![The transmission power versus total circuit power consumption of different RF transceivers [5], [30], [33], [34] when the supply voltage is 3.0 V.](https://www.researchgate.net/profile/Yu-Luo-53/publication/341923896/figure/fig4/AS:965728442277889@1607259011394/The-transmission-power-versus-total-circuit-power-consumption-of-different-RF_Q320.jpg)
Content uploaded by Yu Luo
Author content
All content in this area was uploaded by Yu Luo on Dec 06, 2020
Content may be subject to copyright.
Practical Issues of RF Energy Harvest and Data Transmission in
Renewable Radio Energy Powered IoT
Yu Luo∗, Lina Pu†
∗ECE Department, Mississippi State University, Mississippi State, MS, 39762.
†School of CSCE, University of Southern Mississippi, Hattiesburg, MS 39406.
Abstract— The sustainable Internet of Things (IoT) is becom-
ing a promising solution for green living and smart industries.
In this article, we investigate the practical issues in radio energy
harvesting and data communication systems through extensive
field experiments. A number of important features of energy
harvesting circuits and communication modules, including the
nonlinear energy consumption of the communication system
relative to the transmission power, the wake-up time associated
with the payload, and the system power reduction during contin-
uous data transmission, have been studied. In order to improve
the efficiency of energy harvesting and energy utilization, we
propose a new model to accurately describe the energy harvesting
process and the power consumption for sustainable IoT devices.
Experiments were conducted using commercial RF transceivers
and RF energy harvesters to verify the accuracy of the proposed
model. The experiment results show that the new model matches
the performance of sustainable IoT devices well in real scenarios.
Index Terms—Energy harvesting, renewable RF energy, power
consumption, sustainable Internet of Things
I. INTRODUCTION
Internet of Things (IoT) coupled with wearable technologies
and cyber physical systems are expected to transform our
world in multiple sectors including healthcare, smart city,
agricultural monitoring, public safety, and surveillance. These
applications involve a larger number of sensors and lower-
power actuators. How to power a large number of devices is
a big challenge.
Due to the ever-increasing demand for green and sus-
tainability, the renewable energy harvesting technology has
attracted extensive attention due to its own potential to self-
power a large number of low-power IoT devices. Recently,
many semiconductor modules have been developed by both
academia and industry to efficiently capture renewable radio
frequency (RF) energy radiated from TV towers, WiFi base
stations, and wireless routers [1]–[3]. The received energy can
be stored in batteries, like supercapacitor or small rechargeable
batteries, to power the target systems for control, sensing, and
wireless communications.
One key to IoT technology is wireless communication
capability, which greatly expands the application scope of IoT.
However, the communication module may be the most power-
hungry part of a low-power IoT device. In order to meet
the communication quality requirements under strict energy
constraints, system designers need to know the maximum
achievable throughput of each renewable RF energy powered
device because it determines the optimal performance of IoT
applications. However, to estimate the maximum throughput,
we need to know the amount of energy that can be collected
and the part of the energy to be consumed directly by the RF
transceiver. Due to the high dynamic RF environment and the
complex hardware design of the IoT device, these two are hard
to obtain in the real world .
Modeling the power consumption of communication mod-
ules and the RF energy harvesting process are two important
but challenging tasks, which have not been fully studied in the
literature. First, the system power consumption increases non-
linearly with the transmission power of the transceiver; how to
properly describe their relationship should be given carefully
considerations in the power consumption model. Second, as
will be introduced in this article, the energy overhead involved
in each data transmission is not constant but varies with the
energy level of the battery and the payload of the data packet.
Accurate calculation of the energy overhead with different
protocol configurations and hardware settings is a challenge.
Third, due to the environment-dependent energy conversion
efficiency of the harvesting circuit and the nonlinear battery
charging characteristic, the energy harvesting rate of an RF
energy harvesting device becomes a complex function of the
RF power density and the energy harvesting time. Therefore,
it is challenging to accurately estimate the energy harvesting
rate in a dynamic RF environment.
In order to solve the above problems, we conducted ex-
tensive experiments to study the practical inherited from the
hardware and protocols of renewable RF energy powered
IoT. In the experiment, a Powercast P2110 harvester [4] was
used to scavenge renewable RF energy from surrounding
environments. The collected energy was then stored in a super-
capacitor to power the Microchip ATmega256RFR2 chipset [5]
that served as the core of a low-power IoT device for system
control, environment sensing, and wireless communication.
We ran IEEE standard 802.15.4 with different transmission
powers, data rates, and effective payloads to evaluate the
accuracy of the proposed model under different settings.
Several new important conclusions can be drawn from the
experimental results. First, despite the significant and random
fluctuations in the RF energy intensity, we observed a smooth
battery charging process. This observation provides us an op-
portunity to accurately estimate the amount of energy that can
be received in the future, which is very useful for designing
online power management and data transmission scheduling.
Second, we used IEEE 802.15.4 as an example to measure
the energy overhead involved in data transmissions, including
the protocol overhead and system overhead transiting between
Sleep and Busy_Tx. Third, we modeled the transmission
power as a modified sigmoid function of system power con-
sumption, which is generally applicable to many commercial
1
RF transceivers. This allows us to accurately calculate the
system power consumption at a given transmission power.
Fourth, we verified that the wake-up time of IoT devices is
not a fixed value, but increases linearly with the data payload.
Reducing the packet size can help wake the device from the
sleep mode faster. Finally, we integrated the energy overhead,
the nonlinearity of system power consumption, and the reduc-
ing supply voltage into the energy consumption model of data
transmission. This model allows us to calculate the total energy
consumed by IoT devices on transmitting single or successive
packets, taking into account all the aforementioned practical
issues.
The new model proposed in this paper takes into account
the characteristics of the renewable RF energy, harvester and
transceiver circuits, and communication protocol to accu-
rately describe the energy harvesting process and the power
consumption of RF energy harvesting powered IoT devices.
The results obtained in this article can be used to design
more realistic and efficient power management strategies. The
proposed model can help develop hardware and communica-
tion protocols for sustainable IoT devices, thereby improving
system performance in terms of energy harvesting efficiency,
throughput, and energy utilization efficiency.
The rest of the article is organized as follows. Section II
presents the related work. In Section III, we briefly introduce
the low power density and spectral/spatial heterogeneity of
renewable RF energy. These features prompt us to model the
RF energy harvesting process in Section IV and investigate
the practical issues of data transmission in Sections V and VI.
Specifically, we analyze the energy overhead on single packet
transmission and successive packet transmissions in SectionV.
In SectionVI, we discuss the nonlinearity of the system power,
introduce a unique phenomenon, called the reducing supply
voltage in a renewable energy powered IoT, and propose a
new energy consumption model that incorporates the new
discoveries. Conclusions are presented in Section VII.
II. RE LATE D WOR K
In recent years, many models have been developed to
describe the energy harvesting process and power consumption
of renewable RF energy powered IoT devices [6]. In these
models, it is usually assumed that the received energy are
discrete energy packets with random sizes and arrival time [7],
[8]. With this assumption, an efficient data transmission strat-
egy is converted to a segmentation optimization problem to
find the appropriate transmission power in an irregular energy
tunnel. As introduced in [9], after considering the battery
capacity constraint and the energy harvesting causality, the
profile of the cumulative energy consumption of sustainable
IoT devices should be the tightest string in the energy tunnel
to maximize throughput.
In order to make the data transmission strategy realistic,
more and more hardware characteristics of the RF energy
harvester are considered in recent energy harvesting and power
consumption models. According to the results published in
[10] and [11], the amount of harvested energy depends not
only on the strength of RF power density but also on the
amount of remaining energy level of the battery. In [12], the
authors considered the scenario when an energy harvester only
has an incomplete knowledge about the battery energy level. In
this case, the power management can be modeled as a partially
observable Markov decision process. The energy harvester will
not measure the energy level of the battery directly, but rather
maintain a probability distribution of the remaining energy in
a set of possible states based on the historical observations
of the harvested energy. In [13], the overhead energy and the
energy loss of the battery over time are taken into account
to improve the accuracy of the energy consumption model.
However, the energy consumption of the circuit is simply
treated as a constant rather than an environment-dependent
variable, which in some cases may not match the real scene.
In [14], the authors investigated the impacts of the diode
nonlinearity and parasitic effects in the harvester circuit on
the energy harvesting efficiency. The result reveals that the
intensity of the incident radio waves can significantly affect
the power conversion efficiency of harvesters. In addition,
energy leakage caused by the off current in the circuit and the
self-discharge characteristic of the battery also affect system
performance. The results obtained in [15] show that the impact
of energy leakage on a transmission strategy is equivalent
to adding a constant operation power on the circuit power.
To correctly estimate energy consumption of an IoT device,
the overhead energy consumed by the microprocessor and
associated modules should be considered. As analyzed in [16]
and [17], with the energy overhead taken into account, the
continuous data transmission becomes inefficient, and on-off
transmission strategy is advocated for throughput optimization.
From the literature review, we realized that most of the
existing works are based on assumptions rather than real
experiment data in modeling the energy harvesting and energy
consumption processes of RF energy powered IoT devices.
However, according to the experimental results, we realized
that some assumptions may be inaccurate. Therefore, in this
article, we explore the practical issues in RF energy harvest
and data transmission and propose a new model for renewable
RF energy powered IoT.
III. OVERVIEW OF RENEWABLE RF ENERGY
In this section, we summarize the important features of
renewable RF energy in the outdoor environment that help
model the renewable RF energy harvesting process in practice.
In Fig. 1, we show the spectrograms of renewable RF
energy measured at different locations using the Keysight
N9340B spectrum analyzer and the Keysight N6850A broad-
band omnidirectional antenna. Each measurement took around
15 minutes. The X-axis represents the frequency and the Y-
axis is time, with the trace from the latest sweep displayed at
the bottom of the figure. The strength of RF signal in each
1MHz frequency band is represented by color. As illustrated
in the figure, the power density of radio waves is typically low
(less than −35 dBm/MHz) at most frequencies. Therefore, not
all frequencies of renewable RF energy can be used to power
IoT devices. In certain frequency bands, when the intensity
of the incident power is above −25 dBm (the yellow and
2
-5
-15
-25
-35
-45
-55
-65
Sharpstown, Houston
Mississippi State University
Quail Run, Houston
Westover, Hattiesburg
190 540 740 890 1950 2145 (MHz)
190 540 740 890 1950 2145 (MHz)
101
190 540 740 890 1950 2145 (MHz)95
740 890 1950 2115 (MHz)95 2355 2550
2355 2550
(dBm/MHz)
Time (min)
015
Time (min)
015
Time (min)
015
Time (min)
015
Frequency (MHz)
Fig. 1. Spectrograms of ambient RF energy in outdoor environments.
orange lines in the figure), the strength of RF energy is higher
than the sensitivity of activating the RF harvester [18]. The
spectrograms of RF energy at different locations shown in
Fig. 1 verifies the feasibility of using RF energy harvesting to
power low-power IoT devices.
By comparing the first two spectrograms in Fig. 1, it can
be observed that due to the high propagation attenuation of
radio waves and the non-uniform deployment of RF sources
(e.g., TV tower, radio base station, and WiFi access point), the
power density of renewable RF energy at different locations
in a city is dramestically different. For instance, the average
power density at FM frequency band (around 101 MHz) can
achieve −16.7dBm in Quail Run, but it is only −56.3dBm
in Sharpstown. Furthermore, the RF intensity in the same
location may have significant heterogeneity at different fre-
quencies. For instance, in the campus of Mississippi State
University, the strength of RF signal at 890 MHz cellular band
is above −10 dB, which is 20 dB higher than that measured at
1950 MHz cellular band.
Even if RF energy harvesters have the same distance to an
energy source, their energy harvesting rates can be different.
This is because the propagation attenuation of the radio wave
is not uniform, but is affected by weather, obstacles, and
terrain. In Fig. 2, we use the tool provided in [19] to plot
the spatial distribution of TV signals in two different cities.
As illustrated in the figure, the signal attenuation in the plain
area like Houston is relatively slow and smooth; therefore, an
RF harvester 5 miles from the TV tower can still get −15 dBm
of incident power. In contrast, the spatial distribution of radio
energy fluctuates drastically in the hilly area like Boston. For
an energy harvester 5 miles from the TV tower, the intensity
of the incident power can be as low as −40 dBm.
Based on the above analysis, it can be appreciated that
although we are surrounded by renewable RF energy, their
power density is generally weak and their distribution in
frequency and space may be uneven. The low power density
-15 -30 -45 -75
(dBm)
Houston
Sugar Land
Pasadena
Pearland
Cinco Ranch
Sandy Point
Aldine
Jersey Village
Cypress
Bellaire
Friendswood
Sheldon
League City
Katy
Lochridge Rosharon
Channelview
Crosby
Baytown
La Porte
Manvel
Greatwood
Fairchilds
Alvin
Highlands
Seabrook
Kemah
Santa Fe
Fresno
Mission Bend
Richmond
South Houston
Deer Park
NORTHWEST
HOUSTON
WILLOWBROOK
Pleak
Sienna
Plantation
Needville
Guy
SOURTHWEST
HOUSTON
Meadow
Place
Algoa
Hillcrest
ENERGY
CORRIDOR
5 mi20 mi
Houston
Boston
Newton Brookline
Medford
Needham
Dedham
Framingha
Waltham
Quincy
Lincoln
Lexington
Revere
Cambridge
Somerville
Dover
Norwood
Holliston
Wellesley
Sherborn
Milton
Wayland
Sudbury
Medfield
Westwood Braintree
Canton
Millis
SOUTH BOSTON
WEST ROXBURYMATTAPAN
AUBURNDALE
DORCHESTER
Natick
EAST WALPOLE
Watertown
BACK BAY
Weston
COCHITUATE
Winthrop
Chelsea
Arlington
READVILLE
2 mi5 mi
Boston
Fig. 2. The effect of terrain on RF power density distribution drawn with
TV Fool [19]. Note that the map on the right (Boston) is four times the
map on the left (Houston). The effective radiated power (EPR) and the height
above average terrain (HAAT) of the TV station in Houston are 1kW and
580 m [20], respectively; the television station in Boston has two similar
parameters, respectively 1.35 kW and 390 m [21].
and spectral/spatial heterogeneity of RF energy motivated us
to investigate the practical issues of RF energy harvest and
data transmissions through field experiments. We deployed RF
energy powered wireless devices in the campus of Mississippi
State University and studied the model of radio energy harvest-
ing process and modeled the energy consumptions in outdoor
environments.
IV. RAD IO ENERGY HARVESTING PROC ES S
In this section, we briefly introduce the factors that have
direct impacts on the efficiency of RF energy harvesting and
the charging characteristics of real radio energy harvesters.
A. Factors that Affect Harvesting Efficiency
An RF energy harvester usually consists of three key com-
ponents: an impedance matching circuit, a rectifier, and a bat-
tery [22]. The matching circuit adjusts the output impedance
of the receiving circuit to match the input impedance of the
antenna, aiming to maximize the energy conversion efficiency.
The rectifier converts alternating current (AC) to direct current
(DC) and then boosts the output voltage appropriately to
charge the battery. The collected energy is then stored in the
battery for future usage.
The batteries of radio energy powered IoT devices charge
and discharge more frequently than the batteries in conven-
tional wireless devices. In addition, many IoT devices need to
work reliably in harsh outdoor environments with high or low
temperatures. Therefore, the supercapacitor that has a long
lifespan (100,000 to a million charging cycles) and a wide
range of operating temperatures (−40◦F to 248◦F) is a more
popular choice than a rechargeable battery in sustainable IoT
applications [23]. To this end, we mainly uses supercapacitor
as an example to analyze the energy harvesting process of
radio energy powered IoT devices.
The following three factors have significant impact on the
efficiency of radio energy harvesting:
a) Frequency of radio waves: Radio energy harvester built
with Schottky diodes can achieve more than 50% efficiency
on energy conversion when the frequency of the input
3
energy matches the circuit [24], [25]. However, if the
frequency of the radio energy does not match the resonant
frequency of the antenna, the efficiency will reduce as more
incident radio waves will be reflected back to the air.
b) Intensity of incident power: The harvester circuit is usually
designed to work efficiently within a specific intensity
range of the RF energy [26]. If the incident energy is too
weak and the voltage swing at diodes of the harvester
circuit is below the forward voltage, the diode will be
off and only a small current can flow through the diode,
resulting in low efficiency on energy harvest. If the incident
energy is too strong, the diode is reverse-conducting and
the current under reverse bias becomes larger since the
voltage swing of the diode exceeds the breakdown voltage,
which also leads to low harvesting efficiency.
c) Battery’s energy level: Owing to the battery charing non-
linearity, the amount of energy that can be harvested not
only depends on the intensity of RF energy and harvester
circuit, but is also affected by the energy level of battery
[10], [11]. As analyzed in [10], the charging speed is fast
when the energy level of the battery is about 20%, and
degrades as the energy level increase. Therefore, the current
energy level of the battery needs to be taken into account
to accurately estimate the energy harvesting rate.
It is worth noting that the open-circuit voltage of the RF
energy harvester is not a constant, but varies greatly with the
strength of the incident radio power. This output characteristic
is very different from that of a solar harvester, which maintains
fairly stable open-circuit voltage around 0.89 V when the
illuminance of natural light ranges from 10k to 100k lux
in outdoor use [27]. Denote the open-circuit voltage and the
intensity of the incident power by VOC and Pi, respectively.
In Table I, we use the Powercast P2110 energy harvester as an
example to show how VOC rises as Piincreases. The varying
VOC with respect to Piis a common phenomenon of most
existing RF energy harvesters and similar results have been
reported in [26], [28], [29].
TABLE I
VOC OF P2110 E NE RGY H ARVE ST ER WI TH R ESP EC T TO Pi.
Pi(dBm) -14 -11.3 -8.5 -7 -5 -3 -2
VOC (V) 0.4 0.9 1.6 2 2.6 3.2 4
B. Charging Characteristics of RF Energy Harvester
Due to the low power density of renewable RF energy in
the air, the intensity of the incident RF signal is typically less
than 0dBm, which can provide up to hundreds of microwatts
to a low-power IoT device. However, commercial transceivers1
consume about a few milliwatts of power during data trans-
mission and reception [5], [30]. For this reason, studying the
charging characteristics of RF energy harvester becomes an
important issue for more efficient energy harvest and usage.
1We consider active transceivers in this paper. The passive systems such
as backscattering transmitters and near-field inductively coupled transmitters
that are commonly used in RFID and NFC devices are not included.
Following this concern, we use Powercast P2110 harvester as
an example to investigate the charging characteristics of P2110
through a large number of outdoor experiments. A typical
experiment scenario can be found in Fig. 3.
Fig. 3. Outdoor experiment conducted at Mississippi State University. The
distance between energy source and harvester is about 280 feet.
We measured the strength of ambient RF energy changes
in outdoor environments and illustrated the results in Fig. 4,
where the time interval between adjacent measurements is
5.75 s. The envelop of radio waves is presented to visualize
the upper and lower bounds in the energy strength. As can
be observed from the figure, the instantaneous RF intensity
has random and high fluctuations (over 5to 10 dBm) in each
frequency band. By contrast, the smoothed power intensity and
the envelop of the renewable radio energy change slowly with
time. Similar results are also reported in [1] and [31], where
RF power density was measured in Boston, MA, United States
and London, United Kingdom, respectively. This indicates that
random fluctuations are an inherent feature of renewable radio
energy in outdoor environments.
Intuitively, we could expect randomly varying energy har-
vesting rate since the open-circuit voltage depends on the
energy intensity as measured in Table I. This is the reason
why the arrival of renewable radio energy is modeled as
discrete energy packets of random sizes and arrival time in the
literature. However, we show that our next experiment results
challenge this assumption.
We deployed a Powercast P2110 harvester with a 1dBi
omnidirectional antenna at the campus of Mississippi State
University receiving renewable RF energy radiated from a
880 MHz cellular base station. The density of the RF energy at
the harvester can be found in Fig.1 and Fig.4. Fig.5 shows the
battery charging curves of multiple measurements. Despite the
random fluctuation of RF density, the supercapicitor’s voltage
rises smoothly with time. The charging curve is similar to the
charging process of a capacitor with a stable charging voltage.
In addition, different charging curves have good consistency
4
0 5 10 15
-25
-20
-15
880MHz, Mississippi State University
Amplitude Smooth Envelop
0 5 10 15
-35
-30
-25
Amplitude (dBm)
540MHz, Quail Run, Houston
Amplitude Smooth Envelop
0 5 10 15
Time (minutes)
-20
-15
101MHz, Quail Run, Houston
Amplitude Smooth Envelop
Fig. 4. Fluctuations of RF energy over time.
indicating a stable charging process at different test time.
These observations are counter-intuitive because they show
that even if the intensity of the incident energy fluctuates
significantly in outdoor environments, the battery voltage of
the RF energy harvester will rise steadily over time.
The smooth and consistent charging curve provides an
opportunity for radio energy powered IoT devices to accurately
predict their future energy arrival from a small number of
historical measurements. By simplifying the RF harvester to
a resistor – capacitor (RC) circuit, we can use the capacitor
charging equation to describe the energy harvesting process:
Vcc(t) = VOC 1−e
−t
RC ,(1)
where Vcc(t)is the voltage level of the battery at time t,VOC
is the open-circuit voltage of the energy harvester, Ris the
equivalent impedance of the harvesting circuit, and Cis the
capacitance of the supercapacitor.
Given the charging equation provided in (1), the unknown
parameters Rand VOC can be easily calculated using the
s fitting2. Afterward, the amount of energy stored in the
supercapacitor at time t, which is denoted by Es(t), can be
obtained via Es(t)= 1
2CV 2
cc.
We compare the results of least-squares fitting with the
experiment data in Fig. 5. It can be observed that even if
the intensity of the incident radio wave fluctuated between
-25 dBm and -13 dBm (i.e., top graph of Fig. 4), the voltage
increased smoothly during charging and the three charging
curves of Experiments 1 to 3 match well with the correspond-
ing fitted curve. To evaluate the performance of the curve
fitting, we define the fitting error as the difference between the
estimated battery voltage and the average of the experiment
2VOC depends on the intensity of the incident radio waves and the
efficiency of the harvester circuit. In some energy harvesters, if the battery can
be temporarily disconnected from the main circuit by an electronic switch,
the open-circuit voltage can be directly measured. In this case, we only needs
to calculate R, which significantly reduces the computational complexity of
the least-squares fitting.
0 0.2 0.4 0.6
Time (min)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Supercapacitor voltage (V)
2.2 mF
Experiment 1
Experiment 2
Experiment 3
Curve fitting
0 5 10 15 20
Time (min)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
50 mF
Experiment 1
Experiment 2
Experiment 3
Curve fitting
Fig. 5. Charging curve of a renewable energy harvester in the real world.
The capacitance of supercapacitors: 2.2mF (left), and 50 mF (right).
measurements. The results show that the mean values of the
fitting errors are very small, which are only 5.1×10−3V and
8.4×10−3V when C= 2.2mF and C= 50 mF, respectively.
This experiment result verifies the feasibility of using the least-
squares fitting to estimate the battery voltage that the harvester
can be charged.
It is worth noting that the open-circuit voltage and the
equivalent impedance of the harvesting circuit are affected by
the intensity of the incident energy. In fact, it is difficult to
accurately model the charging process based on the hardware
design of the energy harvester and the dynamic RF environ-
ment. However, with the least-squares fitting, we can treat the
harvester circuit and environment as a black box; the battery
voltage samples are the only information needed to calculate
the charging curve.
V. OVERHEAD ENERGY IN DATA TRANSMISSION
In this section, we will focus primarily on the overhead
energy involved in data transmissions. We use the Microchip
ATmega256RFR2 evaluation board as an example to analyze
the system overhead in single packet transmission and during
successive transmissions. The protocol overhead of IEEE
802.15.4 is also presented.
The ATmega256RFR2 RF transceiver has extremely low
power consumption (i.e., as little as 1.08 µW = 0.6µA×1.8V)3
in the deep sleep mode which makes it a promising hardware
for renewable radio energy powered IoT devices. Therefore,
we use ATmega256RFR2 microcontroller-based IoT device
running the IEEE 802.15.4 wireless communication protocol
as an example to investigate the overhead energy in data
transmissions. Although we use ATmega256RFR2 as an ex-
ample, the conclusions can be easily extended to low-power
IoT devices that are implemented with other hardware and
communication protocols.
A. System Overhead in Single-Packet Transmission
The power consumption of the ATmega256RFR2 microcon-
troller in active mode is at least 7.2mW (4mA×1.8V), which
in most cases is much higher than the energy harvesting rate, as
3The CC2500 low-power 2.4 GHz RF transceiver designed by the TI
company has 1.5µA current when only voltage regulator to digital part and
crystal oscillator running [30].
5
0 1 2 3 4 5 6 7
Time (ms)
0
2
4
6
8
10
12
14
16
18
Current (mA)
01234567
Time (ms)
0
2
4
6
8
10
12
14
16
18
Current (mA)
0 1 2 3 4 5 6 7
Time (ms)
0
2
4
6
8
10
12
14
16
18
Current (mA)
Data transmission
(MSDU = 1 Octet)
Sleep to
Busy_Tx
Busy_Tx to
Sleep
Data transmission
(MSDU = 40 Octets)
Sleep to
Busy_Tx
Busy_Tx to
Sleep
Data transmission
(MSDU = 105 Octets)
Sleep to
Busy_Tx
Busy_Tx to
Sleep
(a) (b) (c)
payload-
dependent
payload-
dependent
payload-
dependent
Fig. 6. Current consumptions of ATmega256RFR2 under different configurations. (a) Ptx = 3.5dBm, Le= 1 octet, and Vcc = 2 V; (b) Ptx = 1.8dBm,
Le= 40 octets, and Vcc = 2.5V; (c) Ptx =−16.5dBm, Le= 105 octets, and Vcc = 3.5V.
described in Section IV. As a result, the radio energy powered
IoT device may not stay active continuously but remain sleep
and occasionally wake up to respond to the sensing and
transmission requests. The high power consumption and low
energy harvesting rate drive us to investigate the detailed
system overhead for the hardware transiting between sleep
mode and active mode.
1) Sleep to Busy_Tx overhead: The wakeup energy
overhead is significant and inevitable. Each time when the IoT
device transits from deep sleep to busy transmission mode, a
non-negligible amount of energy is consumed on the voltage
regulator ramp up, the phase-locked loop (PLL) calibration,
the packet encoding, and starting the radio transceiver. The
current wave of the ATmega256RFR2 board from Sleep to
Busy_Tx is demonstrated in Fig. 6.
The Sleep to Busy_Tx time represented by Ts)tis de-
fined as the time it takes for the IoT device to switch from deep
sleep to data transmission mode. It consists of a significant
constant component for the hardware to ramp up voltage and
to start PLL and transceiver, and a payload dependent interval
to prepare the packet for transmission. Through least-square
fitting, Ts)tcan be estimated as follows4:
Ts)t= 0.004 Le+ 1.395,(2)
where Leis the MSDU payload in terms of octets. The
linear relationship between Sleep to Busy_Tx time MSDU
payload is verified in Fig. 7. Increasing the MSDU payload
from 1octets to 106 octets results in an increased Ts)tfrom
1.4ms to 1.82 ms (i.e., 23% increase).
The average current consumption in the Ts)tinterval is
7.8mA, which does not change with Ts)tor power supply
voltage, Vcc. As a result, the wakeup energy overhead, repre-
sented by Es)t, becomes a linear function of Vcc. According
to (2), we have that
Es)t=CcVccTs)t= 7.8Vcc (0.004Le+ 1.395),(3)
where the units of Es)tand Vcc are µJ and V, respectively.
4The coefficients like 0.004,1.395,7.8, and 192, etc. in (2), (3), (4), (5)
and (6) are calculated through least-square fitting and are only valid to the
specific scenario with settings as described in Section V-A and Section V-B.
Those values need to be modified accordingly when different protocols or
different hardware are used.
0 20 40 60 80 100
MSDU payload (octet)
1.4
1.5
1.6
1.7
1.8
Sleep to Busy_tx time (ms)
Experiment results
Linear fit
Fig. 7. The Sleep to Busy_Tx time as a function of the MSDU payload.
2) Busy_Tx to Sleep overhead: Let Tt)sbe the time it
takes for the device to return from the Busy_Tx mode to the
Sleep mode. As shown in Fig. 6, this transition time, Tt)s,
is nearly constant (i.e., 0.45 ms for ATmega256RFR2) and
does not change with the supply voltage and MSDU payload.
In addition, the average current consumption of the device
during Tt)sis approximately half of the current during data
transmission. If we use Et)sto represent the system energy
consumption during Tt)s, then it can be obtained that5
Et)s=VccTt)sCc
2= 0.225 Vcc Cc,(4)
where the units of Et)s,Vcc, and Ccare µJ, V, and mA,
respectively.
B. System Overhead During Successive Transmissions
If an IoT device is scheduled to send multiple packets in one
active cycle, it will not enter the deep sleep mode until the end
of last packet transmission. To investigate the overhead energy
between adjacent transmissions, we use ATmega256RFR2 as
an example and show the supply current waveform during
consecutive transmissions in Fig. 8.
As shown in the figure, the transceiver will be turned off
at the end of each transmission to save energy. When the
transceiver is off, the microcontroller will be the major source
5Note that when the low-power IoT device is powered with RF energy
harvester, the supply voltage, Vcc, may vary significantly while energy is
consumed on data transmissions. The energy consumption with dynamic Vcc
is calculated in Section VI-C.
6
of energy consumption resulting in 4 mA supply current. A
fixed overhead energy is generated for the device to transit
from Tx_Off to Busy_Tx (i.e., 0.86 ms of time interval and
average 10.25 mA of supply current). This overhead energy is
independent of the transmission power (Ptx), MSDU payload
(Le), and supply voltage (Vcc). Therefore, the total energy
consumed by the microcontroller to switch from Tx_Off to
Busy_Tx mode is 0.86 ms×10.25 mA×Vcc, i.e., 8.8×Vcc
microjoules.
-4 -2 0 2 4 6
Time (ms)
5
10
15
20
Power supply current (mA)
105 octets 50 octets 1 octet
Busy_Tx to Tx_Off
Tx_Off to Busy_Tx
Fig. 8. The current consumption of ATmega256RFR2 during consecutive
packets transmissions. Red curve: Le= 1 octet, Vcc ≡2V, Ptx =
−16.5dBm; black curve: Le= 50 octets, Vcc ≡3.5V, Ptx = 1.2dBm;
Le= 105 octets, Vcc ≡2.5=2.5V, Ptx = 3.5dBm.
Denote the overhead energy between the transmitting packet
jand j+ 1 by Ej
ov. Based on the above analysis, Ej
ov can be
estimated as follows:
Ej
ov =CcVccTt)t= 8.8Vcc .(5)
The units of Ej
t)t,Vend, j
cc , and Cj
cin (5) are µJ, V, and mA,
respectively.
The system overhead varies during consecutive packet trans-
missions, the calculation of which is described as follows:
a) The system overhead sending the first packet is involved
when system transits from Sleep to Busy_Tx, i.e., Es)t.
It can be calculated with (3).
b) The system overhead sending packet 2to N−1is the
overhead between transmitting packet j−1and packet j,
i.e., Ej
ov.Ej
ov is available in (5).
c) The system overhead to transmit the last packet is for the
system to transit from Busy_Tx to Sleep, i.e., Et)s,
which can be calculated with (4).
C. Protocol overhead
In addition to the wakeup energy overhead, the communi-
cation protocols also generate a certain amount of overhead
for packet control, synchronization, frame checking, and en-
cryption.
Taking IEEE 802.15.4 standard [32] that is for low-rate
wireless communication as an example, the 6octets PHY
preamble includes 5octets of synchronization header and
1octet of physical header. This PHY preamble must be
transmitted at a rate of 250 kbps. The PHY payload, which
consists of MAC header (MHR), MAC service data unit
(MSDU), and frame check sequence (FCS), can be sent at
a customized data rate. In this paper, we consider the MAC
layer payload, i.e., the MSDU, as the effective data; all other
functional frames, including the PHY preamble, MHR, and
FCS are overhead as shown in Fig. 9.
PHY preamble
(6 octets)
PHY payload
(Up to 127 octets)
250 kbps rd
MHR
( )
LMHR
MSDU
( )
Le
FCS
(2 octets)
FCF
(2 octets)
SN
(1 octets)
AF
(0~20 octets)
ASH
(0~14 octets)
Fig. 9. The data structure of IEEE 802.15.4.
Denote transmission rate by rtx. In Fig. 6, we have demon-
strated the total transmission time when the length of MSDU
varies among 1Octet, 40 Octets, and 105 Octets respectively
with rtx = 250 kbps. As illustrated in Fig. 6(a), even if the
MSDU payload is only 1octet, the transceiver will take
some time to send the entire packet due to the overhead
frames added by the communication protocol. Let LPPR,
LMH R , and LF CS represent the length of the physical layer
preamble, MAC layer header and FCS, respectively. Denote
the lengths of MAC addressing fields (AF) and auxiliary
security header (ASH) by LAF and LASH , respectively. Then
the total transmission time of the data packet represented by
Ttx can be calculated as follows:
Ttx =8∗LPPR
250 kbps +8∗(LMH R +Le+LF C S )
rtx
= 192 µs +40 bit+8∗(LAF +LAS H +Le)
rtx
.
(6)
We set AF and ASH 6octets and 10 octets, respectively. In this
case, the MSDU payload will be between 0and 106 octets, i.e.,
Le∈[0,106]. The effective data transmission time in this case
changes from 3.6% to 80% of the total transmission time. The
system energy consumption during Ttx relies on the system
power, which further depends on the direct transmission power
as to be analyzed in Section VI-A.
If the data to be sent is greater than the maximum MSDU
payload, the data will be divided into several consecutive
packages. The transceiver will be turned off between the
consecutive transmissions, but the CPU will be kept active to
avoid repeated wake-up overhead. Nevertheless, when switch-
ing between the transceiver off and data transmission modes,
some overhead energy will be generated, which will be studied
in Section VI.
VI. ENERGY CONSUMPTION MODELING
In this section, we utilize the results obtained in previous
sections and propose a comprehensive energy consumption
model for renewable radio energy powered IoT devices. We
first analyze the relationship between the direct transmit power
and system current. After that, a unique feature of the renew-
able radio energy powered IoT, namely the reducing supply
7
voltage during data transmission, is introduced. At last, we
propose a comprehensive energy consumption model.
A. Direct Transmit Power and System Current
Considering the significant power consumption of IoT de-
vices in the active transmission mode and the limited harvest
rate of RF energy harvester, it is essential to understand the
relationship between the direct transmit power and the total
system power consumption for optimal energy utilization. In
the literature, the total system power is usually modeled as the
sum of direct transmit power and circuit power. The circuit
power, which accounts for the power consumption at the
AC/DC converter, the analog radio frequency (RF) amplifier
and etc., is assumed constant [16], [17]. This linear model of
the total system power neglects the dependence of the circuit
power on direct transmit power in the real system. Next, we
present a general power consumption model to describe the
nonlinear relationship between direct transmit power and total
system power for different IoT devices.
0 2 4 6 8 10
Direct transmit power (mW)
30
40
50
60
70
80
90
100
Total system power (mW)
ATmega256RFR2: Real
ATmega256RFR2: Fitted
AT86RF212: Real
AT86RF212: Fitted
CC2500: Real
CC2500: Fitted
CC2430: Real
CC2430: Fitted
Fig. 10. The transmission power versus total circuit power consumption of
different RF transceivers [5], [30], [33], [34] when the supply voltage is 3.0V.
In the experiment, we fix the supply voltage as 3.0V and
measure the total system power, Ps, and the direct transmit
power, Ptx, of ATmega256RFR2 low-power RF transceiver.
The measurement results of ATmega256RFR2 are marked as
Oin Fig. 10. As shown in the figure, Psdoes not increase
linearly with Ptx. Specifically, the system power increases fast
when the direct transmit power is extremely low or very high,
but grows slowly when Ptx is moderate. Since we can consider
Ps−Ptx as the overhead of circuit power, the results suggest
that when Ptx is moderate, a higher percentage of energy
will be contributed to the direct data transmission. Based on
the observation, we construct a modified sigmoid function to
model the system power as follows:
f:Ps=α1+α2ln α3
α4−Ptx
−1,(7)
where α1to α4are scaling and displacement coefficients.
We can observe that the fitted curve and the measurement
results match very well for ATmega256RFR2. In order to
verify the correctness of (7), we presented the relationship of
total system power and direct transmit power of other off-the-
shelf RF transceivers. The measurement results come from the
datasheet provided by the Microchip and TI company [5], [30],
[33], [34]. The modified sigmoid function has good fits to the
measured data for different RF transceivers. This verifies that
the sigmoid function constructed in (7) can accurately describe
the relationship between the system power consumption and
the transmission power of the IoT device. By performing a
least-squares fitting, we can calculate the coefficients α1to α4
and get a general model (7) for system power estimation. Since
the supply voltage is set constant, the supply current, Ccthat is
proportional to Pscan be also presented as a modified sigmoid
function of Ptx similar to (7), i.e., Cc=f(Ptx). Combining (7)
and (6), we can further calculate the total energy consumption
of IoT devices in the active transmission mode.
B. Reducing Supply Voltage and Its Impact
Due to the thin power density of renewable RF energy
and the size as well as weight constraints of low-power IoT
devices, the battery capacity is usually low. As a result, the
battery voltage, which is the also the supply voltage of the
IoT device, may drop considerably in a short time during
data transmissions. For this reason, we investigate how the
varying supply voltage affects the system power consumption
on data transmissions through experiments. Note that, the
energy received by the device during data transmission will
be ignored because the energy harvesting rate in an outdoor
RF environment is at least two orders of magnitude lower than
the system power consumption, as described in Section III.
In the experiments, we use a 1mF capacitor as the battery to
store the energy harvested by the RF harvester, which is then
utilized for data transmission. The payload of the data packet
is 105 Octet and the transmission power is set to be 3.5dBm.
Before the transmission starts, the capacitor is charged to
3.3V. We record the variation of the supply voltage during
the transmission and present the result in Fig. 11. As shown
in the figure, we observe a nearly linear reduction in the supply
voltage. This phenomenon is essentially resulted from the low
battery capacity in renewable radio energy powered IoT and
the high energy consumption on data transmission.
0.228 0.23 0.232 0.234 0.236 0.238
Time (s)
3.2
3.22
3.24
3.26
3.28
3.3
3.32
Supply voltage (V)
Experimental result
Theoretical result
Send first bit
Send last bit
Sleep to
Busy_Tx
overhead
Fig. 11. Supply voltage drops with data transmission.
In Fig. 12, we demonstrate the supply voltage and supply
current of the system during successive data transmissions.
Since a considerable amount of energy is consumed on each
packet transmission, we observe a staircase reduction on the
supply voltage. However, when the voltage drops from 3.3V to
8
0 0.5 1 1.5 2 2.5
Time (s)
2.6
2.8
3
3.2
3.4
Supply voltage (V)
Packet 1
Packet 2
Packet 7
0 0.5 1 1.5 2 2.5
Time (s)
0
5
10
15
20
Supply current (mA)
Fig. 12. Supply voltage and current in successive transmissions.
2.7V, the system current in Busy_Tx mode is nearly constant.
This result suggests that the supply current, Cc, is not affected
by the supply voltage, but only depends on system operation
parameters such as the microprocessor CPU clock rate and
the direct transmit power of RF transceiver. The relationship
between the Ccand Ptx is illustrated in (7).
Therefore, given a fixed transmission power and a data rate,
Ccis considered to be constant and we model the system
power consumption, Ps, as a linear function of Vcc. Reducing
the supply voltage can proportionally decrease the power
consumption of data transmission. In other words, the power
consumption on continuous data transmission is not fixed, but
gradually decreases with time. This phenomenon has received
little attention in existing research on energy modeling and
power management. Next, we will model the reducing supply
voltage as follows.
Assuming the supply voltage before transmitting bit iis Vi
cc,
then the energy consumed on sending the ith data bit will be
ei
s=Vi
cc CcTb,(8)
where Tbis the time transmitting a single bit.
According to the energy storage equation of the capacitor,
the bitwise energy consumption, ei
s, can be also represented
as
ei
s=1
2ChVi−1
cc 2−Vi
cc2i.(9)
Combining (8) and (9), we have that
Vi−1
cc ="Vi
cc2+CcTb
C2#
1
2
+CcTb
C.(10)
Since the product of Cc(mA) and Tb(µs) is in the order of
10−8or lower, which is much smaller than the capacity (µF–
mF) of the supercapacitor used in a sustainable IoT device.
Therefore, we can simplify (10) as
Vi−1
cc ≈Vi
cc +CcTb
C.(11)
The linear function of Vi
cc matches our observation in the
experiments presented in Fig. 11: when the transmission rate
is fixed, the supply voltage drops linearly with time (i.e., the
number of transmitted bits). We use Cc= 16.8mA, Tb=
4µs and C= 1 mF, and draw the theoretical estimation of
Vcc in Fig. 11. We can see that the theoretical result matches
experimental measurements very well.
One consequence of the dropping voltage is that the energy
consumed on sending one bit linearly decreases as the bit
index increases. This phenomenon is a unique feature in
the renewable energy powered IoT system that is resulted
from the much smaller capacity of the supercapacitors than
conventional batteries. When the supply voltage drops below
a threshold, the system will be forced to shut down owing to
the insufficient power supply. The threshold voltage largely
depends on the operational settings of the system, such as
the microprocessor CPU clock rate and the direct transmit
power of RF transceiver. When the supply voltage is above
the threshold voltage, the IoT system will operate normally
but with a decreasing energy consumption rate. Next, we will
use these observations to model the energy consumption on
sending data packets.
C. Energy Consumption on Data Transmission
Based on (8) and (11), we have that the bitwise energy
consumption, e1
s, e2
s, . . . , ei
s, forms an arithmetic sequence,
where
ei
s−ei−1
s=(CcTb)2
C.(12)
Specifically, assuming the MSDU payload of IEEE 802.15.4
protocol is full, the packet length will be 127 octets, i.e., 1016
bits. Consider the scenario when the capacity of the superca-
pacitor is C= 1 mF; transmission power is Ptx = 3.5dBm;
data rate is rtx = 250 kbps; supply current is Cc= 16.8mA;
initial voltage of the supercapacitor is V1
cc = 3.3V. According
to (12), the energy consumption on sending the first bit and the
last bit of the packet are 0.222 µJ, and 0.217 µJ, respectively;
the latter is 2.11% lower than the former. The reducing energy
consumption during data transmission cannot be ignored in
the power management for renewable radio energy powered
IoT systems, especially when successive data transmissions
are scheduled in each active transmission cycle.
The cumulative energy consumed on sending a data packet
of nbits is
Et(n) =
n
X
i=1
ei
s=n e1
s−n(n−1) (CcTb)2
2C.(13)
According to (13), the cumulative energy consumption on
a packet transmission is a parabola that opens downward.
By substituting (7) and (8) into (13), we can calculate the
cumulative energy consumption on data transmission given a
direct transmission power of the RF transceiver.
We show the model of Nconsecutive energy consumption
in Fig. 13. The x-axis, denoted by b(i,j ), is the bit index i
in the PHY payload frame6of the packet j. Let Et(LN
e, N )
represent the accumulative energy consumption on sending
data from b(1,1) to b(LN
e,N). The model presented in Fig. 13
demonstrates Et(LN
e, N )as a function of payload length of
transmitted packets. The Et(LN
e, N )for sending each packet
jis composed of two parts, namely, the energy overhead Ej
oe
6The PHY payload not only includes the MSDU frame of the message but
also consists of MAC header and FCS overhead. The PHY preamble overhead
is calculated in Ej
oe.
9
0
E1
c
E1
oe
E2
oe
E2
c
EN
c
EN
oe
Packet 1 Packet 2 Packet N
…
b
(L1
e,1) b
(L2
e,2) b
(LN
e,2)
Et(1, 1)
Et(L1
e,1)
Et(1, 2)
Et(1, N)
Et(L2
e,2)
Et(LN
e,N)
Conventional model
New model
Fig. 13. Comparison between new and conventional energy consumption
models.
and the energy consumed on sending the effective data Ej
c. The
calculation of Ej
ccan refer to (13). Since the bitwise energy
consumption, ei
s, is not constant but reduces over time, the
gradient of Et(LN
e, N )gradually decreases within each data
transmission.
The energy overhead, Ej
oe, consists of two parts: (a) the sys-
tem overhead and (b) the PHY preamble overhead, as shown
in Fig. 14. Since the PHY preamble overhead, denoted by
E1
p, involves the energy consumed on sending PHY preamble
stream at 250 kbps, we can use (13) to calculate E1
p. Next, we
introduce how to calculate the system overhead energy.
MHR
PHY preamble
(6 octets)
MSDU
FCS
V1, j
cc V2, j
cc
250 kbps rj
d
V3, j
cc
PHY payload
system
overhead
V0, j
cc
Fig. 14. Data structure and voltage notation, where jis the packet index.
After considering the impact of battery voltage reduction,
the wakeup system overhead calculated in (3) can be rewrote
as
Es)t=RTs)t
0CcVcc(t)dt
≈7.8V0,1
cc −3.9Ts)t
C(0.004Le+ 1.395),
(14)
where V0,1
cc is the supply voltage before the system waking
up and the average supply current is assumed to be 7.8mA
as measured in Fig. 6. Similarly, when successive packets are
transmitted, the system overhead on Tx_Off to Busy_Tx
transition can be rewrote as
Ej
t)t=RTt)t
0CcVcc(t)dt ≈8.8V0,j
cc −4.4
C.(15)
Here V0,j
cc is the supply voltage after packet j−1being
transmitted. The average supply current is measured as 10.25
mA and Tt)t= 0.86 ms as shown in Fig. 8.
Note that according to the experiment results presented in
Fig. 6 and Fig. 8, the system overhead is independent of the
direct transmit power (Ptx). Therefore, (14) and (15) generally
applied to Microchip ATmega256RFR2 system regardless of
the transmission rate or direct transmit power of the RF
transceiver. However, the external sensors or peripheral devices
may introduce additional system overhead that needs to be
calculated separately.
Based on the energy storage equation of the capacitor, we
can calculate V2,j
cc , which is also the battery voltage after
transmitting PHY preamble, as follows:
V2,j
cc =sV0,j
cc 2
−2Ej
oe
C.(16)
Then combining (7), (8), (13), and (16), we can calculate the
energy consumption on sending PHY payload.
The energy overhead, Ej
oe, resulted in a step increase in
energy consumption on successive data transmissions. As
illustrated in Fig. 13, the profile of Et(LN, N )becomes a
discontinuous curve.
In Fig. 13, we compare the new energy consumption model
with the conventional one. In the conventional model, the
total system power has typically been modeled as the sum of
direct transmit power and constant circuit power. It assumes
constant Psgiven a fixed rtx [7], [9]; in other words, the
drop in supply voltage is ignored. As a result, the cumulative
energy consumption, Et, is a straight line (i.e., constant slope
as shown in Fig. 13). In reality, with the obvious drop in supply
voltage, Etbecomes a nonlinear function of b(i, j).
To summarize, no matter whether a single packet or suc-
cessive packets are sent, the total energy consumption can be
estimated based on the proposed energy consumption model.
The only system parameters required in the calculation are
the initial supply voltage before the system wakes up (i.e.
V0,1
cc ) and the supply current (i.e., Cc), which depends on the
transmission power, Ptx.Cccan be calculated based on the
modified sigmoid function similar to (7). As shown in Fig. 11,
Vcc decreases linearly with data transmission. Therefore, we
only need to measure the supply voltage once at the beginning
of each active cycle.
Compared to the conventional model, the contributions of
our proposed energy consumption model are threefold. First,
we use modified sigmoid function in (7) to represent the non-
linear relationship between Psand Ptx. This is a general power
consumption model that applies for different IoT devices.
Second, our energy consumption model considers the non-
negligible energy overhead, which results in a step increase
in energy consumption on successive data transmissions. As
illustrated in Fig. 13, the profile of Et(LN
e, N )becomes a dis-
continuous curve. By contrast, the linear energy consumption
in the conventional model ignores the energy overhead. Third,
our new model accounts for the reducing supply voltage that
is resulted from the low energy harvest rate and the smaller
capacity of the supercapacitors than conventional batteries. As
proved by (13), with linearly reducing supply voltage, the
cumulative energy consumption on a packet transmission is
a parabola that opens downward as shown in Fig. 13.
Due to the small capacity of the battery and the low power
density of renewable radio energy, the impact of the overhead
energy and the nonlinearity of Etfunction cannot be ignored,
which brings considerable challenges to the practical power
10
management. How to manage the transmission power and
efficiently utilize the received energy on the renewable radio
energy powered IoT devices will be our future work.
VII. CONCLUSIONS
In this paper, we proposed a new model to describe the
energy harvesting process and the total energy consumption
of communication modules for renewable RF energy powered
IoT devices. All modeling efforts are based on experiment
results to make the proposed model accurate.
As investigated in this paper, despite the significant fluc-
tuations in the intensity of renewable radio energy in the
outdoor environment, the battery charging process remains
stable, which makes the energy harvesting process highly
predictable. To use the received energy efficiently, we used the
Microchip ATmega256RFR2 microcontroller as an example to
analyze the overhead energy generated by the hardware and
communication protocol in the real world. Experimental re-
sults show that the overhead energy, the nonlinearity of system
power, and the reducing supply voltage can significantly affect
the total energy consumption on data transmissions, which
have been carefully considered in the new model.
We believe that the model and results obtained in this
paper can help researchers design more realistic and efficient
power manage methods and data transmission strategies for
renewable radio energy powered IoT devices.
REFERENCES
[1] U. Muncuk, K. Alemdar, J. D. Sarode, and K. R. Chowdhury, “Multi-
band ambient RF energy harvesting circuit design for enabling battery-
less sensors and IoT,” IEEE Internet of Things Journal, vol. 5, no. 4,
pp. 2700–2714, 2018.
[2] S. Kim, R. Vyas, J. Bito, K. Niotaki, A. Collado, A. Georgiadis, and
M. M. Tentzeris, “Ambient RF energy-harvesting technologies for self-
sustainable standalone wireless sensor platforms,” Proceedings of the
IEEE, vol. 102, no. 11, pp. 1649–1666, 2014.
[3] Powercast Corporation, P2110B 915 MHz RF power harvester receiver,
Powercast Corporation, Pittsburgh, PA, U.S.A., December 2016.
[4] ——, Lifetime power energy harvesting development kit for wireless
sensors user’s manual, Powercast Corporation, Pittsburgh, PA, U.S.A.,
July 2015.
[5] Microchip Technology, Atmel ATmega256RFR2 Data Sheet, Microchip
Technology, Chandler, AZ, U.S.A., September 2014.
[6] S. Ulukus, A. Yener, E. Erkip, O. Simeone, M. Zorzi, P. Grover, and
K. Huang, “Energy harvesting wireless communications: a review of
recent advances,” IEEE Journal on Selected Areas in Communications,
vol. 33, no. 3, pp. 360–381, 2015.
[7] X. Chen, W. Ni, X. Wang, and Y. Sun, “Provisioning quality-of-service
to energy harvesting wireless communications,” IEEE Communications
Magazine, vol. 53, no. 4, pp. 102–109, 2015.
[8] J. Yang and S. Ulukus, “Optimal packet scheduling in an energy harvest-
ing communication system,” IEEE Transactions on Communications,
vol. 60, no. 1, pp. 220–230, 2012.
[9] K. Tutuncuoglu and A. Yener, “Optimum transmission policies for
battery limited energy harvesting nodes,” IEEE Transactions on Wireless
Communications, vol. 11, no. 3, pp. 1180–1189, 2012.
[10] Y. Luo, L. Pu, Y. Zhao, W. Wang, and Q. Yang, “A nonlinear recursive
model based optimal transmission scheduling in RF energy harvesting
wireless communications,” IEEE Transactions on Wireless Communica-
tions, 2020.
[11] M. Gorlatova, A. Wallwater, and G. Zussman, “Networking low-power
energy harvesting devices: measurements and algorithms,” IEEE Trans-
actions on Mobile Computing, vol. 12, no. 9, pp. 1853–1865, 2013.
[12] A. Biason and M. Zorzi, “On the effects of battery imperfections in an
energy harvesting device,” in Proceedings of International Conference
on Computing, Networking and Communications (ICNC). IEEE, 2016,
pp. 1–7.
[13] S. Zhao, G. Huang, and Q. Zhu, “Power allocation algorithm for
an energy-harvesting wireless transmission system considering energy
losses,” Algorithms, vol. 12, no. 1, p. 25, 2019.
[14] I. Chaour, A. Fakhfakh, and O. Kanoun, “Enhanced passive RF-DC
converter circuit efficiency for low RF energy harvesting,” Sensors,
vol. 17, no. 3, p. 546, 2017.
[15] B. Devillers and D. G ¨
und¨
uz, “A general framework for the optimization
of energy harvesting communication systems with battery imperfec-
tions,” Journal of Communications and Networks, vol. 14, no. 2, pp.
130–139, 2012.
[16] J. Xu and R. Zhang, “Throughput optimal policies for energy harvesting
wireless transmitters with non-ideal circuit power,” IEEE Journal on
Selected Areas in Communications, vol. 32, no. 2, pp. 322–332, 2014.
[17] O. Orhan, D. Gunduz, and E. Erkip, “Energy harvesting broadband com-
munication systems with processing energy cost,” IEEE Transactions on
Wireless Communications, vol. 13, no. 11, pp. 6095–6107, 2014.
[18] C. R. Valenta and G. D. Durgin, “Harvesting wireless power: survey
of energy-harvester conversion efficiency in far-field, wireless power
transfer systems,” IEEE Microwave Magazine, vol. 15, no. 4, pp. 108–
120, 2014.
[19] TV Fool, “TV signal analysis,” tvfool.com, 2019, [Accessed: February,
2019]. [Online]. Available: http://www.tvfool.com/index.php?option=
com frontpage&Itemid=1
[20] Cavell Mertz & Associates Incorporated, “KETH-TV,” fc-
cinfo.com, 2018, [Accessed: February, 2018]. [Online]. Avail-
able: http://www.fccinfo.com/CMDProEngine.php?sCurrentService=
TV&tabSearchType=Appl&sAppIDNumber=1354719
[21] ——, “WSBK-TV,” fccinfo.com, 2018, [Accessed: February, 2018].
[Online]. Available: http://www.fccinfo.com/CMDProEngine.php?
sCurrentService=TV&tabSearchType=Appl&sAppIDNumber=614548
[22] Y. Luo, L. Pu, G. Wang, and Y. Zhao, “RF Energy Harvesting Wireless
Communications: RF Environment, Device Hardware and Practical
Issues,” Sensors, vol. 19, no. 13, p. 3010, 2019.
[23] K. Hung, C. Masarapu, T. Ko, and B. Wei, “Wide-temperature range
operation supercapacitors from nanostructured activated carbon fabric,”
Journal of Power Sources, vol. 193, no. 2, pp. 944–949, 2009.
[24] H. Sun, Y.-x. Guo, M. He, and Z. Zhong, “Design of a high-efficiency
2.45-GHz rectenna for low-input-power energy harvesting,” IEEE An-
tennas and Wireless Propagation Letters, vol. 11, pp. 929–932, 2012.
[25] S. Agrawal, S. K. Pandey, J. Singh, and M. S. Parihar, “Realization
of efficient RF energy harvesting circuits employing different matching
technique,” in Fifteenth International Symposium on Quality Electronic
Design. IEEE, 2014, pp. 754–761.
[26] P. Nintanavongsa, U. Muncuk, D. R. Lewis, and K. R. Chowdhury,
“Design optimization and implementation for RF energy harvesting
circuits,” IEEE Journal on Emerging and Selected Topics in Circuits
and Systems, vol. 2, no. 1, pp. 24–33, 2012.
[27] Panasonic Solar Amorton Co., Ltd., “Panasonic Amorphous Silicon
Solar Cells Catalog,” https://panasonic.co.jp, 2019, [Accessed:
December, 2019]. [Online]. Available: https://panasonic.co.jp/ls/psam/
en/products/pdf/Catalog Amorton ENG.pdf
[28] G. Papotto, F. Carrara, and G. Palmisano, “A 90-nm CMOS threshold-
compensated RF energy harvester,” IEEE Journal of Solid-State Circuits,
vol. 46, no. 9, pp. 1985–1997, 2011.
[29] H. Jabbar, Y. S. Song, and T. T. Jeong, “RF energy harvesting system
and circuits for charging of mobile devices,” IEEE Transactions on
Consumer Electronics, vol. 56, no. 1, 2010.
[30] Texas Instruments, Low-cost low-power 2.4 GHz RF transceiver, Texas
Instruments, Inc., Dallas, TX, USA, Aug. 2017.
[31] M. Pi˜
nuela, P. D. Mitcheson, and S. Lucyszyn, “Ambient RF energy
harvesting in urban and semi-urban environments,” IEEE Trans. on
Microwave Theory and Techniques, vol. 61, no. 7, pp. 2715–2726, 2013.
[32] Rohde & Schwarz, Generation of IEEE 802.15.4 Signals – Application
Note, Rohde & Schwarz GmbH & Co, Munich, Germany, Dec. 2015.
[33] Microchip Technology, AT86RF212 Data Sheet, Microchip Technology,
Chandler, AZ, U.S.A., September 2018.
[34] Texas Instruments, A True System-on-Chip solution for 2.4 GHz IEEE
802.15.4 / ZigBee, Texas Instruments, Inc., Dallas, TX, USA, Aug. 2017.
11
