This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

Energy Efficient Data and Energy Integrated

Management Strategy for IoT Devices Based on
RF Energy Harvesting
Yang Wang, Kun Yang, Senior Member, IEEE, Weixiang Wan, Yitian Zhang and Qiang Liu, Member, IEEE

Abstract—Considering not only reducing the energy consump-

tion based on the fixed energy source but also capturing new
energy by RF energy harvesting for Internet of Thing (IoT)
T HE IoT devices have been massively deployed nowa-
days. However, mainly limited with the relatively small
battery capacity and the high energy consumption, how to
devices, an integrated energy efficient strategy for IoT devices
is proposed in this paper. On one hand, a node sampling save energy and prolong the battery life has become more
scheduling algorithm based on matrix completion is designed. and more important. In order to solve the problem of energy
All the sampling data is handled as matrix elements. The basic consumption of IoT devices, especially for low-power devices
idea is to reduce the sampling data and then reconstruct the which are commonly battery-powered, the conventional way
complete data set with the technology of matrix completion by is to reduce the energy consumption with various energy-
using the spatial-temporal correlation of all the sampling data.
Thus, the IoT nodes can keep dormant more instead of working saving algorithms. In recent years, some new ideas have
on data sampling. The energy consumption can be significantly emerged. Apart from reducing the energy consumption, many
reduced with very little data loss. On the other hand, an adaptive researchers try to obtain new energy sources by means of en-
RF energy management strategy is introduced. Based on the self- ergy harvesting. For example, conversions from piezoelectric
designed Data and Energy Integrated Network (DEIN) system, energy, solar energy, electromagnetic energy and other ways
both data and energy are transferred between the DEIN Gateway
(DEING) and the DEIN Node (DEINN). With the adaptive energy have been proved effective to power some low-power IoT
management strategy, they both automatically switch between devices [1]. Additionally, another very important way is to
the wireless information transfer (WIT) mode and the Wireless convert the harvested RF energy into the electrical energy that
Energy Transfer (WET) mode. Combining energy saving with can be utilized by IoT devices, which is called Wireless Energy
energy harvesting, the energy efficiency can be greatly enhanced. Transfer (WET) [1].
The proposed integrated solution aims to decrease the energy
consumption of IoT devices and provide them with constant Compared with the energy-saving strategies from the per-
new energy by RF energy harvesting. Thus, their battery lives spective of network levels, such as energy-efficient routing
can be prolonged. Both the effectiveness and the efficiency of schemes, topology control, MAC protocols, it is more intuitive
the proposed integrated strategy have been validated with the to reduce the amount of sampling data and keep the IoT nodes
simulation. dormant as long as possible. As an effective data reduction
Index Terms—energy harvesting, matrix completion, data sam- scheme, the compressed sensing scheme has been widely
pling, DEIN (Data and Energy Integrated Network), energy used in IoT networks to solve energy-saving problems, of
efficiency, IoT which the core is to reconstruct data with as little sampling
data as possible based on the sparsity of sampling data [2].
I. INTRODUCTION Matrix completion, as an extension of the compression sensing
technology, has become an important research field in IoT
networks in recent years. Matrix completion technology is to
Manuscript received July 13, 2020; revised February 28, 2021; accept- process the data collected by nodes as a matrix [3]. Only a set
ed March 16, 2021. This work was supported in part by the National of data needs to be collected, then the complete data matrix
Natural Science Foundation of China (No.61620106011, No.U1705263 and
No.61731006) and the Special Science Foundation of Quzhou (Grant No.: can be accurately reconstructed at the gateway by using the
2020D002). spatial-temporal correlation of the sampling data, the sparse
Y. Wang is with Yangtze Delta Region Institute (Huzhou) & School of Infor- distribution of nodes and the feature of low rank. Based on
mation and Communication Engineering, University of Electronic Science and
Technology of China, Huzhou 313001, China. (e-mail: [email protected]) that, this strategy explores how to increase the sleeping time
K. Yang and Y. Zhang are with School of Computer Science and Electronic of nodes with data reduction and working-slot reduction, but
Engineering, University of Essex, Essex CO4 3SQ, UK. K. Yang is also with without compromising the data accuracy too much.
Yangtze Delta Region Institute (Quzhou), University of Electronic Science and
Technology of China, Quzhou 324000, China. (e-mail: [email protected]; In addition, from the perspective of energy harvesting, this
[email protected]) paper proposes an adaptive RF energy management strategy
W. Wan and Q. Liu are with School of Information and Communication based on the integration of data and energy. The wireless
Engineering, University of Electronic Science and Technology of China,
Chengdu 611731, China. Q. Liu is also with Yangtze Delta Region Institute charging process is divided into two phases: the voltage
(Quzhou), University of Electronic Science and Technology of China, Quzhou boosting stage and the voltage maintaining stage. At the
324000, China. (e-mail: [email protected]; [email protected]) beginning, the receiver is at the boosting stage and is in the
Copyright (c) 2021 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be power-off mode. The transmitter will keep sending the energy
obtained from the IEEE by sending a request to [email protected]. packets to constantly charge the receiver with the highest

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

charging intensity until the receiver has enough energy to start rank minimization problem, which is NP-hard, is shifted to a
working, which is called the maintaining stage. Based on the convex optimization one. Due to lower sampling rate and fewer
voltage feedback mechanism, the transmitter will reduce the packets to transmit, energy consumption using EDCA is signif-
charging intensity to a lower level once the receiver comes to icantly reduced compared with the centralized scheme. Based
the maintaining stage. on the energy harvesting cognitive radio sensor networks, the
The remainder of this paper is organized as follows. Section network utility maximization problem is investigated [18].
II describes the related work on the energy saving efforts Aiming to maximize the network utility by jointly controlling
for IoT devices from different perspectives, various energy the sampling rates and channel access of sensor nodes, under
harvesting technologies and different dimensions of Simul- the energy consumption, channel capacity and interference
taneous Wireless Information and Power Transfer (SWIPT). constraints, a joint channel access and sampling rate control
The overall system architecture and the adaptive RF energy scheme is proposed. Le [19] tries to limit the transmission
transfer strategy are clarified in Section III. In Section IV, the of information over the wireless network and ensure that the
proposed node sampling scheduling algorithm based on matrix sensor node transmits enough information about the host at the
completion is designed and clarified in detail. The designed same time. This article introduces a lossless data compression
prototype and the performance evaluation are presented in algorithm to suit the wireless sensor network and meet the
Section V. Finally, the paper comes to a conclusion at Section transmission requirements.
VI.
B. Energy Harvesting
II. RELATED WORK
Apart from reducing energy consumption of the IoT devices,
A. Energy Saving many new perspectives of obtaining new energy sources have
In order to reach the goal of energy saving, many related been proposed, such as harvesting energy from piezoelectric
strategies and algorithms have been proposed [4]–[8]. Y Li energy, solar energy, electromagnetic energy and RF energy.
[4] presents a novel Media Access Control (MAC) protocol, The piezoelectric material is specially made by using two
which improves the energy efficiency and enhances the latency parallel metal plates filled with the lead zirconate titanate
and jitter. From the perspective of application layer [9]– in between [20], [21]. The electricity is generated due to
[13], much research has been conducted. A non-decreasing the piezoelectric effect by pressing the metal plates, which
and a non-increasing order maps proposed like a breathing converts mechanical energy to electric energy. The silicon-
pattern is proposed by Vijeth [9]. The map is fitted by using based solar cell is the critical material for converting solar
an increasing or decreasing values of transmission power, energy into electric energy [22]. The solar photovoltaic cell
which looks like a digital sequence of values of transmission employed in [23] is to harvest solar energy for WSN nodes.
power. By calculating the practical resource blocks required A method to obtain solar energy by employing a tracking
by IoT devices and the SINR detected by IoT devices, the technique based on gear and spring driver is proposed, which
critical required transmission power over each resource block keeps the solar panel always facing the sun directly. The
assigned to each IoT device will be calculated. J Azevedo inductive charging technology is one of the electromagnetic
[12] implements a sleeping technology for all ZigBee nodes energy harvesting technologies [24], which is widely used
and also includes full function nodes. Except for the ZigBee in the smartphone and smartwatch nowadays. It is typically
end-device, the ZigBee coordinator and router are called full implemented by the coupling circuit, which mainly consists
function devices because they are always in active mode in the of two coils.
practical ZigBee network. A time synchronization mechanism Different from the above three kinds of energy harvesting
is developed to deal with the clock drift of nodes. This kind technologies, the RF energy harvesting technology refers to
of mechanism sets up a certain sleep period and synchronizes harvest energy from RF waves, which is usually applied for
this period to each node. In the clustered CR-IoT networks, far-field charging scenario [1]. For instance, a harvester for
a two-way information exchange dynamic spectrum sensing collecting energy form 2.4 GHz Wi-Fi signal is designed
algorithm to improve energy efficiency for data transmission in [25], which consists of a 2.4 GHz antenna, a matching
in licensed channels is proposed in [14]. Furthermore, in network circuit, a rectifier, a power converter and a recharge-
consideration of the energy consumption in dynamic spectrum able battery. The load of the harvester is the microcontroller
sensing and switching, this paper proposes an energy efficient and the sensors, which derives energy from the rechargeable
optimal transmit power allocation technique to enhance the battery. Apart from collecting RF energy from the ambient
dynamic spectrum sensing and data throughput. radio signals, harvesting energy from the artificial radio signals
Optimization of energy consumption from the perspective will be more efficient. An artificial noise-assisted interference
of node sampling and data processing is also an important alignment scheme with wireless power transfer is proposed in
starting point [15]–[19]. An Efficient Data Collection Ap- [26]. The wireless power of artificial noise and interference
proach (EDCA) for data query in WSNs, which exploits is collected to replenish the batteries at the receivers so that
recent matrix completion techniques, is presented in paper to some extent its battery life can be prolonged. In order to
[15]. A part of nodes from the sensor network are randomly maximize the harvested RF energy, some attempts are made at
selected to sample at each time instance and directly forward the harvester side to increase the energy conversion efficiency.
the data to the sink. Then, to recover the data precisely, the A rectenna designed by using 2 × 2 circuit waveguide array

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

antenna is proposed in [27], which is for harvesting RF energy In the Time Splitting (TS) based receiver, the signal is
from 1.8 GHz GSM signal. Furthermore, the pump-charge split in the time domain by a time switcher for the in-
circuit is designed for capturing more RF energy, which is dependent energy and information receptions [33]. For its
sensitive to minimum -10 dBm. practical implementation, we only need to partition a typical
transmission frame into two parts, namely the energy transfer
C. Simultaneous Wireless Information and Power Transfer sub-frame and the information transfer sub-frame dedicated for
To achieve architecture of the DEIN, the technology of Si- the energy and information receptions respectively. However,
multaneous Wireless Information and Power Transfer (SWIP- having diverse TS factors for each antenna is not quite realistic
T) is critical. In order to deal with the integrated information since this operation may result in the mismatch of the RF
and power based on the SWIPT, the signal of the splitter-based signals received by the independent channels, which may
receiver architecture can be split into spatial, power and time significantly complicate the subsequent signal and energy
domains. After being received by the receiver, the RF signal processing. As a result, all the antennas on the receiver side
is split into two parallel flows. One is for energy reception, are assumed to have identical TS factors. Therefore, for the
where the rectifier plays an essential role in converting the RF implementation of the TS based receiver, we have to jointly
signal to DC before it can be stored in the receiver’s battery. optimize both the power allocation on the transmitter side and
The other is for conventional information reception, where the the TS factors on the receiver side in order to achieve the
RF signal is transformed from the passband to the baseband balance between the information rate and the energy reception.
for further demodulation and decoding operations before it For example, the authors of [34] studied the trade-off between
arrives at the information destination. energy consumption and the transmission delay by proposing a
In the Spatial Splitting (SS) based receiver, the signal is split dynamic algorithm for optimally allocating the transmit power
by allocating the different number of antennas at the receiver and deciding the TS factor.
for the energy and information receptions, respectively [28]. Much research has been conducted on the energy efficiency
For the implementation of the SS based receiver, we have and new energy resources of IoT devices. But if only one
to jointly optimize the antenna selection on the receiver side of the above aspects is considered, hardly can the energy
as well as the transmit power allocation on the transmitter consumption problem of the increasing number of IoT devices
side so as to achieve the maximum information rate subject be fully addressed. Accordingly, this paper proposes a new
to the requirement of the energy reception. For example, perspective of combining energy saving with wireless energy
in [29], by invoking an antenna selection scheme and the transfer, working on the energy efficiency problem with an
interference alignment technique, the authors partitioned the integrated solution.
received signal into two orthogonal spaces in order to achieve
simultaneous information and energy transfer. III. OVERALL SYSTEM ARCHITECTURE AND
In the Power Splitting (PS) based receiver, the signal is split ADAPTIVE RF ENERGY MANAGEMENT STRATEGY
by the power splitters for realizing the independent energy
and information receptions [30]. With the aid of the MIMO A. Overall System Architecture
system, we can realize the transmission of several indepen- The designed DEIN system is depicted in Fig. 1, which
dent information flows and hence substantially increase the consists of several Data and Energy Integrated Network Nodes
information throughput. In order to extract energy from these (DEINNs), one Data and Energy Integrated Network Gateway
independent information flows, each receive antenna should (DEING) and cloud. The proposed DEING in this paper
be equipped with a power splitter. These power splitters are integrates both the common functions of the conventional
capable of adaptively tuning themselves to conceive specific gateway and the wireless energy transmitting function of
splitting factors. As a result, for the implementation of the PS the DEIN transmitter. The DEING starts the DEIN network
based receiver, we have to jointly optimize the splitting factors and maintains it so that the DEINNs can join the network
for each antenna on the receiver side as well as the transmit through the DEING. The DEING is connected to the cloud
power allocation on the transmitter side so as to achieve the via ethernet cable, which connects the DEIN to the Internet.
balance between the information rate and the energy reception. Therefore, DEINNs in the DEIN can be remotely controlled
For example, the authors of [31] focused their attention on and monitored by the network users who are in charge of the
the impact of channel estimation on the performance of DEIN.
integrated data and energy transfer by jointly designing the Both the DEING and the DEINN are designed to work
PS factor as well as the duration of both the training phase with two antennas: an omni-directional antenna responsible
and the transmission phase. A UAV-aided NOMA scheme is for Wireless Information Transfer (WIT) and a directional
proposed to achieve simultaneous wireless information and antenna responsible for Wireless Energy Transfer (WET). As
power transfer (SWIPT) and guarantee the secure transmission for the proposed DEIN system depicted in Fig. 1, the WIT
for ground passive receivers (PRs), in which the nonlinear process and WET process are carried out in a time splitting
wireless energy harvesting model is applied [32]. Each time way. During the WET mode, the directional antenna on the
frame can be divided into two phases. In the first phase, the DEING for RF energy transfer is enabled while the omni-
UAV delivers energy to PRs. In the second phase, the received directional antenna for data communication is disabled. When
power of each PR is split into two parts, one for energy the RF energy reaches the DEINN side, the directional antenna
harvesting (EH) and the other for information decoding (ID). connected to the DEINN’s RF energy harvesting adapter will

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

Request

Respon se
Use r

Cloud

WIT WET Fig. 2. Idle time utilized for transmitting EPs.

Matrix Completion Management

Dat a Signal
The packet format designed for the EP is described in Fig.
RF Energy
Dat a and Ener gy Integrated
Networ k Gateway
3(a). The EP is broadcasted to all the network device without
(DEING) being encrypted, and they can recognize the EP because of
a 2-bytes field in the header, namely, the cluster identifier.
In the payload field, the first byte indicates the charging
status, which is filled with the boosting or the maintaining
Eth ernet Lin k state. In the beginning, the super-capacitor on the battery-
Dat a Lin k free DEINN has a nearly zero-volts initial voltage. When the
RF Energy Link system starts to charge and activate the DEINN, the charging
Dat a and Ener gy Integrated Dat a and Ener gy Integrated
Networ k Nod e Networ k Nod e status field in the EP is filled with the boosting state. After the
(DEINN i) (DEINN j)
DEINN is activated (powered on) and successfully connected
Fig. 1. Overall system architecture.
to the DEING, the charging status field in the EP will be
updated with the maintaining state. The boosting state tells
other network devices not to communicate with the DEINN
absorb energy from the carrier wave, then the RF energy will as it is still in the power-off mode, while the maintaining
be converted into electrical energy and stored for further use. state indicates that the DEINN is alive and the voltage of
Similarly, in the WIT mode, the omni-directional antennas its super-capacitor will be maintained in a suitable range. The
on both the DEING and the DEINN are enabled while the second byte in the payload field indicates the current charging
directional antennas are disabled. intensity, and the third byte counts the number of bytes filled in
The DEINNs are usually wireless IoT devices without the last packet field, namely, the energy packet payload field.
batteries, so they must run in ultra-low power mode. A node The charging intensity is classified into three degrees, and its
sampling scheduling algorithm based on matrix completion degree value determines how many bytes are required in the
and an adaptive RF energy management strategy are imple- packet payload field. As a result, the EP has three different
mented at the DEING, providing a data and energy integrated packet size, and the EP with bigger size requires a longer time
solution for energy efficiency problems of IoT devices. to transmit, which represents the stronger charging intensity.

B. Adaptive RF Energy Management Strategy

1) Utilize the Idle Time for RF Energy Transmission: In
the most of the IoT systems, the sensor data is uploaded at
intervals of several seconds or minutes. Hence, a lot of idle
time is left if the application scenario is a small-scale network
like the proposed DEIN system. Therefore, the idle time can (a) Packet format designed for the EP
be utilized for sending the energy packets (EPs) created in
the DEING periodically, which is depicted in Fig. 2. At the
beginning of time slot tEP , the DEING shifts to RF energy
transmission mode, and an EP is created and then moved to
the RF PA to transmit through the antenna. The value of the
tEP is determined by the duration of sending the EP, which
relies on the EP’s packet size. At the end of tEP , the DEING
(b) Packet format designed for the charging command
shifts back to conventional communication mode. The EPs
are sent in every 10 milliseconds and the rest time is left for Fig. 3. Packet format designed for the EP and the charging command.
transmitting the conventional data packets (DPs).
2) Packet Format Definition: Considering relieving the DE- The packet format designed for the charging command is
ING’s burden of WET which requires relatively high transmis- described in Fig. 3(b). Unlike the EP, the charging command
sion power, the DEIN transmitter (DEINT) is introduced. The packet is not transparent to all the network devices. If the
DEINT is specially designed to send the EPs and receive the charging command packet is transmitted from the DEING to
charging command to change the charging mode. the DEINT, it is used for controlling the charging intensity.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

However, if the charging command packet is sent from the the beginning of the voltage boosting stage, the DEINN is in
RF energy harvester embedded DEINN to the DEING, it is the power-off mode, and the charging command is sent from
for feedback. In the charging command packet, the payload the DEING to instruct the DEINT to broadcast the EPs with
field is designed only with two bytes, which represent the the highest charging intensity at constant intervals. Once the
charging intensity degree and the energy-hungry level of the voltage of the super-capacitor on the DEINN reaches 4.0 V,
battery-less DEINN, respectively. then the voltage regulator on the DEINN is enabled, which
The first byte indicates the degree of charging intensity and powers on the wireless module and its peripherals to activate
the value of the first byte is assigned in the DEING. Before the wireless communication function. Thereafter, the voltage
triggering the RF charging, the DEINT will not send any EP to maintaining stage starts.
the DEINN until the DEINT receives the charging command In the voltage maintaining stage, the processor module
packet from the DEING. Furthermore, which size of EP should on the DEINN detects the super-capacitor’s voltage in every
be created and sent by the DEINT is determined by that one- 5 seconds and grades its energy-hungry level. The energy-
byte field. Moreover, the value of the charging intensity degree hungry level is classified into three levels according to the
will be assigned to the field with the same name in the EP acquired voltage, and the corresponding grading standard is
which is going to be created. The second byte in the payload indicated in Fig. 3(b). The lower the voltage is detected, the
field, the energy-hungry level, is only assigned by the battery- higher the energy-hungry level is classified. Once the energy-
less DEINN. After the DEINN is powered on, it detects the hungry level is confirmed, the value of the charging intensity
voltage of the super-capacitor at regular intervals and grades its degree will be fetched from the latest received EP, which is
energy-hungry level according to the acquired sample values. used for comparing with recent detected energy-hungry level.
As a result, the charging command packet used as feedback If they are equal, the voltage detection process will be executed
contains the latest DEINN’s energy-hungry level, and the value again in the next 5 seconds. Otherwise, their values will be
of the charging intensity degree field is copied from the field assigned to a charging command packet, which will be sent
with the same name in the last received EP. from the DEINN to the DEING as feedback. And then the
DEING will prepare the charging command packet and send
it to the DEINT to update the charging intensity.

IV. NODE SAMPLING SCHEDULING ALGORITHM

BASED ON MATRIX COMPLETION
A. Sampling Model Based on Matrix Completion
In IoT network consisting of N nodes with redundant
deployment, the nodes are randomly distributed and sampled
to monitor the surrounding environment at a certain frequency.
The data collected by all nodes is summarized by the sink
node, so the data has a strong spatial-temporal correlation. A
matrix of sampling data of N × T can be obtained at the
sink node side. As shown by the expression (1), the matrix
elements are represented by Xij . The i is the ID of end nodes
and j represents the slot ID in which the node samples.
 
x1,1 x1,2 · · · x1,T
 x2,1 x2,2 · · · x2,T 
(1)
 
 .. .. . . .. 
 . . . . 
xN,1 xN,2 · · · xN,T
In order to reduce the energy consumption of IoT nodes,
avoid redundant sampling data and ensure the sampling ac-
curacy, during the node sampling process based on matrix
Fig. 4. A designed adaptive RF energy management mechanism. completion, the sink node divides the continuous time into
different task periods, and each task period contains several
3) An Adaptive RF Energy Management Strategy: A de- time slots. For saving energy, the nodes in the network will
signed adaptive energy management strategy is described not perform the sampling task in full saturation, but the sink
in the Fig. 4, which describes the charging process from node randomly selects the nodes in different time slots to
the voltage boosting stage to the voltage maintaining stage. perform sampling tasks. Therefore, the node that does not
The voltage feedback mechanism is applied in the voltage perform the sampling task will remain idle or dormant. After
maintaining stage to achieve the adaptive charging control a task period, the sink node will gather a sampling matrix
to manage the voltage level of the super-capacitor on the with missing values. The value of the node that has not been
battery-less DEINN and keep the DEINN always alive. At selected for sampling is set to 0. Thus, the sampling process

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

can be represented with linear operation A as B = A(X), slots n unchanged, reduce the number of sampling time slots
of which X is the original matrix without data loss and B a and increase the number of dormant time slots b. Thereby,
represents the sampling matrix with data loss collected by the the energy consumption of the end nodes can be reduced.
sink node. For better understanding, A is defined as a matrix
of N × T :
(
1, selected
Q(i, j) = (2)
0, otherwise
If the matrix element Q(i, j) is 0, it means the node i is
not selected as the sampling node in time slot j. Otherwise,
the node is selected. Therefore, on the sink node side, the
sampling data matrix can be presented as B = Q. ∗ X, of
which .∗ represents the product of elements, and the matrix
B maintains the available data in X. Therefore, the matrix
completion problem of the node sampling can be described
as:
Fig. 5. Node sampling time slot selection process.
(
minimize rank(X)
(3) In order to ensure the accuracy of the reconstruction matrix,
s.t PΩ (X) = Q. ∗ X
the number of sampling slots of the node needs to comply with
.. the defined sampling rate requirements. The task period of the
 
 0 . 1 0  node T is set to several minutes. The length of each time slot
 1 0 ··· 0 
is defined as ts and each task period contains m time slots.
Q= . . (4)
 
.. .. 
 .. .. . .  Then the task period can be described as T = ts × m. Within
a task period, the specific time slot that a node is allocated to
 
..
0 1 . 0 can be determined by a random threshold decision method.
It can be seen from the construction of the sampling The threshold of node i in time slot j is defined as τi (j).
matrix model that the purpose of the algorithm is to use Group[u] represents the set of nodes that have not been
as few sampling points as possible to achieve the effect of selected for sampling in the current task period. τi (j) is
energy saving. At the same time, with matrix completion, the directly related to the sampling rate ν of the IoT network.
sampling matrix with data loss can be restored to a complete Within each task cycle, the sink node will generate a random
sampling matrix. value ω(0 < ω < 1) for the node. If the random value is less
than the threshold, the node is selected as the sampling node
in the current time slot and removed from the unsampled node
B. Node Sampling Time Slot Selection Algorithm
group. In the remaining time slots of the current task cycle,
From the perspective of node task time slots and matrix this node will go to sleep to save energy. Then, in the next
sparsity, the node sampling method designed in this section task cycle, all nodes will be regrouped into Group[u] . The
divides the node sampling process into task cycles. Each node i ∈ Group[u] will repeat the same sampling process as
task cycle contains m time slots, and the node in each task the previous task cycle. The sampling threshold τi (j) of node
cycle can only sample and transmit once at most, so the i in time slot j is described as follows:
sampling rate of the entire network is ν = 1/m. It is assumed
that a network consists of N nodes. After all nodes are ν
deployed and connected to the sink node, the network is τi (j) = , i ∈ Group[u] (5)
1 − ν × ((j − 1)mod m)
successfully established. Each node is assigned with a time
slot for sampling data transmission within a task period. The The proposed algorithm is described in Algorithm 1.
sampling data is defined as Xij , of which the i indicates ID The specific steps of the node sampling time slot selection
of the corresponding node and j corresponds to the time slot algorithm are described as follows:
in which the node samples. Within every task cycle, each end • Step 1: The network is initiated and all end nodes are
node samples once and the sink node will receive N sampling waken up. The sink node initializes the task period T
data. It is assumed that it is an ideal sampling process and there and size of the time slot ts according to the networking
is no error in the sampling data. The node will be dormant situation and sets the sampling rate ν of the network.
in the rest time, which greatly reduces the frequency of data At the same time, the unselected node array Group[u] is
transmission, and the dormant state will also reduce the node’s initialized to ensure that the nodes only sample once per
energy consumption to the minimum standard. As shown in task cycle.
Fig. 5 about the node sampling time slot selection process, 0 • Step 2: Each node in the IoT network is polled, and the
represents the node is in a dormant state, and 1 represents the current node generates a random value ω(0 < ω < 1).
node is working on data sampling. The node sampling time • Step 3: The random number ω is compared with the
slot selection algorithm aims to keep the number of total time time slot selection threshold τi (j). If ω is less than τi (j),

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

Algorithm 1 Node Sampling Time Slot Selection Algorithm Algorithm 2 Matrix Reconstruction Algorithm Based on
Input: ν, m, ts , T Singular Value Threshold
Output: matrix Q Input: matrix X, threshold τ , soft threshold operator Dτ (X),
1: initialize: Fill all nodes into the group Group[u] the number of singular values sk , iterative step sequence
2: for each node i do δk , initial value Y0
3: ω = RandomN umber(0, 1) Output: matrix B
ν T
4: if ω < τi (j) = , i ∈ 1: Initialize Xm×n = Um×m Σm×n Vn×n
1 − ν × ((j − 1)mod m) Pmin(m,n)
Group[u] then 2: Decompose X by X  = i=1 σi ui viT 
select node i and drop node i out of Group[u] 1
5: 3: Dτ (X) = arg minX τ kXk∗ + kX − Y k2F
6: if (node i is in Group[u] ) then 2
7: Qij = 1 kPΩ (X k − B)k
4: while >ε do
8: end if kPΩ (B)kF
5: while all singular values >τ do
9: Qij = 0
6: increase sk
10: end if
7: end while
11: Qij = 0
8: Update δk
12: end for
9: X k = Dτ (Y k−1 )
10: Y k = Z k−1 + δk PΩ (B − X k )
11: end while
current data should be sent to the sink node, and this
node should be removed from the array Group[u] . If ω
is greater than τi (j), the node does not sample and goes
to sleep. matrix with the order n × n which is the conjugate transpose
• Step 4: After the sink node receives data from the of the matrix V . Σ represents a diagonal matrix with the
corresponding node in the corresponding time slot, it order m × n consisting of p (p = min(m, n)) singular values
creates a feature matrix Q and a node sampling matrix, arranged in descending order. Σii = σi is the singular value
which are used to reconstruct the complete data set. of the diagonal matrix Σ, σ1 ≥ σ2 ≥ ... ≥ 0. The rest
elements of Σ are 0. The column vector of U is defined
as ui , 1 ≤ i ≤ m. The column vector of V is defined as
C. Matrix Reconstruction Algorithm Based on Singular Value vj , 1 ≤ j ≤ m. The ith element of the diagonal line of Σ is
Threshold defined as σk , 1 ≤ k ≤ min(m, n). The matrix singular value
With the random sparse selection of the sampling slots decomposition can be described as the form of summation as
of the nodes, the dormant time can be extended while the follows:
sampling time can be reduced. Thus, the energy consumption
of the nodes can be reduced. However, at the same time, min(m,n)
X
an incomplete sampling sequence from the end node will X= σi ui viT (6)
be received at the sink node. In order to provide users i=1
with complete sampling data, it is necessary to reconstruct a Using the kernel norm heuristic algorithm [35], the approx-
complete sampling matrix based on the existing incomplete imate rank of the matrix can be minimized to decompose the
sampling matrix. On the one hand, it can meet the needs matrix. The rank minimization problem of the matrix restora-
of energy saving. On the other hand, it can also ensure the tion process can be transformed into the optimal approximate
integrity and accuracy of the sampling data. solution of its kernel norm. With the minimized kernel norm,
For the low-rank matrix, the singular values of the matrix the rank minimization operation can be performed on a matrix
can reflect the main content of the original matrix. In many with sufficiently low rank.
cases, the sum of the first 10% or even 1% of the singular (
values of the matrix can occupy more than 99% of the sum minimize kXk∗
of all the singular values. The singular value of the matrix (7)
s.t PΩ (X) = B
can be obtained by the matrix singular value decomposition
(SVD). Accordingly, the idea of restoring a matrix with matrix By continuously updating the matrix to reduce the di-
singular value decomposition is to reconstruct an incomplete mension, the secondary singular values are reduced to 0.
matrix with a large dimension and sparse element distribution Finally, a low-rank matrix approximation is obtained. For the
into a matrix with a low rank. The process of reconstructing reconstruction of the matrix with missing values, it is to use the
a matrix using singular value thresholds is the process of singular value feature of the matrix to reversely transform the
matrix low rank approximation based on matrix singular value singular value decomposition process until the error between
decomposition. The matrix reconstruction algorithm based on the reconstructed matrix and the original observation is less
singular value threshold is designed and the pseudo code is than the predetermined threshold, then the resulting low-rank
described in Algorithm 2. approximate matrix is the completed matrix. Therefore, in
A matrix X of m × n is defined. U represents the unitary each round of process of matrix singular value decomposition,
matrix with the order m × m, and V T represents the unitary updating and dimension reduction, secondary singular values

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

are removed. The definition of secondary singular values V. PROTOTYPE AND EVALUATION
depends on a certain threshold. With each round of matrix
A. Prototype
singular value decomposition, the singular values beyond the
threshold are reset to 0. This replacement update process is The proposed DEIN system is self-designed and a solid
the process of singular value threshold algorithm (SVT) and prototype has been implemented. The proposed DEING in this
its reverse operation can reconstruct the complete approximate paper integrates both the common functions of the convention-
primitive matrix. al gateway and the wireless energy transmitting function of the
With the expression (7), the optimal approximate solution DEIN transmitter. But for simplicity and convenience of imple-
of the matrix can be solved using the minimized kernel norm. menting the solid prototype, the wireless energy transmitting
With the singular value threshold algorithm, the problem can function is separated. The wireless communication technology
be converted into: chosen in the DEIN system is ZigBee which is a worldwide
 standard based on IEEE 802.15.4. Energy saving is one of the
minimize τ kX k + 1 kXk2 most important features of ZigBee.
∗ F
2 (8) 1) The Data and Energy Integrated Network Gateway: The
s.t P (X) = B Ω DEIN gateway is mainly structured by three parts, which are
For the matrix X = U ΣV T , the essence is to apply the ethernet module, USB circuitry and the peripheral circuitry.
soft threshold operator to the singular value of the matrix The principal part is the ethernet module with a high-speed
through this operation in order to reduce the singular value of MCU which is the AR9331 from Atheros Communications,
the matrix and reset the singular value less than the threshold Inc. A fit Linux system, namely, the OpenWrt, is embedded
τ to 0. The threshold τ > 0 and the soft threshold operator in the ethernet module, which embraces the ethernet protocol
Dτ (X) meets: stack. The ethernet port on the gateway is responsible for
ethernet communication. The second part is the USB circuitry,
  which consists of a USB hub circuitry and four extended USB
1 2 ports. The primary IC utilized for the USB hub circuitry is the
Dτ (X) = arg minX τ kXk∗ + kX − Y kF (9)
2 USB2514B, a USB hub controller, which supports both USB
The threshold τ , iterative step sequence δk and initial value 2.0 and 1.1 standard to compromise the low-speed ZigBee
Y0 are defined. It iterates from the initial value Y0 until the dongle. Furthermore, helped by the driver subjects to the
iteration stop criterion is reached. The specific process can be USB 1.1 standard installed in the OpenWrt system, the USB
described as: circuitry provides a PnP manner to the ZigBee dongle. The
( peripheral circuity includes the Ethernet and GPIO ports. The
X k = Dτ (Y k−1 ) smart gateway is connected to a network switch through the
(10) Ethernet port and cable, and the GPIO ports are left for future
Y k = Z k−1 + δk PΩ (B − X k )
extension application.
In the process of iterative reconstruction of the matrix, Two main parts, the ZigBee circuity and the USB-to-
it firstly needs to estimate Dτ and initialize the number of serial converter, structure the ZigBee dongle. In the ZigBee
singular values sk . This is to ensure that during the iterative circuity, the primary IC is the CC2530F256, a chip with the
updating process, there is a singular value which is less than ZigBee protocol stack installed, which is programmed as a
the value of τ so that the dimension of the matrix can be ZigBee coordinator. The chosen USB-to-serial converter is
reduced. When the algorithm is running, if there is no singular PL-2303HX, which converts the data from the CC2530F256’s
value less than τ , then the number of singular values should UART pins to the differential pairs of the USB port and vice
be increased accordingly until the singular value smaller than versa. The USB data speed is limited by the UART speed,
τ appears. Then the matrix reconstruction iteration step size δ which only complies with the USB 1.1 standard.
(δ ∈ (0, 2)) is chosen. Final solution obtained by the iterative 2) The Data and Energy Integrated Network Transmitter:
operation can eventually converge to the solution of expression The DEINT depicted in Fig. 6 is structured with the DC
(11) by choosing a larger threshold τ . The expression (11), of supplier, ZigBee module, RF path selector, RF PA and two
which  means fixed tolerance, can be used to confirm the antennas. For the DC supplier, the chosen IC is the MP1584,
criterion of the end of matrix iteration reconstruction. a step-down switching regulator to generate 5 V and up to 3
A constant DC power, which is a specific choice for the RF
kPΩ (X k − B)k
≤ε (11) PA. The rest of the IC requiring 3.3 V power is supplied by
kPΩ (B)kF another step-down DC-DC converter, the MCP16311/2. The
With the sampling matrix reconstruction process based on ZigBee module integrates the CC2530F256, a radio function
matrix completion, the sampling matrix with missing values IC with ZigBee standard protocol stack installed, which is
can be effectively restored. Consequently, accurate environ- configured as a ZigBee router by software. The AS179-92LF is
mental sampling information can still be acquired while the chosen as the RF path selector, which is a single-pole, double-
sampling amount is reduced. This can not only greatly reduce throw switch. The single input pin of the RF path selector is
the energy consumption of the nodes, but also reduce the connected to the unique RF interface of the ZigBee module,
pressure of data collection on the sink node. Accordingly, the and its switching is controlled by the GPIO port of the ZigBee
incidence of communication conflicts can be reduced as well. module. The SZM-2066Z is utilized as the RF PA to increase

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

the transmission power up to 33.5 dBm for sending the EPs.

Fig. 7. The data and energy integrated network node.

Fig. 6. The data and energy integrated transmitter. is connected to the VBAT OK pin. After the voltage of the
super-capacitor rises to 4.0 V, the logic 1 is output from the
Base on the ZigBee standard protocol stack running in the VBAT OK pin to enable the voltage regulator until the super-
CC2530F256, the additional procedure is added to control capacitor drops to 3.5 V, and the TPS73101 is configured by
the RF path selector and the RF PA for transmitting the RF the external registers to produce a stable 2.8 V. The ZigBee
energy. During the RF energy transmission mode, the RF PA module is programmed as a ZigBee end-device to collect the
is enabled, and the DEINT itself starts to create EPs. Then, the temperature sensor data then send the data to the ZigBee
RF path selector is switched to move the created EPs to the coordinator (ZigBee dongle) periodically. Meanwhile, it also
RF PA with 2 Watt transmission power. Thereafter, the EPs monitors and sends the voltage of the super-capacitor regularly
are transmitted through the 14 dBi directional antenna. While to the ZigBee coordinator.
in the communication mode, the path to RF PA is blocked,
and the conventional ZigBee data packets are sent through the
omni-directional antenna directly.
3) The Data and Energy Integrated Network Node: The
battery-less DEINN depicted in Fig. 7 is also designed with
two antennas. The omni-directional antenna on the ZigBee
module is used for fetching data from both DPs and EPs,
while the directional antenna is only used for extracting the
energy from the EPs. The impedance match circuit is deployed
to maximize the RF energy from the antenna to the PCB trace.
The Schottky diode, the HSMS-2862, is utilized for building
the rectifier. The rectifier is designed as a four-stage voltage
doubler, which converts the RF waveform to high enough DC
voltage to active the BQ25504, the chosen IC for the power
management circuitry. When the BQ25504 is activated by
the DC voltage coming from the rectifier, it starts to boost
the input voltage to charge the connected super-capacitor.
Meanwhile, the BQ25504 draws the energy from the super- Fig. 8. Charging and discharging process of the super-capacitor.
capacitor to the VSTOR pin, namely the output pin to drive the
load. In other words, the voltage at the VSTOR pin represents 4) Validation of Wireless Energy Transfer: The fundamen-
the voltage of the super-capacitor. The VBAT OK pin, the tal function of the solid prototype is the wireless energy
digital logic output, outputs logic 1 when the voltage of the transfer. In order to invalidate the effectiveness of wireless
super-capacitor is in the range from 3.9 to 2.5 V. The voltage energy transfer, an instrument named as 34465A which is
range is hardware-programmed by deploying the registers with a digital multimeter with high levels of accuracy, speed,
different values in the power management circuitry. and resolution is utilized. With the assistance of 34465A,
The TPS73101 is deployed to regulate the voltage from the a complete charging and discharging curve can be drawn,
power management circuitry to supply constant voltage for indicating the variation trend of the voltage of the energy
the ZigBee module and the temperature sensor. Its voltage- storage unit, which is actually a super-capacitor in the solid
input pin is connected to the VSTOR pin, and the enable pin prototype. The measurement with the voltage variation of the

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

10

super-capacitor is shown in Fig. 8. If the θr is greater than 99% for the first 10%, it means
As illustrated in Fig. 8, the voltage of the super-capacitor that the original matrix is low rank and can be recovered with
is 0 V, indicating there is no energy stored in the DEINN. matrix completion algorithm.
Once the WET process starts in a high intensity mode, the In this paper, the temperature and humidity data set sampled
voltage keeps increasing until it reaches the upper threshold by the open source sensor node of Intel Berkeley Lab is used.
(4.0 V) which is set in advance. Then, the DEINN starts All 54 sensor nodes are selected to verify whether its low rank
working and the WET process changes to a lower intensity can be used to further verify the matrix completion algorithm.
mode. Consequently, the voltage of the super-capacitor starts The idea is to create a node-slot matrix and meet the number
decreasing. Once it reaches the lower threshold (3.5 V), the condition of random matrix samples needed to recover the
WET process repeats energy transmitting in the same intensity unknown matrix. For environment monitoring applications, it
as the beginning. Meanwhile, the voltage feedback mechanism is feasible to reduce the node sampling time slot in the network
is validated as well. to save node battery energy and reconstruct the complete
sampling data with matrix completion technology.
B. Performance Evaluation 2) Energy consumption evaluation of node sampling time
As for performance evaluation, consider now that the energy slot scheduling scheme: The core goal of this solution is to
arrives (is harvested) stochastically at the transmitter as a reduce the number of node samples in the network, increase
stationary and ergodic random process. And it is assumed that the sleep time of the node and reduce the energy consumption
the cumulative energy expended cannot exceed the cumulative of the node. In this scheme, the sampling time of a node is
energy harvested at each channel use [36]. Thus, the harvested divided into time slots and each time slot represents a sampling
energy is sufficient to support the energy consumption. This point of the node. In the conventional sampling mode, all
experimental data set is from the open source data set of Intel nodes in the network need to sample each time slot and send
Berkeley Lab, containing 54 sensor nodes deployed in the data to the sink node, which will not only cause network
laboratory. The performance evaluation is verified with the congestion but also bring a lot of energy consumption. Due to
MATLAB simulation verification platform. the full sampling coverage, nodes often need to be deployed
1) Verification of low rank of sampling matrix: Due to the redundantly, so a large amount of redundant data will also
non-uniformity of node deployment, cross-redundancy may appear during sampling.
occur in the monitoring area of the node, so the sampling data The energy consumption of the node sampling time slot
of different nodes are spatially correlated. In environmental scheduling scheme proposed in this paper is compared with
monitoring, data such as temperature and light intensity will the conventional sampling scheme to evaluate the energy
not suddenly change to ensure the time correlation of the consumption characteristics and energy saving effect of the
sampling data at each monitoring point. The premise of proposed scheme. The experimental data set is selected from
recovering the sampled data in the network using matrix the humidity data of 50 nodes in Intel Lab within 1 hour.
completion theory is that the sampling matrix is low rank or The time slot is set to 1 second. According to the ener-
close to low rank. gy consumption test of the end node, the sampling energy
Judging whether the matrix is low rank depends on whether consumption of each sampling time slot is 204 mA · h. For
its rank r satisfies the condition r  min{m, n}. For a the conventional sampling scheme, the nodes will sample in
common matrix, calculating its rank size is a difficult problem. every time slot. Fig. 9 shows the node energy consumption
Using singular value decomposition (SVD), the best approx- comparison between the node sampling time slot scheduling
imation of the matrix with rank r can be obtained, as shown scheme and the conventional mode sampling scheme at a fixed
in expression (12). Singular value decomposition can ensure sampling rate of 0.2. It can be seen from the figure that
the amount of key information of the original matrix. the proposed scheme has a significant energy saving effect
compared with the conventional sampling scheme.
n According to different application requirements, the sam-
Xm×n = Um×r Σr×r Vn×r (12)
pling rate requirements of nodes in the IoT network are
The threshold θk is defined as below.
different. The energy consumption and accuracy of the cor-
Sk Σki=1 σi2 responding node’s sampling time slot scheduling scheme are
θk = = m (13) also different. As the sampling rate increases, the energy
S Σi=1 σi2
consumption of the node sampling slot scheduling scheme
Sk is the sum of squares of the first k larger singular values also gradually increases. From an overall perspective, the node
of diagonal matrix Σ. S is the sum of squares of all singular sampling slot scheduling scheme proposed in this paper has
values of diagonal matrix Σ. If σ is given, k can be obtained. different sampling rates. Because the number of sampling slots
The best approximate matrix of the matrix X can be obtained of the node is directly reduced, the number of times the node
as well. For low-rank matrices, only 10% of the diagonal transmits data to the sink node is reduced and the sleep time
matrix Σ is needed to restore 99% of the information. Then is also significantly improved. Thus, it has good energy saving
θk is defined as the ratio of the sum of squares of the first k effect. The lower the sampling rate is, the better the energy
singular values to the sum of squares of all singular values: saving effect is. However, considering the error in the process
Sk of reconstruction the sampling matrix by the algorithm, the
θk = ≥ 90% (14) sampling rate cannot be selected too low. The reconstruction
S

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

11

When choosing an energy-saving scheduling scheme, the node

sampling time slot scheduling scheme proposed in this paper
has certain advantages.

VI. CONCLUSION
This paper proposes an energy efficient integrated solution,
combining energy saving with wireless energy transfer. From
the aspect of energy saving, a node sampling scheduling
algorithm based on matrix completion is proposed. It reduces
the sampling data and enables the IoT devices to sleep more
but not to lose data because of the reconstruction capability
of the matrix completion. As for wireless energy transfer, this
paper designs an adaptive energy management strategy. Then
the IoT devices can still work normally and communication
with the gateway even in a battery-free mode. The simulation
results have shown both the effectiveness and the efficiency of
Fig. 9. Node energy consumption of node sampling time slot scheduling the integrated solution.
scheme.
As for the future work, the one-to-many topology of the
DEIN system will be further studied, especially for the battery-
accuracy of the sampling matrix with different sampling rates free IoT devices. Based on that, how to achieve an optimal
is evaluated below. scheme of the resource allocation in aspects of energy, time,
3) Data reconstruction error assessment: The reconstruc- space, radio frequency, etc. is also an interest.
tion accuracy of the node sampling matrix is also very
important for the node sampling time slot scheduling. The R EFERENCES
Normalized Mean Absolute Error (NMAE) value is used [1] J. Hu, K. Yang, G. Wen, and L. Hanzo, “Integrated data and energy com-
to evaluate the recovery accuracy. The absolute difference munication network: A comprehensive survey,” IEEE Communications
Surveys & Tutorials, vol. 20, no. 4, pp. 3169–3219, 2018.
between the element entries of the estimated matrix and the [2] L. Zhang and D. Guo, “Virtual full duplex wireless broadcasting via
original matrix is evaluated. The definition is: compressed sensing,” IEEE/ACM Transactions on Networking, vol. 22,
no. 5, pp. 1659–1671, 2014.
[3] K. Xie, L. Wang, X. Wang, G. Xie, G. Zhang, D. Xie, and J. Wen,
Σu,v:M (u,v)=0 |X(u, v) − Xest (u, v)| “Sequential and adaptive sampling for matrix completion in network
N M AE = (15) monitoring systems,” in 2015 IEEE Conference on Computer Commu-
Σu,v:M (u,v)=0 |X(u, v)|
nications (INFOCOM), 2015, pp. 2443–2451.
Xest is the estimated matrix, X is the original incomplete [4] Y. Li, H.Shi, and S. Zhang, “An energy-efficient mac protocol for
wireless sensor network,” in 2010 3rd International Conference on
matrix, and M is the sampling matrix. Advanced Computer Theory and Engineering(ICACTE), vol. 4, 2010,
pp. V4–619–V4–623.
[5] M. Abderrahim, H. Hakim, H. Boujemaa, and F. Touati, “Energy-
efficient transmission technique based on dijkstra algorithm for de-
creasing energy consumption in wsns,” in 2019 19th International
Conference on Sciences and Techniques of Automatic Control and
Computer Engineering (STA), 2019, pp. 599–604.
[6] S. Alwakeel, A. Prasetijo, and N. Alnabhan, “An adaptive energy-
saving routing algorithm for mobile wireless sensor networks,” in 2015
International Conference on Electrical and Information Technologies
(ICEIT), 2015, pp. 104–108.
[7] X. Song, “An energy-saving geographic routing in wireless sensor
network,” in 2015 International Conference on Intelligent Networking
and Collaborative Systems, 2015, pp. 390–393.
[8] C. H. Fai and H. Rudolph, “New energy efficient routing algorithm for
wireless sensor network,” in TENCON 2015 - 2015 IEEE Region 10
Conference, 2015, pp. 1–5.
[9] V. J. Kotagi, R. Thakur, S. Mishra, and C. S. R. Murthy, “Breathe to
save energy: Assigning downlink transmit power and resource blocks to
lte enabled iot networks,” IEEE Communications Letters, vol. 20, no. 8,
pp. 1607–1610, 2016.
[10] K. C. Jeong, W. S. Choi, and S. G. Choi, “A wireless ap power saving
algorithm by changing operating mode and altering transmission power
Fig. 10. Comparison of errors in different matrix reconstruction schemes. in ieee 802.11 wlan,” in 2014 International Conference on Information
and Communication Technology Convergence (ICTC), 2014, pp. 9–10.
[11] M. Lpez, J. Sabater, M. Daemitabalvandani, J. Sabater, J. M. Gmez,
As can be seen from Fig. 10, the matrix reconstruction error M. Carmona, and A. Herms, “Software management of power consump-
of the node sampling time slot scheduling scheme decreases tion in wsn based on duty cycle algorithms,” in Eurocon 2013, 2013,
with the increase of the sampling rate. In addition, it can also pp. 399–406.
[12] A. Joaquim, S. Filipe, M. Rodrigues, and L. Aguiar, “Sleeping zigbee
be seen that the accuracy is slightly higher for the scheme networks at the application layer,” IET Wireless Sensor Systems, vol. 4,
proposed in this paper when the sampling rate is very low. no. 1, pp. 35–41, 2013.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3068040, IEEE Internet of
Things Journal

12

[13] R. Nagareda, A. Hasegawa, T. Shibata, and S. Obana, “A proposal of [33] G. Huang, Q. Zhang, and J. Qin, “Joint time switching and power allo-
power saving scheme for wireless access networks with access point cation for multicarrier decode-and-forward relay networks with swipt,”
sharing,” in 2012 International Conference on Computing, Networking IEEE Signal Processing Letters, vol. 22, no. 12, pp. 2284–2288, 2015.
and Communications (ICNC), 2012, pp. 1128–1132. [34] Y. Dong, M. J. Hossain, and J. Cheng, “Joint power control and time
[14] J. Adu Ansere, G. Han, H. Wang, C. Choi, and C. Wu, “A reliable energy switching for swipt systems with heterogeneous qos requirements,” IEEE
efficient dynamic spectrum sensing for cognitive radio iot networks,” Communications Letters, vol. 20, no. 2, pp. 328–331, 2016.
IEEE Internet of Things Journal, vol. 6, no. 4, pp. 6748–6759, 2019. [35] R. Benjamin, F. Maryam, and P. Parrilo, “Guaranteed minimum-rank
[15] J. Cheng, H. Jiang, X. Ma, L. Liu, L. Qian, C. Tian, and W. Liu, solutions of linear matrix equations via nuclear norm minimization,”
“Efficient data collection with sampling in wsns: Making use of ma- SIAM review, vol. 52, no. 3, pp. 471–501, 2010.
trix completion techniques,” in 2010 IEEE Global Telecommunications [36] S. Ulukus, A. Yener, E. Erkip, O. Simeone, M. Zorzi, P. Grover, and
Conference GLOBECOM 2010, 2010, pp. 1–5. K. Huang, “Energy harvesting wireless communications: A review of
[16] C. Habib, A. Makhoul, R. Darazi, and C. Salim, “Self-adaptive data recent advances,” IEEE Journal on Selected Areas in Communications,
collection and fusion for health monitoring based on body sensor vol. 33, no. 3, pp. 360–381, 2015.
networks,” IEEE Transactions on Industrial Informatics, vol. 12, no. 6,
pp. 2342–2352, 2016. Yang Wang Yang Wang has received the B.Eng.
degree in integrated circuit design and integrated
[17] G. Samir and I. Tomasz, “Prediction-based monitoring in sensor net-
system from Nantong University, Jiangsu, China,
works: taking lessons from mpeg,” ACM SIGCOMM Computer Com-
in 2016. He is currently pursuing the Ph.D. degree
munication Review, vol. 31, no. 5, pp. 82–98, 2001.
with the School of Information and Communication
[18] J. Ren, Y. Zhang, R. Deng, N. Zhang, D. Zhang, and X. Shen, “Joint Engineering, University of Electronic Science and
channel access and sampling rate control in energy harvesting cognitive Technology of China (UESTC), Sichuan, China. His
radio sensor networks,” IEEE Transactions on Emerging Topics in research mainly focuses on Internet of Things and
Computing, vol. 7, no. 1, pp. 149–161, 2019. the Data and Energy Integrated Network (DEIN).
[19] T. L. Le and M. Vo, “Lossless data compression algorithm to save energy
in wireless sensor network,” in 2018 4th International Conference on
Green Technology and Sustainable Development (GTSD), 2018, pp. 597– Kun Yang Kun Yang received his PhD from the
600. Department of Electronic & Electrical Engineering
[20] A. Somov, Z. J. Chew, T. Ruan, Q. Li, and M. Zhu, “Poster abstract: of University College London (UCL), UK. He is
Piezoelectric energy harvesting powered wsn for aircraft structural health currently a Chair Professor in the School of Comput-
monitoring,” in 2016 15th ACM/IEEE International Conference on er Science & Electronic Engineering, University of
Information Processing in Sensor Networks (IPSN), 2016, pp. 1–2. Essex, leading the Network Convergence Laboratory
[21] S. Partovi, V. Rashidian, B. Mahmood, A. Isa, and A. Younes, “Fabri- (NCL), UK. He is also an affiliated professor at
cation of a simple and easy-to-make piezoelectric actuator and its use UESTC, China. Before joining in the University
as phase shifter in digital speckle pattern interferometry,” Journal of of Essex at 2003, he worked at UCL on several
Optics, vol. 48, no. 2, pp. 272–282, 2019. European Union (EU) research projects for several
[22] A. Joshi, A. Khan, and A. SP, “Comparison of half cut solar cells with years. His main research interests include wireless
standard solar cells,” in 2019 Advances in Science and Engineering networks and communications, IoT networking, data and energy integrated
Technology International Conferences (ASET), 2019, pp. 1–3. networks, and molecular communications. He manages research projects
[23] M. Mangrulkar and S. G. Akojwar, “A simple and efficient solar energy funded by various sources such as UK EPSRC, EU FP7/H2020 and industries.
harvesting for wireless sensor node,” in 2016 Second International Con- He has published 150+ journal papers and filed 10 patents. He serves on the
ference on Research in Computational Intelligence and Communication editorial boards of both IEEE and non-IEEE journals. He is a Senior Member
Networks (ICRCICN), 2016, pp. 95–99. of IEEE (since 2008) and a Fellow of IET (since 2009).
[24] H. Rahmani and A. Babakhani, “A 434 mhz dual-mode power harvesting Weixiang Wan Weixiang Wan received his B.Eng.
system with an on-chip coil in 180 nm cmos soi for mm-sized implants,” degree from the School of Electrical Information
in 2018 IEEE/MTT-S International Microwave Symposium - IMS, 2018, Engineering, Dalian Jiaotong University, Liaoning,
pp. 1130–1133. China, in 2017. He has received the MSc degree
[25] T. Vamsi, K. Bryce, R. Benjamin, N. Saman, G. Shyamnath, and with the School of Information and Communication
J. Smith, “Powering the next billion devices with wi-fi,” in Proceedings Engineering, University of Electronic Science and
of the 11th ACM Conference on Emerging Networking Experiments and Technology of China (UESTC), Sichuan, China, in
Technologies, 2015, pp. 1–13. 2020. His research mainly focuses on the energy-
[26] N. Zhao, Y. Cao, F. R. Yu, Y. Chen, M. Jin, and V. C. M. Leung, efficient solution in the Internet of Things.
“Artificial noise assisted secure interference networks with wireless
power transfer,” IEEE Transactions on Vehicular Technology, vol. 67,
no. 2, pp. 1087–1098, 2018. Yitian Zhang Yitian Zhang received the B.Eng.
[27] A. Munir, M. P. K. Praja, and M. S. Arifianto, “Circular waveguide degree from a joint education of Beijing Institute
array rectenna equipped with pump-charge and dc-dc converter circuits Technology, China and University of Central Lan-
for 1.8ghz rf energy harvesting application,” in 2019 IEEE-APS Topical cashire, UK. He received the MSc degree from
Conference on Antennas and Propagation in Wireless Communications Coventry University, UK and received the PhD
(APWC), 2019, pp. 1–4. degree from University of Essex, UK. His current
[28] S. Zhao, Q. Li, Q. Zhang, and J. Qin, “Antenna selection for simulta- research interests embrace high efficiency wireless
neous wireless information and power transfer in mimo systems,” IEEE network, Internet of Things and data and energy
Communications Letters, vol. 18, no. 5, pp. 789–792, 2014. integrated networks.
[29] B. Koo and D. Park, “Interference alignment and wireless energy transfer
via antenna selection,” IEEE Communications Letters, vol. 18, no. 4, pp.
548–551, 2014. Qiang Liu Qiang Liu received his B. S. (1996), M.
S. (2000) and Ph.D. (2012) degrees from University
[30] Z. Zhou, M. Peng, Z. Zhao, W. Wang, and R. S. Blum, “Wireless-
of Electronic Science and Technology of China
powered cooperative communications: Power-splitting relaying with
(UESTC) respectively. After graduating from M.S.
energy accumulation,” IEEE Journal on Selected Areas in Communi-
study in 2000, he has worked with School of Com-
cations, vol. 34, no. 4, pp. 969–982, 2016.
munication and Information Engineering, UESTC,
[31] X. Zhou, “Training-based swipt: Optimal power splitting at the receiver,” China, and is an associate professor now. He had
IEEE Transactions on Vehicular Technology, vol. 64, no. 9, pp. 4377– worked in the University of Essex in U.K. (2012.12-
4382, 2015. 2013.12) and the University of California, Davis in
[32] W. Wang, J. Tang, N. Zhao, X. Liu, X. Y. Zhang, Y. Chen, and U.S.A (2017.8-2018.8) as a visitor scholar. His re-
Y. Qian, “Joint precoding optimization for secure swipt in uav-aided searches focus on wireless sensor networks, Internet
noma networks,” IEEE Transactions on Communications, vol. 68, no. 8, of Things, broadband wireless networks and molecular communication.
pp. 5028–5040, 2020.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Linkoping University Library. Downloaded on June 23,2021 at 08:28:48 UTC from IEEE Xplore. Restrictions apply.