sensors

Article
Energy Efficiency Optimization for SWIPT-Enabled IoT
Network with Energy Cooperation
Yang Cao 1, *, Ye Zhong 2 , Chunling Peng 2 , Xiaofeng Peng 2 and Song Pan 2

1 Periodical Agency, Chongqing University of Technology, Chongqing 400054, China

2 School of Electrical and Electronic Engineering, Chongqing University of Technology,
Chongqing 400054, China; [email protected] (Y.Z.); [email protected] (C.P.);
[email protected] (X.P.); [email protected] (S.P.)
* Correspondence: [email protected]

Abstract: As an advanced technology, simultaneous wireless information and power transfer (SWIPT),
combined with the internet of things (IoT) devices, can effectively extend the online cycle of the
terminal. To cope with the fluctuation of energy harvesting by the hybrid access points (H-AP), the
energy cooperation base station is introduced to realize the sharing of renewable energy. In this paper,
we study the SWIPT-enabled IoT networks with cooperation. Our goal is to maximize the energy
efficiency of the system, and at the same time, we need to meet the energy harvesting constraints, user
quality of service (QoS) constraints and transmission power constraints. We jointly solve the power
allocation, time switching and energy cooperation problems. Because this problem is a nonlinear
programming problem, it is difficult to solve directly, so we use the alternating variable method,
the iterative algorithm is used to solve the power allocation and time switching problem, and the
matching algorithm is used to solve the energy cooperation problem. Simulation results show that
the proposed algorithm has obvious advantages in energy efficiency performance compared with the
comparison algorithm. At the same time, it is also proved that the introduction of energy cooperation
technology can effectively reduce system energy consumption and improve system energy efficiency.

Citation: Cao, Y.; Zhong, Y.; Peng, C.;

Keywords: SWIPT; IoT; energy efficiency; power allocation; time switching; energy cooperation
Peng, X.; Pan, S. Energy Efficiency
Optimization for SWIPT-Enabled IoT
Network with Energy Cooperation.
Sensors 2022, 22, 5035. https://
1. Introduction
doi.org/10.3390/s22135035
The development of 5G technology has contributed to the rapid spread of the internet
Academic Editor: Javier M. Aguiar
of things (IoT). As an important application scenario for 5G, IoT connects a large number
Received: 10 June 2022 of physical objects such as wearable devices, smart home sensors, industrial sensors and
Accepted: 30 June 2022 agricultural sensors wirelessly to provide various services such as healthcare, smart indus-
Published: 4 July 2022 try and smart agriculture [1,2]. According to Cisco, the number of global IoT devices in
IoT networks will reach 14.7 billion by 2023 [3], which means that the number of wireless
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
devices and traffic demand will grow at an explosive rate, and it is clear that simply using
published maps and institutional affil-
macro base stations in cellular networks to connect this huge number of devices is not
iations.
enough. To address this problem, dense hybrid access point (H-AP) deployment is seen as
a promising technology to meet the quality of service (QoS) needs of devices [4].
The dense deployment of H-APs increases the energy consumption of communication
systems, and in the context of a dual carbon strategy, green solutions need to be developed
Copyright: © 2022 by the authors. to reduce network-wide energy consumption. To address this challenge, a large number of
Licensee MDPI, Basel, Switzerland. scholars have explored efficient energy utilization and solutions, mainly in two dimensions:
This article is an open access article “open source” and “cost reduction”. Specifically, the introduction of renewable energy, i.e.,
distributed under the terms and “open source”, such as solar and wind energy, to reduce energy consumption on the grid.
conditions of the Creative Commons Efficient energy management strategies, i.e., “cost reduction”, such as wireless resource
Attribution (CC BY) license (https:// management strategies or H-AP sleeping, control the transmitting power of base stations
creativecommons.org/licenses/by/
to reduce system energy consumption.
4.0/).

Sensors 2022, 22, 5035. https://doi.org/10.3390/s22135035 https://www.mdpi.com/journal/sensors

Sensors 2022, 22, 5035 2 of 21

Energy harvesting technology, as an environmentally friendly and economically

friendly technology, has been widely used in the communications sector. For example,
about 2/3 of the H-APs deployed by China Mobile in Tibet are powered by renewable
energy [5]. Huawei has designed solar cellular H-APs around the world to the tune of
20 million kWh [6]. A number of studies have been carried out for communication networks
with energy harvesting by combining both “open source” and “cost reduction” dimensions.
The authors of [7] propose a dynamic energy-aware power allocation algorithm based on
Lyapunov optimization to maximize system throughput. The work in [8] considers the
mixed integer programming problem of user association and power allocation to improve
the system energy efficiency using Lagrangian algorithms. The authors of [9] propose a
joint optimized power allocation and energy management maximization energy efficiency
method based on the Lyapunov framework for dense base station networks with energy
harvesting, which improves the overall system throughput and optimizes the system en-
ergy efficiency. In [10], a generalized Benders decomposition method based on Lagrange
multipliers is proposed for dormant networks with energy harvesting and base stations
to optimize the transmit power consumption and reduce the energy consumption of the
system. However, the renewable energy harvested by base stations in different climates
and geographical locations varies significantly due to its stochastic and unstable nature.
To address the unevenness of harvested energy, various approaches have been pro-
posed, mainly including two major approaches, energy storage systems and energy collab-
oration [11]. The first option requires large capacity battery storage and there are losses
in the use of batteries, and the deployment of large quantities can lead to high costs and
environmental pollution. On the other hand, energy cooperation is an important technology
based on the development of smart grids, and is considered to be an effective solution
to improve the energy utilization of the system energy cooperation technology is based
on smart grids as a carrier, through its bi-directional power flow devices to effectively
solve the problems of H-AP energy shortage and uneven distribution of renewable energy
sources [12,13], and has been studied by scholars in conjunction with energy cooperation
networks. In [14], a deep reinforcement learning algorithm is proposed to improve the
total system throughput by offline training, taking into account the transmit power, energy
harvesting and battery capacity constraints in a fused energy cooperation IoT downlink
scenario. The authors of [15] investigate the problem of minimizing energy consumption in
H-AP networks with caching and propose a low-complexity hierarchical solution algorithm
that effectively reduces system power consumption by optimizing terminal bandwidth
and energy cooperation mechanisms. The authors of [16] propose a hybrid energy ratio
allocation algorithm, which effectively improves the impact of renewable energy on the
communication system and reduces the energy cost of the system.
Another important constraint on the development of IoT is the power supply to the
terminals. Wired and battery power cannot effectively solve the problem of energy shortage
in terminals, so it is particularly important to study the maintenance of the sustainable
operation of terminals. Energy harvesting technology is considered an important means
of reducing system energy consumption and extending the operation of devices in order
to extend the operating time of terminals on the network. Traditional renewable energy
sources such as wind and solar are intermittent and unreliable, and a large number of
terminals are deployed indoors where solar energy supply is not applicable [17]. In recent
years, a large number of scholars have studied the use of simultaneous wireless information
and power transfer (SWIPT) technology to solve the energy shortage problem of wireless
communication equipment nodes. The characteristics of this technology are to make full
use of the characteristics of radio frequency (RF) signals that carry data information and
electromagnetic energy at the same time, and to extend the network operation cycle of
communication devices by harvesting the surrounding energy for wireless charging while
realizing wireless information transmission [18,19]. To further improve the performance
of SWIPT-enabled IoT, many studies have been conducted extensively from the scheme
of resource allocation. The authors of [20] jointly optimized the transmit power, power
Sensors 2022, 22, 5035 3 of 21

split ratio and subcarrier allocation to enhance the rate of IoT networks. In [21], a penalty
function algorithm is proposed for energy-carrying communication networks to jointly
optimize transmit power and collect energy to reduce system energy consumption. The
authors of [22] investigate the problem of maximizing the throughput of IoT devices and
propose a Lagrangian-based algorithm for this gradient, which jointly allocates spectrum
and transmit power to improve the total system throughput. The authors of [23] address
the energy efficiency optimization problem of optimized networks and propose an iterative
algorithm based on the Karush–Kuhn–Tucker (KKT) condition to combine the transmit
power and power coefficient, and its algorithm improves the energy efficiency performance
of the system. In [24], a multi-objective energy-carrying network optimization problem
was investigated, and the multi-objective problem was converted into a single-objective
problem by using the defined equivalent sum-rate method for solving the problem, and
the scheme optimized both system throughput and system energy consumption. In [25],
a Dinkelbach-based two-layer iterative algorithm was proposed to jointly optimize the
time-switching and power allocation problems. A particle swarm algorithm was proposed
in [26] to optimize the rate of the SWIPT-enabled network.
Driven by the carbon-peaking and carbon-neutrality strategic goals, communication
systems will move towards a “greener” direction, where balancing the requirements of
low power consumption and high speed is crucial. The authors of [7–10] focus on the opti-
mization of resource allocation with energy harvesting, which addresses the optimization
of base station energy consumption and does not consider terminal standby time. The
authors of [14] and others demonstrate IoT networks with energy cooperation, low power
consumption and good performance. In [20–26] studied SWIPT networks optimized the
on-net operation of terminals without introducing energy cooperation to save the power
consumption of the H-AP. The above literature shows that SWIPT and energy cooperation
technologies can effectively reduce system power consumption, however, the energy effi-
ciency performance of SWIP-enabled IoT with energy cooperation is still unknown, and
as people attach importance to the green network, energy efficiency becomes more and
more important. Therefore, this paper aims to maximize the energy efficiency of the system
and realize the design of a green communication system. Inspired by the aforementioned
literature, we consider IoT networks in using energy cooperation and SWIPT to optimize
the energy efficiency of the system as a goal. The main contributions of this paper are
summarized as follows:
• We consider a downlink transmission model for SWIPT-enabled IoT with energy coopera-
tion. A resource allocation problem is proposed that considers the quality of service (QoS)
constraints for users, energy harvesting constraints, and jointly optimizes the power
allocation, time switching coefficients and energy cooperation problems to maximize
system energy efficiency as the optimization objective. Considering that the optimization
problem is a mixed-integer non-linear programming problem that is difficult to solve
directly, we consider decomposing the problem into three sub-problems of lower com-
plexity, namely the power allocation, time switching coefficient and energy cooperation
problems. We propose a two-stage algorithm for solving the problem
• The first-stage algorithm is used to solve the power allocation and time-switching
problems. The first-stage algorithm is a two-level iterative algorithm that the power
allocation coefficient and the time switching coefficient are separated using the fixed
variable method. In the outer layer, the power allocation solution is obtained using
the Dinkelbach method iteratively. In the inner layer of the algorithm, the Dinkelbach
method is used again to solve for the time switching coefficient under a fixed power
allocation. The power allocation and time switching resolution are obtained by several
iterations. Finally, in the second stage of the algorithm, the matching theory is used to
obtain the resolution of energy cooperation.
• Our results show that our proposed algorithm has higher energy efficiency compared
to the comparison algorithm. The system with SWIPT has good energy efficiency
performance and can effectively extend the terminal on-grid operation cycle. In
Sensors 2022, 22, 5035 4 of 21

addition, the simulations show that energy cooperation can effectively reduce the
energy consumption of the system. This demonstrates the performance benefits of
integrating SWIPT and energy cooperation technologies in the IoT network.
The remainder of the paper is structured as follows. Section 2 presents the system-
based model and the energy efficiency maximization problem modeling. In Section 3,
we propose a two-layer algorithm to find the optimal joint power allocation and time-
switching allocation. Section 4 investigates the use of a matching algorithm to solve
the energy cooperation problem. Numerical results are given in Section 5 to prove the
theoretical results. Finally, we conclude the paper in Section 6.

2. System Model
2.1. Transmission Model
As shown in Figure 1, we consider a SWIPT-enabled IoT system with energy co-
operation, consisting of hybrid access points (H-AP) and a smart grid with downlink
transmission, where the H-AP is fed by a mix of smart grid and renewable energy sources.
The system consists of m H-AP and j terminal. Let m ∈ {1, 2, 3, · · · , M } denote the set
of H-APs. Let j ∈ {1, 2, 3, · · · , N } denote the set of terminals. Each terminal contains
information decoding and RF energy harvesting circuits. Considering the low cost and easy
implementation of the time-switching method circuit, we distinguish between information
and energy signals through the time-switching method. It is assumed that the H-AP is
capable of serving multiple terminals simultaneously, the terminals associate with the near-
est H-AP, the H-AP improves the spectrum efficiency of the system by sharing the entire
transmission band, and multiple terminals under a single H-AP service use orthogonal
spectrum resources, only one H-AP can be associated with a terminal. It is also assumed
that all H-APs and terminals have perfect channel state information (CSI). h The isignal Sm
transmitted by the m-th H-AP can be expressed as Sm = Pjm sm , with E |sm | H = 1, Pjm
p

indicates the transmitted power of the m-th H-AP. When the terminal is associated with the
m-th H-AP, the received signal at the terminal can be expressed as

M
0
q  q 
ym = h jm Pjm sm + ∑0 hm
jm Sm0 Pj0 m0 sm0 + v0 (1)
m0 =1,m 6=m
p
where, v0 denotes additive Gaussian white noise, and h jm Pjm sm denotes the accepted
M +1 0
 q 
useful signal. ∑ hmjm S m 0 Pj 0 m0 sm0 indicates received co-channel interference sig-
m0 =1,m0 6=m
 2
nal. The channel gain consists of path loss and Rayleigh fading, denoted as h jm  = d− β g,
g denotes Rayleigh fading, d− β denotes the path loss model, β denotes the path loss factor
and d denotes the distance from the terminal to the H-AP. Pjm indicates the transmit power
from the H-AP m to the terminal j. h jm denotes the channel gain of the terminal j associated
0
m denotes the transmit power received by the terminal j from other
with the H-AP m, Pjm
0
H-APs and hm jm denotes the channel gain of other H-AP interference.
γ jm is the signal interference noise ratio (SINR) of the terminals. Since the H-AP
spectrum is shared within the network, there is mutual interference between the terminal
links, and the SINR of the terminals is expressed as
 2
Pjm h jm 
γ jm = (2)
M N  0 2

∑ ∑ hm jm  Pj0 m0 + σ
2
m0 =1,m0 6=m j0 =1
 Sensors 2022, 22, 5035 5 of 21

 2
where σ2 is the noise power. Pjm h jm  is the strength of the useful signal received by
M N  0 2

terminal j associated with H-AP m. ∑ ∑ hmjm  Pj0 m0 indicates that terminal j is
m0 =1,m0 6=m j0 =1
 2
subject to co-channel interference from terminals under other H-APs. Let h jm  = Gjm ,
 2
 m0  m0 . The signal interference noise ratio (SINR) is expressed as
h jm  = Gjm

Pjm Gjm
γ jm = (3)
M N 0
m P 0 0 + σ2
Sensors 2022, 22, x FOR PEER REVIEW
∑ ∑ Gjm jm 5 of 21
m0 =1,m0 6=m j0 =1

Aggregator
HAP

Smart grid
Energy harvesting link
IoT device Communication link
Interference link
ID Renewable energy
 Receiver
Grid energy
Time
HAP Switching Solar energy
EH
IoT device Receiver Wind energy
1

Figure 1. System
Figure model.
1. System model.

is the signal
 jm According tointerference noise ratio
Shannon’s formula, the(SINR) of the terminals.
transmission Since the
rate of terminal j isH-AP
givenspec-
by
trum is shared within the network, there is mutual interference between the terminal
links, and the SINR of the terminals is = τj W logas
R jexpressed 2 (1 + γ jm ) (4)

where W is the bandwidth of the system. τj is expressed

2 as the transmission time allocated
to the information time slot.   Pjm hjm
jm M N
2 (2)
2.2. Energy Model   hmjm Pjm   2
m 1, m  m j  1
Assuming that all terminals collect energy from the RF signal using time-switching
2
techniques
where  2 is the andnoise
perform SWIPT
power. Pjm techniques,
h jm is the the terminals
strength split
of the the received
useful signal from
signal received by the
H-AP into two parts: in the first time slot M
forN information transfer and in the second time
2
slot for energy harvesting. Where is expressed hmjmasPthe transmission time allocated to the
terminal j associated with H-AP m. τj
information time slot and 1 − τj is expressed
 
m 1, m  m j  as
j m indicates that terminal j is
1 the time portion of the energy harvesting
time slot. For the RF harvesting model of the IoT terminal, the paper uses 2the widely
subject to co-channel interference from terminals under other H-APs. Let h jm  G jm ,
used linear energy harvesting model [25]. The energy harvesting circuit of the terminal is
2
h mjm capable
 G mjm . of
The signal interference
converting noise
the received powerratiosignal (SINR) and is expressed as power signal into DC
the interfering
power for use by the terminal. For the presence of multiple H-APs in an IoT system, the
RF energy collected by the terminal Pjmof G jmthe superposition of the power emitted by
 jm  consists
M N
multiple H-APs. The SWIPT-enabled terminalsm are used2 for energy harvesting within (3)time
slot 1 − τj . The energy collected byma  G jm Pj m  
single
1, m  m j  1
terminal is expressed as

According to Shannon’s formula, the transmission

M rate of terminal j is given by
ESj = (1 − τj )η j ∑ Gjm Pjm (5)
Rj   jW log2 (1m=1jm ) (4)

where W is the bandwidth of the system.  j is expressed as the transmission time allo-
cated to the information time slot.

2.2. Energy Model

Assuming that all terminals collect energy from the RF signal using time-switching
Sensors 2022, 22, 5035 6 of 21

where 0 < η j < 1, η j denotes the conversion efficiency of the energy harvesting. It is
assumed that the conversion efficiency of all the terminals of the system energy harvesting
is the same, satisfying η j = η (∀ j).
In IoT systems with energy harvesting, each H-AP is equipped with an energy har-
vesting device. Due to the uneven distribution of renewable energy density and differences
in transmitting power, some H-APs do not harvest enough energy to maintain their own
standby power. Some H-APs harvest too much energy. To avoid wasting renewable energy,
we have introduced energy cooperation technology, which is an important solution for
effective energy dispatch. Through aggregators in the smart grid, the excess energy is
transferred to the more power-consuming H-APs, effectively increasing the utilization of
renewable energy. During the energy cooperation, the renewable energy received by H-AP
M M
is ∑ Tm0 m , ∑ Tmm0 denotes the renewable energy transferred out of the H-AP. Where
m =1 m =1
α ∈ [0, 1] denotes the efficiency factor of the H-AP energy transfer. For a single H-AP, the
available renewable energy is expressed as

M M
RE
Em = Em + α ∑ Tm0 m − ∑ Tmm0 (6)
m =1 m =1

where Tmm0 represents energy transferred to other H-APs, Tm0 m represents energy received
from other H-APs and Em RE is the renewable energy collected by the H-AP.

Typically, the power consumption of a conventional wireless communication system

is defined as the following linear model [25], and the power consumption of an H-AP in a
communication system is defined as the following model

PW = ζPm + PmC (7)

where ζ is the power amplification factor of the signal. PmC is the standby power Consump-
tion of the H-AP, including power consumption such as baseband signal processing and
cooling system.
In IoT network, the circuit power consumption of SWIPT-enabled terminals is not
considered because of their low device power consumption. The total energy consumption
of the system is expressed as

M M M N
∑ Gtotal = ∑ PW − ∑ RE
Em − ∑ Esj (8)
m =1 m =1 m =1 j =1

2.3. Problem Formulation

The constraints in this paper include QoS constraints for terminals, maximum transmit
power constraints for H-APs, and maximizing the total energy efficiency of the IoT system
under energy harvesting constraints. According to [8], the energy efficiency (EE) of a
system is defined as the ratio of the total achievable rate to the total power consumption.
The optimization problem can then be expressed as

N M
P1 : Max : EE( P, τ, T ) = ∑ R j / ∑ Gtotal
P,τ,T j =1 m =1
s.t. C1 : R j ≥ Rmin
N
C2 : ∑ Esj ≥ Emin
j =1
N
(9)
C3 : ∑ Pjm ≤ max
Pm
j =1
C4 : Tm0 m ∩ Tmm0 = ∅
C5 : 0 ≤ Pm 0 ≤ Tmm0
C6 : 0 ≤ τj ≤ 1
Sensors 2022, 22, 5035 7 of 21

where P = [ Pjm ], τ = [τj ], T = [ Tmm0 ]. C1 is the terminal minimum QoS requirement

constraint. C2 represents the total terminal energy collection threshold under a single H-AP.
C3, C5 denotes the transmit power constraint. C4 denotes that transfer of energy and reception
cannot occur simultaneously. c6 denotes the constraint for the time switching factor.
We can observe that the P1 problem is a complex fractional form, and the energy
cooperation problem is an integer programming problem, then the P1 problem is a mixed-
integer non-linear optimization problem that is difficult to solve directly. We consider
reducing the problem to three less complex sub-problems, namely power allocation, time
switching and energy cooperation. In addition the two sets of variables for the power
allocation P and time switching coefficients τ, which are mutually coupled, as in the
scheme in the literature [12], can be solved for the multi-variable problem by considering
the remaining variables as constants and how to solve the remaining variables, and finally
by an iterative scheme of alternating variables to obtain the resolution of the multi-variables.
Because the power allocation P and time switching coefficients τ are coupled, we propose to
solve the problem in the first stage using an iterative algorithm. In contrast, the subproblem
of energy cooperation is uncoupled from the two sets of variables mentioned above, and
for this reason we consider proposing a matching algorithm in the second stage to find the
resolution of energy cooperation.

3. Joint Power Distribution and Time-Switching Control Algorithms

This section may be divided by subheadings. It should provide a concise and precise
description of the experimental results, their interpretation, as well as the experimental
conclusions that can be drawn.

3.1. Power Distribution Problem

In this section, we propose control algorithms for joint power allocation and time
switching. From the P1 solution objective formulation, it is clear that optimizing the
transmit power and time switching problems is the focus of ensuring terminal QoS and
terminal standby time and improving system energy efficiency. Under the subproblem
of ground-given energy collaboration. We have adopted the resource allocation scheme
in [25], which is applied to the case of a single H-AP. We have expanded the scheme and
applied it to multiple H-APs. we develop a two-layer iterative algorithm to solve for the
power allocation and time-switching coefficients. Firstly, in the outer iteration, the iterative
algorithm is used to solve for the transmit power given a time factor. In the inner layer of
the algorithm, the transmit power is fixed and then the time switching coefficient is solved.
The specific algorithmic analysis of the solution process is shown below.
A. Power allocation method under timed switching allocation
Given the two sets of variables for the time switching factor τ and energy cooperation
T, only one set of variables for the transmit power needs to be solved. The original problem
P1 is downscaled to solve a one-dimensional power distribution problem, then the P1
problem is reformulated as

N M
P2 : Max : EE( P) = ∑ R j / ∑ Gtotal
j =1 m =1 (10)
s.t. C1, C2, C3, C5

As can be seen from the P2 problem, the problem is a fractional objective function, mak-
ing the problem non-linear and difficult to solve directly. First, we need to determine the
non-concave nature of the optimization problem and then adopt the appropriate solution.
The first step is to determine the non-concave nature of the constraints. Condition C1 can be
M N 0 Rmin M N 0
m P 0 0 + σ2 + P G − 2 m P 0 0 + σ2 ) ≥
converted to ∑ ∑ Gjm jm jm jm τW ( ∑ ∑ Gjm jm
m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1
0. As can be seen from the inequality of the transformation deformation, the constraint is
Sensors 2022, 22, 5035 8 of 21

that the feasible domain of the resolution is a convex set. Similarly, the analytic feasible
region of the constraint C2, C3, C5 is also a convex set.
N M
Since EE( P) = ∑ R j / ∑ Gtotal is fractional, we first prove the concavity of the
j =1 m =1
N N
numerator ∑ R j of the objective function. Then, the first order derivative of ∑ R j with
n =1 n =1
respect to Pjm is expressed as follows:

N N M N 0 M N 0
m P 0 0 + σ2 ) − P G m
∂ ∑ Rj W ∑ τj Gjm ( ∑ ∑ Gjm jm jm jm ∑ ∑ Gjm
n =1 j =1 m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1
= · M N M N
(11)
∂Pjm ln 2 m0 P 0 0 m0 P 0 0
( ∑ ∑ Gjm jm + σ2 + Pjm Gjm )( ∑ ∑ Gjm jm + σ2 )
m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1

Then, the second order derivative with respect to Pjm is expressed as follows:
" #
N N M N 0 M N 0
m P 2 m
∂2 ∑ Rj W ∑ τj Gjm ( ∑ ∑ Gjm j0 m0 + σ )− Pjm G jm · ∑ ∑ Gjm
n =1 j =1 m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1
∂Pjm Plm =− ln 2 ·" #2
M N 0 M N 0
m P m P
( ∑ ∑ Gjm 2
j0 m0 + σ + Pjm G jm )( ∑ ∑ Gjm 2
j0 m0 +σ )
" m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1 #
M N M N M N M N
m0 + G )( m 0 2 m 0 m0 P 2 +P G )
( ∑ ∑ Gjm jm ∑ ∑ Gjm Pj0 m0 +σ )+ ∑ ∑ Gjm ·( ∑ ∑ Gjm 0
j m 0 + σ jm jm (12)
m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1
× " #2
M N M N
m0 P m0 P
( ∑ ∑ Gjm 2
j0 m0 + σ + Pjm G jm )( ∑ ∑ Gjm 2
j0 m0 +σ )
0 0 0
m =1,m 6=m j =1 0 0 0
m =1,m 6=m j =1

∀ j, l = 1, 2, · · · N
N
∂2 ∑ R j
n =1
Let Hj = ∂Pjm 2
, according to the above formula, we get

N
∂2 ∑ R j 
n =1 Hj , j ≤ l
= (13)
∂Pjm Plm Hl , otherwise

Then, the Hessian matrix with respect to variables expressed as

···
 
H1 H1 H1
 H1 H2 ··· H2 
H= . (14)
 
.. .. .. 
 .. . . . 
H1 H2 ··· HN

Then, the opposite matrix of the Hessian matrix is Q = − H, then the j-th order
principal subformula of the matrix is expressed as

− H1 − H1 ··· − H1
 
 − H1 − H2 ··· − H2 
Qj = 
 
.. .. .. .. 
 . . . . 
 − H1 − H2 · · · − HN (15)

 − H1, j = 1
= N
 − H1 j∏

=2
( H j −1 − H j ), 2 ≤ j ≤ N

According to the formula, we can see that the power and channel is constantly greater
than 0, and from the matrix properties can be obtained Hj−1 − Hj ≥ 0. Any j-order
sequential principal subformula of matrix Q, Q j ≥ 0. It is possible to obtain Q ≥ 0, H ≤ 0.
Sensors 2022, 22, 5035 9 of 21

N
It follows that the Hessian matrix H of ∑ R j , with respect to the variable Pjm , is a semi-
n =1
N
negative definite matrix. It follows that ∑ R j is a concave function with respect to the
n =1
variable Pjm . It follows from Shannon’s theorem that the communication rate is positive,
N
thus it is shown that ∑ R j is a non-negative concave function with respect to the variable
n =1
Pjm . Similarly, the denominator of the objective function is non-negative. In summary, it
N M
is shown that the optimization objective function EE( P) = ∑ R j / ∑ Gtotal is a concave
n =1 m =1
fractional programming problem with respect to the transmit power P.
It can be seen that the objective function is a fractional programming problem, which
is difficult to solve directly, and the Dinkelbach algorithm [26] has been widely used
with solving non-linear fractional optimization problems. According to the nature of the
Dinkelbach algorithm [27], the objective function needs to be transformed into the form of
subtracting the numerator from the denominator. According to the Dinkelbach method,
we need to introduce a parameter e = EE( P), and the P2 problem is converted into the
following form
N M
P2.1 : Max : EE( P) = ∑ R j − e· ∑ Gtotal
n =1 m =1 (16)
s.t. C1, C2, C3, C5

Proposition 1. Assuming that the optimal transmit power of the H-AP is P∗ , the e∗ is an optimal res-
N M
olution of problem P2.1 for which the sufficient conditions are Max: F (e) = ∑ R j − e· ∑ Gtotal = 0.
n =1 m =1

Proof. The proposition is a classical conclusion in generalized dispersion planning, the

proof of which has been proved in [28], and the proof process is not described in this
paper. The above proposition provides the theoretical support for the transformation of the
optimization problem P2.1. Thus, an approximate equivalent problem for problem P2.1
can be obtained by iteration. the Dinkelbach algorithm requires several iterations to obtain
the resolution of the problem, where the power allocation and energy efficiency resolution
at the t-th iteration are P(t) and e(t), respectively. Needs to be satisfied in the t-th iteration
N M
F (e(t) ) = ∑ R j ( P(t) ) − e(t) · ∑ Gtotal ( P(t) ) ∼
= 0. The Dinkelbach parameter e is updated by
j =1 m =1
iterations until the convergence condition is met and the iteration is exited. From the above
N M
proof it can be seen that EE( P) = ∑ R j − e· ∑ Gtotal is a convex optimization problem
j =1 m =1
and its solution function and constraints satisfy the scope of application of the Lagrangian
dual method, so we can use the Lagrangian dual method to solve its optimization problem
Sensors 2022, 22, 5035 10 of 21

 
N Pjm Gjm
L( P, µ, ν, ψ) = ∑ τj W log2 1 +
 
M N 
m0 P
j =1 ∑ ∑ Gjm j0 m0 +σ
2
m0 =1,m0 6=m j0 =1 !
M N M
−e ∑ (ζPm + PmC − Em ) − ∑ (1 − τj )η j ∑ Gjm Pm
m =1 j =1 m =1
  
Pjm Gjm (17)
−µ j  Rmin − τj W log2 1 +
  
M N 
m0 P
∑ ∑ Gjm j0 m0 +σ
2
m0 =1,m0 6=m j0 =1
!
N M
−νj Emin − ∑ (1 − τj )η j ∑ Gjm Pm
j =1 m =1
!
N
max , µ ≥ 0, ν ≥ 0, ψ ≥ 0
−ψ ∑ Pjm − Pm
j =1

where denote the Lagrange multipliers of the constraints, respectively, and the pair-
wise function expressions are

g(µ, ν, ψ) = maxL( P, µ, ν, ψ) (18)

The pairwise optimization problem expression for the problem is

ming(µ, ν, ψ)
(19)
s.t. µ ≥ 0, ν ≥ 0, ψ ≥ 0

In this section, we use a gradient descent-based algorithm to obtain a power allocation

solution to the pairwise optimization problem by multiple iterations, and we need to find
the first-order derivative of the Lagrangian function [29], whose derivative is

N M N 0 M N 0
m P 2 m
W ∑ τj Gjm ( ∑ ∑ Gjm j0 m0 + σ )− Pjm G jm ∑ ∑ Gjm
j =1 m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1
∂L
∂Pjm = ln 2 · M N M N
m0 P m0 P
∑( ∑ Gjm 2
j0 m0 + σ + Pjm G jm )( ∑ ∑ Gjm 2
j0 m0 +σ )
0 0
m =1,m 6=m j =10 0 0
m =1,m 6=m j =1 0
M N M N
m0 P 2 m0
Gjm ( ∑ ∑ Gjm j0 m0 + σ )− Pjm G jm ∑ ∑ Gjm
µ j Wτj m0 =1,m0 6=m j0 =1 m0 =1,m0 6=m j0 =1 (20)
+ ln 2 · M N M N
m0 P m0 P
( ∑ ∑ Gjm 2
j0 m0 + σ + Pjm G jm )( ∑ ∑ Gjm j0 m0 +σ )
2
0 0 0
m =1,m 6=m j =1 m 0 = 1,m 0 6 = m j 0 = 1
!  
N M M
−e ζ − ∑ (1 − τj )η j ∑ Gjm Pm + νj (1 − τj )η j ∑ Gjm − ψ
j =1 m =1 m =1

Based on the Lagrangian derivative of the dual function, we give an updated formula
for the power distribution, expressed as follows:
!+
∂L
Pjm (t + 1) = Pjm (t) + β (t) (21)
∂Pjm

To ensure that the iteration values converge, we update the step size to satisfy
β ( t −1)
β(t) = t−1 . We use the subgradient method to update the Lagrange multipliers of
Sensors 2022, 22, 5035 11 of 21

the constraints [25,30]. The subgradient formulation of the dual function g(µ, ν, ψ) is given
by  
∂g Pjm Gjm
= τj W log2 1 +  − Rmin
 
∂µ j M N
m0 P
∑ ∑ Gjm j0 m0 +σ
2
m0 =1,m0 6=m j0 =1

∂g N M (22)
∂νj = ∑ (1 − τj )η j ∑ Gjm Pm − Emin
j =1 m =1
N
∂g
= max
Pm − ∑ Pjm
∂ψj
j =1

The Lagrange multiplier update equation is as follows:

δg +
 
µ j (t + 1) = µ j (t) + δ· δµ
j
δg +
 
νj (t + 1) = νj (t) + δ· δν (23)
j
δg +
 
ψ(t + 1) = ψ(t) + δ· δψ

where δ is the iteration step. 

3.2. Time Switching Problem

This section analyzes the time-switching scheme in detail. In the inner layer of the
algorithm, the transmit power is fixed and then the time switching coefficients are solved
by iteration. After fixing the transmit power P, the time switching coefficients τ are a set of
unknown variables to be solved, then the problem is reformulated as

N M
P2.2 : Max : EE(τ ) = ∑ R j / ∑ Gtotal
j =1 m =1 (24)
s.t. C1, C2, C6

To simplify the expression of the formula we make

 
Pjm Gjm
A = W log2 1 +
 
M N 
m0 P
∑ ∑ Gjm j0 m0 +σ
2
m0 =1,m0 6=m j0 =1
M N M (25)
B = ∑ (ζPm + PmC − Em ) − η j ∑ ∑ Gjm Pm
m =1 j =1 m =1
N M
C= η j ∑ ∑ Gjm Pm
j =1 m =1

It can be seen that the problem is a fractional programming problem with time coef-
ficients, and as above, we also use the Dinkelbach method for the fractional problem, for
which the problem is reformulated after the Dinkelbach treatment as

N N
P2.3 Max : F (τ ) = ∑ Aτj − λ( B + C ∑ τj )
j =1 j =1
(26)
s.t. Aτj ≥ Rmin
C (1 − τj ) ≥ Emin
Sensors 2022, 22, 5035 12 of 21

It is clear that the problem is a one-time programming problem with a one-time

switching coefficient τ, and it is only necessary to prove whether the function is monotonic
to find the resolution of the time switching factor. The derivative of the problem P2.3 is

∂F (τ )
= A−λC (27)
∂τ
If the derivative is positive, then the function is monotonically increasing, then the time
switching coefficient τ for its positive direction of the boundary value, if the negative, the
function is monotonically decreasing, then the minimum value of the boundary value for the
negative direction. The time switching coefficient discriminant is expressed as follows:
 n o
 max 0, 1 − Emin , i f A−λC ≥ 0
C
τ∗ n o (28)
 min 1, Rmin , i f A−λC < 0
A

We also consider a special scenario. When the terminal is close to the H-AP, the
throughput is almost similar. In this case, we can think that the time switching coefficients
are the same, and the computational complexity can be reduced. Finally, we obtain the
resolution of the power distribution and the time switching coefficients by iteration of the
alternating variables method. The two-level iterative algorithm for power allocation and
time switching is shown in Algorithm 1.

Algorithm 1. Two-layer iteration for power allocation and time switching

1: Input: transmit power P, time switching coefficient τ, Lagrange multiplier µ, ν, ψ, Maximum
number of iterations Tout , Update step β(t),δ(t), QoS threshold Rmin , threshold Emin number of
iterations t = 1, and ε;
2: Output: energy efficiency EE( P, τ )
3: for j = 1 : 1 : N
4: for 1 < t < Tout
5: update transmit power Pjm (t + 1) according to (21)
6: update Lagrange multiplier µ, ν, ψ according to (23)
7: for 1 < t < Tout
8: update time switching coefficient τ according to (28)
9: end for
10: if EE(t + 1)− EE(t + 1) ≤ ε exit loop;
11: break
12: end if
13: end for
14: end for

4. Energy Cooperation Programmer

In this section, we investigate IoT systems with energy cooperation scenarios. We
develop a many-to-many matching algorithm to solve the renewable energy dispatch
problem. As can be seen from the system diagram, energy cooperation is done through
aggregators in the smart grid, which act as intermediaries between the base stations and
the grid, so that the grid operator charges a fee for the energy exchange through the
aggregators, but it is lower than the actual fee, due to the fact that the renewable energy is
acquired by the base stations [31]. The issue of cost is beyond the scope of this paper. In
this paper, we only consider the scheduling of incoming energy from the perspective of
energy consumption and quality of service optimization.
When given two sets of variables using the transmit power P and time switching
coefficients τ, only one set of variables for the energy cooperation T needs to be solved.
The original optimization problem P1 can be reformulated as optimization problem P4,
which is formulated as follows:
P3 : maxEE( T )
(29)
s.t. C4, C5
Sensors 2022, 22, 5035 13 of 21

The energy cooperation subproblem is an integer combinatorial optimization problem.

In this paper, we consider the use of many-to-many matching theory to solve it. Matching
theory is one of the effective solution tools for studying decentralized resource allocation
and can transform the resource allocation problem into a simple distributed problem. In
existing studies, the many-to-many matching theory has been used to solve optimization
problems related to wireless networks [15,32,33], demonstrating that matching theory has
the characteristics of fast convergence and stable configuration results.
We adopt the allocation scheme based on the matching theory in [15,32]. We extend
the scheme to the IoT network and adjust the preference according to the characteristics of
the IoT network. According to the matching theory allocation scheme,
 RE the H-APs are first
divided into two sets of categories, where the set M+ = m+ ∈ M Em

− PW > 0 denotes
the set of H-APs that have excess energy  while satisfying their own
power consumption,
both in terms of energy output. M− = m− ∈ M  Em
 RE
− PW < 0 indicates a collection of
H-APs that do not collect enough energy to sustain their own power consumption. Based
on the principle of matching bilateral benefits [15], the utility functions (preference degrees)
corresponding to each other between the two types of H-APs are established, and setting a
suitable preference degree function can effectively reduce the loss of collected energy and
improve the energy utilization of the system. The matching diagram of its two types of
collections. Each H-AP in set M+ corresponds to all H-APs in set M− and has a preference
list corresponding to it. Similarly, the base stations in set M− have their own preference for
the H-APs in set M+ . The premise of matching is to first match the base station with the
corresponding base station according to its preference to complete the energy cooperation.
In the process of energy cooperation, the transmission efficiency is mainly related
to the resistance value of the power line, the greater the resistance value, the greater the
energy loss, the loss of energy is expressed as follows:

Eloss = I 2 R(l ) (30)

where I is the current in the transmission line, R(l ) is the total resistance of the power line,
R(l ) = ρl, ρ is the resistance factor and l is the length of the power line [34]. It can be seen
that the lost energy is positively related to the length of the power line. The transmission
efficiency αmm0 from H-AP m+ to H-AP m− is expressed as

Tmm0 − Eloss
αmm0 = (31)
Tmm0

where Tmm0 indicates the renewable energy allocated by H-AP m+ to H-AP m− . There is
a difference in transmission efficiency αmm0 between the two H-APs due to the different
lengths of the power lines. Set M+ prefers to transfer energy to a base station with higher
transmission efficiency to reduce energy losses, so the preference of H-APs in Set M+ over
those in Set M− is expressed as

p( M+ , M− ) = αmm0 , m− ∈ M− (32)

When H-AP m− sends an energy request to the H-APs in Set M+ , the H-APs in set
M+ will select the H-APs with the highest ranking according to the preference ranking in
(32) and accept its request, passing the energy to H-APs m− .
H-Aps M− within the set prefer H-APs with more energy remaining in set M+ , as this
reduces the number of passes and responses from the H-AP and the H-AP is able to obtain
energy faster. The preference of the H-APs in set M− for the H-APs in set M+ is expressed as

p( M− , M+ ) = Em
RE
− ζPm − PmC , m+ ∈ M+ (33)

H-AP m− is ranked according to the preference of (33) and H-AP m− selects the H-AP
with the highest preference in set m+ .
Sensors 2022, 22, 5035 14 of 21

Based on the previous analysis, we propose a solution of joint power allocation,

time switching and energy cooperation. The specific solution process is summarized in
Algorithm 2.
Algorithm 2. Joint power allocation, time switching and energy cooperation algorithm
0 : Input: transmit power P, time switching coefficient τ, Lagrange multiplier µ, ν, ψ, maximum
number of iterations Tout , update step β(t), δ(t), QoS threshold Rmin , threshold Emin number of
iterations t = 1, and ε;
1 : Output: energy efficiency EE( P, τ, T )
2 : According to Algorithm 1, obtain transmit power P, time switching τ solution
3 : for m = 1 : 1 : M
4 : Calculate the preference of all the H − AP in set M+ according to (32), and rank them
5 : Calculate the preference of all the H − AP in set M− according to (33), and rank them
6 : H − AP m+ selects the H − AP with the largest preference in set M− to complete the energy
cooperation.
7 : if the set M− or M+ is the empty set, exit loop;
8: break
9: end if
10: end for

Complexity and Convergence Analysis

The joint algorithm consists of a two-stage algorithm. Based on the computational
complexity of the Dinkelbach method O( ε12 log(K )) [25]. Both the inner and outer iteration
processes use the Dinkelbach method. Thus, the computational complexity of the proposed
two-layer iterative algorithm is approximately O( ε12 ε12 log(K )). The complexity of the
1 2
matching algorithm is related to the length of the two sets, with a complexity of O( M1 M2 ),
the complexity of the joint algorithm is O( ε12 ε12 log(K ) + ( M1 M2 )).
1 2

5. Performance Analysis
This section verifies the effectiveness of the algorithm through simulation. It is as-
sumed that there are 5 H-APs in the network. The cell range of the H-APs is 100 × 100 m2 ,
the terminals N = 10 is uniformly distributed within a 10m radius of the H-AP with a termi-
nal RF energy conversion rate of 0.5 [25]. The communication rate threshold is 10 Mbit/s,
terminal energy collection thresholds is 0.01 mW. The channel fading model contains
Rayleigh fading and path loss, the channel gain is denoted as d− β g, where d− β is the path
loss, β = 2, and g is the small-scale fading, generated by the Rayleigh distribution, with a
mean difference of 0 and a variance of 1 [19]. The static power consumption is 6 W and the
energy transfer efficiency is rand (0.7–0.9) [12]. The joint optimization problem solved by
the proposed algorithm following algorithms to compare the performance of each aspect:
the rate-maximization (Max-rate) algorithm, which optimizes the power allocation and
time switching [25]. The wireless portable energy resource optimization algorithm, which
optimizes power allocation and time switching using particle swarm optimization (PSO)
algorithms [26]. The energy cooperation part adopts the matching algorithm in this paper.
The simulation parameters are shown in Table 1.

Table 1. System Parameters.

Parameter Value
System bandwidth 10 MHz
Noise power density −174 dBm/Hz
Max transmit power of H-AP 30 dBm
transmit power generation factor 1
ζ = 38%
H-AP Energy collection 4–10 W
 System bandwidth 10 MHz
Noise power density −174 dBm/Hz
Max transmit power of H-AP 30 dBm
1
transmit power generation factor  =
Sensors 2022, 22, 5035 38% 15 of 21
H-AP Energy collection 4–10 W

Figure
Figure 2 shows
2 shows thetheiterative convergence
iterative convergence of of
thethe algorithm.
algorithm.It can
It canbe be
seen
seenfrom
from thethe
figure that the algorithm proposed in this paper is the fastest in terms
figure that the algorithm proposed in this paper is the fastest in terms of convergence of convergence
speed
speedandandalso obtains
also thethe
obtains highest
highestenergy
energy efficiency
efficiency performance.
performance. AsAscan bebe
can seen from
seen from thethe
algorithm flow diagram, the algorithm in this paper is given certain initial
algorithm flow diagram, the algorithm in this paper is given certain initial values in the values in the
setting
settingof of
parameters,
parameters, such
suchasasenergy
energyefficiency
efficiencyandandfiring
firingpower,
power, such
such that
that the initial values
the initial val-
ues willreduce
will reducethe thenumber
numberof ofiterations
iterations to
to some
some extent.
extent. The Theparticle
particleswarm
swarm algorithm,
algorithm, onon
thethe
other hand, searches from a global resolution, so the search resolution
other hand, searches from a global resolution, so the search resolution requires a large requires a large
number
number ofofiterations
iterationstotocomplete,
complete, and therefore iterative
and therefore iterativeconvergence
convergencebecomes
becomes slower.
slower. This
This suggests that, for this system, our proposed algorithm, has some
suggests that, for this system, our proposed algorithm, has some performance advantage performance ad-
vantage
in termsin terms of convergence
of convergence speed.speed.

Figure 2. Convergence
Figure performance
2. Convergence of of
performance different algorithm.
different algorithm.

Figure
Figure 3 shows
3 shows thethecurve
curveofofthe
thenumber
numberof ofterminals
terminalsversus
versus energy
energy efficiency.
efficiency. AsAs can
canbe
beseen
seenfrom
fromthethefigure,
figure,ourourproposed
proposed algorithm
algorithm achieves
achieveshigher
higherenergy
energyefficiency compared
efficiency com-
to the
pared PSO
to the PSOalgorithm
algorithm and the
and the maximum
maximumrate ratealgorithm. Thisisisbecause
algorithm. This becauseour our algorithm
algorithm
achieves a higher resolution of power allocation during the convergence
achieves a higher resolution of power allocation during the convergence iterations, which iterations, which
effectively suppresses the co-channel interference to its users, which
effectively suppresses the co-channel interference to its users, which results in a higher results in a higher
throughput
throughput forfor
thethe users,
users, andand at the
at the same
same time,
time, thethe system
system consumes
consumes lessless transmit
transmit power.
power.
The particle swarm algorithm, on the other hand, tends to fall into localized
The particle swarm algorithm, on the other hand, tends to fall into localized resolution and resolution and
does
does notnot have
have high
high search
search accuracy.
accuracy. TheThe maximum
maximum rate
rate algorithm,
algorithm, although
although ableable
to to obtain
obtain
higher rates, consumes more transmit power, resulting in a less energy efficient system. On
the other hand, it can be seen from the figure that the use of energy cooperation techniques
can improve the energy efficiency of the system compared to scenarios where no energy
cooperation techniques are used, because the excess renewable energy is fully utilized and
the consumption of the grid is reduced. As can be seen from the figure, the algorithm of
this paper can be effectively applied to a multi-terminal scenario.
Figure 4 shows the curve of the effect of the number of H-APs on energy consump-
tion. From the figure, it can be seen that as the number of H-APs increases, the energy
consumption of the system also increases, which is due to the increase in static power
consumption of the H-APs. The algorithm in this paper and the PSO algorithm both use
energy cooperation technology, which makes full use of renewable energy to reduce the
system energy consumption. H-APs that do not use energy cooperation will consume
more energy. The maximum rate algorithm has the highest energy consumption because,
in order to obtain a higher throughput, the transmit power is high, which leads to more
energy consumption of the system on the grid. The algorithm proposed in this paper
consumes less energy than the PSO algorithm because the algorithm achieves a resolution
that is closer to the optimal solution through multiple iterations and consumes less transmit
power. As a result, the algorithm is more suitable for multi-H-AP scenarios.
 higher rates, consumes more transmit power, resulting in a less energy efficient system. On
the other hand, it can be seen from the figure that the use of energy cooperation techniques
can improve the energy efficiency of the system compared to scenarios where no energy
cooperation techniques are used, because the excess renewable energy is fully utilized and
Sensors 2022, 22, 5035 the consumption of the grid is reduced. As can be seen from the figure, the algorithm of16
this
of 21
paper can be effectively applied to a multi-terminal scenario.

Figure 3. Energy efficiency versus number of terminals for different algorithms.

Figure 4 shows the curve of the effect of the number of H-APs on energy consump-
tion. From the figure, it can be seen that as the number of H-APs increases, the energy
consumption of the system also increases, which is due to the increase in static power
consumption of the H-APs. The algorithm in this paper and the PSO algorithm both use
energy cooperation technology, which makes full use of renewable energy to reduce the
system energy consumption. H-APs that do not use energy cooperation will consume
more energy. The maximum rate algorithm has the highest energy consumption because,
in order to obtain a higher throughput, the transmit power is high, which leads to more
energy consumption of the system on the grid. The algorithm proposed in this paper con-
sumes less energy than the PSO algorithm because the algorithm achieves a resolution
that is closer to the optimal solution through multiple iterations and consumes less trans-
mit power. As a result, the algorithm is more suitable for multi-H-AP scenarios.
Figure 3. 3.
Figure Energy efficiency
Energy versus
efficiency number
versus ofof
number terminals forfor
terminals different algorithms.
different algorithms.

Figure 4 shows the curve of the effect of the number of H-APs on energy consump-
tion. From the figure, it can be seen that as the number of H-APs increases, the energy
consumption of the system also increases, which is due to the increase in static power
consumption of the H-APs. The algorithm in this paper and the PSO algorithm both use
energy cooperation technology, which makes full use of renewable energy to reduce the
system energy consumption. H-APs that do not use energy cooperation will consume
more energy. The maximum rate algorithm has the highest energy consumption because,
in order to obtain a higher throughput, the transmit power is high, which leads to more
energy consumption of the system on the grid. The algorithm proposed in this paper con-
sumes less energy than the PSO algorithm because the algorithm achieves a resolution
that is closer to the optimal solution through multiple iterations and consumes less trans-
mit power. As a result, the algorithm is more suitable for multi-H-AP scenarios.

Figure
Figure 4. System
4. System energy
energy consumption
consumption versus
versus thethe number
number of of H-APs
H-APs forfor different
different algorithms.
algorithms.

Figure 5 shows the relationship between the number of H-APs and energy efficiency.
As can be seen from the figure, the energy efficiency decreases as the number of H-APs
increases. This is due to the fact that the standby energy consumption of the system
increases as the number of H-APs increases, resulting in a decrease in energy efficiency.
On the other hand, the algorithm proposed in this paper outperforms other algorithms
in terms of energy efficiency. This is because our proposed algorithm achieves higher
throughput and lower energy consumption in the power allocation problem, and therefore
higher energy efficiency performance, which indicates that our proposed algorithm can be
applied to multi-H-APs scenarios.
Figure 6 shows the effect of QoS on the energy efficiency of the system. The graph
shows that as the QoS increases, the energy efficiency of the system decreases. This is
because, in order to ensure the QoS constraint, the H-APs needs to increase the transmitting
power to ensure the QoS demanded by terminals with poor channel conditions, which
Figure 4. System energy consumption versus the number of H-APs for different algorithms.
increases the energy consumption of the system and consequently decreases the energy
efficiency. The graph shows that the energy efficiency of the system decreases more slowly
when the QoS is increased compared to other algorithms, which indicates that this paper
has a higher resolution accuracy in solving the power allocation and therefore achieves a
higher energy efficiency than other algorithms.
 increases. This is due to the fact that the standby energy consumption of the system in-
creases as the number of H-APs increases, resulting in a decrease in energy efficiency. On
the other hand, the algorithm proposed in this paper outperforms other algorithms in
terms of energy efficiency. This is because our proposed algorithm achieves higher
throughput and lower energy consumption in the power allocation problem, and there-
Sensors 2022, 22, 5035 fore higher energy efficiency performance, which indicates that our proposed algorithm17 of 21
can be applied to multi-H-APs scenarios.

Figure 5. energy efficiency versus the number of H-APs for different algorithms.

Figure 6 shows the effect of QoS on the energy efficiency of the system. The graph
shows that as the QoS increases, the energy efficiency of the system decreases. This is
because, in order to ensure the QoS constraint, the H-APs needs to increase the transmit-
ting power to ensure the QoS demanded by terminals with poor channel conditions,
which increases the energy consumption of the system and consequently decreases the
energy efficiency. The graph shows that the energy efficiency of the system decreases
more slowly when the QoS is increased compared to other algorithms, which indicates
that this paper has a higher resolution accuracy in solving the power allocation and there-
fore achieves a higher energy efficiency than other algorithms.
Figure
Figure5.5.energy efficiency
Energy versus
efficiency the
versus number
the ofof
number H-APs for
H-APs different
for differentalgorithms.
algorithms.

Figure 6 shows the effect of QoS on the energy efficiency of the system. The graph
shows that as the QoS increases, the energy efficiency of the system decreases. This is
because, in order to ensure the QoS constraint, the H-APs needs to increase the transmit-
ting power to ensure the QoS demanded by terminals with poor channel conditions,
which increases the energy consumption of the system and consequently decreases the
energy efficiency. The graph shows that the energy efficiency of the system decreases
more slowly when the QoS is increased compared to other algorithms, which indicates
that this paper has a higher resolution accuracy in solving the power allocation and there-
fore achieves a higher energy efficiency than other algorithms.

Figure 6.6.
Figure energy efficiency
Energy versus
efficiency the
versus QoS
the forfor
QoS different algorithms.
different algorithms.

Figure 7 shows the curve of the effect of energy harvesting on the energy efficiency of
the system. As can be seen from the figure, the energy efficiency of the system decreases
as the energy harvesting constraint increases. This is because, in order to satisfy the
energy harvesting constraint, on the one hand, the H-APs needs to transmit more power to
satisfy the constraint, which increases the energy consumption of the system and leads to a
decrease in energy efficiency. On the other hand, in order to satisfy the energy collection
constraint, the time for the energy collection part will increase, while the time for the
message decoding will decrease accordingly, which will increase the amount of energy
collected by the terminal, but at the same time will decrease the throughput of the terminal.
Figure 6. energy efficiency
The introduction versus the
of the SWIPT QoS for different
technology, algorithms.the standby time of the terminal,
which increases
the effect of this technology on the energy efficiency of the system is negligible, is very
promising. There is a trade-off between throughput and terminal standby time, and in
practical scenarios the values can be set according to the different terminal categories.
Figure 8 gives the curve of the effect of the number of H-APs on the energy consump-
tion of the system. As can be seen from the graph, as the number of H-APs increases, the
energy consumption of the system also increases. This is because the energy consumption
increases as the standby power consumption of the H-APs increases, and although energy
harvesting techniques are introduced, they are not yet able to balance the standby energy
consumption. In this section, we mainly show the performance comparison between our
proposed energy cooperation algorithm and the comparison algorithm DES [15]. It can
 Figure 7 shows the curve of the effect of energy harvesting on the energy efficiency
of the system. As can be seen from the figure, the energy efficiency of the system decreases
Sensors 2022, 22, 5035 as the energy harvesting constraint increases. This is because, in order to satisfy the energy 18 of 21
harvesting constraint, on the one hand, the H-APs needs to transmit more power to satisfy
the constraint, which increases the energy consumption of the system and leads to a de-
crease
be seeninfrom
energy efficiency.
the figure On algorithm
that our the other hand,
performsin order
bettertodue
satisfy the
to the energy
impact of collection
the trans-
mission efficiency that we prioritize in setting the matching preference, which,for
constraint, the time for the energy collection part will increase, while the time to athe mes-
certain
sage decoding will decrease accordingly, which will increase the amount
extent, reduces the energy loss during transmission. On the other hand, the graph shows of energy col-
lected by the terminal, but at the same time will decrease the throughput
that the two sets of algorithms that introduce energy cooperation significantly outperformof the terminal.
Thescenario
the introduction of the
without SWIPTcooperation
energy technology,in which
termsincreases
of energy theconsumption.
standby time of theenergy
The termi-
nal, the effecttechnique
cooperation of this technology
makes fullon theofenergy
use efficiency
renewable energy ofsources
the system
andisavoids
negligible, is very
the waste of
promising. There is a trade-off between throughput and terminal standby
excess energy. Therefore, the introduction of energy cooperation techniques has a positive time, and in
practical scenarios the values can be set according
effect on the energy consumption of the communication system. to the different terminal categories.

Sensors 2022, 22, x FOR PEER REVIEW 19 of 21

algorithm with
Figure 7. The performance of the proposed algorithm with different
different collection
collection threshold.
threshold.

Figure 8 gives the curve of the effect of the number of H-APs on the energy consump-
tion of the system. As can be seen from the graph, as the number of H-APs increases, the
energy consumption of the system also increases. This is because the energy consumption
increases as the standby power consumption of the H-APs increases, and although energy
harvesting techniques are introduced, they are not yet able to balance the standby energy
consumption. In this section, we mainly show the performance comparison between our
proposed energy cooperation algorithm and the comparison algorithm DES [15]. It can be
seen from the figure that our algorithm performs better due to the impact of the transmis-
sion efficiency that we prioritize in setting the matching preference, which, to a certain
extent, reduces the energy loss during transmission. On the other hand, the graph shows
that the two sets of algorithms that introduce energy cooperation significantly outperform
the scenario without energy cooperation in terms of energy consumption. The energy co-
operation technique makes full use of renewable energy sources and avoids the waste of
excess energy. Therefore, the introduction of energy cooperation techniques has a positive
effect on the energy consumption of the communication system.

Figure
Figure 8. System
8. System energy
energy consumption
consumption versus
versus thethe number
number of of H-AP
H-AP forfor different
different algorithms.
algorithms.

Figure
Figure 9 9shows
showsthe theeffect
effectofof the
the number
number ofofterminals
terminalson onthe
theenergy
energycollected. It isIt clear
collected. is
from the figure that as the number of terminals increases, the amount
clear from the figure that as the number of terminals increases, the amount of energy col- of energy collected
also also
lected increases. It is clear
increases. that based
It is clear on the
that based onPSO
the algorithm
PSO algorithmits obtained energy
its obtained collection
energy col- is
significantly better than the algorithm proposed in this paper, this is because
lection is significantly better than the algorithm proposed in this paper, this is because the the particle
swarmswarm
particle algorithm has a larger
algorithm has a value
largerofvalue
powerof allocation, althoughalthough
power allocation, the impact
theon the energy
impact on
consumption and energy efficiency of the H-APs is negative, this
the energy consumption and energy efficiency of the H-APs is negative, this increases increases the amount
the
of energy collected, which is positive for the standby time of the terminals. The present
amount of energy collected, which is positive for the standby time of the terminals. The
algorithm, on the other hand, mainly optimizes the energy efficiency of the system and
present algorithm, on the other hand, mainly optimizes the energy efficiency of the system
therefore obtains a smaller resolution of the power allocation, which is able to suppress
and therefore obtains a smaller resolution of the power allocation, which is able to sup-
co-channel interference and increase the throughput, but the amount of energy collected
press co-channel interference and increase the throughput, but the amount of energy col-
lected by the terminal is then reduced. Additionally, for different systems under the sys-
tem, it is necessary to weigh the H-AP power consumption and terminal standby time and
set the energy collection constraint of the terminal for different needs.
 lection is significantly better than the algorithm proposed in this paper, this is because the
particle swarm algorithm has a larger value of power allocation, although the impact on
the energy consumption and energy efficiency of the H-APs is negative, this increases the
amount of energy collected, which is positive for the standby time of the terminals. The
Sensors 2022, 22, 5035 present algorithm, on the other hand, mainly optimizes the energy efficiency of the system 19 of 21
and therefore obtains a smaller resolution of the power allocation, which is able to sup-
press co-channel interference and increase the throughput, but the amount of energy col-
lected
by bythethe terminal
terminal is then
is then reduced.
reduced. Additionally,
Additionally, for different
for different systems
systems under
under the sys-it is
the system,
tem,necessary
it is necessary
to weigh the H-AP power consumption and terminal standby time andand
to weigh the H-AP power consumption and terminal standby time set the
set the energy collection
energy collection constraint
constraint of the
of the terminal
terminal for different
for different needs.
needs.

Figure
Figure 9. Terminal
9. Terminal collects
collects energy
energy versus
versus the the number
number of terminals
of terminals for different
for different algorithms.
algorithms.

6. Conclusions
In this paper, we study the SWIPT-enabled IoT network with energy cooperation. We
develop a mathematical model with energy efficiency as the optimization objective, while
needing to satisfy quality of service and minimum energy harvesting constraints. The
problem is nonlinear and difficult to solve directly. We propose an iterative algorithm to
solve the problem of power allocation, time switching and energy cooperation. Simulation
results show that. Our proposed algorithm outperforms the comparison algorithm in terms
of energy efficiency performance. Moreover, this algorithm has good performance for multi
H-APs and multi terminal scenarios. In addition, simulation shows that SWIPT technology
can effectively extend the operation cycle of the terminal, and the energy cooperation
technology can effectively reduce the system energy consumption, which is positive for the
development of green communication.
Our algorithm can be extended to other networks with energy collection, especially
low-power terminal device networks, such as the current research focus on 5G networks
with NOMA, or heterogeneous network systems, where the energy efficiency of the system
can be effectively improved using our proposed algorithm for systems with multiple H-APs.
In future work, there is still room for improvement. In this paper, we consider the case of
having perfect channel conditions; according to the literature [34], the case of imperfect
channel conditions information can cause interruptions and rate degradation, for such
problems still need further analysis and proposed solutions. On the other hand, with the
rapid development of smart grids and energy cooperation technology involving the trading
of energy in order to weigh the interests of energy intermediaries and communication
operators, further research is needed to solve it.

Author Contributions: Conceptualization, Y.C., Y.Z. and C.P.; methodology, Y.Z. and C.P.; software,
Y.Z.; validation, Y.C.; formal analysis, Y.C.; data curation, Y.Z.; writing—original draft preparation,
Y.Z.; writing—review and editing, Y.C., Y.Z., C.P., X.P. and S.P.; visualization, Y.C. All authors have
read and agreed to the published version of the manuscript.
Funding: This work was partially supported by The Science and Technology Bureau Project of
Chongqing (No. cstc2019jcyj-msxmX0233, No. cstc2017shmsA40019), The Education Commission
Project of Chongqing (No. KJQN201901125), The Science and Technology Research Program of
Sensors 2022, 22, 5035 20 of 21

Chongqing Municipal Education Commission (No. KJQN201901103), Innovation Research Group

of Universities in Chongqing (No. CXQT21035), The Bureau of Science and Technology Project
of Banan District (No. 2019TJ07), Research Start-up Fund of Chongqing University of Technology
(2019ZD127), science and technology research youth project of Chongqing Education Commission
(Kjqn202101124), and achievements of the action plan for high quality development of graduate
education of Chongqing University of Technology (gzlcx20222024, gzlcx20223090).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Fedullo, T.; Morato, A.; Tramarin, F.; Rovati, L.; Vitturi, S. A Comprehensive Review on Time Sensitive Networks with a Special
Focus on Its Applicability to Industrial Smart and Distributed Measurement Systems. Sensors 2022, 22, 1638. [CrossRef] [PubMed]
2. Stankovic, A.J. Research directions for the Internet of Things. IEEE Internet Things J. 2014, 1, 3–9. [CrossRef]
3. Cisco Annual Internet Report (2018–2023) White Paper. Available online: https://www.cisco.com/c/en/us/solutions/collateral/
executiveperspectives/annual-internet-report/white-paper-c11-741490.html (accessed on 15 May 2022).
4. Guo, H.; Zhang, J.; Liu, J.; Zhang, H. Energy-aware computation offloading and transmit power allocation in ultradense IoT
networks. IEEE Internet Things J. 2019, 6, 4317–4329. [CrossRef]
5. Niu, Z.; Zhou, S.; Sun, Y. Green communication and networking for Carbon-peaking and Carbon-neutrality: Challenges and
solutions. J. Commun. 2022, 43, 1–14.
6. Huawei. Mobile Networks Go Green. Available online: https://www.huawei.com/en/news/2016/8/huawei-sustainability-
report676 (accessed on 15 April 2022).
7. Xu, B.; Chen, Y.; Carrión, R.J.; Loo, J.; Vinel, A. Energy aware power control in energy cooperation aided millimeter wave cellular
networks with renewable energy resources. IEEE Access. 2017, 5, 432–442. [CrossRef]
8. Zhang, H.; Huang, S.; Jiang, C.; Long, K.; Leung, M.C.; Poor, V.H. Energy efficient user association and power allocation
in millimeter-wave-based ultra-dense networks with energy harvesting base stations. IEEE J. Sel. Areas Commun. 2017, 35,
1936–1947. [CrossRef]
9. Xu, B.; Zhu, P.; Li, J.; Wang, D.; You, X. Joint Long-Term Energy Efficiency Optimization in C-RAN With Hybrid Energy Supply.
IEEE Trans. Veh. Technol. 2020, 69, 11128–11138. [CrossRef]
10. Alqasir, A.; Kamal, A.E. Cooperative Small Cell HetNets with Dynamic Sleeping and Energy Harvesting. IEEE Trans. Green
Commun. Netw. 2020, 4, 774–782. [CrossRef]
11. Rahbar, K.; Chai, C.C.; Zhang, R. Energy cooperation optimization inmicrogrids with renewable energy integration. IEEE Trans.
Smart Grid. 2018, 9, 1482–1493. [CrossRef]
12. Xu, B.; Chen, Y.; Carrion, J.R.; Zhang, T. Resource allocation in energy-cooperation enabled two-tier NOMA HetNets toward
green 5G. IEEE J. Sel. Areas Commun. 2017, 35, 2758–2770. [CrossRef]
13. Sheng, M.; Zhai, D.; Wang, X.; Li, Y.; Shi, Y.; Li, J. Intelligent Energy and Traffic Coordination for Green Cellular Networks with
Hybrid Energy Supply. IEEE Trans. Veh. Technol. 2017, 66, 1631–1646. [CrossRef]
14. Li, Y.; Zhao, X.; Liang, H. Throughput Maximization by Deep Reinforcement Learning with Energy Cooperation for Renewable
Ultradense IoT Networks. IEEE Internet Things J. 2020, 7, 9091–9102. [CrossRef]
15. Yin, F.; Zeng, M.; Zhang, Z.; Liu, D. Coded Caching for Smart Grid Enabled HetNets with Resource Allocation and Energy
Cooperation. IEEE Trans. Veh. Technol. 2020, 69, 12058–12071. [CrossRef]
16. Han, D.; Li, S.; Chen, Z. Hybrid Energy Ratio Allocation Algorithm in a Multi-Base-Station Collaboration System. IEEE Access
2019, 7, 147001–147009. [CrossRef]
17. Guntupalli, L.; Gidlund, M.; Li, Y.F. An on-demand energy requesting scheme for wireless energy harvesting powered IoT
networks. IEEE Internet Things J. 2018, 5, 2868–2879. [CrossRef]
18. Feng, Y.; Cao, Y. Achievable Rate Maximization for Multi-Relay AF Cooperative SWIPT Systems with a Nonlinear EH Model.
Sensors 2022, 22, 3041. [CrossRef]
19. Zhang, R.; Xiong, K.; Guo, W.; Yang, X.; Fan, P.; Letaief, K.B. Q-Learning-Based Adaptive Power Control in Wireless RF Energy
Harvesting Heterogeneous Networks. IEEE Syst. J. 2021, 15, 1861–1872. [CrossRef]
20. Lu, W.; Si, P.; Huang, G.; Han, H. SWIPT cooperative spectrum sharing for 6G-enabled cognitive IoT network. IEEE Internet
Things J. 2021, 8, 15070–15080. [CrossRef]
21. Mao, S.; Leng, S.; Hu, J.; Yang, K. Power Minimization Resource Allocation for Underlay MISO-NOMA SWIPT Systems. IEEE
Access 2019, 7, 17247–17255. [CrossRef]
22. Sun, W.; Song, Q.; Zhao, J. Adaptive Resource Allocation in SWIPT-Enabled Cognitive IoT Networks. IEEE Internet Things J. 2022,
9, 535–545. [CrossRef]
Sensors 2022, 22, 5035 21 of 21

23. Masood, Z.; Park, H.; Jang, H.S. Optimal Power Allocation for Maximizing Energy Efficiency in DAS-Based IoT Network. IEEE
Syst. J. 2021, 15, 2342–2348. [CrossRef]
24. Tang, J.; Yu, Y.; Liu, M. Joint Power Allocation and Splitting Control for SWIPT-Enabled NOMA Systems. IEEE Trans. Wirel.
Commun. 2020, 19, 120–133. [CrossRef]
25. Tang, J.; Luo, J.; Liu, M. Energy Efficiency Optimization for NOMA With SWIPT. IEEE J. Sel. Top. Signal Processing 2019, 13,
452–466. [CrossRef]
26. Acosta, M.; Moreta, C.; Koo, I. Joint Power Allocation and Power Splitting for MISO-RSMA Cognitive Radio Systems with SWIPT
and Information Decoder Users. IEEE Syst. J. 2020, 15, 5289–5300. [CrossRef]
27. Dinkelbach, W. On nonlinear fractional programming. Manag. Sci. 1967, 13, 492–498. [CrossRef]
28. Schaible, S. Fractional Programming I, Duality. Manag. Sci. 1976, 22, 858–867. [CrossRef]
29. Fan, F.; Ye, G.; Zhang, H.; Cheng, J.; Leung, M.C.V. Energy-Efficient Joint User Association and Power Allocation in a Heteroge-
neous Network. IEEE Trans. Wirel. Commun. 2020, 19, 7008–7020.
30. Fang, F.; Zhang, H.; Cheng, J.; Roy, S.; Leung, M.C.V. Joint user scheduling and power allocation optimization for energy-efficient
NOMA systems with imperfect CSI. IEEE J. Sel. Areas Commun. 2017, 35, 2874–2885. [CrossRef]
31. Reyhanian, N.; Maham, B.; Mansouri, S.V.; Tushar, W.; Yuen, C. Game-theoretic approaches for energy cooperation in energy
harvesting small cell networks. IEEE Trans. Veh. Technol. 2017, 66, 7178–7194. [CrossRef]
32. Gu, Y.; Saad, W.; Bennis, M.; Debbah, M.; Han, Z. Matching theory for future wireless networks: Fundamentals and applications.
IEEE Commun. Mag. 2015, 53, 52–59. [CrossRef]
33. Zhao, J.; Liu, Y.; Chai, K.K.; Chen, Y.; Elkashlan, M. Many-to-many matching with externalities for device-to-device communica-
tions. IEEE Wireless Commun. Lett. 2017, 6, 138–141. [CrossRef]
34. Yang, Z.; Ding, Z.; Fan, P.; Karagiannidis, K.G. On the performance of non-orthogonal multiple access systems with partial
channel information. IEEE Trans. Commun. 2016, 64, 654–667. [CrossRef]