SPECIAL SECTION ON ADVANCED COMMUNICATIONS AND NETWORKING TECHNIQUES

FOR WIRELESS CONNECTED INTELLIGENT ROBOT SWARMS

Received December 3, 2019, accepted December 16, 2019, date of publication December 24, 2019, date of current version January 3, 2020.
Digital Object Identifier 10.1109/ACCESS.2019.2961977

Energy Efficiency Characterization in

Heterogeneous IoT System With
UAV Swarms Based on Wireless
Power Transfer
YUANYUAN YAO 1 , (Member, IEEE), ZHENGYU ZHU 2,3 , (Member, IEEE),
SAI HUANG 4 , (Member, IEEE), XINWEI YUE 1 , (Member, IEEE),
CHUNYU PAN 1 , (Member, IEEE), AND XUEHUA LI 1
1 Schoolof Information and Communication Engineering, Beijing Information Science and Technology University, Beijing 100101, China
2 Schoolof Information Engineering, Zhengzhou University, Zhengzhou 450001, China
3 Zhengzhou University Research Institute of Industrial Technology Company, Ltd., Zhengzhou 450001, China
4 Key Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunications, Beijing 100876, China

Corresponding authors: Sai Huang (huangsai@bupt.edu.cn) and Zhengyu Zhu (zhuzhengyu6@gmail.com)

This work was supported in part by the Natural Science Foundation of Beijing Municipality under Grant 19L2022, Grant 4202046,
Grant 4204099, and Grant KZ201911232046, in part by the National Natural Science Foundation of China under Grant 61801052 and
Grant 61801434, in part by the Science and technology Project of Beijing Municipal Education Commission under Grant
KM202011232002 and KM202011232003, in part by the Key Research and Cultivation Project at Beijing Information Science and
Technology University under Grant 5211910926 and Grant 5211910924, in part by the Supplementary and Supportive Project for Teachers
at Beijing Information Science and Technology University under Grant 5029011103 and Grant 5111911147, in part by the Zhengzhou
Municipal Science Technology Innovation Project under Grant 2019CXZX0037, in part by the China Postdoctoral Science Foundation
under Grant 2018M642784, in part by the Scientific and Technological Key Project of Henan Province under Grant 192102310178, and in
part by the Key Laboratory of Dynamic Cognitive System of Electromagnetic Spectrum Space (Nanjing Univ. Aeronaut. Astronaut.),
Ministry of Industry and Information Technology under Grant KF20181901.

ABSTRACT An unmanned aerial vehicle (UAV) swarm together with a large-scale heterogeneous Internet
of Things (IoT) network consisting of macrocells and energy-constrained IoT transmitters (IoT-Ts) is
investigated. The UAVs are utilized as flying robot swarms that intelligently transfer energy to the energy-
constrained IoT-Ts on the ground. Each IoT-T has an associated IoT device (IoT-D) that is placed at a fixed
distance in a random direction. The transmission probability of the energy-constrained IoT-Ts is derived
by considering one-slot charging and two-slot charging according to three dimensional (3D) locations,
respectively. The coverage probability of each type of IoT-D is investigated. The energy efficiency is derived
by considering the transmission power of the active IoT-Ts and the effect of the association biasing factor,
and the energy efficiency is also maximized by deploying the optimal density of IoT-Ts. Simulation results
are examined to validate the accuracy of our theoretical analysis. Results illustrate the insightful effects of
the network parameters, and the helpful guidelines for practical UAV swarms and IoT system design.

INDEX TERMS Unmanned aerial vehicle swarms, heterogeneous IoT networks, wireless power transfer,
stochastic geometry, energy efficiency.

I. INTRODUCTION Typically, IoT systems consist of small, energy-constrained

With the development of the fifth generation wireless com- nodes, and these nodes cannot continue working for a long
munication technology, the Internet of Things (IoT) system periods of time and at long distances [5]. In some special
has recently been integrated widely into the field of wire- scenarios, such as forest fire warning and pipeline inspection,
less communications [1]–[3]. The IoT aims to provide the the batteries of these IoT nodes cannot be manually replaced
interconnection of all objects anytime and anywhere [4]. or recharged; thus the system performance is limited due to
the energy constraints of the IoT device [6].
The associate editor coordinating the review of this manuscript and The development of unmanned aerial vehicle (UAV)
approving it for publication was Jinming Wen . swarms is recognized as an innovative approach to

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 967
 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

complement existing IoT networks [7]–[10]. The introduc- rescue. The UAV swarms employ a decentralized method
tion of UAV swarms has several advantages compared with to deploy the searching paths on the basis of the location
traditional wireless networks on the ground. First, they can information. Liu et al. in [27] propose that using the tools
be quickly deployed and configured [11], and transfer energy of stochastic geometry [28], [29], the UAV swarms can be
to the IoT nodes in some special scenarios on the ground; modeled as homogeneous Poisson point processes (HPPPs)
Second, they are highly controllable and have low-cost fea- in 3D space. However, the detailed derivation process is not
tures [12]; Moreover, with the high probability of enjoying a illustrated with stochastic geometry. Although some works
line-of-sight (LoS) link, UAV swarms can have a relatively are related to the UAV swarms and ground wireless net-
good channel state to complete wireless power transfer [13]. working, few studies considered the UAV swarm-assisted
Terrestrial IoT nodes can transmit/receive data by using wireless powered large-scale heterogeneous IoT network and
the harvested power, and the system can be named as a applied stochastic geometry to 3D analysis. Most related
UAV-assisted wireless powered heterogeneous IoT network. works ignore the impact of air-ground energy transmission
For terrestrial heterogeneous IoT networks, the dimension from UAV swarms to IoT nodes in the system.
of energy should be considered, from which the heteroge- This study considers a UAV-assisted heterogeneous IoT
neous IoT network faces certain challenges, such as sup- system consisting of a swarm of UAVs, a tier of macro base
porting high network throughput, high energy efficiency, and stations (MBSs), and multi-tier of heterogeneous IoT sys-
good quality-of-service requirement. The macrocells together tems. The UAVs are utilized as flying robot swarms that intel-
with multi-tier IoT transmitters (IoT-Ts) are considered as ligently transfer energy to the energy-constrained IoT-Ts on
a key technology to overcome these challenges for future the ground, when the energy-constrained IoT-Ts are located
wireless networks. IoT-Ts have small coverage area and inside an energy harvesting zone. Each energy-constrained
require low transmission power, making them suitable for IoT-T has an associated IoT-D located at a specific distance
deployment in the special areas [6]. Therefore, the density of in a random direction. The two types of IoT-D in the system
the IoT-Ts and the transmission power should be considered are as follows: 1) the associated IoT-Ds that the associated
to improve the energy efficiency of the heterogeneous IoT IoT-Ts are active (have a communication request) and 2) the
system. associated IoT-Ds that have a communication request, and
the associated IoT-Ts are inactive, together with the ordi-
A. RELATED WORK nary IoT-Ds requiring communications. The transmission
Wireless power transfer (WPT) recently received consider- probability of the energy-constrained IoT-Ts is derived by
able attention for recharging devices because of the large- considering one-slot and two-slot charging, respectively. The
scale wireless IoT nodes in the IoT system. It enables the coverage probabilities of every type of IoT-Ds are analyzed
IoT nodes to collect energy from the radio-frequency (RF) of and derived, the network throughput and energy efficiency
the surrounding transmitters, which is beneficial to prolong are both characterized by considering the effect of the IoT-T
the period of manual battery replacement, and effectively density, the association biasing factor, and the transmission
saves energy cost [14]–[17]. The feasibility of WPT and power of the active IoT-Ts. The numerical results and sim-
the wireless powered communication network (WPCN) have ulations are examined to provide insightful guideline on the
been investigated in the recent works [18]–[20]. However, various system parameters for the design in practice.
the number of IoT sensor nodes is large in some special The main contributions of this work are summarized as
scenarios, such as forest fire warning and pipeline inspection. follows:
These nodes are always distributed in a wide area, making it 1) We propose a UAV-assisted heterogeneous IoT system;
difficult to deploy a swarm of power beacons for transferring The UAV swarms intelligently transfer energy to the
energy to the energy-constrained IoT nodes. Thus, charging energy-constrained IoT-Ts on the ground, and evaluate
the batteries of the IoT nodes remains an urgent problem. The the transmission probability of the energy-constrained
UAV with energy harvesting can be viewed as a key solution IoT-Ts with different cases;
to overcome these problems [21]. 2) We consider the impact of the density of the UAV
The research on UAV-enabled wireless powered hetero- swarms and the transmission power of the IoT-Ts on
geneous IoT network is still in the preliminary stage, and the activity of the IoT-Ts in each tier;
most existing works focus on studying a single UAV that 3) We analyze the energy efficiency by considering the
transmits energy to terrestrial IoT-Ds [22]–[24]. Huang et al. effect of the IoT-T density and the association biasing
in [6] investigate a joint optimization of power allocation and factor. The energy efficiency is also maximized by
trajectory design of a single UAV to provide infrastructure- deploying the optimal density of IoT-Ts.
less IoT services. The objective is to maximize the minimum The remainder of this paper is organized as follows.
energy harvested by the large-scale terrestrial distributed IoT Section II describes the system model and performance met-
nodes. Xu et al. in [25] investigate a UAV transferring energy rics. The transmission probability of the IoT-Ts is investigated
to the terrestrial devices. in Section III. Coverage probability of the two types of users
In [26], a novel method of swarm robot task planning based are presented in Section IV. Section V studies the network
on UAV and ground coordination is proposed for emergency throughput and the energy efficiency. Numerical results and

968 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

FIGURE 1. An illustration of heterogeneous IoT network. FIGURE 2. Harvesting zone of the UAV.

The ordinary IoT-D are geographically situated accord-

simulations are conducted in Section VI. Finally, Section VII ing to a HPPP with density µo . Hence, the users in the
concludes the paper. heterogeneous IoT network can be classified in two types:
Notations: Throughout this paper, we use E [·] to denote the associated IoT-Ds are preferentially served by their own
the expectation operator. Pr () stands for the abbreviation for IoT-Ts; the ordinary IoT-Ds can be served by any MBSs or
probability. active IoT-Ts. All the IoT-Ds have a communication request
with probability ωreq .
II. SYSTEM MODEL Each UAV in the swarms is assumed to be associated with
A. NETWORK MODEL a 3D cone which is named as harvesting zone to delivers RF
Let us consider a UAV-assisted downlink heterogeneous IoT energy to the energy-constrained IoT-T, as shown in Fig. 2,
system, as shown in Fig. 1, which is composed of a swarm hmax is defined as the busbar which is the maximum dis-
of UAVs and a large-scale HetNet that consists of a tier tance of the harvesting zone from the UAV to the energy-
of MBSs, denoted by tier 0, along with K tiers of energy- constrained IoT-T. The dotted line on the ground is the bottom
constrained IoT-Ts, denoted by K= {1 · · · K }. The UAVs of the harvesting zone which is a disk with radius rh , where
are utilized as flying robot swarms that intelligently trans- h2max = h2uav + rh2 , and the five-pointed star representing
fer energy to the energy-constrained IoT-Ts on the ground. the UAV is the vertical mapping from the air to the ground.
Moreover, the UAVs are distributed according to a HPPP 8u Therefore, the horizontal distance between the i-th UAV to
with density µu , and have a fixed transmission power Pu and the j-th IoT-T in tier k on the horizontal plane, is equivalent to
a single antenna. It is noted that we assume the UAVs are the distance between the five-pointed star to the IoT-T, which
deployed at a fixed altitude huav for avoiding the frequent is denoted by ri,kj in Fig. 2. For the simplicity of theoretical
ascending or descending caused by diverse terrains. analysis, we refer to ‘‘cone bottom‘‘ as ‘‘harvesting zone’’ in
On the other hand, All the MBSs and the energy- the sequel.
constrained IoT-Ts in the same tier are assumed to deployed
according to independent HPPPs 8m and 8k , with density B. AIR TO GROUND CHANNEL MODEL
µm and µk , k ∈ K, (µm  µk ). Each energy-constrained As discussed in [30], the energy-constrained IoT-Ts on the
IoT-T in tier k supports its transmission solely with RF energy ground receive two groups of signals experiencing the line-of-
harvesting from the UAVs, and has its associated IoT-D to sight (LoS) and non-line of sight (NLoS). Due to the different
serve primarily, each associated IoT-D is located at a distance occurrence probabilities of LoS or NLoS links, the channel
of da,k away in a random direction, and the associated IoT-Ds power gain from the i-th UAV to the j-th IoT-T in tier k is
in tier k follow a HPPP with density µk . Time is partitioned written as
into slots with unit duration. In each time slot, an energy- ( −α
h j LoS,
constrained IoT-T in tier k becomes active only when 1) its gi,kj = i,k−α (1)
associated IoT-D has a communication request; 2) the battery χhi,kj NLoS,
of the energy-constrained IoT-T is fully charged by the flying
where
UAVs. We further assume that MBSs consume power from q
the grid, and use the same power Pm for data transmissions. hi,kj = xu,i − xkj

2
+ yu,i − ykj

2
+ h2uav . (2)
When the batteries are fully charged, the energy-constrained
IoT-Ts transmits signal with the maximum power, and the Each UAV
i ∈ 8u has a 3D coordinate, that is
maximum transmit power in the tier k is Pk . Moreover, it is xu,i , yu,i , huav . Each energy-constrained IoT-T j ∈ 8k has
a 2D coordinate on the horizontal plane, that is xkj , ykj .


noted that Pm  Pk .

VOLUME 8, 2020 969

 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

Therefore, the distance between the i-th UAV to the j-th As shown in Fig. 2, the distance between an energy-
IoT-T in tier q k on the
horizontal plane can be given by constrained
q IoT-T and its nearest UAV is denoted as hi,kj =
2
2
2 2
ri,kj = xu,i − xkj + yu,i − ykj , which is shown ri,kj + huav , where ri,kj ≤ rh . Therefore, the average
in Fig. 2. χ is the additional loss to the free space propagation energy harvested by an energy-constrained IoT-T in a time
loss which is caused by the NLoS link, and α is the air-to- slot is denoted as
ground path loss exponent.
It is worth highlighting that the probability of receiving Pslot = ηPu gi,kj
 
LoS and NLoS groups is vastly higher than that of multi- = ηPu Pr (LoS, θ)×h−α
i,kj +Pr (NLoS, θ)×χhi,kj , (7)
−α
path fading [5]. Therefore, the impact of multipath fading
is neglected. The probability of having LoS connections is where η (0 < η < 1) is the harvesting efficiency.
dependent on the propagation environment and the elevation At each time slot, an energy-constrained IoT-T becomes
angle θ , i.e., the sight of the
! IoT-T to the UAV, which is given active only when 1) its associated IoT-D has a communication
request, the probability is denoted as ωreq ; 2) the battery of
by θi = tan−1 q huav
2 +h2
. the energy-constrained IoT-T is fully charged by the flying
ri,k uav
UAVs, the probability is denoted as ωt . As the point process
j
The probability of the LoS connections with an elevation
angle θi can be approximated to a simple modified Sigmoid of the UAVs 8u changed independently over different time
function (S-curve), which is given by slots, the events that an energy-constrained IoT-T has been
fully charged in one slot, and that its associated IoT-D has
1
Pr (LoS, θi ) = , (3) a communication request in next time slot are independent.
1 + ς exp (−β [θi − ς]) Consequently, the active probability of the IoT-Ts denoted by
where ς and β are S-curve parameters and related to the ωf is given by
propagation environment (woods, rural, dense urban, etc.).
ωt = ωreq · ωf . (8)
Accordingly, the NLoS probability are linked as
The calculation of ωf will be derived in Section III.
Pr (LoS, θi ) = 1 − Pr (LoS, θi ) . (4)
Similar to the assumptions in [32], we assume that the
Finally, the channel power gain from the i-th UAV to the MBSs and the active IoT-Ts in tier k, k ∈ (0 ∪ K), are
j-th IoT-T in tier k can be written as distributed according to a new HPPP 8 ek with density
q −α (
gi,kj = Pr (LoS, θi ) × ri,kj 2 + h2uav (a) ωt × µk , k ∈ K,
µk = (9)
q −α µm , k = 0.
+ Pr (NLoS, θi ) × χ ri,kj 2 + h2uav . (5)
D. CELL ASSOCIATION
C. ENERGY HARVESTING MODEL By the thinning property [33], the associated IoT-Ds that have
The RF energy harvester in each energy-constrained IoT- communication request in tier k follow a HPPP with density
T is deployed with a power converting module, which ωreq · µk ; and the ordinary IoT-Ds that have communication
transforms the energy of the RF signals received from the request follow a HPPP with density ωreq · µo ; respectively.
UAVs into direct current (DC) power. Such module in real- With the above analysis, as seen in Fig. 3, the associated
ity have certain sensitivity requirements, such as the har- IoT-Ds that have communication request are divided into two
vested power should be larger than a predefined threshold, categories: some of them are served by their own associated
so that the energy harvesting module can receive energy effi- IoT-Ts; the remaining IoT-Ds are served by other active
ciently. Thus, the harvesting zone is defined. As mentioned in IoT-Ts or MBSs since their associated IoT-Ts are inactive.
Section II-A and Fig. 2, we have defined the harvesting zone Therefore, the locations of the two type of the associated
as a disk with radius rh centered at each mapping node of the IoT-Ds that have communication request in tier k are sub-
ject to two HPPPs with density ωf · ωreq µk and 1 − ωf ·


UAV (the five-pointed star in Fig. 2).
Accordingly, an energy-constrained IoT-T can harvest RF ωreq µk . The associated IoT-Ds (have communication request)
energy from its nearest UAV supposing it is located in the that are served by their own IoT-Ts are referred to as pre-
UAV’s harvesting zone. Otherwise, the power harvested by an ferred associated IoT-Ds (PAIDs), the remaining IoT-Ds that
energy-constrained IoT-T located outside any harvesting zone have communication request are referred to as free access
is too small to trigger the energy harvesting module, and is IoT-Ds (FAIDs). All the FAIDs can be modeled by a HPPP
assumed to be neglected. Similar with [31], it is assumed that with density
the harvesting zones of different UAVs do not overlap most K
X
of the time. The probability that an energy-constrained IoT-T µo = 1 − ωf · ωreq µk + ωreq · µo .


(10)
located in a harvesting zone is denoted by ωh . Therefore,
e
k=1
we have
We consider a cell association scheme with a biasing factor
ωh = 1 − e−πrh µu .
2
(6) Ak (Ak > 0), where increasing Ak means the MBSs or the

970 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

with density,
(
(a) ωt × µk , k ∈ K,
µk = (15)
µm = µm 1 − snum0 (0) , k = 0.
e

e
In addition, if the association biasing factor Ak in tier k
FIGURE 3. Two types of IoT-Ds.
is larger than that in tier 0, more FAIDs will connect to the
(a)
IoT-Ts, thus resulting in a smaller e µ0 and allowing more
active IoT-Ts in tier k will effectively serve more FAIDs and MBSs to get into the sleep mode.
offload more data from other tiers, and is named by cell range
expansion [34]. When Ak = 1, the FAIDs are connected to E. PERFORMANCE METRICS
the MBSs or the active IoT-Ts in tier k with the largest receive In this paper, we utilize the following metrics to evaluate the
power. Following [35], we have the following Lemma. performance.
Lemma 1: The probability that a typical FAID connects to
a MBSs or a tier-k IoT-T is 1) COVERAGE PROBABILITY
The coverage probability is defined as the probability that a
(a) 2
µk (Pk Ak ) α randomly selected IoT-D that has communication request can
5k = , k ∈ (0 ∪ K) , (11)
K
P (a) 2 successfully receive data at its target rate. Specifically, given
µi (Pi Ai ) α
the total bandwidth W , the number of requesting IoT-Ds that
i=0
connected to a MBS or IoT-T N , the signal-to-interference
where Pk = Pm , when k = 0. ratio (SIR), and a corresponding target T , the coverage prob-
Let Xk denote the distance between the typical FAID and ability is given by
its serving MBS or IoT-T in tier k. The PDF of the distance 
W

Xk is given as follows [36] 9 = Pr log2 (1 + SIR) ≥ T . (16)
N
Lemma 2: The PDF sXk (x) of the distance Xk between the
FAID and its serving MBS or IoT-T in tier k is 2) ENERGY EFFICIENCY
 2
 The energy efficiency is the amount of data that can be
(a) K α
2π µk  X (a) Pi Ai
 
transmitted successfully to the IoT-Ds of all tiers with unit
sXk (x) = x exp −π µi x 2  . (12)

5k Pk Ak power, which is measured by bps/Joule. Let the total energy
i=0
consumption denote as Qc , the network throughput is denoted
Note that the total interference of a typical FAID received as Ct , the energy efficiency is given by
is from all IoT-Ts or MBSs in tier i which are located
µ·9 ·T
P
out ofa circle Ct
 centered at the typical FAID with radius
−α
Hc = = . (17)
Qc Qc
dj = x PPki Ai
Ak .
Denote the random variable numk as the number of FAIDs The symbols and notations utilized in the paper are listed
associated with a typical MBS or IoT-T in tier k, the proba- in Table 1.
bility mass function (pmf) of numk is given by [33],
III. TRANSMISSION PROBABILITY OF THE IoT-TS
snumk (n) From (8), the transmission probability of the IoT-Ts ωt
= Pr {numk = n} depends on ωf and ωreq . In this section, ωf will be derived
!n !−(n+3.5) using the Markov chain (MC) model. For convenience,
3.53.5 0 (n + 3.5) e
µo 5k µo 5k we assume that the active IoT-Ts will transmit all the stored
.
e
= (a)
3.5 + (a)
0 (3.5) n! µk µk energy in one time slot, and we focus the mapping node
(13) of a typical UAV on the ground at the origin, as shown
in Fig. 2, from (7), when the IoT-T is located at the edge of
It is worth highlighting if any FAID is connected to a MBS, a harvesting zone, the harvested power from the typical UAV
then the MBS is active, i. e., the MBS transmits signal and is minimized, which can be expressed as
consumes energy, which is given by 1 − snum0 (0). Accord- q −α
ingly, the transmitting MBSs form a new HPPP with density Pslot = ηPu ×
min 2 2
rh + huav × γ, (18)
µm = µm 1 − snum0 (0) , where


e
!−3.5 where γ = (Pr (LoS, θ) + χ Pr (NLoS, θ)), and Pminslot ≤ Pslot .
µo 5k Therefore, the battery of an energy-constrained IoT in tier k,
snum0 (0) = 1 + ,
e
(a)
k = 0. (14) k ∈ (0 ∪ K), can be fully charged during one time slot if
3.5µk
0 ≤ Pk ≤ Pmin slot , which is named as one-slot charging. In
the same way, if Pminslot < Pk ≤ 2 Pslot , the battery is fully
min
Based on the above analysis, the locations of the active
e(a)
MBSs and IoT-Ts in tier k can be re-modeled as a HPPP 8 charged within at most two time slots, this case is named as
k

VOLUME 8, 2020 971

 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

TABLE 1. Symbol notations.

FIGURE 4. The partitioned harvesting zone of two-slot charging.

B. TWO-SLOT CHARGING
slot < Pk ≤ 2 Pslot , the battery is fully charged within at
If Pmin min

most two time slots. As shown in Fig. 4, the harvesting zone

is divided into two regions, R1 and R1 , respectively. Let the
node Mu denote a typical UAV that mapping on the ground;
and Tk denote a typical energy harvesting IoT-T in tier k. R1
two-slot charging. Hence, the transmission probability of the is a circle centered at Mu with radius r1 , and R2 is an annulus
IoT-Ts ωt can be derived. However, if Pk > 2 Pmin slot , the case centered at Mu with radii 0 < r1 < rh .
is called multi-slot charging, which is out of the scope of It is assumed that when the IoT-T in tier k is located at the
this paper. Since most of the derivations in this part follow edge of the range R1 , it can just be exactly fully charged, thus
a similar derivation method with that in [31], we omit the we have
detailed proofs of the propositions in the sequel. q −α
Pk = ηPu × r12 + h2uav × γ, (21)
A. ONE-SLOT CHARGING
If 0 ≤ Pk ≤ Pmin
slot , the battery is fully charged within one time r1 can be derive from (21) as
slot. Thus the power condition of the IoT-Ts can be marked as
v
two grade {0, 1}, which are corresponding to the power level
u P − α2
u
0 and Pk , respectively. Thus, the state transition probability r1 =
t k
− h2uav . (22)
matrix denoted as M1 is given by ηPu γ

1 − ωh ωh
 
The average power harvested by Tk from the nearest UAV in
M1 = . (19)
ωreq 1 − ωreq one slot is shown in (7). If Tk is located in the region R1 ,
it can be fully charged; else if Tk is located inside the region
Therefore, ωt is obtained from the steady-state convergence R2 , it will be charged to a range [0.5Pk , Pk ).
theorem of MC, as given in the following proposition. Thus the power condition of the IoT-Ts can be marked
Proposition 1: If 0 ≤ Pk ≤ Pmin slot , the transmission proba- as three grade {0, 1, 2}, which are corresponding to the
bility of a typical IoT-T in tier k is power level [0, 0.5Pk ), [0.5Pk , Pk ) and Pk , respectively.
ωh Thus, the state transition probability matrix denoted as M2
ωt = ωf ωreq = ωreq is given by
ωh + ωreq
1 − ωh ωh − ω1 ω1
 
1 − e−πrh µu
2

= ωreq . (20) M2 =  0 1 − ωh ωh  (23)

1 − e−πrh µu +ωreq
2
ωreq 0 1 − ωreq .
From (20), it can be observed that the transmission prob-
ability of a typical IoT-Ts in tier k depends on µu , rh , and The probability of ω1 is derived similarly with (6) as
ωreq , but has no dependence on Pk . This is due to that once
ω1 = 1 − e−πr1 µu
2
an IoT-Ts located inside a harvesting zone, it can be ensured
to be fully charged within one time slot under the proposed
!
− 2
α

Pk
ωh −π −h2uav µu
condition 0 ≤ Pk ≤ Pminslot , with ωf = ωh +ωreq .
ηPu γ
= 1−e . (24)

972 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

   
slot < Pk ≤ 2 Pslot , the transmission
Proposition 2: If Pmin where κ (α) = π 0 1 + α2 0 1 − α2 , and α > 2 with
min

probability of a typical IoT-T in tier k is R∞

0(x) = 0 t x−1 e−t dt indicating the Gamma function, and
ωt = ωf ωreq snumk (n) in given in (13).
ωh Proof: See Appendix A. 
=   ωreq . (25)
ωh + ωreq 2ωhω−ωh
1

It is worth highlighting from (25) that, ωt is a decreasing B. COVERAGE PROBABILITY OF THE FAIDS
function of Pk as r1 is a decreasing function. The reason is that We consider the coverage probability of the FAIDs that are
increasing Pk resulting to a smaller region R1 , which means randomly located in the heterogeneous IoT system, which can
that most of the IoT-Ts need to be fully charged within two be either an associated IoT-D whose associated IoT-T is not
time slots, i,e., the probability ωf becomes smaller. Hence, active or an ordinary IoT-D, and are served by MBSs or active
the transmission probability of a typical IoT-T in tier k, ωt IoT-Ts. Since a typical FAID can associate with only one tier,
becomes smaller. the coverage probability that a typical FAIDs connects with
the k th tier is obtained as
IV. COVERAGE PROBABILITY
In this section, we derive the coverage probability of both the 9cov,k
FAID

PAIDs and FAIDs. Note that the PAIDs are served by their  
associated IoT-Ts, and FAIDs can be served by the MBSs Wk 
FAID

FAID
= Pr log2 1 + SIRk ≥ Tt
or the active IoT-Ts. Therefore, we compute the coverage 1{k∈ 0∪K} + numk
probability of the PAIDs and FAIDs separately. h   i
= E Pr SIRFAID
k ≥ 4 FAID
t (num k ) |num k , (29)
A. COVERAGE PROBABILITY OF THE PAIDS
In a large-scale heterogeneous IoT system, compared to (1{k∈ 0∪K } +numk )TtFAID
the aggregated interference, the thermal noise is always be where 4FAID
t (numk )
= 2 Wk − 1, Tt FAID is
neglected [34]. Therefore, we investigate the SIR by con- the target SIR. Different with (27), the total bandwidth is
sidering that all MBSs and the active IoT-Ts coexist on the divided by 1{k∈ 0∪K } + numk , for k = 0, it is numk , and
frequency spectrum. The received SIR of a typical PAIDs for k ∈ K, it is (numk + 1). SIRFAID k is the SIR at the
that located at x0 and served by its associated IoT-T in tier-k typical FAID (located at x00 ) at a distance x from the con-
located at yk,0 is written as nected MBS or IoT-T in tier k (located at y0k,0 ), which is
−α denoted as
Pk gx0 ,yk,0 da,k
SIRPAID = . (26)
k K −α Pk gx00 ,y0k,0 x −α
SIRFAID −α ,
P P
Pj gx0 ,yj,i yj,i − x0

k = (30)
K
e(a) \y

j=0 i∈8 P P 0 0
Pj gx0 0 ,yj,i yj,i − x0
k,0

j 0

From (16), the condition coverage probability of a PAIDs e(a) \y0

j=0 i∈8
j k,0
that is served by its associated IoT-T in tier-k is
  where gx00 ,y0k,0 is the channel power gain from the link x00 to
Wk  
9cov,k = Pr
PAID
log2 1 + SIRk PAID
≥ Tt PAID
y0k,0 . x is the distance between the node x00 and y0k,0 . Then We
1 + numk
h   i have the following Result 2 as follows.
= E Pr SIRPAIDk ≥ 4PAID
t (numk ) |numk , (27) Result 2: The coverage probability that a typical FAIDs
connects with the k th tier is,
(1+numk )TtPAID
where 4t
PAID (numk ) = 2 Wk −1, Tt PAID is the target
SIR, Wk is the total available bandwidth. It is worth noting 9cov,k
FAID

that Wk is divided by the associated PAID and some FAIDs  K

(a) 2
connected to the corresponding IoT-T in tier k. By assuming µi (Pi Ai ) α
P
∞ Z ∞
that the number of the IoT-Ds, numk , and SIRPAID are inde- X i=0
k = snumk (n)  2π x

2
pendent, the coverage probability can be obtained in Result 1  0 (Pk Ak ) α
n=1
as follows.
Result 1: The coverage probability of a PAIDs that is
 
served by its associated IoT-T in tier-k is K  2 
α

X (a) Pj Aj
∞

K
!2
α
×exp−π x 2 µj M 4FAID
t (numk ), α, 
(1+numk )TtPAID
e
X X (a) Pk Ak
9cov,k
PAID
= exp − µj 2
e Wk −1 j=0

n=0 j=0   
K 2
α

2 ! X (a) Pj Aj
α µj x 2  dx  ,

Pj × exp −π (31)
2
× da,k κ (α) snumk (n) , (28) Pk Ak
Pk j=0

VOLUME 8, 2020 973

 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

where energy efficiency expression in (35), we take a average load


Aj
 approximation, where the number of IoT-Ds served by a MBS
M 4FAID
t (num k ), α, or an active IoT-T, which is denoted as numk , is approximated
Ak

  2 −1
 by its mean value
A α
24FAID
t (numk ) Akj µ o 5k
,
e
= E [numk ] = (a)
(36)
 α−2 µk
e
Ak

2 2
×2 F1 1, 1 − ; 2 − ; −4t FAID
(numk ) , (32) and we assume that Pr (numk = E [numk ]) = 1. In this case,
α α Aj
the energy efficiency in Result 3 is approximated as (37),
and 2 F1 [., .; .; .] is the Gauss hypergeometric function [33]. 2
as shown at the bottom of the next page, where 4 eα =
P
Proof: See Appendix B.  ! !
µo 51 Tt µo 5k Tt
1+ 1{k∈ 0∪K } +
e e
(a) W1 2 (a) Wk
µ1 µk
V. ENERGY EFFICIENCY 2 e
− 1, and 4
eα
F,k = 2
e
− 1.
The network throughput is defined as the total amount of
the signal that is successfully received by the PAIDs and VI. NUMERICAL RESULTS
the FAIDs, from the expressions in (11), (28), and (31). The In this section, we evaluated some numerical results of the
network throughput can be obtained as transmission probability and the network throughput, and
K K give some interpretations based on our theoretical analysis.
Unless otherwise specified, we set the transmission power of
X X
Ct = Tt PAID ωt µk 9cov,k
PAID
+ Tt FAID 9cov,k
FAID
5k
k=1 k=0
the UAVs is Pu = 20 W, and the density is µu = 5 × 10−2
K
! nodes/m2 ; path-loss exponent α = 4, the energy harvesting
efficiency η = 0.8, the transmission power of the MBSs and
X
1 − ωf · ωreq µk + ωreq · µo . (33)


×
k=1 IoT-Ts is Pm = 43 dBm, and P1 = 20 dBm, respectively. The
target SIR Tt = 256 × 103 bits/sec, the distance between the
The total energy consumption of the network includes the
IoT-T and its associated IoT-D da = 5 m, the probability that
power transmitted by the active MBSs, and the power of the
the IoT-Ds require communications ωreq = 0.9, the density
active IoT-Ts which is transmitted by the UAVs. Note that
of the MBSs in tier 0 and the IoT-Ts in tier 1 is µm = 10−6
the energy for the operations during the sleep mode and the
nodes/m2 , and µ1 = 2 × 10−4 nodes/m2 , respectively. The
circuit power is not considered in this work due to that they
density of the ordinary IoT-Ds is µo = 5 × 10−4 nodes/m2 ,
are relatively small. Thus the total energy consumption can
and the association biasing factor in tier 0 and 1 are A0 = 1,
be denoted as
and A1 = 30, respectively. The height of the UAVs
Qc = µm 1 − snum0 (0) Pm + ωt · µ1 P1 . is huav = 4 m.


(34)
For different tier of the active IoT-Ds (PAIDs and FAIDs,
and k ∈ K), we assume the target rate thresholds are unified A. PERFORMANCE OF THE TRANSMISSION
as Tt . Using the definition of (17), we have the following PROBABILITY
Result. Fig. 5 shows the transmission probability of the IoT-Ts ωt
Result 3: The explicit solution of the energy efficiency is versus the transmission power P1 , for different values of the
given by (35), as shown at the bottom of this page. communication request probability of an IoT-D ωreq . It is
for the convenience of analysis, we consider a two-tier IoT observed that ωt is a fix value when 0 ≤ Pk ≤ Pmin slot , while
network on the ground with one tier of MBSs, k = 0, Pmin
slot < P k ≤ 2 P min , ω is increasing with the transmission
slot t
and one tier of energy-constrained IoT-Ts, k = 1. For the power P1 , which are consistent with the simulation results.

Hc
  
K ∞ K 2  2
P α
(a) α 2

j
X X X
= Tt ωf ωreq µk exp − µj
e 4PAID
t (numk ) da,k κ (α) snumk (n)
Pk
k=1 n=0 j=0
 K

(a) 2
µi (Pi Ai ) α
P  
∞ Z ∞
K X K  2 2 !
Pj α Aj (a) Aj
α
   
(a)

i=0
X X
+ 2π x exp −π x 2 µj M 4FAID (num ), α, + µ  dx 

2
e t k j
Ak Ak
 
 0 (Pk Ak ) α Pk 
k=0 n=1 j=0
K
!!
X
×Tt snumk (n) 5k × 1 − ωf · ωreq µk + ωreq · µo / µm 1 − snum0 (0) Pm + ωt · µ1 P1 .




(35)
k=1

974 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

FIGURE 5. Transmission probability versus the transmission power of FIGURE 6. Transmission probability versus the density of UAVs and the
IoT-T in tier 1. transmission power of IoT-T in tier 1.

From (20) and (25), it can be observed that ωt has no relation-

ship with P1 for one-slot charging, due to the fact that once an
IoT-Ts located inside a harvesting zone, it can be ensured to
be fully charged within one time slot; and for two-slot charg-
ing, increasing P1 resulting to a smaller region R1 , which
means that most of the IoT-Ts need to be fully charged within
two time slots, thus the probability ωf becomes smaller.
and ωt becomes smaller, accordingly. Furthermore, with (20)
and (25), we can easily obtain that ωt is increasing with the
communication request probability of an IoT-D ωreq .
Fig. 6 demonstrates how the density of the UAVs µu and
the transmission power of the IoT-Ts P1 would affect the
transmission probability ωt . The observations are as follows.
First, the effective of P1 is consistent with that in Fig. 5.
Second, for a constant P1 , ωt is firstly monotonically increas-
ing with the density of the UAVs µu and then converge to FIGURE 7. The network throughput versus the density of IoT-Ts in tier 1.

a constant as µu increases, the reason is that the increased

density of the UAV will cause more IoT-Ts to be fully charged the transmission power P1 , and the association biasing factor
for data transmission, and ωt will increase first, and when A1 . The observations are given as follows. First, the sim-
the density of the UAVs exceeds a threshold, the density ulation curves are consistent with our analytical results for
of the active IoT-Ts will be saturated, thus ωt will remain the different values of the density µ1 ; Second, the network
stable and convergence to a fix value. The results can provide throughput is monotonically increasing with µ1 , the reason is
useful insights and guidelines for the UAV deployments in that as µ1 increases, the number of FAIDs are increases due
special areas. to the increase number of associated IoT-Ds whose IoT-Ts are
inactive, accordingly, from (11), the density of active IoT-Ts
B. PERFORMANCE OF NETWORK THROUGHPUT that can be used for dividing the load increases. This can
AND ENERGY EFFICIENCY reduce the density of transmitting MBSs, which correspond-
Fig. 7 shows the network throughput given in (33) as a func- ingly leads to less interference from tier 0 to the IoT-Ds and
tion of the density of the IoT-Ts, µ1 , under different values of higher network throughput. Third, for different transmission

  
1 2
α

ec = Tt ωf ωreq µ1 exp −
X (a) e
2 Pj
H µj 4
e P da κ (α)
α 2 
P1
j=0
 
1 Z 1  2  2 !
∞ Pj α
  2 A

A α
(a) (a) j (a) j
X X
+ 2π xµk exp −πx 2 µj M 4
e e , α,
F,k
α
+ µj  dx
Pk Ak Ak
k=0 0 j=0
1 − ωf · ωreq µ1 + ωreq · µo / µm 1 − snum0 (0) Pm + ωt · µ1 P1 .






× Tt (37)

VOLUME 8, 2020 975

 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

FIGURE 8. The energy efficiency versus the density of IoT-Ts in tier 1. FIGURE 9. The energy efficiency versus the association biasing factor in
tier 1.

power P1 , there are two cases to be analyzed, which is

one-slot charging (P1 = 24 dBm), and two-slot charging values of the transmission power P1 , and the association
(P1 = 27 dBm). Compare with two-slot charging, the IoT-Ts biasing factor A1 . The observations are given as follows. The
that located in the UAV’s harvesting zone can be fully charged energy efficiency is firstly increasing with the density µ1 , and
within one time slots with one-slot charging, thus result in a then beginning to degrade when µ1 exceeds a threshold. The
higher probability ωt and 51 , and most IoT-Ds will connect to reason is that as µ1 increases, the network throughput corre-
tier 1, which can similarly leads to less interference from tier spondingly increases as shown in Fig. 7. However, the energy
0 to the IoT-Ds and higher network throughput. Furthermore, consumption is also continue to increase with µ1 . Once µ1
it is also observed that choosing a larger association biasing exceeds a threshold, the number of the MBSs that in sleep
factor in tier 1, A1 = 100, which means that associating most mode is saturated, and the number of active IoT-Ts is continue
IoT-Ds to tier 1, can improve the network throughput. to increase, the growing energy consumption cannot continue
Fig. 8 shows the energy efficiency given in (35) as a to maintain the increasing energy efficiency, and thus will
function of the density of the IoT-Ts, µ1 , under different be decreased. Moreover, reducing the transmission power

h   i
9cov,k
PAID
= E Pr SIRPAID
k ≥ 4PAID
t (numk ) |numk
   
  −α
 
  Pk gx0 ,yk,0 da,k  
= E Pr  ≥ 4PAID (numk ) |numk 
   
K t
 P P −α  
  Pj gx0 ,yj,i yj,i − x0  
e(a) \yk,0
j=0 i∈8
j
K
   
P P −α
Pj gx0 ,yj,i yj,i − x0
e(a) \y
   
  j=0 i∈8  
j k,0
= E Pr gx0 ,yk,0 ≥ 4PAID (numk ) |numk 
   
−α t
  Pk da,k  
   

     
K
(a) 4PAID (num ) X
k
= E E exp − t −α Ij  |Ij  |numk 
Pk da,k j=0
  !  
K
Y 4 PAID (num )
k
= E E  exp − t −α Ij |Ij  |numk 
j=0
Pk d a,k
 !
∞ K
X Y 4 PAID (num )
k
= EIj  exp − t −α Ij snumk (n) . (38)
n=0 j=0
P k d a,k

976 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

P1 is satisfying to improve the energy efficiency, this can both characterized by considering the effect of the associa-
be explained as that smaller P1 is associated with one-slot tion biasing factor, and the transmission power of the active
charging, which results to a higher density of active IoT-Ts. IoT-Ts. The results can provide insightful guideline on the
Finally, choosing a larger association biasing factor in tier network parameters and the deployment for practical UAV
1 A1 = 100 can enhance the energy efficiency, which is swarms and IoT system.
consistent with the trend of the network throughput.
Fig. 9 demonstrates the energy efficiency as a function of APPENDIX A
the association biasing factor in tier 1 A1 . We have the follow- PROOF OF RESULT 1
ing observations. The energy efficiency is initially decreases From (27), we derive the coverage probability of a PAIDs that
with small A1 due to the overloading of the data in tier 1, is served by its associated IoT-T in tier-k as (38), as shown at
the coverage probability of the PAIDs and FAIDs are both the bottom of the previous page.
decreasing with A1 , thus result to the degrade of the energy The (a) in (38) follows from
efficiency; As A1 become larger, the energy efficiency starts X −α
Pj gx0 ,yj,i yj,i − x0 .

to increase since the number of active MBSs are decreased, Ij = (39)
thus the interference from MBSs to the IoT-Ds is decreas- i∈8
(a)
e \yk,0
j
ing. However, the increase will be diminish and the energy
efficiency converges to a fixed value, the reason is that all
  
4PAID
t (numk )
the IoT-Ds will be offloaded to tier 1 as the increase of Note that EIj exp − P d −α Ij are Laplace trans-
k a,k
A1 , and all the MBSs will eventually become sleep mode forms of the random variables Ij , with input parameter
state. Furthermore, larger ωreq will result in a larger ωt , and 4PAID
t (numk )
. According to the result in [37], the Laplace trans-
−α
more IoT-Ds can be served, thus result to a larger energy Pk da,k
efficiency. form of the random variables Ij is obtained by
!
4PAID
t (numk )
VII. CONCLUSION LIj −α
Pk da,k
In this paper, we investigate a UAV-assisted downlink het-  2 !
erogeneous IoT system consisting of macrocells and IoT- 2 Pj α
(a) α

= exp −e µj 4PAID
t (numk ) da,k 2
κ . (40)
Ts with energy harvesting. The transmission probability of Ps
the energy-constrained IoT-Ts is derived by considering one-
slot and two-slot charging, and the HPPP models for the 3D Putting (40) into (38), and combining with (13), the cov-
spatial distributions of the UAV swarms, MBSs, and IoT-Ts. erage probability of a PAIDs that is served by its associated
Energy efficiency is maximized by optimizing the IoT-T IoT-T in tier-k is obtained after some algebraic calculation.
density. The network throughput and energy efficiency are This completes the proof of Result 1.

   

Pk gx00 ,y0k,0 x −α
   
   
9cov,k
FAID
= E Pr  ≥ 4 FAID
(num ) |num
   
K t k  k 
 P −α  
Pj gx00 ,y0j,i y0j,i − x00
P
   
e(a) \y0
j=0 i∈8
j k,0
K
  −α  
Pj gx00 ,y0j,i y0j,i − x00
P P
e(a) \y0
   
  j=0 i∈8  
j k,0
= E Pr gx00 ,y0k,0 ≥ 4FAID (numk ) |numk 
   
t
  Pk x −α  
   

     
K
4 FAID (num ) X
k
= E E exp − t Ij0  |Ij0  |numk 
Pk x −α
j=0
   
K
4 FAID (num ) 
Z ∞ 
k 0 
Y
= E EIj0  exp − t Ij |numk  sXk (x) dx
0 Pk x −α
j=0
 
∞ K
∞Y 4 (num )
Z  FAID 
k
X
=  LIj0 t
sXk (x) dx  snumk (n). (41)
0 Pk x −α
n=1 j=0

VOLUME 8, 2020 977

 Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

APPENDIX B [14] Z. Zhu, S. Huang, Z. Chu, F. Zhou, D. Zhang, and I. Lee, ‘‘Robust designs
PROOF OF RESULT 2 of beamforming and power splitting for distributed antenna systems with
wireless energy harvesting,’’ IEEE Syst. J., vol. 13, no. 1, pp. 30–41,
We derive the coverage probability 9cov,k
PAID as follows, starting
Mar. 2019.
with the SIRk FAID given in (30). We have (41), as shown at the [15] Y. Wang, M. Hua, Z. Liu, D. Zhang, B. Ji, and H. Dai, ‘‘UAV-based mobile
wireless power transfer systems with joint optimization of user scheduling
bottom of the previous page.
and trajectory,’’ Mobile Netw. Appl., pp. 1–15, 2019.
Using the same method as the proof Result 1, and as [16] Z. Chu, Z. Zhu, and J. Hussein, ‘‘Robust optimization for AN-
mentioned before, the total interference of a typical FAID Aided transmission and power splitting for secure MISO SWIPT
received is from all IoT-Ts or MBSs in tier i which are located system,’’ IEEE Commun. Lett., vol. 20, no. 8, pp. 1571–1574,
Aug. 2016.
out of a circle centered at the typical FAID with radius dj = [17] Z. Chu, F. Zhou, P. Xiao, Z. Zhu, D. Mi, N. Al-Dhahir, and
 −α
x PPki Ai
Ak . Then, we have R. Tafazolli, ‘‘Resource allocation for secure wireless powered inte-
grated multicast and unicast services with full duplex self-energy recy-
cling,’’ IEEE Trans. Wireless Commun., vol. 18, no. 1, pp. 620–636,
4FAID (numk )
 
t Jan. 2019.
LIj [18] X. Lu, P. Wang, D. Niyato, D. I. Kim, and Z. Han, ‘‘Wireless charg-
Pk x −α ing technologies: Fundamentals, standards, and network applications,’’
 2  ! IEEE Commun. Surveys Tuts., vol. 18, no. 2, pp. 1413–1452, 2nd Quart.,
α Aj

(a) Pj
= exp −πe µj M 4t
FAID
(numk ), α, x2 . 2016.
Pk Ak [19] Z. Zhu, Z. Chu, F. Zhou, H. Niu, Z. Wang, and I. Lee, ‘‘Secure beamform-
ing designs for secrecy MIMO SWIPT systems,’’ IEEE Wireless Commun.
(42) Lett., vol. 7, no. 3, pp. 424–427, Jun. 2018.
  [20] Z. Zhu, Z. Chu, Z. Wang, and I. Lee, ‘‘Outage constrained robust beam-
A
where M 4FAIDt (numk ), α, Akj is given by (32). Putting forming for secure broadcasting systems with energy harvesting,’’ IEEE
Trans. Wireless Commun., vol. 15, no. 11, pp. 7610–7620, Nov. 2016.
(12) and (42) into (41), 9cov,k
FAID can be obtained. This com-
[21] X. Pan, C. Yan, and J. Zhang, ‘‘Nonlinearity-based single-channel
pletes the proof of Result 2. monopulse tracking method for OFDM-aided UAV A2G communica-
tions,’’ IEEE Access, vol. 7, pp. 148485–148494, 2019.
[22] Z. Xue, J. Wang, G. Ding, H. Zhou, and Q. Wu, ‘‘Maximization of
REFERENCES data dissemination in UAV-supported Internet of Things,’’ IEEE Wireless
[1] S. Cho, K. Lee, B. Kang, K. Koo, and I. Joe, ‘‘Weighted harvest-then- Commun. Lett., vol. 8, no. 1, pp. 185–188, Feb. 2019.
transmit: UAV-enabled wireless powered communication networks,’’ IEEE [23] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, ‘‘Mobile unmanned
Access, vol. 6, pp. 72212–72224, 2018. aerial vehicles (UAVs) for energy-efficient Internet of Things communi-
[2] J. Zhang, S. Chen, R. G. Maunder, R. Zhang, and L. Hanzo, ‘‘Regularized cations,’’ IEEE Trans. Wireless Commun., vol. 16, no. 11, pp. 7574–7589,
zero-forcing precoding-aided adaptive coding and modulation for large- Nov. 2017.
scale antenna array-based air-to-air communications,’’ IEEE J. Sel. Areas [24] W. Feng, J. Wang, Y. Chen, X. Wang, N. Ge, and J. Lu, ‘‘UAV-aided
Commun., vol. 36, no. 9, pp. 2087–2103, Sep. 2018. MIMO communications for 5G Internet of Things,’’ IEEE Internet Things
[3] J. Zhang, T. Chen, S. Zhong, J. Wang, W. Zhang, X. Zuo, R. G. Maunder, J., vol. 6, no. 2, pp. 1731–1740, Apr. 2019.
and L. Hanzo, ‘‘Aeronautical ad hoc networking for the Internet-above- [25] J. Xu, Y. Zeng, and R. Zhang, ‘‘UAV-enabled wireless power trans-
the-clouds,’’ Proc. IEEE, vol. 107, no. 5, pp. 868–911, May 2019. fer: Trajectory design and energy region characterization,’’ in Proc.
[4] L. Xie, J. Xu, and R. Zhang, ‘‘Throughput maximization for UAV-enabled IEEE Globecom Workshops (GC Wkshps), Singapore, Dec. 2017,
wireless powered communication networks,’’ IEEE Internet Things J., pp. 1–7.
vol. 6, no. 2, pp. 1690–1703, Apr. 2019. [26] M. Li, W. Yang, X. Yi, Y. Wang, and J. Wang, ‘‘Swarm robot task planning
[5] F. Huang, J. Chen, H. Wang, G. Ding, Y. Gong, and Y. Yang, ‘‘Multiple- based on air and ground coordination for disaster search and rescue,’’
UAV-assisted SWIPT in Internet of Things: User association and power J. Mater. Eng., vol. 55, no. 11, pp. 1–9, Jun. 2019.
allocation,’’ IEEE Access, vol. 7, pp. 124244–124255, 2019. [27] Y. Liu, Z. Qin, Y. Cai, Y. Gao, G. E. Li, and A. Nallanathan, ‘‘UAV
[6] F. Huang, J. Chen, H. Wang, G. Ding, Z. Xue, Y. Yang, and F. Song, ‘‘UAV- communications based on non-orthogonal multiple access,’’ Sep. 2018,
assisted SWIPT in Internet of Things with power splitting: Trajectory arXiv:1809.05767. [Online]. Available: https://arxiv.org/abs/1809.05767
design and power allocation,’’ IEEE Access, vol. 7, pp. 68260–68270, [28] D. Stoyan, W. Kendall, and J. Mecke, Stochastic Geometry and Its Appli-
2019. cations, 2nd ed. Hoboken, NJ, USA: Wiley, 1996.
[7] C. Yan, L. Fu, J. Zhang, and J. Wang, ‘‘A comprehensive survey [29] F. Baccelli and B. Błaszczyszyn, ‘‘Stochastic geometry and wireless net-
on UAV communication channel modeling,’’ IEEE Access, vol. 7, works: Volume II applications,’’ in Foundations and Trends in Networking,
pp. 107769–107792, 2019. vol. 4, nos. 1–2, pp. 1–312, 2010.
[8] Z. Xue, J. Wang, G. Ding, and Q. Wu, ‘‘Joint 3D location and power [30] A. Al-Hourani, S. Kandeepan, and S. Lardner, ‘‘Optimal LAP altitude
optimization for UAV-enabled relaying systems,’’ IEEE Access, vol. 6, for maximum coverage,’’ IEEE Wireless Commun. Lett., vol. 3, no. 6,
pp. 43113–43124, 2018. pp. 569–572, Dec. 2014.
[9] J.-K. Zhang, S. Chen, R. G. Maunder, R. Zhang, and L. Hanzo, ‘‘Adaptive [31] S. Lee, R. Zhang, and K. Huang, ‘‘Opportunistic wireless energy harvest-
coding and modulation for large-scale antenna array-based aeronautical ing in cognitive radio networks,’’ IEEE Trans. Wireless Commun., vol. 12,
communications in the presence of co-channel interference,’’ IEEE Trans. no. 9, pp. 4788–4799, Sep. 2013.
Wireless Commun., vol. 17, no. 2, pp. 1343–1357, Feb. 2018. [32] X. Song, C. Yin, D. Liu, and R. Zhang, ‘‘Spatial throughput characteriza-
[10] C. Xu, T. Bai, J. Zhang, R. Rajashekar, R. G. Maunder, Z. Wang, tion in cognitive radio networks with threshold-based opportunistic spec-
and L. Hanzo, ‘‘Adaptive coherent/non-coherent spatial modulation aided trum access,’’ IEEE J. Sel. Areas Commun., vol. 32, no. 11, pp. 2190–2204,
unmanned aircraft systems,’’ IEEE Wireless Commun., vol. 26, no. 4, Nov. 2014.
pp. 170–177, May 2019. [33] Y. Yao, X. Chen, X. Li, and X. Song, ‘‘Hybrid small cell base station
[11] Y. Zeng, R. Zhang, and T. J. Lim, ‘‘Wireless communications with deployment in heterogeneous cellular networks with wireless power trans-
unmanned aerial vehicles: Opportunities and challenges,’’ IEEE Commun. fer,’’ Inf. Sci., no. 454, pp. 16–29, Jul. 2018.
Mag., vol. 54, no. 5, pp. 36–42, May 2016. [34] P.-S. Yu, J. Lee, T. Q. S. Quek, and Y.-W. P. Hong, ‘‘Traffic offload-
[12] H. Wang, G. Ding, F. Gao, J. Chen, J. Wang, and L. Wang, ‘‘Power control ing in heterogeneous networks with energy harvesting personal cells-
in UAV-supported ultra dense networks: Communications, caching, and network throughput and energy efficiency,’’ IEEE Trans. Wireless Com-
energy transfer,’’ IEEE Commun. Mag., vol. 56, no. 6, pp. 28–34, Jun. 2018. mun., vol. 15, no. 2, pp. 1146–1161, Feb. 2016.
[13] Y. Zeng and R. Zhang, ‘‘Energy-efficient UAV communication with tra- [35] S. Singh, H. S. Dhillon, and J. G. Andrews, ‘‘Offloading in heterogeneous
jectory optimization,’’ IEEE Trans. Wireless Commun., vol. 16, no. 6, networks: Modeling, analysis, and design insights,’’ IEEE Trans. Wireless
pp. 3747–3760, Jun. 2017. Commun., vol. 12, no. 5, pp. 2484–2497, May 2013.

978 VOLUME 8, 2020

Y. Yao et al.: Energy Efficiency Characterization in Heterogeneous IoT System With UAV Swarms Based on WPT

[36] H.-S. Jo, Y. J. Sang, P. Xia, and J. G. Andrews, ‘‘Heterogeneous cellular XINWEI YUE received the degree from the Yellow
networks with flexible cell association: A comprehensive downlink SINR River Conservancy Technical Institute, in 2007,
analysis,’’ IEEE Trans. Wireless Commun., vol. 11, no. 10, pp. 3484–3495, the M.S. degree from Henan Normal University,
Oct. 2012. in 2013, and the Ph.D. degree in communica-
[37] M. Haenggi, J. G. Andrews, F. Baccelli, O. Dousse, and M. Franceschetti, tion and information system from Beihang Uni-
‘‘Stochastic geometry and random graphs for the analysis and design versity (BUAA), Beijing, in 2018. He has been a
of wireless networks,’’ IEEE J. Sel. Areas Commun., vol. 27, no. 7, Lecture with the School of Information and
pp. 1029–1046, Sep. 2009.
Communication Engineering, Beijing Informa-
tion Science and Technology University (BISTU),
Beijing, since 2018, and promoted to Associate
Professor, in 2019. His research interests include 5G/6G networks, wireless
YUANYUAN YAO received the Ph.D. degree in
communications theory, non-orthogonal multiple access, cooperative net-
information and communication engineering from
works, and physical layer security.
the Beijing University of Posts and Telecommu-
nications, Beijing, China, in 2017. Since 2017,
she has been with the School of Information and
Communication Engineering, Beijing Informa-
tion Science and Technology University, Beijing,
as an Associate Professor. Her research interests
include stochastic geometry and its applications in
large-scale wireless networks, energy harvesting,
and wireless communication. CHUNYU PAN received the Ph.D. degree with
the School of Information and Communication
Engineering, Beijing University of Posts and
Telecommunications, Beijing, China. She has been
ZHENGYU ZHU received the B.S. degree from
a Lecturer with the School of Information and
Henan University, in 2010, and the Ph.D. degree
Communication Engineering, Beijing Information
from Zhengzhou University, China, in 2017.
Science and Technology University, Beijing, since
He is currently with the School of Information
July 2019. Her research interests include software-
Engineering, Zhengzhou University, and with
defined cellular networks, heterogeneous wire-
the Zhengzhou University Research Institute of
less networks, resource allocation and mobility
Industrial Technology Company Ltd., Zhengzhou,
management in cellular networks.
China. His research interests include information
theory and signal processing for wireless com-
munications, such as MIMO wireless networks,
physical layer security, wireless cooperative networks, convex optimization
techniques, and energy harvesting communication systems.

SAI HUANG is currently working with the

Department of Information and Communication
Engineering, Beijing University of Posts and
Telecommunications, as a Lecturer. He serves as XUEHUA LI received the Ph.D. degree in telecom-
an Academic Secretary with the Key Laboratory munications engineering from the Beijing Univer-
of Universal Wireless Communications, Ministry sity of Posts and Telecommunications, Beijing,
of Education, China. He is the Reviewer of inter- China, in 2008. She is currently a Professor and
national journals, such as the IEEE TRANSACTIONS the Deputy Dean of the School of Information
ON VEHICULAR TECHNOLOGY, the IEEE WIRELESS and Communication Engineering with the Beijing
COMMUNICATIONS LETTERS, the IEEE TRANSACTIONS Information Science and Technology University,
on COGNITIVE COMMUNICATIONS AND NETWORKING and the international confer- Beijing. She is also a Senior Member of the Beijing
ences, such as IEEE ICC and IEEE GLOBECOM. His research directions Internet of Things Institute. Her research interests
are machine learning assisted intelligent signal processing, statistical spec- are in the broad areas of communications and
trum sensing and analysis, fast detection and depth recognition of universal information theory, particularly the Internet of Things, and coding for mul-
wireless signals, millimeter wave signal processing, and cognitive radio timedia communications systems.
networks.

VOLUME 8, 2020 979