be
dynamically controlled to fit this requirement.
constraint and the total energy constraint as
0 ≤ p
in
≤ P
max
, ∀i, n, (2)
p
in
τ
n
≤ E
max
, ∀i, n, (3)
where p
in
is the transmit power of device i in cluster n. Note
that the transmission duration time is usually short in practice
due to the energy constraint. Therefore, we assume that p
in
maintains a constant during τ
i
.
As discussed above, in IoT applications, the sensing data
volume may dynamically change. Therefore, we introduce the
budget power constraint P
budget
to simply represent the current
data sensing status. If there is only a little data to be reported,
P
budget
is set small for a longer life. Otherwise, P
budget
should
be set large enough for faster data reporting. We set
M
i=1
p
in
≤ P
budget
, ∀n. (4)
Different from previous studies, we consider a more prac-
tical channel model for all device-UAV links, by taking into
account both the large-scale channel fading and the small-
scale channel fading. As τ
i
is short, we assume the opening
angle of UAV with respect to the coordinate axis x is φ
n
for
cluster n when it is scheduled. Then the coordinates of UAV
would be (r
U
cos(φ
n
), r
U
sin(φ
n
), h
U
). Accordingly, we have
the distance between device i in cluster n and the mobile UAV
as
d
in
=
h
2
U
+ r
2
in
+ r
2
U
− 2r
in
r
U
cos
φ
n
− θ
in
. (5)
We consider both the line-of-sight (LOS) and non-line-of-sight
(NLOS) channel elements, the large-scale path loss between
device i in cluster n and the mobile UAV can be modeled
as [10], [12]
L
dB
in
=
A
1 + ae
−b(ρ
in
−a)
+ B
in
, (6)
where
A = η
LOS
− η
NLOS
, (7)
B
in
= 20log
10
(d
in
) + 20log
10
(
4πf
c
) + η
NLOS
, (8)
ρ
in
=
180
π
arcsin(
h
U
d
in
), (9)
and f is the carrier frequency, c is the speed of light. η
LOS
,
η
NLOS
, a and b are constants related to the propagation
environments [10], [12]. Consequently, the large-scale channel
fading between device i in cluster n and the mobile UAV is
derived as
L
in
= 10
−
L
dB
in
10
. (10)
Taking the Rayleigh small-scale channel fading into account,
one may obtain the channel coefficient between device i in
cluster n and the mobile UAV as
h
in
= L
1/2
in
s
in
, (11)
where s
in
∈ C
M×1
, the entries of which are independent and
identically distributed (i.i.d.) variables according to CN (0, 1).
Here, we use the Rayleigh fading to present the severe
4
propagation environment of IoT applications. For a more
comprehensive study, the Rician fading [22], [23], as well as
the Nakagami fading model [12] can be adopted.
Denoting the transmit signal of device i in cluster n by x
in
,
we have
E{|x
in
|
2
} = p
in
. (12)
Based on the previous definitions, the received signal vector
at the UAV from the device cluster n can be expressed as
y
n
= H
n
x
n
+ n
n
, (13)
where
H
n
= [h
1n
, h
2n
, ..., h
Mn
], (14)
x
n
= [x
1n
, x
2n
, ..., x
Mn
]
T
, (15)
and n
n
represents the additive white Gaussian noise with
independent entries distributed according to CN (0, σ
2
).
For brevity, we set
H
n
= S
n
L
n
, (16)
where S
n
and L
n
represent the small-scale and the large-scale
channel coefficients, respectively, and
S
n
= [s
1n
, s
2n
, ..., s
Mn
], (17)
L
n
=
L
1/2
1n
.
.
.
L
1/2
Mn
. (18)
As S
n
varies fast and is difficult to obtain in practice,
we maximize the data collection efficiency of the considered
system using only the slowly-varying large-scale CSI. In the
following, both the transmission duration time and the transmit
power of a ll devices for the whole flight will be optimized
under realistic constraints of IoT applications.
III. OPTIMIZED UAV-AIDED MIMO COMMUNICATIONS
In this part, we first formulate a whole-trajectory-oriented
optimization problem to maximize the data collection effi-
ciency of UAV-aided MIMO communications. Then, we solve
the non-convex problem in an iterative way, leading to an
efficient resource allocation algorithm. The convergence of the
algorithm is theoretically proved.
A. Problem Formulation
We maximize the data collection efficiency of the con-
sidered IoT model, subject to the UAV flying duration time
constraint, the peak transmit power constraint and the total
energy constraint of each device, as well as the budget power
constraint of each device cluster. The problem is formulated
as
max
{τ
n
},{p
in
}
N
n=1
τ
n
R
n
(19a)
s.t.
N
n=1
τ
n
≤ T, (19b)
τ
n
≥ 0, ∀i, (19c)
0 ≤ p
in
≤ P
max
, ∀i, ∀n, (19d)
p
in
τ
n
≤ E
max
, ∀i, ∀n, (19e)
M
i=1
p
in
≤ P
budget
, ∀n, (19f)
where R
n
denotes the achievable rate of device cluster n when
it is scheduled. Since the small-scale CSI is unknown, R
n
is
calculated by taking expectation with respect to S
n
as
R
n
=
E
{S
n
}
log
2
det(I
M
+
1
σ
2
(S
n
L
n
)P
n
(S
n
L
n
)
H
)
, (20)
where
P
n
=
p
1n
.
.
.
p
Mn
. (21)
This problem is difficult to solve. On one hand, the objective
function is implicit due to the expectation operator. On the
other hand, the transmission duration time and transmit power
are coupled with each other, rendering the problem non-
convex, due to the constraints in (19e) [24].
B. Iterative Solution
The key difficulty of solving the problem in (19) lies in the
coupling relationship between {τ
n
} and {p
in
}. Accordingly,
we solve the problem in an iterative way, inspired by the idea
of divider and conquer.
Let s ≥ 1 denote the iterative step number. If {τ
s−1
n
} is a
feasible point of the problem, given {τ
s−1
n
}, we can simplify
the problem as N subproblems as
max
{p
in
}
R
n
(22a)
0 ≤ p
in
≤ P
max
, ∀i, (22b)
p
in
τ
s−1
n
≤ E
max
, ∀i, (22c)
M
i=1
p
in
≤ P
budget
, ∀n, (22d)
n = 1, 2, ...N. (22e)
Each subproblem is much simplified. However, it is still
challenging due to the complicated objective function.
By adopting the random matrix theory [25], we derive an
approximation for R
n
as
R
n
≈
¯
R
n
=
M
i=1
log
2
(1 +
1
σ
2
L
in
p
in
ϖ
−1
M)
+M log
2
(ϖ
n
) − M log
2
e
1 − ϖ
−1
n
, (23)
