sensors

Article
Residual Energy Estimation-Based MAC Protocol for Wireless
Powered Sensor Networks
Sol-Bee Lee , Jung-Hyok Kwon and Eui-Jik Kim *

School of Software, Hallym University, 1 Hallymdaehak-gil, Chuncheon 24252, Gangwon-do, Korea;

[email protected] (S.-B.L.); [email protected] (J.-H.K.)
* Correspondence: [email protected]; Tel.: +82-33-248-2333

Abstract: This paper presents a residual energy estimation-based medium access control (REE-MAC)
protocol for wireless powered sensor networks (WPSNs) composed of a central coordinator and
multiple sensor devices. REE-MAC aims to reduce overhead due to control messages for scheduling
the energy harvesting operation of sensor devices and provide fairness for data transmission oppor-
tunities to sensor devices. REE-MAC uses two types of superframes that operate simultaneously in
different frequency bands: the wireless energy transfer (WET) superframe and wireless information
transfer (WIT) superframe. At the beginning of each superframe, the coordinator estimates the
change in the residual energy of individual sensor devices caused by their energy consumption
and energy harvesting during the previous superframe. It then determines the devices’ charging
priorities, based on which it allocates dedicated power slots (DPSs) within the WET superframe. The
simulation results demonstrated that REE-MAC exhibits superior performance for the harvested
energy, average freezing time, and fairness to existing representative WPSN MAC protocols.





 Keywords: energy harvesting; Internet of Things; out-of-band approach; residual energy estimation;
Citation: Lee, S.-B.; Kwon, J.-H.; wireless power transfer; wireless powered sensor network
Kim, E.-J. Residual Energy
Estimation-Based MAC Protocol for
Wireless Powered Sensor Networks.
Sensors 2021, 21, 7617. https:// 1. Introduction
doi.org/10.3390/s21227617
Recent advances in radio frequency (RF)-based wireless energy transfer (WET) tech-
niques enable battery-powered sensor devices to receive energy remotely without time
Academic Editors: Slim Naifar,
and space constraints on ambient resources such as solar, thermal, wind, and vibration, en-
Olfa Kanoun and Carlo Trigona
abling perpetual operations. Thus, wireless sensor networks (WSNs) with RF-based WET—
wireless powered sensor networks (WPSNs)—are considered one of the most promising
Received: 19 October 2021
Accepted: 13 November 2021
technologies for a sustainable Internet of Things [1–9]. In the WPSN, a power station
Published: 16 November 2021
wirelessly transfers energy to sensor devices that use the harvested energy to transmit their
collected information to a fusion center [10–12]. The power station and fusion center may
Publisher’s Note: MDPI stays neutral
be included in one device or separated into different devices.
with regard to jurisdictional claims in
Although such WPSNs are expected to have a potentially infinite network life, they
published maps and institutional affil- cannot always prevent short-term disconnections due to temporary energy shortages
iations. of some sensor devices. Such short-term disconnections cause an imbalance in trans-
mission opportunities between sensor devices, resulting in an unfairness problem for
WPSNs [13–16]. Moreover, in contrast to the traditional WSNs, in the WPSN, when sensor
devices operate, wireless information transfer (WIT) and WET must be jointly considered.
Copyright: © 2021 by the authors.
Consequently, it is necessary to design an appropriate medium access control (MAC)
Licensee MDPI, Basel, Switzerland.
protocol for WPSNs.
This article is an open access article
Many studies have been conducted to design an efficient MAC protocol for WPSNs.
distributed under the terms and In [17–19], the authors proposed a MAC protocol based on carrier-sense multiple access
conditions of the Creative Commons with collision avoidance (CSMA/CA) for WPSNs. Sensor devices access the channel com-
Attribution (CC BY) license (https:// petitively to conduct energy harvesting and data transmission; their channel access priori-
creativecommons.org/licenses/by/ ties are probabilistically differentiated by the backoff duration and inter-frame space (IFS),
4.0/). determined by the remaining energy. In [20–22], the authors proposed the time-division

Sensors 2021, 21, 7617. https://doi.org/10.3390/s21227617 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 7617 2 of 21

multiple access (TDMA)-based MAC protocol, in which a central coordinator allocates time
resources for energy harvesting and data transmission considering the remaining energy of
the sensor devices and the energy consumption required for data transmission. The sensor
devices harvest energy in the dedicated time slots and transmit data.
Cho et al., used both TDMA and CSMA/CA methods to support energy harvesting
and data transmission for two types of traffic patterns: periodic and non-periodic [23].
In [23], a coordinator allocates the dedicated TDMA time slots for energy harvesting and
data transmission within a superframe to sensor devices that generate periodic traffic
requiring transmission reliability on-demand. In contrast, energy harvesting and data
transmission for non-periodic traffic are performed in the rest of the superframe using the
CSMA/CA scheme. The studies in [17–23] commonly suffer from network performance
degradation due to a lack of bandwidth resources from performing both WET and WIT
operations within the same frequency band.
In [24–26], the authors used an out-of-band WET approach where sensor devices
perform WET and WIT operations independently in different frequency bands, mitigating
the effect of bandwidth limitation on WIT operation and improving network performance.
However, their research suffers from high overhead due to the exchange of numerous con-
trol messages required to schedule WET and WIT operations in separate frequency bands.
Furthermore, the WET scheduling method used in [24–26] depends on simple criteria
such as the distance to the sensor device and the energy required to transmit the data
packet, increasing the difference in the residual energy between the sensor devices in the
network. This difference causes an imbalance in transmission opportunities between sensor
devices, resulting in an unfairness problem for network performance. This unfairness
problem also applies to the in-band WET approach of [17–23].
This paper proposes a residual energy estimation-based MAC (REE-MAC) protocol,
with two advantages for WPSNs composed of a central coordinator and multiple sensor
devices. First, REE-MAC increases the residual energy of individual sensor devices by
reducing overhead due to control messages for scheduling the energy harvesting operation
of sensor devices. The coordinator numerically estimates the residual energy of individual
sensor devices rather than exchanging numerous control messages. Second, REE-MAC
establishes fairness among the data transmission opportunities for sensor devices. The
coordinator allocates WET slots within the superframe to the sensor device by comprehen-
sively considering the distance, harvested energy, and consumed energy for individual
sensor devices. Accordingly, the residual energy of the sensor devices in the network is
maintained at a similar level.
To this end, REE-MAC uses two types of superframes that operate simultaneously
in different frequency bands: WET superframe and WIT superframe. In the WET super-
frame, a power transmitting unit (PTU) serving as a central coordinator supplies power to
power receiving units (PRUs) (i.e., sensor devices) using the TDMA scheme. In the WIT
superframe, multiple PRUs compete to transmit data packets to the PTU using CSMA/CA.
At the beginning of each superframe, the PTU estimates the residual energy of individual
PRUs changed due to their energy consumption and energy harvesting during the previous
superframe. The PTU then determines the PRUs’ charging priorities, based on which it
allocates dedicated power slots (DPSs) within the WET superframe.
We performed an experimental simulation to verify the superiority of REE-MAC over
FF-WPT [25] and HE-MAC [19], which are the representative MAC protocols for WPSNs
of out-of-band and in-band WET approaches, respectively. The results demonstrated
that REE-MAC achieves 18.08% and 145.60% higher average harvested energy, 81.03%
and 64.21% shorter average freezing time, and 100.49% and 135.56% higher fairness than
FF-WPT and HE-MAC, respectively.
The rest of this paper is organized as follows. In Section 2, we present a system
model for REE-MAC. In Section 3, the detailed operation of REE-MAC is described. The
simulation configuration and results are presented in Section 4. Finally, Section 5 concludes
this paper.
 Sensors 2021, 21, x FOR PEER REVIEW 3 of 22

The rest of this paper is organized as follows. In Section 2, we present a system model
Sensors 2021, 21, 7617 for REE‐MAC. In Section 3, the detailed operation of REE‐MAC is described. The simula‐3 of 21
tion configuration and results are presented in Section 4. Finally, Section 5 concludes this
paper.
2. System Model
2. System Model
Figure
Figure 11 illustrates
illustratesthe
thesystem
systemarchitecture
architectureof of
thethe considered
considered WPSN,
WPSN, which
which consists
consists
of a single PTU and multiple PRUs. The PRUs are deployed within the
of a single PTU and multiple PRUs. The PRUs are deployed within the transmission range transmission
range
of the of the At
PTU. PTU.
theAt the request
request of thethe
of the PRU, PRU,
PTU the PTUtransfers
either either transfers power
power to the PRUto the PRU or
or re‐
receives
ceives data from the PRU, for which the PTU and PRU are equipped with two types of of
data from the PRU, for which the PTU and PRU are equipped with two types
transceivers:
transceivers: aa power
powertransceiver
transceiver(P‐Tx/Rx)
(P-Tx/Rx)forfor WET
WET andand a data
a data transceiver
transceiver (D-Tx/Rx)
(D‐Tx/Rx) for for
WIT. The P-Tx/Rx and D-Tx/Rx in a device operate independently in different
WIT. The P‐Tx/Rx and D‐Tx/Rx in a device operate independently in different frequency frequency
bands
bands and interact
interactusing
usinginternal
internalsignaling—enabling
signaling—enabling thethe WET
WET andand
WITWIT of REE-MAC
of REE‐MAC to to
be
be performed simultaneously.
performed simultaneously.

Figure 1.
Figure 1. System
Systemarchitecture
architectureofofREE‐MAC.
REE-MAC.

Furthermore, the
Furthermore, theP‐Tx/Rx
P-Tx/Rx andand
D‐Tx/Rx of PTU
D-Tx/Rx always
of PTU keep their
always keepradio
theiron, while
radio on,forwhile
for the PRU, only its D-Tx/Rx is always on. The PRU keeps its P-Tx/Rx turned aon to
the PRU, only its D‐Tx/Rx is always on. The PRU keeps its P‐Tx/Rx turned on to receive
beacon aand
receive power
beacon only
and in WET
power slots
only allocated
in WET slotstoallocated
it and turns
to ititand
off in other
turns it WET
off inslots
othertoWET
prevent unnecessary energy consumption due to idle listening. Both the PTU and PRU
slots to prevent unnecessary energy consumption due to idle listening. Both the PTU
have two antennas. The PTU has a directional antenna with a fixed beamwidth for WET
and PRU have two antennas. The PTU has a directional antenna with a fixed beamwidth
and an omnidirectional antenna for WIT. We consider an adaptive array smart antennas
for WET and an omnidirectional antenna for WIT. We consider an adaptive array smart
as a directional antenna, which adjusts the beam direction by adapting the phase distri‐
antennas as a directional antenna, which adjusts the beam direction by adapting the phase
bution of its antenna array elements. The PTU can transfer power to multiple PRUs using
distribution of its antenna array elements. The PTU can transfer power to multiple PRUs
such a directional antenna. The PRU has two omnidirectional antennas for WET and WIT,
using such a directional antenna. The PRU has two omnidirectional antennas for WET and
respectively.
WIT, Inrespectively.
REE‐MAC, for determining when and how long to transfer power to individual
PRUs,Inthe
REE-MAC, for determining
PTU estimates the power thatwheneachand
PRU howcanlong to transfer
receive per unit power to individual
time, derived by
PRUs, the PTU
Equation (1). estimates the power that each PRU can receive per unit time, derived by
Equation (1).
2
   λ 2

Pr ,iPr,iPt= PtrG t Gr  di , di −α ,
Gt G (1) (1)
 4  4π
where PPr,ir ,i isisthe
where thepower thatthe
power that i-thPRU
thei‐th PRU receives
receives perper second
second and and Ptthe
Pt is is the transmission
transmission
power of the PRU. G
power of the PRU. Gtt and G and G r are
are the antenna gain of the PTU and the PRU,respectively.
the antenna gain of the PTU and the PRU, respec‐
r
We consider a flat-top radiation pattern as the directional antenna model of the PTU [27];
tively. We consider a flat‐top radiation pattern as the directional antenna model of the
thus, Gt is the same as 2π/θ, where θ is the beamwidth of the directional antenna of the
PTU [27]; thus, G t is the same as 2 /  , where  is the beamwidth of the directional
PTU. In contrast, because the PRU receives power from the PTU using its omnidirectional
antenna of
antenna, Grthe PTU. In
is equal to contrast,
one [28].because theαPRU
λ, di , and referreceives power from the
to the wavelength, PTU using
distance its the
between
PTU and i-th PRU, and path loss exponent, respectively.

3. Design of REE-MAC
Figure 2 illustrates a superframe structure of the REE-MAC. The REE-MAC main-
tains a dual superframe structure—WET and WIT superframes—operated in separated
frequency bands. In the WET superframe, the PTU transfers power to the PRUs according
to their dedicated schedule using the TDMA scheme. The PRUs use the harvested energy
to exchange control messages and data packets using contention-based CSMA/CA in the
 length, distance between the PTU and i‐th PRU, and path loss exponent, respectively.

3. Design of REE‐MAC
Figure 2 illustrates a superframe structure of the REE‐MAC. The REE‐MAC main‐
Sensors 2021, 21, 7617 tains a dual superframe structure—WET and WIT superframes—operated in separated 4 of 21
frequency bands. In the WET superframe, the PTU transfers power to the PRUs according
to their dedicated schedule using the TDMA scheme. The PRUs use the harvested energy
to exchange control messages and data packets using contention‐based CSMA/CA in the
WITsuperframe.
WIT superframe.The The WET
WET superframe
superframe is divided
is divided intointo multiple
multiple equally
equally sizedsized
DPSs,DPSs,
each each
allocatedtotoananindividual
allocated individual PRU
PRU forfor
its its exclusive
exclusive power
power reception.
reception.

Superframe duration

DPS #1 DPS #2 DPS #3 DPS #4 DPS #5 DPS #6 DPS #7 DPS #8 DPS #9 DPS #10
... DPS #n

WET
superframe

Subslot 2
Subslot 1 Subslot 3
(P-Tx/Rx
(Beacon) (WET)
switching)

Beacon period

WIT
Data communication period
superframe

Figure 2. WET and WIT superframe structure.

Figure 2. WET and WIT superframe structure.
The DPS consists of three subslots: a beacon subslot, a P‐Tx/Rx switching subslot, and
The DPS consists of three subslots: a beacon subslot, a P-Tx/Rx switching subslot,
a WET subslot. In the beacon subslot, the PTU or PRU beacon is broadcasted. The PTU
and a WET subslot. In the beacon subslot, the PTU or PRU beacon is broadcasted. The
beacon is transmitted in the beacon subslot of the first DPS in the WET superframe,
PTU beacon is transmitted in the beacon subslot of the first DPS in the WET superframe,
through which the PTU informs the PRUs of the allocation result for the DPSs in the WET
through which the PTU informs the PRUs of the allocation result for the DPSs in the WET
superframe. In the beacon subslot of the remaining DPSs in the WET superframe, the
superframe.
PRUs request In the beacon
power transfersubslot of the remaining
by transmitting DPSs intothe
a PRU beacon theWET
PTU.superframe,
In the P‐Tx/Rx the PRUs
request power transfer by transmitting a PRU beacon to the PTU.
switching subslot, the PTU obtains the direction of the PRU from the received PRU bea‐ In the P-Tx/Rx switching
subslot,
con, and thethe PRU
PTU switches
obtains the the direction
mode of its of P‐Tx/Rx
the PRUfrom fromTxthe to received PRU beacon,
Rx for harvesting the en‐ and the
PRU
ergy. switches the mode of its P-Tx/Rx from Tx to Rx for harvesting the energy.
Thedirection
The direction ofofthe thePRUPRU indicates
indicates thethe phase
phase of the of the
PRUPRU beaconbeacon
signalsignal transmitted
transmitted
fromthe
from thePRU,
PRU,through
throughwhich whichthe thePTU
PTUcan canchange
changethe thebeam
beamdirection
directionofofitsitsantenna
antennato‐ toward
the corresponding
ward the corresponding PRU PRU to to
transfer
transferpower
power to it.it.InInthe
theWETWET subslot,
subslot, the PTU
the PTU transfers
transfers
powertotothe
power theallocated
allocated PRU.
PRU. The TheWITWIT superframe
superframe consists
consists of a of a beacon
beacon periodperiod
and aanddataa data
communicationperiod.
communication period.InIn thethe beacon
beacon period
period in the
in the WIT WIT superframe,
superframe, the the
PTUPTU broadcasts a
broadcasts
abeacon
beacon including
including the theWITWITsuperframe
superframe parameters
parameters through
through D-Tx/Rx.
D‐Tx/Rx. The PTUThe PTU and PRUs
and PRUs
compete
competefreely
freelywith
witheach
each other
otherforfor
channel
channel access using
access the CSMA/CA
using the CSMA/CA scheme.scheme.
Figure
Figure3 3illustrates
illustrates thethe
overall
overalloperation
operation of the REE‐MAC.
of the REE-MAC. The solid and dashed
The solid and dashedlines lines
represent the operation of the devices in WET and WIT superframes,
represent the operation of the devices in WET and WIT superframes, respectively. These respectively. These
two
twosuperframes
superframesare areexpressed
expressedtotooverlap
overlapeacheachother
otherbecause
becausethe theWET
WETand andWITWITopera‐
operations
tions
are performed simultaneously in different frequency bands. At the beginning of
are performed simultaneously in different frequency bands. At the beginning ofthe the WIT
WIT superframe,
superframe, the PTU
the PTU firstfirst broadcasts
broadcasts a beacon
a beacon usingusing D‐Tx/Rx.
D-Tx/Rx. UponUpon receiving
receiving the thebeacon
beacon
from thefrom thethe
PTU, PTU,
PRU thetransmits
PRU transmitsenergy energy harvesting
harvesting information,
information, suchsuchas theasnode
the nodeidentifier
identifier (ID), residual
(ID), residual energy, maximum
energy, maximum batterybattery capacity,
capacity, and distance
and distance from from
the PTU,the PTU,
to the to PTU.
the PTU.
This transmission of energy harvesting information for each PRU is performed only
once throughout the entire network lifetime. The PTU and PRUs then exchange control
messages and data packets using CSMA/CA. At the beginning of the WET superframe,
the PTU performs the DPS allocation, which consists of three operations: calculating the
number of DPSs required for individual PRUs to be fully charged, calculating the number
of DPSs to be allocated to individual PRUs, and determining the charging priority for
individual PRUs.
After the DPS allocation is completed, the PTU broadcasts the PTU beacon to notify
PRUs of the scheduled WET superframe structure in the beacon subslot of the first DPS
in the WET superframe. Then, in the beacon subslot in the subsequent DPSs allocated to
individual PRUs, the PRU sends its PRU beacon to the PTU. With the PRU beacon, the
PTU acquires the direction of the PRU and transfers power to the corresponding PRU in
the WET subslot.
Sensors 2021, 21, 7617 5 of 21
Sensors 2021, 21, x FOR PEER REVIEW 5 of 2

Figure 3. Overall operation Figure 3. Overall operation of REE‐MAC.

of REE-MAC.

In estimating This
the residual
transmissionenergy of individual
of energy harvesting PRUs, the PTU
information forfirst
eachcalculates the
PRU is performed only
number of beacon,
once successful,
throughoutcollided,
the entire and idle slots
network in theThe
lifetime. previous
PTU andWITPRUssuperframe. The contro
then exchange
messages
beacon slot indicates andused
a slot data bypackets
the PTUusingandCSMA/CA. At the beginning
PRU to transmit and receive of the
the WET superframe
beacon.
the PTU performs the DPS allocation, which consists of three
The successful slot indicates a slot used for successful transmission between the PTU and operations: calculating th
number of DPSs required for individual PRUs to be fully charged,
PRU. The collided slot indicates a slot in which collision occurs due to the data packets calculating the numbe
simultaneouslyoftransmitted
DPSs to befrom allocated
two orto more
individual
PRUs.PRUs, andslot
The idle determining the charging
is one in which no PRUs priority fo
individual PRUs.
have transmitted the data packets due to the random backoff. It is assumed that n PRUs
in the network alwaysAfterhave
the DPS
the allocation
data packets is completed,
to transmit theinPTU broadcastscondition.
a saturation the PTU beacon
The to notify
PRUs of the scheduled WET superframe structure in the beacon
PTU maintains the node ID of PRUs (i.e., ID), the distance from itself to each PRU obtained subslot of the first DPS in
the WET superframe. Then, in the beacon subslot in the subsequent
from the received PRU beacons (i.e., D), the neighbor’s node ID of each PRU (i.e., Nbr(i) ), DPSs allocated to
individual PRUs, the PRU sends its PRU beacon to the
and the number of data packets received from each PRU during the previous superframe PTU. With the PRU beacon, th
PTU acquires the direction of the PRU and transfers power to the corresponding PRU in
(i.e., Npkt), which are represented by the matrices, as in Equation (2).
the WET subslot.
In estimating the residual energy of individual PRUs, the PTU first calculates th
number of beacon, successful, collided, and idle slots in the previous WIT superframe
The beacon slot indicates a slot used by the PTU and PRU to transmit and receive th
Sensors 2021, 21, 7617 6 of 21

h i
ID = id(1) , id(2) , · · · , id(i) , · · · , id(n) , 0 < i ≤ n
h i
D = d (1) , d (2) , · · · , d ( i ) , · · · , d ( n ) , 0 < i ≤ n
h i (2)
Nbr(i) = id(i,1) , id(i,2) , · · · , id(i,l ) , 0 < i ≤ n, 0 ≤ l ≤ (n − 1)
h i
Npkt = npkt(1) , npkt(2) , · · · , npkt(i) , · · · , npkt(n) , 0 < i ≤ n,

where id(i) is the node ID for the i-th PRU, n is the number of PRUs in the network, d(i) is
the distance from the PTU to the i-th PRU, id(i,l ) is the node ID of the l-th neighbor PRU of
the i-th PRU, and npkt(i) is the number of data packets received from the i-th PRU during
the previous WIT superframe.
The number of beacon and successful slots in the previous WIT superframe can be
calculated using Lbeacon and Npkt, where Lbeacon is the length of the beacon subslot of DPS.
We calculate the number of collided and idle slots by considering the collision and idle
probabilities in the remaining slots except for beacon and successful slots in the previous
WIT superframe as pcol and pidle , respectively [29]. pcol is the probability that two or more
PRUs transmit the data packet in a randomly chosen slot excluding beacon and successful
slots in the previous WIT superframe, as defined by Equation (3).
n  
n
pcol = ∑ k
τ k (1 − τ ) n − k , (3)
k =2

where τ is the probability that the PRU transmits a data packet in a randomly chosen
slot excluding beacon and successful slots in the previous WIT superframe. pidle is the
probability that no PRU transmits a data packet in a randomly chosen slot, excluding
beacon and successful slots in the previous WIT superframe, as defined by Equation (4).

pidle = 1 − pcol (4)

The total number of slots in the previous WIT superframe, ntotalSlots , can then be
calculated by Equation (5).
ntotalSlots = LdataSF /L BP , (5)
where L BP is the slot length, which is the same as a unit backoff period. The numbers of
beacon, successful, collided, and idle slots in the previous WIT superframe, nbeaconSlots ,
nsuccSlots , ncolSlots , and nidleSlots , can be calculated by Equations (6)–(9), respectively.

nbeaconSlots = Lbeacon /L BP (6)

!
n
nsuccSlots = Lsucc ∑ n pkt,i /L BP (7)
i =1

ncolSlots = pcol (ntotalSlots − nbeaconSlots − nsuccSlots ) (8)

nidleSlots = pidle (ntotalSlots − nbeaconSlots − nsuccSlots ), (9)
where Lsucc is the time taken for successful transmission between the PTU and PRUs.
Figure 4 illustrates the timing diagrams for the successful transmission and collision. Ldata ,
LSIFS , L ACK , and L DIFS are the length of a data packet, a short inter-frame space (SIFS), an
acknowledgment (ACK), and a distributed inter-frame space (DIFS), respectively.
 nidleSlots  p idle  ntotalSlots  nbeaconSlots  nsuccSlots  , (9)

where Lsucc is the time taken for successful transmission between the PTU and PRUs.
Figure 4 illustrates the timing diagrams for the successful transmission and collision. Ldata
Sensors 2021, 21, 7617 7 of 21
, LSIFS , LACK , and LDIFS are the length of a data packet, a short inter‐frame space (SIFS),
an acknowledgment (ACK), and a distributed inter‐frame space (DIFS), respectively.

Ldata LACK

Data LSIFS ACK LDIFS

(a)
Ldata

Data LDIFS

(b)

Figure Timingdiagram:
Figure 4. Timing diagram:(a)
(a)successful
successful transmission,
transmission, (b)(b) collision.
collision.

The PTU
PTU then calculatesEEbeacon
thencalculates beacon,i
,i
Esucc,i
, E, succ ,i
, E,colE,i col,i
, and EidleE, iidle,i
, and whichwhich
are are
the the
amounts amountsof of
energy consumed by the i-th
energy consumed by the i‐th PRU in the beacon, successful, collided, and idle slots in thein the
PRU in the beacon, successful, collided, and idle slots
previous WIT superframe,
previous WIT superframe,respectively,
respectively,considering
considering thethe
timingtiming diagrams.
diagrams. Ebeacon,i
Ebeacon is theis the
,i
energy consumed by the i-th PRU to receive the beacon, as defined by Equation (10).
energy consumed by the i‐th PRU to receive the beacon, as defined by Equation (10).
Ebeacon,i
Ebeacon ,i
 nbeaconSlots
= nbeaconSlots
Erx LBP , Erx L BP , (10) (10)

where EErxrx isisthe

where the energy
energy consumed
consumedperpersecond
secondbybya a
PRU
PRUwhen
whenits its
D‐Tx/Rx
D-Tx/Rxis inisthe
inRx
the Rx
state. EEsucc,i
state. succ , i
is
is the
the energy
energy consumed
consumed by
by the
the i-th
i‐th PRU
PRU during
during successful
successful slots
slots inin the
the previous
previ‐
WIT superframe,
ous WIT superframe, as defined by by
as defined Equation (11).
Equation (11).
E
Esucc
succ,i
,i
=
EsuccTx  EsuccRx
EsuccTx,i
,i +E  EsuccIdle+
, i succRx,i
,
, i EsuccIdle,i , (11) (11)

where EEsuccTx ,i isis the

where theenergy
energyconsumed consumed byby thethesuccessful transmission
successful transmission of the
of i‐th
the PRU
i-th PRU
succTx,i
when its
when its D-Tx/Rx
D‐Tx/Rx isisininthe theTx Txstate.
state.EsuccRx
EsuccRx,i
,i
is the energy
is the energyconsumed
consumed by the successful
by the successful
transmission
transmission of of neighbors
neighborsofofthe i-thPRU
thei‐th PRU when
when itsits D-Tx/Rx
D‐Tx/Rx is inisthe
in Rx
thestate.
Rx state. Specifically,
Specifically,
EEsuccRx,i
succRx , i
is
is the
the energy
energy consumption
consumption of of
thethe
i‐th i-th
PRUPRUin in successful
successful slotsslots
whenwhen
it it overhears
overhears
the data packets transmitted by other PRUs in
the data packets transmitted by other PRUs in the previous WIT superframe.the previous WIT superframe.
Furthermore, E
Furthermore, EsuccIdle
succIdle,i
,i
isis the
the energy
energy consumed
consumed by by
thethe successful
successful transmission
transmission of the
of the
PRUs not adjacent to the i-th
PRUs not adjacent to the i‐th PRU when the D‐Tx/Rx is in an idle state. EsuccTx ,i can be be
PRU when the D-Tx/Rx is in an idle state. E succTx,i can
calculated by Equation (12).
calculated by Equation (12).
EsuccTx,i
E succTx , i  E(ELtx LdataE +L ErxLEACK L+ Eidle
=nn pkt,i
pkt , i tx data rx ACK   , + LDIFS )),
 L( LSIFS
idle SIFS DIFS (12) (12)

where EEtxtx and

where andEidle
Eidle are
arethe
theenergy
energyconsumed
consumed per second by
per second bythe
thePRU
PRUwhen
whenitsitsD‐Tx/Rx
D-Tx/Rx is
in the Tx and idle states, respectively. EsuccRx,i can be calculated by Equation (13).
is in the Tx and idle states, respectively. EsuccRx ,i can be calculated by Equation (13).


EsuccRx ,i  nsuccNbr ,i Erx Ldata  Eidle LACK  Eidle  LSIFS  LDIFS 
EsuccRx,i = nsuccNbr,i ( Erx Ldata + Eidle L ACK + Eidle ( LSIFS
,
+ L DIFS )),
 (13)
(13)

where nsuccNbr,i is the total number of data packets successfully transmitted to the PTU by
neighbor PRUs of the i-th PRU. EsuccIdle,i can be calculated by Equation (14).
   
EsuccIdle,i = nsuccSlots − n pkt,i Ldata − (nsuccNbr,i Ldata ) Eidle L BP (14)

Ecol,i is the energy consumed by the i-th PRU during collided slots in the previous
WIT superframe, as defined by Equation (15).

Ecol,i = EcolTx,i + EcolRx,i + Ecol Idle,i , (15)

where EcolTx,i is the energy consumed by the data packet transmission of the i-th PRU
when one or more other PRUs transmits the data packet simultaneously. EcolRx,i is the
energy consumed by the i-th PRU due to the collision caused by two or more PRUs
including at least one neighbor. Specifically, EcolRx,i is the energy consumption of the i-th
PRU when it overhears the data packets transmitted by two or more PRUs, including
Sensors 2021, 21, 7617 8 of 21

at least one neighbor, simultaneously. Ecol Idle,i is the energy consumed by the i-th PRU
when two or more PRUs, excluding itself and its neighbors, transmit the data packet
simultaneously. In this case, two or more PRUs outside the transmission range of the
i-th PRU transmit the data packet simultaneously. EcolTx,i , EcolRx,i , and Ecol Idle,i can be
calculated by Equations (16)–(18), respectively.
!
n −1  
n−1
EcolTx,i = ncolSlots ( Etx Ldata + Eidle L DIFS ) τ ∑ τ k (1 − τ ) n − k −1 , 1 ≤ k ≤ ( n − 1 ) (16)
k =1
k
!
n −1  
n−1
EcolRx,i = ncolSlots ( Erx Ldata + Eidle L DIFS )(1 − τ ) ∑ k
τ (1 − τ ) n − k −1
− Ecol Idle,i , 2 ≤ k ≤ (n − 1) (17)
k =2
k
n−nnbr,i −1   !
n − nnbr,i − 1
Ecol Idle,i = ncolSlots Eidle (1 − τ )(1 − τ ) nnbr,i
∑ k
τ k , 2 ≤ k ≤ (n − nnbr,i − 1) (18)
k =2

Finally, Eidle,i is the energy consumed by the i-th PRU during the idle slots in the
previous WIT superframe when its D-Tx/Rx is in an idle state, as defined by Equation (19).

Eidle,i = nidleSlots Eidle L BP (19)

For estimating the residual energy of the i-th PRU, the PTU must calculate not only
the energy consumed by the i-th PRU but also the energy harvested by the i-th PRU (i.e.,
Erx,i ) in the previous WET superframe, as defined by Equation (20).
n DPS,i
Erx,i = ∑ EDPS,i , (20)
k =1

where n DPS,i is the number of DPSs allocated to the i-th PRU in the previous WET su-
perframe. EDPS,i is the energy harvested by the i-th PRU during one DPS, as defined by
Equation (21).
EDPS,i = ηPr,i LWET , (21)
where η is the energy harvesting efficiency of the PRU, Pr,i is the power received per
second of the i-th PRU (refer to Equation (1) in Section 2), and LWET is the length of the
WET subslot.
Consequently, the residual energy of the i-th PRU (i.e., Eres,i ) can be represented by
Equation (22).

Eres,i = Eres,i − Ebeacon,i − Esucc,i − Ecol,i − Eidle,i + Erx,i (22)

Based on the knowledge of the residual energy estimation for individual PRUs, the
PTU performs the DPS allocation at the beginning of every WET superframe. Algorithm
1 presents the DPS allocation procedure, which consists of three operations: (1) the cal-
culation of the number of DPSs required for individual PRUs to be fully charged, (2)
the calculation of the number of DPSs to be allocated to individual PRUs, and (3) the
determination of charging priority for individual PRUs.
Sensors 2021, 21, 7617 9 of 21

Algorithm 1. DPS allocation

INITIALIZE NDPS to [],NsortedDPS to [],IDPS to [],IstartDPS to [],SDPS to 0,
1:
nreqDPS,i to 0, cnt to 2, n avaDPS to ntotalDPS − 1, Emax,i to 1
2: Calculation of the number of DPSs required for /*individual PRUs to be fully charged */
3: FOR each PRU, i, i ∈ [1, n]
4: nreqDPS,i ← ( Emax,i − Eres,i )/EDPS,i
5: SDPS ← SDPS + nreqDPS,i
6: ENDFOR
7: /* Calculation of the number of DPSs to be allocated to individual PRUs */
8: FOR each PRU, i, i ∈ [1, n]
9: n DPS,i ← round(n avaDPS (nreqDPS,i /SDPS ))
10: NDPS [i ] ← n DPS,i
11: ENDFOR
12: /* Determination of the charging priority for individual PRUs */
13: [NsortedDPS , IDPS ] ← sort(NDPS ,0 descend0 )
14: FOR each PRU, i, i ∈ [1, n]
15: IstartDPS [IDPS [i ]] ← cnt
16: cnt ← cnt + NsortedDPS [i ]
17: ENDFOR
18: RETURN NDPS and IstartDPS

In the algorithm, the PTU initializes the attributes and variables (i.e., IDPS , IstartDPS ,
NDPS , NsortedDPS , SDPS , nreqDPS,i , cnt, n avaDPS , and Emax,i ), where IDPS and IstartDPS are
the attributes to store the indices of the PRUs, NDPS and NsortedDPS are the attributes to
track the number of DPSs allocated to them, and SDPS , nreqDPS,i , cnt, n avaDPS , and Emax,i
are the variables for counting the number of DPSs. IDPS includes the indices of elements
in NDPS before NDPS is sorted in descending order as NsortedDPS . IstartDPS includes the
starting indices of DPSs allocated to individual PRU in the WET superframe. NDPS includes
the numbers of DPSs allocated to PRUs. NsortedDPS is NDPS sorted in descending order
according to the values of elements included in NDPS . IDPS , IstartDPS , NDPS , and NsortedDPS
are represented by one-dimensional arrays. SDPS is the sum of the number of DPSs
required by PRUs, nreqDPS,i , which is the number of DPSs required for the i-th PRU to be
fully charged from the current residual energy to the maximum battery capacity. cnt is
a counter value used to calculate the starting index of DPSs allocated to individual PRU
in the WET superframe. n avaDPS is the number of available DPSs in the WET superframe
excluding the first DPS and DPSs already allocated to PRUs, initialized to ntotalDPS − 1.
ntotalDPS is the total number of DPSs in the WET superframe.
In the first operation, the PTU obtains the number of DPSs required for individual
PRUs to be fully charged (i.e., nreqDPS,i ), calculated using the energy required for each
PRU’s battery to be charged to its maximum battery capacity (i.e., Emax,i − Eres,i ) and the
energy harvested during one DPS (i.e., EDPS,i ) (line 4). The PTU then calculates the sum of
the number of DPSs required by all PRUs (i.e., SDPS ) (line 5).
In the second operation, the PTU obtains the number of DPSs to be allocated to
individual PRUs (i.e., n DPS,i ) according to the ratio of the number of DPSs required by each
PRU to the sum of the number of DPSs required by all PRUs (i.e., nreqDPS,i /SDPS ) (line 9).
Accordingly, the PTU maintains the list of the number of DPSs to be allocated to individual
PRUs (i.e., NDPS [i ]) (line 10).
Finally, the PTU determines the charging priority for individual PRUs by sorting
NDPS [i ] in descending order according to the number of DPSs (line 13). The starting index
of the DPSs in the WET superframe allocated to each PRU is calculated according to the
charging priority for individual PRUs (line 15). Consequently, the PRU with low residual
energy can perform the energy harvesting ahead of other PRUs.
After completing the DPS allocation algorithm, the PTU includes the number of DPSs
(i.e., NDPS ) and the starting index (i.e., IstartDPS ) of DPSs allocated to PRUs in the PTU
Sensors 2021, 21, 7617 10 of 21

beacon and broadcasts the PTU beacon. Based on the results of DPS scheduling, the PTU
transfers power to the PRUs, and the PRUs perform energy harvesting.

4. Performance Evaluation
We evaluated the performance of REE-MAC using experimental simulations with the
MATLAB simulator. The simulation results were compared with those of FF-WPT [25] and
HE-MAC [19]. FF-WPT is an out-of-band solution that transfers power to devices using a
different frequency band separated from that used to transmit data packets. In contrast,
HE-MAC is an in-band solution that performs both power transfer and data transmission
within the same frequency band. In the following subsections, we present in detail the
simulation setup and configuration and the simulation results.

4.1. Simulation Configuration

In the simulation, we considered a WPSN consisting of one PTU and multiple PRUs.
We assumed that each PRU is randomly deployed within the communication range of
the PTU, set to 4 m. We further assumed that the PRU always has the data packets to
transmit to the PTU. In the simulation, the number of PRUs varies from 2 to 20. The
performance of REE-MAC was compared with those of FF-WPT [25] and HE-MAC [19]
in terms of average harvested energy, average consumed energy, average freezing time,
residual energy distribution, throughput distribution, and fairness indices for residual
energy and throughput.
We investigated fairness by considering the freezing state, which indicates a state of
the device in which the PRU cannot transmit the data packets due to a lack of residual
energy. In the simulation, the PRU enters the freezing state when the residual energy is
less than a pre-configured energy threshold. Then, if it exceeds a pre-configured active
threshold, it is released from the freezing state. The active threshold indicates the minimum
residual energy required for the PRU to resume the data packet transmission in the freezing
state. It is set larger than the energy threshold to prevent the PRUs from entering the
freezing state again immediately after they are released from the freezing state.
In FF-WPT, the WET operation is performed according to TDMA-based round-robin
scheduling, while the WIT operation is performed using the CSMA/CA scheme. In FF-
WPT, the interval required for exchanging control messages was set to 300 µs. Furthermore,
the number of energy frames of each PRU is determined considering packet size and
harvested energy per second. In HE-MAC, the arbitration inter-frame spaces (AIFSs) for
PTU and PRU (i.e., AIFSPTU and AIFSPRU ) were set to 50 and 70 µs, respectively. The
simulation was iterated 50 times. The detailed simulation parameters are listed in Table 1.
Sensors 2021, 21, 7617 11 of 21

Table 1. Simulation parameters.

Parameter Value Parameter Value

Number of PRUs 2–20 d 0–4 m
Maximum battery
Data rate 2 Mbps 1 mJ
capacity
WET superframe 1s Initial residual energy 0.6 mJ
WIT superframe 1s Active threshold 0.6 mJ
DPS length 10 ms Energy threshold 0.1 mJ
Number of DPSs 100 CWmin 31
Beacon subslot 40 µs CWmax 1023
P-Tx/Rx switching
10 µs Gt 12
subslot
WET subslot 9950 µs Gr 1
RTS 20 bytes Ptx 31.47 mA
CTS 14 bytes Prx 26.94 mA
ACK 14 bytes Pidle 0.00156 mA
SIFS 10 µs Pt 3000 mW
DIFS 50 µs α 2.7
Beacon 15 bytes η 0.85
Packet size 100–200 bytes Slot length 20 µs

4.2. Simulation Results

Figure 5a,b illustrate the variations in the average harvested energy for 100- and
200-byte packets, respectively. The average harvested energy decreases as the number of
PRUs in the network increases due to decreased energy harvested by individual PRUs per
WET superframe. In both figures, REE-MAC exhibits a higher average harvested energy
compared with both FF-WPT and HE-MAC. Regarding the DPS allocation operation, REE-
MAC requires a shorter time overhead than FF-WPT. In REE-MAC, the PTU performs
DPS allocation by estimating the residual energy for individual PRUs and informs the
PRUs of the number of allocated DPSs through a PTU beacon at the beginning of the WET
superframe. Therefore, REE-MAC entails only time overhead for sending one PTU beacon.
2021, 21, x FOR PEER REVIEW 12 of 22
Sensors 2021, 21, 7617 12 of 21

(a)

(b)
Figure 5. Average harvested
Figure energy:
5. Average (a) packet
harvested size of
energy: (a)100 bytes;
packet (b)ofpacket
size size of
100 bytes; (b)200 bytes.
packet size of 200 bytes.

As presented inIn Table

contrast, FF-WPT
2, the average incurs a longenergy
harvested time overhead
of REE-MACbecause,
andin FF-WPT,
FF-WPT arethe PTU needs
almost identical, even if the packet size increases from 100 to 200 bytes, because theitself and each
to exchange control messages with all PRUs to obtain the distance between
PRU.
packet size does Furthermore,
not HE-MAC
affect the average exhibitsenergy
harvested the lowest average
of PRUs harvested energy
in out-of-band ap- because, in
HE-MAC,
proaches. However, the exchange
in HE-MAC, the of controlharvested
average messages,energy
the transmission of data packets,
decreases slightly when and energy
harvesting
the packet size increasesare all performed
from 100 to 200inbytes the same frequency
because band.
the proportion of the energy
As presented
harvesting time within the WET in Table 2, the average
superframe harvested
is reduced. energy of REE-MAC
Quantitatively, the averageand FF-WPT are
almost identical, even if the packet size increases from
harvested energy of REE-MAC is 17.79% and 18.38% higher than that of FF-WPT when 100 to 200 bytes, because the packet
size does not affect the average harvested energy of
the PRUs transmit the 100- and 200-byte packets, respectively. Moreover, the average PRUs in out-of-band approaches.
However,
harvested energy in HE-MAC,
of REE-MAC the average
is 132.15% harvested
and 159.04% energy
higher decreases
than that ofslightly
HE-MAC when the packet
size increases from 100 to 200 bytes because
when the PRUs transmit the 100- and 200-byte packets, respectively. the proportion of the energy harvesting time
within the WET superframe is reduced. Quantitatively, the average harvested energy of
REE-MAC is 17.79% and 18.38% higher than that of FF-WPT when the PRUs transmit
the 100- and 200-byte packets, respectively. Moreover, the average harvested energy of
REE-MAC is 132.15% and 159.04% higher than that of HE-MAC when the PRUs transmit
the 100- and 200-byte packets, respectively.
Sensors 2021, 21, 7617 13 of 21

Table 2. Average harvested energy (mJ).

Packet Number of PRUs

Protocol
Size 2 4 6 8 10 12 14 16 18 20
REE-MAC 2.65 1.56 1.03 0.76 0.61 0.52 0.48 0.41 0.36 0.31
100 bytes FF-WPT 2.57 1.47 0.95 0.67 0.51 0.41 0.36 0.30 0.25 0.21
HE-MAC 1.31 0.72 0.48 0.35 0.28 0.24 0.20 0.18 0.16 0.14
REE-MAC 2.65 1.56 1.03 0.76 0.61 0.52 0.46 0.41 0.36 0.31
200 bytes FF-WPT 2.57 1.47 0.95 0.67 0.51 0.41 0.36 0.30 0.25 0.21
HE-MAC 1.20 0.65 0.44 0.32 0.25 0.21 0.18 0.16 0.14 0.13

Figure 6a,b illustrate the variations in the average consumed energy for 100- and
200-byte packets, respectively. In REE-MAC, the average consumed energy decreases as
the number of PRUs increases because the number of data packets transmitted by PRUs
gradually decreases due to the increase in collisions and backoff delay. Consequently, the
number of transmissions of PRUs is reduced, reducing the energy consumed by PRUs. In
FF-WPT, as in REE-MAC, the average consumed energy tends to decrease overall as the
number of PRUs increases.
Furthermore, FF-WPT exhibits an average consumed energy similar to REE-MAC.
However, in some sections (i.e., when the number of PRUs is 14 to 18), the average
consumed energy of FF-WPT slightly increases. As the number of PRUs increases, the
energy each PRU can harvest decreases, and accordingly, the number of PRUs entering
the freezing state increases. Therefore, a relatively small number of PRUs transmit data
packets to the PTU, and the average consumed energy can increase due to the reduced
contention level.
HE-MAC exhibits lower average consumed energy compared with both REE-MAC
and FF-WPT. In HE-MAC, the PRUs have relatively few transmission opportunities for data
packets because both WET and WIT operations are performed within the same frequency
band. Therefore, a small number of data packet transmissions reduces the energy consumed
by the PRU. Quantitatively, when the PRUs transmit 100- and 200-byte packets, the average
consumed energy of REE-MAC is 7.79% and 8.29% higher than that of FF-WPT, respectively.
Moreover, it is 43.78% and 43.74% higher compared with HE-MAC, respectively.
Figure 7a,b illustrate the variations in the average freezing time for 100- and 200-byte
packets, respectively. The freezing time is the time the PRU is in the freezing state. As the
number of PRUs increases, the energy harvested by each PRU decreases, and accordingly,
the average freezing time of PRUs is highly likely to increase. REE-MAC exhibits a shorter
average freezing time than both FF-WPT and HE-MAC. In REE-MAC, the DPS allocation is
performed considering the residual energy of individual PRUs. Accordingly, in REE-MAC,
PRUs with less residual energy are allocated more DPSs within the superframe. Therefore,
all PRUs in the network maintain similar residual energy, and the time they are in the
freezing state is relatively short.
 and 200-byte packets, respectively. In REE-MAC, the average consumed energy de-
creases as the number of PRUs increases because the number of data packets transmitted
by PRUs gradually decreases due to the increase in collisions and backoff delay. Conse-
quently, the number of transmissions of PRUs is reduced, reducing the energy consumed
Sensors 2021, 21, 7617
by PRUs. In FF-WPT, as in REE-MAC, the average consumed energy tends to decrease 14 of 21
overall as the number of PRUs increases.

14 of 22

Furthermore, FF-WPT exhibits an average consumed energy similar to REE-MAC.

wever, in some sections (i.e., when the number of PRUs is 14 to 18), the average con-
med energy of FF-WPT slightly increases. As the number of PRUs increases, the energy
h PRU can harvest decreases, and accordingly, the number of PRUs entering the
ezing state increases. Therefore, a relatively small number of PRUs transmit data
ckets to the PTU, and the average consumed energy can increase(a) due to the reduced
ntention level.
HE-MAC exhibits lower average consumed energy compared with both REE-MAC
d FF-WPT. In HE-MAC, the PRUs have relatively few transmission opportunities for
a packets because both WET and WIT operations are performed within the same
quency band. Therefore, a small number of data packet transmissions reduces the en-
y consumed by the PRU. Quantitatively, when the PRUs transmit 100- and 200-byte
ckets, the average consumed energy of REE-MAC is 7.79% and 8.29% higher than that
FF-WPT, respectively. Moreover, it is 43.78% and 43.74% higher compared with
-MAC, respectively.
Figure 7a and 7b illustrate the variations in the average freezing time for 100- and
-byte packets, respectively. The freezing time is the time the PRU is in the freezing
te. As the number of PRUs increases, the energy harvested by each PRU decreases, and
ordingly, the average freezing time of PRUs is highly likely to increase. REE-MAC
hibits a shorter average freezing time than both FF-WPT and HE-MAC. In REE-MAC,
DPS allocation is performed considering the residual energy of individual PRUs.
cordingly, in REE-MAC,Sensors
PRUs2021,
with
21, xless
FORresidual energy are allocated more DPSs
PEER REVIEW
hin the superframe. Therefore, all PRUs in the network maintain(b) similar residual en-
y, and the time they are in the freezing state is relatively short.
Figure 6. Average consumed
Figure energy:
6. Average (a) packet
consumed size of(a)
energy: 100 bytes;size
packet (b) of
packet size of(b)
100 bytes; 200 bytes.size of 200 bytes.
packet

(a) (b)
Figure
Figure 7. Average freezing time: 7. Average
(a) packet size freezing time:(b)
of 100 bytes; (a)packet
packetsize
sizeofof200
100bytes.
bytes; (b) packet size of 200 bytes.

When the sizes of data packets transmitted by PRUs are 100 and 200 bytes,
erage freezing time of REE-MAC is almost zero until the numbers of PRUs are 10
respectively. However, if the numbers of PRUs exceed 10 and 12, the average f
Sensors 2021, 21, 7617 15 of 21

When the sizes of data packets transmitted by PRUs are 100 and 200 bytes, the
average freezing time of REE-MAC is almost zero until the numbers of PRUs are 10 and 12,
respectively. However, if the numbers of PRUs exceed 10 and 12, the average freezing time
of REE-MAC increases as the number of PRUs increases. FF-WPT consistently exhibits a
longer average freezing time than both REE-MAC and HE-MAC because, in FF-WPT, the
PTU transfers power to PRUs according to the distance without considering the residual
energy of PRUs. Furthermore, individual PRUs harvest smaller energy due to the exchange
of control messages required for WET operation. Accordingly, in FF-WPT, PRUs enter the
freezing state more frequently.
When the number of PRUs is more than 12, the average freezing time of FF-WPT
increases gradually. When the number of PRUs is increased to more than 12, the PRUs enter
the freezing state more quickly due to the decrease in harvested energy. Therefore, it takes
longer for the residual energy of PRUs in the freezing state to reach the active threshold.
In HE-MAC, the PRU occupying the channel uses the harvest-then-transmit scheme.
Therefore, the PRU first harvests the energy required to transmit the data packet and then
uses it to transmit the data packet. Other PRUs maintain an idle state to minimize energy
consumption. Consequently, the average freezing time of HE-MAC is shorter than that of
FF-WPT. Quantitatively, when the PRUs transmit 100- and 200-byte packets, the average
freezing time of REE-MAC is 72.03% and 90.04% shorter than that of FF-WPT, respectively.
It is also 47.26% and 81.15% shorter than that of HE-MAC.
Figure 8a,b illustrate the variations in the residual energy distribution of individual
PRUs for 100- and 200-byte packets, respectively. Both figures indicate the residual energy
distribution of individual PRUs in a specific round of the experimental simulation. In
REE-MAC and FF-WPT, when the packet size increases, the residual energy of individual
PRUs decreases due to an increase in the consumed energy.
In HE-MAC, the number of dead PRUs increases. A dead PRU indicates a PRU with
a residual energy of zero. In REE-MAC, the difference between the residual energy of
individual PRUs is slight compared with both FF-WPT and HE-MAC. REE-MAC enables
PRUs to maintain similar residual energy through the DPS allocation considering the
residual energy. In FF-WPT, the fluctuation in the residual energy distribution of individual
PRUs is larger than that of REE-MAC because the DPSs are allocated considering only
the distance between the PTU and individual PRUs. In HE-MAC, the residual energy of
individual PRUs is almost zero except when the number of PRUs is 2 because the PRUs
consume additional energy in addition to data packet transmission, and the residual energy
of the PRUs gradually decreases. Consequently, in HE-MAC, more dead PRUs occur
compared with REE-MAC and FF-WPT.
Figure 9a,b illustrate the variations in the throughput distribution of individual PRUs
for 100- and 200-byte packets, respectively. Both figures indicate the throughput distribu-
tion of individual PRUs in a specific round of the experimental simulation. The fluctuation
of the throughput distribution increases as the number of PRUs in the network increases.
Moreover, when the packet size increases, the throughput of individual PRUs increases due
to the decrease in the backoff delay. REE-MAC exhibits a constant throughput distribution
regardless of the number of PRUs, compared with FF-WPT and HE-MAC. In REE-MAC,
all PRUs in the network maintain similar throughput performance.
From the results of Figures 7 and 8, in REE-MAC, the PRUs maintain the shortest
freezing time, on average, and dead PRUs with zero residual energy rarely occur. Therefore,
compared with FF-WPT and HE-MAC, the PRUs using REE-MAC can have relatively
even transmission opportunities. However, in FF-WPT and HE-MAC, the transmission
opportunity is biased toward some PRUs as the number of PRUs in the network increases.
Accordingly, in FF-WPT and HE-MAC, the throughput distribution for specific PRUs
becomes severely concentrated.
 Sensors 2021, 21, x FOR PEER REVIEW 16 of 22
Sensors 2021, 21, 7617 16 of 21

0.5

0
2 4 6 8 10 12 14 16 18 20

0.5

0
2 4 6 8 10 12 14 16 18 20

0.5

0
2 4 6 8 10 12 14 16 18 20
Number of PRUs

(a)

0.5

0
2 4 6 8 10 12 14 16 18 20
1

0.5

0
2 4 6 8 10 12 14 16 18 20

0.5

0
2 4 6 8 10 12 14 16 18 20
Number of PRUs

(b)
Figure
Figure8.8.Residual
Residualenergy
energydistribution of individual
distribution PRUs:
of individual (a) packet
PRUs: size of
(a) packet 100ofbytes;
size (b) packet
100 bytes; (b) packet
size of 200 bytes.
size of 200 bytes.

In HE‐MAC,
Figure 10a,bthe number the
illustrate of dead PRUsindex
fairness increases. A dead PRU
for residual indicates
energy a PRU
for 100- andwith
200-byte
apackets,
residualrespectively.
energy of zero. In REE‐MAC, the difference between the residual
The fairness index for residual energy (Fres ) can be calculatedenergy of by
individual PRUs is
Equation (23) [30]. slight compared with both FF‐WPT and HE‐MAC. REE‐MAC enables
PRUs to maintain similar residual energy through  theDPS allocation considering the re‐
n
∑ xi2 energy distribution of individual
sidual energy. In FF‐WPT, the fluctuation in the residual
Fres = thei DPSs
PRUs is larger than that of REE‐MAC because n ,are allocated considering only the (23)
distance between the PTU and individual PRUs. xi2
n ·In∑HE‐MAC, the residual energy of indi‐
vidual PRUs is almost zero except when the number i =1 of PRUs is 2 because the PRUs
where n is the number of PRUs and xi is the fairness parameter, which represents the
residual energy of the i-th PRU.
 tion of individual PRUs in a specific round of the experimental simulation. The fluctuation
of the throughput distribution increases as the number of PRUs in the network increases.
Moreover, when the packet size increases, the throughput of individual PRUs increases
due to the decrease in the backoff delay. REE‐MAC exhibits a constant throughput distri‐
Sensors 2021, 21, 7617 bution regardless of the number of PRUs, compared with FF‐WPT and HE‐MAC. In REE‐17 of 21
MAC, all PRUs in the network maintain similar throughput performance.

0.4

0.2

0
2 4 6 8 10 12 14 16 18 20

0.4

0.2

0
2 4 6 8 10 12 14 16 18 20

0.4

0.2

0
2 4 6 8 10 12 14 16 18 20
Number of PRUs

(a)

0.4

0.2

0
2 4 6 8 10 12 14 16 18 20

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10

0.4

0.2

0
2 4 6 8 10 12 14 16 18 20
Number of PRUs

(b)
Figure
Figure9.9.Throughput
Throughputdistribution
distributionofof
individual PRUs:
individual (a)(a)
PRUs: packet sizesize
packet of 100 bytes;
of 100 (b) (b)
bytes; packet sizesize of
packet
of 200 bytes.
200 bytes.

In REE-MAC, the fairness index for residual energy decreases and then increases
again as the number of PRUs increases. For 100- and 200-byte packets, when the number
of PRUs is 10 and 14 or less, respectively, the fairness index for residual energy of REE-
MAC decreases as the number of PRUs increases. As the number of PRUs in the network
increases, the difference in the residual energy between individual PRUs increases due to
a decrease in the energy harvested by individual PRUs and the difference in the energy
consumed by individual PRUs (refer to Figure 8).
In contrast, when the numbers of PRUs are larger than 10 and 14, respectively, the
fairness index for residual energy of REE-MAC increases as the number of PRUs increases.
As the number of PRUs increases, the energy harvested by individual PRUs decreases,
and the number of PRUs in the freezing state increases accordingly. PRUs in the freezing
state perform only WET operations until their residual energy reaches the active threshold.
Therefore, as the residual energy of many PRUs in the freezing state approaches the active
threshold, the fairness index for residual energy increases.
  2 
  xi 
Fres   n  ,
i
(23)
n   xi2
i 1

Sensors 2021, 21, 7617 18 of 21

where n is the number of PRUs and x i is the fairness parameter, which represents the
residual energy of the i-th PRU.

(a)

(b)
Figure 10. Fairness
Figure index
10. Fairness for residual
index energy:
for residual (a) packet
energy: sizesize
(a) packet of 100 bytes;
of 100 (b) packet
bytes; sizesize
(b) packet of 200 bytes.
of 200 bytes.

However, in FF-WPT and HE-MAC, the fairness index for residual energy decreases
as the number of PRUs in the network increases. As depicted in Figure 7, the gap in
transmission opportunities between PRUs increases as the average freezing time of PRUs
increases. This gap increases the difference between the energy consumed by individual
PRUs. Therefore, the fairness index for residual energy gradually decreases as the difference
between the residual energy of individual PRUs increases. Quantitatively, when the PRUs
transmit 100- and 200-byte packets, the fairness index for residual energy of REE-MAC
is 95.34% and 116.23% higher than that of FF-WPT, respectively. It is also 165.88% and
276.00% higher compared with that of HE-MAC, respectively.
Figure 11a,b illustrate the fairness index for throughput for 100- and 200-byte packets,
respectively. The fairness index for throughput (Fth ) can be calculated by Equation (24).
 
n
2
∑ yi
i
Fth = n , (24)
n · ∑ y2i
i =1
se the average freezing time of PRUs increases due to the decrease in the energy
vested by individual PRUs.
HE-MAC exhibits a higher fairness index for throughput than FF-WPT regardless of
packet size. In HE-MAC, PRUs that are not in the freezing state maintain similar re-
Sensors
ual energy 2021, 21,
through the7617
harvest-then-transmit scheme, and thus they have a relatively 19 of 21
ilar transmission opportunity. Quantitatively, when the PRUs transmit 100- and
byte packets, the fairness index for throughput of REE-MAC is 98.58% and 91.80%
her than that of FF-WPT, respectively. It is also 44.46% and 55.91% higher compared
where n is the number of PRUs and yi is the fairness parameter indicating the number of
h HE-MAC, respectively. data packets that the i-th PRU transmits to the PTU. (a)

(a) (b)
Figure 11.
Figure 11. Fairness index for throughput: (a) Fairness index
packet size for bytes;
of 100 throughput: (a) packet
(b) packet size ofsize
200 of 100 bytes; (b) packet size of 20
bytes.

In each case5.ofConclusions
100- and 200-byte packets, REE-MAC maintains the fairness index for
throughputs of 0.924 and 0.956 or higher, respectively, regardless of the number of PRUs.
This paper presents the REE-MAC protocol for WPSNs, which aims to
As the number of PRUs increases, the energy harvested by individual PRUs decreases and
overhead due to control messages for scheduling the WET operation and prov
the difference in energy consumed by individual PRUs increases. Nevertheless, in REE-
MAC, the residual energy of PRUs is maintained similar to each other due to DPS allocation
considering the residual energy of individual PRUs. Accordingly, individual PRUs achieve
a high fairness index for throughput by having a similar transmission opportunity.
In contrast, FF-WPT exhibits a lower fairness index for throughput compared with
both REE-MAC and HE-MAC. From the results in Figure 7, in FF-WPT, PRUs maintain
the longest freezing time on average. The difference in throughput performance between
PRUs in FF-WPT becomes significant because the increase of the freezing time causes the
transmission opportunities to be biased to some PRUs. The fairness index for throughput
of FF-WPT decreases as the number of PRUs increases regardless of the packet size because
the average freezing time of PRUs increases due to the decrease in the energy harvested by
individual PRUs.
(b) HE-MAC exhibits a higher fairness index for throughput than FF-WPT regardless
of the packet size. In HE-MAC, PRUs that are not in the freezing state maintain similar
re 11. Fairness index for throughput: (a) packet size of 100 bytes; (b) packet size of 200 bytes.
residual energy through the harvest-then-transmit scheme, and thus they have a relatively
similar transmission opportunity. Quantitatively, when the PRUs transmit 100- and 200-
onclusions
byte packets, the fairness index for throughput of REE-MAC is 98.58% and 91.80% higher
This paper presents the REE-MAC than protocol forFF-WPT,
that of WPSNs,respectively.
which aims Ittoisreduce
also 44.46% and 55.91% higher compared with
rhead due to control messages for scheduling the WET
HE-MAC, respectively. operation and provide fair-

5. Conclusions
This paper presents the REE-MAC protocol for WPSNs, which aims to reduce over-
head due to control messages for scheduling the WET operation and provide fairness for
data transmission opportunities to the sensor devices. REE-MAC achieves low overhead by
numerically estimating the residual energy of individual PRUs without exchanging control
messages. Furthermore, in REE-MAC, the PTU allocates the DPSs inversely proportional
to the residual energy of individual PRUs, so that all PRUs in the network maintain similar
residual energy. Thereby, it minimizes the energy depletion of some PRUs and provides
Sensors 2021, 21, 7617 20 of 21

individual PRUs with a fair data transmission opportunity. At the beginning of each
superframe, the PTU calculates the consumed and harvested energy of individual PRUs
and then estimates their residual energy. It then performs the DPS allocation based on the
results of the residual energy estimation.
We conducted an experimental simulation to evaluate the performance of REE-MAC
under the environment of changing network size and packet size. The results demon-
strate that REE-MAC uses the residual energy estimation to reduce unnecessary waste of
bandwidth due to the exchange of control messages, increasing the energy harvested by
individual PRUs. Moreover, REE-MAC prevents the DPSs from biased allocation to some
PRUs, reducing the freezing time of the PRUs due to lack of energy.
These operations of REE-MAC give similar transmission opportunities to PRUs in
the network, ensuring higher fairness compared with FF-WPT and HE-MAC in terms of
residual energy and throughput. On average, REE-MAC achieves 18.08% and 145.60%
higher energy harvested, 81.03% and 64.21% shorter average freezing time, 105.79% and
220.94% higher fairness index for residual energy, and 95.19% and 50.18% higher fairness
index for throughput, compared with FF-WPT and HE-MAC, respectively.

Author Contributions: Conceptualization, S.-B.L. and E.-J.K.; methodology, S.-B.L., J.-H.K. and
E.-J.K.; software, S.-B.L., J.-H.K. and E.-J.K.; validation, S.-B.L. and E.-J.K.; formal analysis, S.-B.L. and
E.-J.K.; investigation, J.-H.K. and E.-J.K.; resources, E.-J.K.; data curation, J.-H.K.; writing—original
draft preparation, S.-B.L., J.-H.K. and E.-J.K.; writing—review and editing, S.-B.L., J.-H.K. and E.-J.K.;
visualization, S.-B.L.; supervision, E.-J.K.; project administration, E.-J.K.; funding acquisition, J.-H.K.
and E.-J.K. All authors have read and agreed to the published version of the manuscript.
Funding: This research was supported by Basic Science Research Program through the National Re-
search Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2020R1I1A3052733 and
NRF-2019R1I1A1A01059787). This work was also supported by the NRF grant funded by the Korea
government (MSIT) (NRF-2021R1C1C2095696).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No new data were created or analyzed in this study. Data sharing is
not applicable to this article.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Choi, K.W.; Ginting, L.; Rosyady, P.A.; Aziz, A.A.; Kim, D.I. Wireless-powered sensor networks: How to realize. IEEE Trans. Wirel.
Commun. 2017, 16, 221–234. [CrossRef]
2. Perera, T.D.P.; Jayakody, D.N.K.; Sharma, S.K.; Chatzinotas, S.; Li, J. Simultaneous wireless information and power transfer
(SWIPT): Recent advances and future challenges. IEEE Commun. Surv. Tutor. 2018, 20, 264–302. [CrossRef]
3. Setiawan, D.; Aziz, A.A.; Kim, D.I.; Choi, K.W. Experiment, modeling, and analysis of wireless-powered sensor network for
energy neutral power management. IEEE Syst. J. 2018, 12, 3381–3392. [CrossRef]
4. Bi, S.; Ho, C.K.; Zhang, R. Wireless powered communication: Opportunities and challenges. IEEE Commun. Mag. 2015, 53,
117–125. [CrossRef]
5. Ramezani, P.; Jamalipour, A. Toward the evolution of wireless powered communication networks for the future Internet of Things.
IEEE Netw. 2017, 31, 62–69. [CrossRef]
6. Du, R.; Shokri-Ghadikolaei, H.; Fischione, C. Wirelessly-powered sensor networks: Power allocation for channel estimation and
energy beamforming. IEEE Trans. Wirel. Commun. 2020, 19, 2987–3002. [CrossRef]
7. Du, R.; Xiao, M.; Fischione, C. Optimal node deployment and energy provision for wirelessly powered sensor networks. IEEE J.
Sel. Areas Commun. 2019, 37, 407–423. [CrossRef]
8. López, O.L.A.; Alves, H.; Souza, R.D.; Montejo-Sánchez, S.; Fernández, E.M.G.; Latva-Aho, M. Massive wireless energy transfer:
Enabling sustainable IoT towards 6G era. IEEE Internet Things J. 2021, 8, 8816–8835. [CrossRef]
9. Clerckx, B.; Huang, K.; Varshney, L.R.; Ulukus, S.; Alouini, M.-S. Wireless power transfer for future networks: Signal processing
machine learning computing and sensing. IEEE J. Sel. Top. Signal Process. 2021, 15, 1060–1094. [CrossRef]
10. Chu, Z.; Zhou, F.; Zhu, Z.; Hu, R.Q.; Xiao, P. Wireless powered sensor networks for Internet of Things: Maximum throughput and
optimal power allocation. IEEE Internet Things J. 2018, 5, 310–321. [CrossRef]
Sensors 2021, 21, 7617 21 of 21

11. Mai, V.V.; Shin, W.-Y.; Ishibashi, K. Wireless power transfer for distributed estimation in sensor networks. IEEE J. Sel. Top. Signal
Process. 2017, 11, 549–562. [CrossRef]
12. Hong, Y.-W.P.; Hsu, T.-C.; Chennakesavula, P. Wireless power transfer for distributed estimation in wireless passive sensor
networks. IEEE Trans. Signal Process. 2016, 64, 5382–5395. [CrossRef]
13. Shi, H.; Prasad, R.V.; Onur, E.; Niemegeers, I.G.M.M. Fairness in wireless networks: Issues measures and challenges. IEEE
Commun. Surv. Tutor. 2014, 16, 5–24. [CrossRef]
14. Lu, X.; Wang, P.; Niyato, D.; Kim, D.I.; Han, Z. Wireless networks with RF energy harvesting: A contemporary survey. IEEE
Commun. Surv. Tutor. 2015, 17, 757–789. [CrossRef]
15. Lei, M.; Zhang, X.; Ding, H.; Yu, B. Fairness-aware resource allocation in multi-hop wireless powered communication networks
with user cooperation. Sensors 2018, 18, 1890. [CrossRef]
16. Zhu, Z.; Peng, J.; Gu, X.; Li, H.; Liu, K.; Zhou, Z.; Liu, W. Fair resource allocation for system throughput maximization in mobile
edge computing. IEEE Access 2018, 6, 5332–5340. [CrossRef]
17. Naderi, M.Y.; Nintanavongsa, P.; Chowdhury, K.R. RF-MAC: A medium access control protocol for re-chargeable sensor networks
powered by wireless energy harvesting. IEEE Trans. Wirel. Commun. 2014, 13, 3926–3937. [CrossRef]
18. Iqbal, A.; Kim, Y.; Lee, T.-J. Access mechanism in wireless powered communication networks with harvesting access point. IEEE
Access 2018, 6, 37556–37567. [CrossRef]
19. Ha, T.; Kim, J.; Chung, J.-M. HE-MAC: Harvest-then-transmit based modified EDCF MAC protocol for wireless powered sensor
networks. IEEE Trans. Wirel. Commun. 2018, 17, 3–16. [CrossRef]
20. Ju, H.; Zhang, R. Throughput maximization in wireless powered communication networks. IEEE Trans. Wirel. Commun. 2014, 13,
418–428. [CrossRef]
21. Hadzi-Velkov, Z.; Nikoloska, I.; Chingoska, H.; Zlatanov, N. Opportunistic scheduling in wireless powered communication
networks. IEEE Trans. Wirel. Commun. 2017, 16, 4106–4119. [CrossRef]
22. Yao, Q.; Huang, A.; Shan, H.; Quek, T.Q.S.; Wang, W. Delay-aware wireless powered communication networks—Energy balancing
and optimization. IEEE Trans. Wirel. Commun. 2016, 15, 5272–5286. [CrossRef]
23. Cho, S.; Lee, K.; Kang, B.J.; Joe, I. A hybrid MAC protocol for optimal channel allocation in large-scale wireless powered
communication networks. EURASIP J. Wirel. Commun. Netw. 2018, 2018, 1–13. [CrossRef]
24. Iannello, F.; Simeone, O.; Spagnolini, U. Medium access control protocols for wireless sensor networks with energy harvesting.
IEEE Trans. Commun. 2012, 60, 1381–1389. [CrossRef]
25. Xia, M.; Aissa, S. On the efficiency of far-field wireless power transfer. IEEE Trans. Signal Process. 2015, 63, 2835–2847. [CrossRef]
26. Nobar, S.K.; Niya, J.M.; Tazehkand, B.M. Performance analysis of cognitive wireless powered communication networks under
unsaturated traffic condition. IEEE Trans. Green Commun. Netw. 2020, 4, 819–831. [CrossRef]
27. Kwon, J.-H.; Kim, E.-J. Asymmetric directional multicast for capillary Machine-to-Machine using mmWave communications.
Sensors 2016, 16, 515. [CrossRef]
28. Cai, L.X.; Cai, L.; Shen, X.; Mark, J.W. REX: A randomized exclusive region based scheduling scheme for mmWave WPANs with
directional antenna. IEEE Trans. Wirel. Commun. 2010, 9, 113–121. [CrossRef]
29. Bianchi, G. Performance analysis of the IEEE 802.11 distributed coordination function. IEEE J. Sel. Areas Commun. 2000, 18,
535–547. [CrossRef]
30. Babu, A.V.; Jacob, L. Fairness analysis of IEEE 802.11 multirate wireless LANs. IEEE Trans. Veh. Technol. 2007, 56, 3073–3088.
[CrossRef]