On Battery Recovery Effect in Wireless Sensor Nodes

Abstract

With the perennial demand for longer runtime of battery-powered Wireless Sensor Nodes (WSNs), several techniques have been proposed to increase the battery runtime. One such class of techniques exploiting the battery recovery effect phenomenon claims that performing an intermittent discharge instead of a continuous discharge will increase the usable battery capacity. Several works in the areas of embedded systems and wireless sensor networks have assumed the existence of this recovery effect and proposed different power management techniques in the form of power supply architectures (multiple battery setup) and communication protocols (burst mode transmission) in order to exploit it. However, until now, a systematic experimental evaluation of the recovery effect has not been performed with real battery cells, using high-accuracy battery testers to confirm the existence of this recovery phenomenon. In this article, a systematic evaluation procedure is developed to verify the existence of this battery recovery effect. Using our evaluation procedure, we investigated Alkaline, Nickel-Metal Hydride (NiMH), and Lithium-Ion (Li-Ion) battery chemistries, which are commonly used as power supplies for Wireless Sensor Node (WSN) applications. Our experimental results do not show any evidence of the aforementioned recovery effect in these battery chemistries. In particular, our results show a significant deviation from the stochastic battery models, which were used by many power management techniques. Therefore, the existing power management approaches that rely on this recovery effect do not hold in practice. Instead of a battery recovery effect, our experimental results show the existence of the rate capacity effect, which is the reduction of usable battery capacity with higher discharge power, to be the dominant electrochemical phenomenon that should be considered for maximizing the runtime of WSN applications. We outline power management techniques that minimize the rate capacity effect in order to obtain a higher energy output from the battery.

Full text

00
On Battery Recovery Effect in Wireless Sensor Nodes
Swaminathan Narayanaswamy, TUM CREATE Limited
Steffen Schlueter, TUM CREATE Limited
Sebastian Steinhorst, TUM CREATE Limited
Martin Lukasiewycz, TUM CREATE Limited
Samarjit Chakraborty, Technical University of Munich
Harry Ernst Hoster, Lancaster University
With the perennial demand for longer runtime of battery-powered Wireless Sensor Nodes (WSNs), several
techniques have been proposed to increase the battery runtime. One such class of techniques exploiting the
battery recovery effect phenomenon claims that performing an intermittent discharge instead of a continu-
ous discharge will increase the usable battery capacity. Several works in the areas of embedded systems and
wireless sensor networks have assumed the existence of this recovery effect and proposed different power
management techniques in the form of power supply architectures (multiple battery setup) and communica-
tion protocols (burst mode transmission) in order to exploit it. However, until now, a systematic experimental
evaluation of the recovery effect has not been performed with real battery cells, using high accuracy bat-
tery testers to confirm the existence of this recovery phenomenon. In this paper, a systematic evaluation
procedure is developed to verify the existence of this battery recovery effect. Using our evaluation proce-
dure we investigated Alkaline, Nickel-Metal Hydride (NiMH) and Lithium-Ion (Li-Ion) battery chemistries,
which are commonly used as power supplies for WSN applications. Our experimental results do not show
any evidence of the aforementioned recovery effect in these battery chemistries. In particular, our results
show a significant deviation from the stochastic battery models, which were used by many power manage-
ment techniques. Therefore, the existing power management approaches that rely on this recovery effect
do not hold in practice. Instead of a battery recovery effect, our experimental results show the existence of
the rate capacity effect, which is the reduction of usable battery capacity with higher discharge power, to be
the dominant electrochemical phenomenon that should be considered for maximizing the runtime of WSN
applications. We outline power management techniques that minimize the rate capacity effect in order to
obtain a higher energy output from the battery.
CCS Concepts: •General and reference →Experimentation; •Networks →Sensor networks;
•Hardware →Batteries; Wireless devices;
General Terms: Design, Performance, Experimentation, Measurement
Additional Key Words and Phrases: Batteries, wireless sensor nodes, recovery effect, battery modeling,
power management, battery operated electronics.
ACM Reference Format:
Swaminathan Narayanaswamy, Steffen Schlueter, Sebastian Steinhorst, Martin Lukasiewycz, Samar-
jit Chakraborty, and Harry Ernst Hoster, 2016. On Battery Recovery Effect in Wireless Sensor Nodes. ACM
Trans. Des. Autom. Electron. Syst. 00, 00, Article 00 ( 2016), 28 pages.
DOI: http://dx.doi.org/10.1145/0000000.0000000
This work was financially supported by the Singapore National Research Foundation under its Campus for
Research Excellence and Technological Enterprise (CREATE) programme.
Author’s addresses: S. Narayanaswamy (corresponding author) and S. Steinhorst and M. Lukasiewycz,
Embedded Systems Department, TUM CREATE Limited, Singapore; corresponding author’s email:swami-
nathan.narayana@tum-create.edu.sg. S. Schlueter, Department of Electrochemistry and New Materials,
TUM CREATE Limited, Singapore; S. Chakraborty, Institute for Real-time Computer Systems, Technical
University of Munich, Germany; H. E. Hoster, Chemistry Department, Lancaster University, United King-
dom.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights for components of this work owned
by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or repub-
lish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request
permissions from permissions@acm.org.
© 2016 ACM. 1084-4309/2016/-ART00 $15.00
DOI: http://dx.doi.org/10.1145/0000000.0000000
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:2 S. Narayanaswamy et al.
E1E2E3
Vcut-off
Energy [Wh]
Voltage [V]
Intermittent discharge
(Stochastic
battery models)
Continuous
discharge
Intermittent discharge
(Electrochemical behavior)
Fig. 1: Motivational example depicting the contradiction between stochastic battery
models and electrochemical cell behavior. Stochastic battery models compare un-
equal intermittent and continuous discharge input patterns (continuous discharge has
higher average power than intermittent discharge), which results in a higher energy
output1E3from the battery for an intermittent discharge compared to the energy out-
put E2of the continuous discharge. By contrast, the electrochemical cell behavior sug-
gests that a fair, iso-energy input pattern (both continuous and intermittent discharge
patterns with equal average power), will provide a reduced energy output E1for the
intermittent discharge compared to the energy output E2of the continuous discharge.
This observation is validated by our experimental analysis in Section 5.
1. INTRODUCTION
Recent advancement in the field of wireless communication has enabled the wide-
range application of Wireless Sensor Nodes (WSNs) in many real world applications
such as environmental monitoring, medical equipment, smart buildings and indus-
trial applications [Pellegrini et al. 2006], [Akyildiz et al. 2002]. WSNs measure the
environmental data through various sensors, process it and communicate the data to
a base station. Their ability to communicate with other WSNs to form a wireless ad
hoc network has enabled the widespread application of these nodes. In general, WSNs
are powered using batteries and they are deployed in remote places, having very min-
imal human interaction. Constant replacement of batteries is not possible in certain
applications and therefore it is necessary for these WSNs to maximize their runtime.
Several solutions to increase the runtime of WSNs are available in the literature,
ranging from hardware to software level. On the hardware level, optimized circuit
designs with low power consumption and techniques to harvest energy from environ-
mental sources such as solar, wind, etc. were proposed. On the other hand, several
software techniques such as Dynamic Voltage and Frequency Scaling (DVFS) and
Dynamic Power Management (DPM) focus on power saving by turning off inactive re-
sources. Several Media Access Control (MAC) layer protocols and data management
techniques have been analyzed in literature to reduce power consumption of WSNs.
A different approach to increase the runtime of WSNs by exploiting nonlinear prop-
erties of battery cells, such as the recovery effect, has also been considered in the lit-
1Here the energy output of the battery is used as a figure of merit to compare intermittent and continuous
discharge techniques. The energy output is directly proportional to the runtime of the battery if the battery
is discharged in a constant power discharge mode.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:3
erature. There is an existing belief that by performing an intermittent discharge with
in-between idle periods, the usable capacity of the battery could be increased because
the active materials inside the cell are self-replenished and they recover charge dur-
ing the idle periods [Chiasserini and Rao 2001b]. This concept of recovery effect was
explored in literature through stochastic Markov chain models depicting the battery
discharge process. The battery recovery effect is modeled as a backward transition edge
in the Markov chain battery model [Chiasserini and Rao 2001b]. To exploit the battery
recovery effect, several power management techniques in the domain of task schedul-
ing, protocol designs and power supply architectures have been proposed ([Dhanaraj
et al. 2005], [Jayashree et al. 2004], [Benini et al. 2001a], [Benini et al. 2001b], [Jonger-
den et al. 2010]). However, the electrochemical behavior of a cell, as confirmed by our
experimental analysis, shows that an intermittent discharge will provide less energy
output compared to the continuous discharge of the battery performed with equivalent
average power.
Fig. 1 presents a motivating example of this contradiction between stochastic battery
models and the electrochemical cell behavior confirmed by our experimental analysis.
The stochastic battery models obtain an energy output E3by performing an intermit-
tent discharge that consists of a series of TON and TOFF pulses with a power Ppeak
applied during the TON period. During the TOFF period, the battery is idled with 0
power applied. This energy output is compared with the continuous discharge of the
cell performed with the same peak power Ppeak and, according to the stochastic bat-
tery models, it will provide a reduced energy output E2as shown in Fig. 1a. However,
this comparison is not fair, since the average power of the intermittent discharge and
the continuous discharge is not the same (Pintermittent
avg < P continuous
avg ), due to the rest
periods (TOFF) in the intermittent discharge pattern. A fair, iso-energy comparison be-
tween intermittent and continuous discharge patterns must extract the same amount
of energy from the battery, i.e., the average discharge power of the intermittent and
continuous discharge patterns must be equal (Pintermittent
avg =Pcontinuous
avg ). According to
the electrochemical cell behavior confirmed by our experimental results in Section 5,
the intermittent discharge provides a reduced energy output E1(Fig. 1) compared to
continuous discharge for a fair, iso-energy input pattern of intermittent and contin-
uous discharges having equal average power. Until now, neither a clear explanation
nor a systematic experimental evaluation of the cell behavior during an intermittent
discharge is available. The recovery effect stochastic battery models are not validated
with experimental results. In order to analyze the characteristics of the cell during an
intermittent discharge and verify the existence of such a battery recovery effect, a sys-
tematic evaluation is performed in this paper on real battery cells using a standardized
measurement setup.
Contributions and organization of the paper. The major contributions of this pa-
per are as follows:
— For the first time, we developed a standardized evaluation procedure to verify the
existence of battery recovery effect and tested three different battery chemistries
(alkaline, NiMH and Li-Ion) using a high accuracy battery tester. Our experimental
results do not show any evidence of the existence of a battery recovery effect and
therefore the existing power management techniques that rely on the recovery effect
phenomenon are not usable in practice. All measurement raw data are uploaded
in an online repository and made publicly accessible for modeling and verification
purposes.
— Moreover, we provide a detailed explanation of why a charge recovery is not possible
from an electrochemical perspective. Instead of a charge recovery effect, we identify
the rate capacity effect, which is defined as the reduction in available battery capacity
if the discharge rate is increased, as a dominant electrochemical phenomenon that
should be considered for maximizing the battery runtime of WSNs.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:4 S. Narayanaswamy et al.
— Upon identifying the rate capacity effect as the dominant electrochemical phe-
nomenon, we outline power management techniques to minimize it and increase the
runtime of the WSNs.
We provide a comprehensive overview of existing contributions that have utilized
the battery recovery effect and proposed different power management techniques in
Section 2. In Section 3, we explain an existing stochastic battery model using the re-
covery effect that is commonly referred in the literature for proposing different power
management techniques. We highlight the key points of the model along with the un-
derlying assumptions to the real battery behavior. In addition, we explain in detail
the terminologies that are often used in the literature to compare the gain in energy
output and runtime extension of the sensor nodes. In Section 4, we present a general
block diagram of a typical WSN and discuss the individual modules in detail. We an-
alyze the power supply configurations and the communication modes of the WSN that
have an impact on the energy output of the battery based on which we formulate our
evaluation procedure in Section 5.
Our proposed evaluation procedure for verifying the existence of the recovery effect
in batteries is described in Section 5. Moreover, in this section we provide a detailed
analysis of the experimental results obtained from tests performed according to the
evaluation procedure using a high accuracy battery tester. Our experimental results
do not show any existence of recovery effect, which is a clear deviation from state-of-
the-art stochastic battery models. In contrast, our experimental results identify the
rate capacity effect as the dominant electrochemical phenomenon which needs to be
taken into consideration while designing power management techniques for WSN.
In Section 6, we elucidate the various electrochemical reactions that take place
inside the battery during an intermittent discharge. We clearly explain what self-
replenishment of active materials inside a battery means from an electrochemical
perspective and why a charge recovery effect does not hold in practice. Based on our
experimental analysis in Section 5 and the electrochemical explanation of battery be-
havior in Section 6, in Section 7 we suggest necessary modifications that are required
to be made in the existing stochastic models to be used in practice. Moreover, we out-
line both hardware and software based power management approaches to extend the
battery runtime by minimizing the rate capacity effect. Finally, Section 8 summarizes
the main findings of the paper and recapitulates our future research directions.
2. RELATED WORK
In this section, existing contributions that analyze, model or exploit the recovery effect
behavior of batteries are outlined.
This section is organized into four parts as follows:
— Protocols and design optimization
— Recovery effect models
— Experimental evaluations
— Interpretation of electrochemical literature for power management
Protocols and design optimization. Several works assumed the existence of bat-
tery recovery effect and proposed different MAC layer protocols to increase the run-
time of the WSN. For example in [Dhanaraj et al. 2005] and [Jayashree et al. 2004],
the recovery effect phenomenon is exploited by scheduling the wake-up/sleep time of
WSNs appropriately. Communication data traffic control techniques were developed
in [Dasika et al. 2004] and [Chiasserini and Rao 2001b] to exploit the recovery ef-
fect by optimized discharge profiles of the battery. Similar scheduling algorithms are
proposed in [Chenfu et al. 2015] in the area of wireless body sensors, where an im-
provement of 70 % is reported by exploiting the battery recovery effect. On the other
hand, multiple-battery power supply architectures and battery scheduling schemes are
proposed in [Benini et al. 2001a], [Benini et al. 2001b] to exploit the battery recovery
effect. In [Chiasserini and Rao 2001a], [Jongerden et al. 2010], scheduling algorithms
are proposed to exploit the recovery effect phenomenon by switching between batteries
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:5
to draw power in a multi-battery power supply architecture. They claim that schedul-
ing between the batteries to draw power provides longer runtime compared to the
constant parallel-connected battery power supply architecture because of the charge
recovery effect. However, all the above mentioned works do not evaluate or prove the
existence of the battery recovery effect, but they rely on stochastic battery models from
literature using recovery effect as a basis of their work.
Stochastic battery models. Several stochastic models for the battery recovery effect
are available in literature. A consolidation of various battery models is presented in
[Jongerden and Haverkort 2009]. In [Chiasserini and Rao 1999a], [Chiasserini and Rao
1999b] and [Sarkar and Adamou 2003] the dynamics of cell behavior for an intermit-
tent discharge current profile were captured using a Markov chain model. Similarly,
[Rong and Pedram 2006] proposed a continuous-time Markovian decision processes
for modeling the recovery effect behavior of the battery. Although the methodology for
shaping the communication data traffic using the above mentioned battery models is
valid, the underlying assumption of charge recovery in these models is not formalized.
Instead of proving the existence of the charge recovery, these models assume that a
battery recovers charge when it is allowed to rest. A detailed explanation of a com-
monly used stochastic battery models using recovery effect is presented in Section 3.1.
Experimental evaluations. Another class of work focuses on verifying the existence
of the recovery effect by performing experimental tests on battery cells. In [Castillo
et al. 2004], a relay switch controlled by a computer is used to perform intermittent
and continuous discharge experiments on four different battery chemistries (alkaline,
Nickel Cadmium (Ni-Cd), NiMH and Li-Ion) to verify the existence of recovery effect.
Their results indicate that the recovery effect is only prevalent in alkaline cell chem-
istry, whereas the other battery types did not show any charge recovery effect. More-
over, in [Chau et al. 2010], two commercial WSNs were used to generate discharge
patterns with varying active/sleep durations, on a standard 600 mA h NiMH cell. With
intermittent discharge, an increase of 30 to 45% of normalized battery runtime com-
pared to a continuous discharge was reported. Both of the above-mentioned works
compare the normalized runtime of an intermittent discharge test with the total run-
time of the continuous discharge test for proving the existence of the battery recovery
effect. Fig. 2 better explains the implications of this comparison. Existing works per-
form intermittent discharge experiments with the test pattern shown in Fig. 2a, in
which the battery is discharged with a power Ppulse for a time period TON followed by a
rest period of TOFF with 0 power applied. The normalized runtime of this test pattern,
sum of TON, till the battery is fully discharged is compared with the total runtime of
the continuous discharge shown in Fig. 2b, where the battery is discharged with the
same peak power Ppulse. This comparison is unfair since the discharge pattern shown
in Fig. 2b has a higher average value than the discharge pattern in Fig. 2a, due to
the rest periods TOFF in the later case. Conversely, comparing the total runtime of
the continuous discharge pattern shown in Fig. 2c with the normalized runtime ob-
tained from Fig. 2a will lead to a fair comparison. An even fairer analysis for verifying
the existence of the battery recovery effect involves comparing intermittent discharge
tests performed with different values of TON and TOFF. In Section 5, we developed an
evaluation procedure considering different ratios of TON and TOFF for an intermittent
discharge, to verify the existence of battery recovery effect. More explanation regard-
ing the normalized runtime and its effect on estimating the total battery capacity is
presented in Section 3.2.
Interpretation of electrochemical literature for power management. There
are several works in the electrochemical domain that explain the benefits of an inter-
mittent discharge in certain battery chemistries. For example, [LaFollette 1995] and
[Nelson et al. 1997] explain the design and development of a novel battery based on
lead-acid chemistry that is capable of withstanding high pulse power. Both the above-
mentioned works explain a well-known electrochemical behavior of lead-acid battery
chemistry towards intermittent discharge. Even though lead-acid cells show an im-
proved performance with an intermittent load, due to their huge size and weight, they
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:6 S. Narayanaswamy et al.
0
Ppulse
Time [s]
Power [mW]
0
Ppulse
Time [s] 0
Pavg
Time [s]
TON TOFF
(a) Intermittent discharge (b) Continuous discharge
with peak power
(c) Continuous discharge
with average power
Fig. 2: Intermittent and continuous discharge patterns. (a) Intermittent discharge with
peak power Ppulse applied during TON and 0 power applied during TOFF. (b) Continu-
ous discharge with same peak power (Ppulse) of the intermittent discharge in (a). (c)
Continuous discharge with average power Pavg of the intermittent discharge in (a).
are not generally used as power supply for WSN applications. However, this specific
behavior of lead-acid cells has been generalized to other battery chemistries in certain
literature as explained previously in this section. Based on this incorrect generaliza-
tion, several stochastic Markov chain battery models and power management tech-
niques have been proposed. Furthermore, the basis for the chemical explanation of
the recovery effect in most existing contributions is primarily based on [Fuller et al.
1994], where the voltage relaxation phenomena (the relaxation of the cell voltage to-
wards equilibrium after a charge or discharge pulse) in lithium-ion-insertion cells was
discussed. While this paper does not mention any form of recovery effect, it has been
incorrectly cited as a reference for the existence of recovery effect by the stochastic
battery models mentioned in this section.
Summary. In summary, several power management techniques in the form of commu-
nication data scheduling and power supply architecture are available in the literature.
However, these works are based on stochastic battery models that are not experimen-
tally validated and assume that a battery recovers charge when it is allowed to rest.
Moreover, the results from discharge experiments performed in the literature are mis-
interpreted and claim the existence of the recovery effect. In addition, the well-known
findings in the electrochemical sources pertaining to a specific battery chemistry are
generalized to other battery chemistries and used as a source for the existence of the
recovery effect. Therefore, in this paper, we perform a systematic experimental evalu-
ation on real battery cells using a standardized measurement setup for evaluating the
benefits in terms of energy output obtained from an intermittent discharge compared
to a continuous discharge.
3. MODELS AND TERMINOLOGIES
In this section, we explain a stochastic battery discharge model using recovery effect
that is commonly referred in power management literature. Later, we again discuss
this model in conjunction with our experimental evaluations, in order to understand
which aspect of the model deviates from our experimental results. We then further, in
Section 7, update this model so that it confirms to our experimental results, in which
case the model gets transformed into a battery discharge model and no longer shows
any recovery effect. In addition, we provide a detailed explanation on normalized run-
time of an intermittent discharge, which is used to calculate the gain in runtime ob-
tained due to recovery effect.
3.1. Existing Stochastic Battery Model Using Recovery Effect
Fig. 3, shows an existing stochastic battery discharge model using recovery effect as
proposed in [Chiasserini and Rao 2001b], which is widely utilized ([Chowdhury and
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:7
01....... N-1 N
P
i=1
ai
P
i=2
ai
r1(k)
P1(k)
a1
rN−1(k)
PN−2(k)
a1
a2
a1
rN(k)
PN−1(k)
a2
P
i=N
ai
Fig. 3: Markov chain model of a battery discharge process capturing the recovery effect
behavior [Chiasserini and Rao 2001b].
Chakrabarti 2005], [Nuggehalli et al. 2006]) for shaping communication data traffic
in WSNs. The battery discharge process is represented in states N,N−1,..., 0. The
basic amount of charge that is drained from a cell in one time slot is defined as one
charge unit. Each fully charged cell is assumed to have a theoretical capacity equal to
Tcharge units and a nominal capacity equal to Ncharge units. The nominal capacity
represents the charge that could be drawn from the cell using a continuous discharge
and it is less than the theoretical capacity.
The fully charged state of the cell is Nand state 0represents the completely dis-
charged state of the cell. The cell discharge is depicted as a stochastic process as shown
in Fig. 3, that starts from state Nand terminates either when state 0is reached or the
theoretical capacity Tis exhausted. Therefore, in each time slot depending upon the
communication data packet arrival, the cell discharges icharge units to process the
data and move from state zto state z−iwith i<z≤N. The probability for a data
packet to arrive in a time slot is defined as ai.
During the rest periods of an intermittent discharge, the cell shall remain in the
same state or recover one charge unit as shown in Fig. 3 depending upon the probabil-
ity Pj(k)given by:
Pj(k) =











a0e−(N−j)αN−αC(k), j = 1, ...., N −1
k= 0, ...., Γ1
a0e−(N−j)αN−ΓαC(k), j = 1, ...., N −1
Γc≤k≤Γc+1
c= 1, ...., cmax −1
with cmax being the number of discharge phases, Γcmax =Tand αN,αCdepends upon
the recovery capability of the battery. The probability to recover charge is modeled as
a decreasing exponential function of the cell State-of-Charge (SoC), since the ability to
recover charge decreases at low SoC values and the exponential decay coefficient (αN,
αC) takes different values depending upon the discharge capacity [Chiasserini and Rao
2001b]. On the other hand, the probability to remain in the same state is defined as:
rj(k) = a0−Pj(k), j = 1, ...., N −1; k= 0, ...., T
rN(k) = a0, k = 0, ...., T
The gain in runtime due to battery recovery effect is calculated as:
G=mp
N(1)
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:8 S. Narayanaswamy et al.
Fig. 4: Measured runtime of a commercial NiMH cell for different TON/TOFF ratios of
intermittent discharge and continuous discharge at equivalent average power.
with mpbeing the average number of charge units drained from the cell with an in-
termittent discharge and Nis the average number of charge units discharged by the
cell from a continuous discharge. The model is simulated with different data packet
arrival processes such as Bernoulli-driven discharge demand and truncated Poisson
distribution and the results in [Chiasserini and Rao 2001b] show that the gain due to
recovery effect is more pronounced for a burst packet transmission process, since the
idle time is more in a burst transmission compared to the periodic data transmission.
This observation is due to the edges Pj(k)in Fig. 3, which indicates that irrespec-
tive of the battery chemistry, if a battery is allowed to rest after a pulse, the active
materials inside the battery are self-replenished due to the diffusion process and re-
cover charge. This assumption is also used to model the recovery effect as a battery
transition system in [Boker et al. 2014]. In contrast to these results, [Rao et al. 2005]
observed that the battery-aware task scheduling techniques considering the recovery
effect phenomenon, are ineffective compared to the energy optimization techniques,
that focus only on optimizing the actual charge that is delivered to the load by the bat-
tery. While their observation holds for both very fine-grained (less than 10 ms) and very
coarse-grained tasks (greater than 30 min), it is based on an alternative interpretation
of the high-level battery model from [Rakhmatov and Vrudhula 2003]. However, both
the existence of these recovery edges and the findings in [Rao et al. 2005] in a real bat-
tery behavior are not experimentally validated or explained from an electrochemical
perspective.
Our experimental evaluations in Section 5 characterize the battery behavior during
an intermittent discharge and show that the energy output obtained from an intermit-
tent discharge is less when compared to the energy output obtained from a continuous
discharge. This is a clear contradiction to the assumption of the existence of charge
recovery edges by the stochastic models. Furthermore, we will explain in Section 6
what self-replenishment of active materials inside the battery implies and why such
self-replenishment is not feasible from an electrochemical perspective.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:9
3.2. Normalized Runtime of an Intermittent Discharge
In this section, we explain what a normalized runtime of an intermittent discharge
means and why comparing it with the total runtime of a continuous discharge is not
valid.
A single pulse of an intermittent discharge is characterized by an active period TON
during which the cell is discharged with a peak power Ppulse, followed by a rest period
TOFF during which no power is applied as shown in Fig. 2a. The normalized runtime
for this discharge pattern is calculated as follows ([Chau et al. 2010]):
Normalized runtime = Tpulse ·TON
TON +TOFF
(2)
with Tpulse being the measured total battery runtime of the intermittent discharge
pattern shown in Fig. 2a and the ratio of TON
TON+TOFF is called as duty-cycle rate of an
intermittent discharge. Fig. 2b shows a continuous discharge of a battery performed
with the same peak power Ppulse of the intermittent discharge, which provides a total
battery runtime of Tcont. In [Chau et al. 2010], the normalized runtime calculated as
per the Eq. (2) is compared with the total runtime (Tcont) obtained from the continuous
discharge test performed with the same peak power of the intermittent discharge.
This is not a valid comparison because the average power drawn from the battery
cell in the intermittent discharge case is reduced by the rest periods (TOFF). On the
other hand, the total runtime (Tpulse) of the intermittent discharge performed with
peak power Ppulse must be compared with the total runtime (Tavg) of the continuous
discharge (shown in Fig. 2c) performed with equivalent average power Pavg computed
as follows:
Pavg =Ppulse ·TON
TON +TOFF
(3)
with Ppulse,TON and TOFF being the peak power, active and rest periods of the
intermittent discharge, respectively. For example, an intermittent discharge with
TON =TOFF =50 s and a peak power (Ppulse) of 40 mW has an average power (Pavg)
of 20 mW as per Eq. (3). Therefore, comparing the normalized runtime of the intermit-
tent discharge (Fig. 2a) with the total runtime of the continuous discharge performed
with same peak power (Fig. 2b) results in an overestimation of the battery capacity.
Instead, the total runtime of the intermittent discharge has to be compared with the
total runtime of the continuous discharge performed with equivalent average power
(Fig. 2c). This is experimentally verified in Fig. 4 for the case of a commercial NiMH
cell with different values of TON and TOFF for intermittent discharge and correspond-
ing continuous discharge tests performed with equivalent average power.
4. ARCHITECTURE AND OPERATING MODES OF WSN
In this section, a detailed analysis of the building blocks of a WSN and their existing
power management approaches are presented. Understanding the WSN architecture
and its working behavior enables to analyze the impact on the energy output obtained
from the battery. Moreover, it facilitates to clearly interpret our evaluation procedure
and results explained in Section 5 for experimentally analyzing the existence of the
battery recovery effect. For example, the test patterns in Sets 1, 2 and 3 described in
Table I (Section 5 on page 13) are equivalent to periodically switching the batteries to
power the WSN in a multiple-battery setup as explained in Section 4.2. Comparing the
results of these tests with the output of Set 4 in Table I corresponds to evaluating the
different power supply configurations shown in Fig. 6. Similarly, the different commu-
nication patterns (periodic vs. burst) of WSNs explained in Section 4.2, are evaluated
by comparing the output of tests with equal (TON/TOFF )in Sets 1, 2 and 3 described
in Table I. Therefore, a detailed explanation regarding the different operating modes
of the WSN is required to understand our experimental setup and results explained in
Section 5.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:10 S. Narayanaswamy et al.
Battery B
DC-DC
Sensors MCU Tx/Rx
=
=
−+
Fig. 5: Functional block diagram of a typical WSN [Raghunathan et al. 2002]. Mea-
sured sensor data is processed by the MCU and transmitted through the transceiver.
The DC-DC converter provides a constant regulated power from the battery B.
4.1. Architecture of a WSN
The general architecture of a typical WSN is shown in Fig. 5 and can be classified into
four modules as:
— Sensing module
— Computation module
— Communication module
— Power supply module
Sensing module. The sensing module consists of different types of sensors which es-
tablish connection between the WSN and the environment. They could be classified as
either passive sensors such as temperature and humidity, which consume less power,
or active sensors such as image recorders and Global Positioning System (GPS), which
are large power consumers. Turning off sensors during inactive periods reduces their
power consumption and enables them to achieve longer battery runtime.
Computation module. The computation module consists of the MCU, which controls
all other blocks in the WSN. It receives input data from the sensing module, processes
it and transmits the information through the communication module. Different power
saving techniques such as DPM, DVFS exist in literature [Choi and Cha 2010] to re-
duce the power consumption in MCUs. Most MCUs provide low power modes of oper-
ation to reduce the power consumption of the WSN during sleep state.
Communication module. The communication module consists of the wireless radio
which enables the communication between other WSNs or to the base station. Power
savings in wireless radios can be done by duty cycling their operation with appro-
priate wake-up and sleep times. Several communication protocols such as ASLEEP
[Anastasi et al. 2009], S-MAC [Ye et al. 2004], B-MAC [Polastre et al. 2004] and DS-
MAC [Peng et al. 2004] allow duty cycling of the wireless radio in order to reduce the
energy consumption. On the other hand, in-network data processing techniques such
as data compression or data aggregation [Mo et al. 2011] and energy efficient com-
munication data routing mechanisms as in [Junyoung et al. 2009] help to reduce the
energy consumption of the wireless radio.
Power supply module. The power supply module is comprised of the battery and
the DC-DC converter. The DC-DC converter provides constant regulated power for the
operation of other WSN modules. The converter could be either step-up (Boost), step-
down (Buck or Linear) or step-up/down (Buck-boost) converter [Erickson and Maksi-
movic 2001]. From a system designer’s point of view, most batteries are considered as a
black box, providing constant voltage until their end-of-life. In contrast to this, the out-
put voltage of most batteries declines continuously with discharge. In such cases, a DC-
DC converter plays a vital role by providing constant regulated supply voltage taking
into consideration the decreasing battery output voltage [Min et al. 2001], [Sinha and
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:11
Po
B1
Po/N
B2
Po/N
BN
Po/N
...
−+
(a) Parallel configuration
Po
B1
Po
s1
B2
0
s2
BN
0
sN
...
−+
(b) Switching configuration
Fig. 6: Multiple battery power supply architectures for WSNs. (a) Cells are connected
in parallel to share the power delivered to the WSN, configuration A. (b) Cells are
periodically switched to provide power to the WSN, configuration B.
Chandrakasan 2001] and [Benini et al. 2003]. In general, it is a popular belief among
most designers that adding a DC-DC converter will reduce the overall efficiency of
the system due to the internal losses in the converter. However, it was experimentally
proven in [Day 2009] that by adding a DC-DC converter there could be a significant
increase in the battery runtime, even after considering internal losses of the converter.
Two systems (system 1 and system 2) were experimentally verified in [Day 2009] to
analyze benefits of powering a MCU (MSP430FG4168) with a DC-DC converter. In
system 1, the MCU was powered directly from two series-connected AA alkaline cells.
In system 2, a DC-DC converter (TPS780xx) was used to provide a constant supply
voltage of 2.2V to power the MCU from the battery. At the end of the experiment, sys-
tem 2 with the DC-DC converter operated for 298 hours, whereas system 1 operated
for a duration of 233 hours. An increased runtime of 30% was achieved, even after
considering efficiency losses of the DC-DC converter. This is due to the fact that the
current consumption of a typical MCU increases linearly with the supply voltage and
therefore the MCU in system 1 (which is directly powered from the battery) consumed
higher currents initially when the battery was fully charged. However, in system 2,
the DC-DC converter maintained a constant 2.2V supply voltage, thereby reducing
the current consumption of the MCU compared to system 1.
Implications of DC-DC converter on our experiments. The addition of a DC-DC
converter modifies the discharge profile of the battery by increasing the current drawn
at lower cell voltage to maintain a constant power. As a result, the battery is discharged
with constant power. Hence, all our experiments in Section 5 are performed with con-
stant power discharge mode in order to make the results reproducible for most WSN
and portable computer applications. In applications where a DC-DC converter is not
involved, the magnitude of the discharge current is proportional to the supply voltage
and it behaves like a constant resistance load to the battery.
4.2. Power Supply Configurations and Communication Modes
In this section, we analyze two power supply configurations (constant parallel-
connected and switched) that have an impact on the energy delivered to the system.
Moreover, we discuss two different communication modes (periodic and burst trans-
mission) of the WSN, in case of single battery powered applications. Based on these
two test cases, we formulate an evaluation procedure in Section 5 to verify the exis-
tence of battery recovery effect.
Power supply configurations. In applications where weight or size of the WSN is
not a critical parameter, more than one battery cell can be used as power supply as
shown in Fig. 6. Implementations using multiple batteries as power supply are ex-
plored extensively in the literature [Dhanaraj et al. 2005], [Benini et al. 2001b], [Chi-
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:12 S. Narayanaswamy et al.
asserini and Rao 2001a], [Qing et al. 2000], [Lahiri et al. 2002]. It is claimed in the lit-
erature that periodically switching between the batteries to power the WSN as shown
in Fig. 6b provides a gain in runtime due to the recovery effect compared to the con-
stant parallel-connected setup shown in Fig. 6a. However, this claim is only based on
stochastic battery models explained in Section 3 and no experimental evaluation was
performed on batteries considering a multiple battery setup. In Fig. 6a, hereafter re-
ferred to as configuration A, cells of equal capacity are connected in parallel and the
input power required by the WSN is shared equally by all of them. If Pois the power
required by the WSN, then the power delivered by individual cells Piis given by
Pi=Po
N(4)
with Nbeing the number of cells connected in parallel. In Fig. 6b, hereafter referred
to as configuration B, cells are periodically switched to power the WSN. Therefore,
pulses of power Poare periodically drawn from each cell with a certain rest period.
This operation mode is experimentally analyzed in Section 5.
Communication modes. In general, WSNs are operated in sleep mode for most of the
time to reduce the energy consumption from the battery. They are periodically woken
up to transmit the processed data or process the received data. The average end-to-end
power consumption of a wireless radio is [Shih et al. 2004]:
Pradio =Ntx[Ptx (Ton−tx +Tst) + Pout ·Ton−tx ] + Pbb−tx
+Nrx[Prx (Ton−rx +Tst)] + Pbb−rx (5)
with Ntx/rx being the average number of times per second the transceiver is used,
Ptx/rx is the power consumption of the transceiver, Pout is the output transmit power,
Ton−tx/rx is the on-time of the transceiver, Tst is the start-up time of the transceiver
and Pbb−tx/rx is the average power consumption of the baseband block.
Every transceiver (transmitter and receiver) device has a nonzero start-up time (Tst)
during which no data is transmitted or received. Power consumption of the wireless
radio at low data rates is dominated by the power consumption during this start-up
period of the transceiver. Therefore, the wake-up/sleep time of the WSN (Ton−tx/rx) has
to be chosen appropriately, considering the packet size to minimize start-up transient
losses in the transceiver. The WSN can either transmit the processed data periodically
as shown in Fig. 7a or buffer the data and transmit it as a burst (Fig. 7b). It is claimed
in literature that by allowing the battery to rest long enough during an intermittent
discharge (as in burst transfer mode), the total runtime of the battery is increased
due to charge recovery effect. To investigate the effect of both operating scenarios on
batteries, experiments were performed in Section 5, with different periods of ON and
OFF times ranging from 0.5s to 50 s.
5. EVALUATION PROCEDURE AND RESULT ANALYSIS
In this section, we present the systematic evaluation procedure formulated to verify
the battery recovery effect. A detailed overview of the high accuracy battery tester used
to implement our evaluation procedure is provided. As already mentioned before, the
results obtained from our experimental evaluations do not show any evidence for ex-
istence of recovery effect. By contrast, our results show that the rate capacity effect to
be the dominant electrochemical phenomenon that should be considered for developing
power management techniques for WSNs.
5.1. Evaluation Procedure
The systematic evaluation procedure shown in Table I enables to experimentally verify
the recovery effect behavior on batteries under different discharge load patterns. Sets 1
to 3 consist of intermittent discharge tests with different values of TON,TOFF and Set 4
represents the corresponding continuous discharge tests performed with equivalent
average power of intermittent discharge computed according to Eq. (3). TON and TOFF
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:13
0
Pout
Time [s]
Power [mW]
0
Pout
Time [s]
Tst Ton−tx/rx
Pbb−tx/rx
Tst
10 ·Ton−tx/rx
Pbb−tx/rx
(a) Periodic transfer (b) Burst transfer
Fig. 7: Power consumption of a WSN during communication. (a) Periodic data transfers
with start-up power at the beginning of every transmission. (b) Burst data transfer
where the start-up power is required for a single time.
represent active and sleep states of the WSN, respectively. The discharge power value
shown in Table I was only applied to the cell during TON period and during TOFF period
the cell was completely isolated, with zero power applied. The operating voltage of most
commonly used WSNs (TelosB, TMote [Werner et al. 2006] and Mica Motes [Horton
et al. 2002]) are in the range of 3.3V and their overall current consumption is around
25 mA. Therefore, a power consumption value of 80 mW was chosen upon multiplying
the operating voltage and the total nominal current consumption of these WSNs. In
addition, the batteries used in these nodes are either two series-connected alkaline or
NiMH cells, or a single Li-Ion cell. Therefore, the discharge power was halved in case
of tests performed with alkaline and NiMH cells and full discharge power was applied
in case of Li-Ion cells, since single cells were used for our experimental analysis. More-
over, the active and sleep periods in WSNs can vary widely depending upon the appli-
cation scenario. For example, in environment monitoring applications the TOFF periods
could be very large ranging from minutes to hours. Therefore, in our evaluation pro-
cedure we have considered a wide spread of TON and TOFF values as shown in Table I.
TON[s] TOFF [s] TON/TOFF Power at TON [mW]
Alkaline
& NiMH Li-Ion
Set 1
50 50 1:1
50 100 1:2 40 80
50 200 1:4
Set 2
5 5 1:1
5 10 1:2 40 80
5 20 1:4
Set 3
0.5 0.5 1:1
0.5 1 1:2 40 80
0.5 2 1:4
Set 4
20 40
Continuous 13.3 26.6
8 16
Table I: Evaluation procedure for verifying the existence of the battery recovery effect
in WSN applications. Set 1, 2, 3 are intermittent discharge tests with different val-
ues of TON and TOFF and Set 4 is the corresponding continuous discharge test with
equivalent average power.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:14 S. Narayanaswamy et al.
Computer with
measurement software
Battery tester
Cell holder
Cells
Fig. 8: BaSyTec CTS battery tester setup along with a computer running the mea-
surement software for the systematic experimental evaluation of the battery recovery
effect.
Energy output of all tests was recorded to evaluate the existence of battery recovery
effect. The energy output of the battery is directly proportional to the runtime since
the battery is discharged in a constant power discharge mode.
Verification of power supply configurations. The multiple-battery power supply
architectures explained in Section 4.2 are experimentally verified using the evalua-
tion procedure in Table I. By comparing the energy output obtained from intermittent
discharge tests performed in Sets 1, 2 and 3 to the energy output obtained from con-
tinuous discharge tests done in Set 4, the effectiveness of the two power supply config-
urations could be experimentally evaluated. For example, the test done in Set 4 with a
continuous power discharge of 8 mW is equivalent to configuration A, shown in Fig. 6a
with N= 5. Similarly, the test performed in Set 1 with TON =50 s and TOFF =200 s is
equivalent to configuration B, shown in Fig. 6b with N= 5. By comparing the energy
output of the battery obtained from both tests, an analysis of which power supply ar-
chitecture provides higher energy output for N= 5 could be performed and thereby
the existence of recovery effect in a multiple-battery setup could be experimentally
verified. Similar comparisons could be made for N= 2 and N= 3 from tests performed
with different ratios of TON and TOFF according to Table I.
Verification of communication modes. The effect of varying the wake-up/sleep
time of the WSN on batteries as explained in Section 4.2 is experimentally verified by
comparing the energy output obtained from tests done with equal TON/TOFF ratios in
all sets from Set 1 to 3. For example, the test performed in Set 3 with TON =TOFF =0.5s
is equivalent to a periodic transmission of data as shown in Fig. 7a. Correspondingly,
the test case TON =TOFF =5s in Set 2 is equivalent to the WSN buffering the input
data and transmitting it in bursts, as shown in Fig. 7b. For a qualitative comparison
between periodic data transfer mode and burst data transfer mode, the start-up time
(Tst) of the transceiver and the baseband power (Pbb−tx/rx ) are not considered in the
evaluation procedure. However, for optimization of the data packet size to minimize
energy consumption, start-up time and baseband power should be taken into account.
By comparing the energy output obtained from these tests, the influence of varying
wake-up/sleep times of the WSN on the battery power supply could be experimentally
analyzed.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:15
Fig. 9: Continuous and intermittent dis-
charge of the alkaline batteries with
(TON/TOFF )ratio of (a) 1:1, (b) 1:2 and (c)
1:4. Discharge powers are indicated along
with TON and TOFF times in square brack-
ets.
Table II: Energy output ob-
tained from alkaline battery
with our evaluation procedure.
Power in mW
[Time in s]
Energy
Output
[Wh]
40 [50] / 0 [50] 1.43
40 [5] / 0 [5] 1.44
40 [0.5] / 0 [0.5] 1.49
20 1.5
40 [50] / 0 [100] 1.46
40 [5] / 0 [10] 1.48
40 [0.5] / 0 [1] 1.486
13.3 1.53
40 [50] / 0 [200] 1.48
40 [5] / 0 [20] 1.49
40 [0.5] / 0 [2] 1.487
8 1.6
5.2. Battery Tester
All tests listed in Table I were performed using a high accuracy BaSyTec CTS battery
tester (BaSyTec GmbH, Germany) [BaSyTec Tester 2013]. The BaSyTec CTS system
is capable of accurately characterizing battery cells up to a maximum charge and dis-
charge current of 5A. Advantages of this tester are the precise control of applied power
or current, high accuracy, high resolution measurements and fast data acquisition of
current, voltage, temperature and time. Furthermore, the implementation of various
test cases is easy and flexible, reducing the risk of unforeseen human errors. An exam-
ple test setup is shown in Fig. 8, containing a 4-channel BaSyTec CTS battery tester, a
notebook running the measurement software and the cell holder containing the cells.
5.3. Analysis of Experimental Results
In this section, experimental results from our evaluation procedure of the recovery
effect are presented and analyzed. All tests were performed on three different com-
mercially available battery chemistries (alkaline, NiMH and Li-Ion), that are suitable
power sources for WSN applications. Results of each individual chemistry are plotted
separately and explained in the remainder of this section.
All cells were discharged according to the evaluation procedure in Table I, until their
voltage reached the cut-off value, specified in their data sheet. In order to detect possi-
ble manufacturing variances, all tests were performed on two cells of the same batch.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:16 S. Narayanaswamy et al.
Fig. 10: Magnification of the discharge profile shown in Figure 9a at a high SoC (I) and
low SoC (II).
The maximum detected variance between two cells for the same test pattern was in
the range of 1.5%. Data was recorded in the BaSyTec measurement software for a step
change of 1mV in battery voltage. All measurement raw data of approximately 5 GB
of size is available for reference and modeling purposes at [Recovery Effect Results
2015].
Alkaline battery. AAA alkaline batteries from Varta (powerone) [Varta Alkaline
2003] were used for our evaluation. Energy output obtained from various ratios of
(TON/TOFF )is plotted in Fig. 9, where Fig. 9 a, b and c correspond to (TON/TOFF)ratios
of 1:1, 1:2 and 1:4, respectively. Values of TON and TOFF along with their corresponding
discharge power are indicated in the graph and explained in Table I. The figure also
includes the energy output obtained from the continuous discharge tests performed
with equivalent average power for each specified (TON/TOFF)ratio. All tests were per-
formed until the cell voltages reached their cut-off values of 0.9V as specified in their
datasheet. The energy output obtained from the battery for each test case is provided
in Table II.
Fig. 10 shows the magnification of the discharge profile of Fig. 9a at two specific
SoC regions (high and low). At high SoC values, the difference in cell voltage between
TON and TOFF of an intermittent discharge is less compared to that at low SoC values.
As a result, the cell voltage profile becomes broader towards the end of discharge. Due
to the high sampling rate of the tester, the cell voltage looks like a thick broad line
at low SoC values in Fig. 9. From our experimental results for alkaline cells shown in
Table II, we observe that the energy output obtained from intermittent discharge tests
for different values of TON and TOFF is less compared to the energy output obtained
from continuous discharge tests performed with equivalent average power. This indi-
cates that there is no existence of charge recovery effect and therefore the existing
power management techniques that rely on the recovery effect phenomenon are not
applicable in practice. Based on this observation, in Section 7.1 we provide necessary
amendments for the existing stochastic battery models that rely on the recovery effect.
From Table II, the energy output obtained from continuous discharge tests of 8mW
and 20 mW are 1.6W h and 1.5W h, respectively. Even at these investigated lower
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:17
Fig. 11: Discharge profiles of alkaline batteries measured at different constant power
rates as indicated. A higher power rate reduces the usable energy output of the cell.
power rates, we observe a significant change in the energy output obtained from the
battery. This is primarily attributed to the rate capacity effect, which states that in-
creasing the rate of discharge decreases the capacity and energy output obtained from
the battery (a detailed explanation of the rate capacity effect is provided in Section
6). The rate capacity effect can be observed from the measurements shown in Fig. 11,
where higher energy output can be obtained by discharging the cell at lower discharge
powers. Therefore, from our experimental analysis on alkaline cells, we identify that
the rate capacity effect to be the dominant electrochemical phenomenon governing
the energy output obtained from the battery. Henceforth, in Section 7.2, we outline
hardware and software based power management techniques to obtain higher energy
output from the battery by minimizing the rate capacity effect.
NiMH and Li-Ion batteries. Results for NiMH (Panasonic, HHR-75AAA/HT) [Pana-
sonic NiMH 2000] and Li-Ion (GSP062530) [GSP Li-Ion 2012] cells for (TON/TOFF )ra-
tio of 1:1 are shown in Fig. 12a and 12b, respectively. The cut-off voltage of NiMH cells
is 0.9V and the cut-off voltage of the investigated Li-Ion cell is 3V. For both battery
chemistries no change in energy output was observed for intermittent and continuous
discharge tests performed as per Table I. The discharge profiles of (TON/TOFF)ratios
1:2 and 1:4 did not show any variation and therefore are not included here, whereas
the measurement data is available in the online repository [Recovery Effect Results
2015]. In comparison to the alkaline cells, the experimental results of NiMH and Li-Ion
batteries as shown in Fig. 12a and 12b, respectively, do not show any significant devi-
ation in energy output between continuous discharge tests and intermittent discharge
tests for various (TON/TOFF )values. The reason for this behavior is that the designs of
the NiMH and Li-Ion cells are capable of withstanding higher power rates than those
commonly experienced in commercial WSN applications. Therefore, they do not show
a significant rate capacity effect in the investigated power range mentioned in Table I.
Nevertheless, 30 % to 40 % improvements obtained in energy output due to intermittent
discharges, as claimed by existing works referenced in Section 2, are in clear contra-
diction with our experimental results. Small variations in the energy output observed
are primarily attributed to manufacturing variances of cells and ambient temperature
fluctuations of ±3◦C during the experiment. Therefore, in these battery chemistries,
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:18 S. Narayanaswamy et al.
(a) NiMH (b) Li-Ion
Fig. 12: Continuous and intermittent discharge of NiMH and Li-Ion cells with
(TON/TOFF )ratio of 1:1. Discharge powers are indicated along with TON and TOFF
times in square brackets.
no potential charge recovery effect was observed within the operating power range of
WSNs.
6. ELECTROCHEMICAL EFFECTS IN BATTERY CELLS
In this section, we elucidate the underlying electrochemical phenomena of batteries,
which are mainly responsible for our results obtained in Section 5.3. Moreover, we
provide clear reasoning of why a charge recovery cannot take place in real battery cells
as claimed in the literature. In addition, the fundamental operations of an electro-
chemical cell along with a focus on overpotential and rate capacity effect are explained.
6.1. Electrochemical Cell
Batteries are electrochemical storage devices, which implies a chemical reaction cou-
pled with an electron transfer. The schematic in Fig. 13 gives an overview of the ba-
sic components inside a battery. During discharge, shuttle ions (M+) are oxidized at
the anode (Eq. (6)) and release electrons (e−), which travel through the outer circuit
to power the load. Oxidized shuttle ions inside the battery move through the elec-
trolyte to the cathode side, where they are reduced by electrons coming through the
load (Eq. (7)). This happens spontaneously because the cathode material is chosen
such that it forms a chemically favorable (very negative Gibbs free energy) reaction
product with the metal anions. The separator prevents flow of electrons through the
electrolyte, forcing them through the outer circuit in order to power the load.
Anode:M→M++e−(Oxidation)(6)
Cathode :M++e−→M(Reduction)(7)
Self-replenishment of active materials. While other effects such as temperature
changes might have an impact in the energy output obtained from the battery, the
so-called recovery effect is often explained as self-replenishment of the active mate-
rials during an intermittent discharge by most of the existing works referenced in
Section 2. From an electrochemical point of view, this means, that the oxidized shuttle
ion (M+) would travel back to the active side of the anode and be reduced again by
an electron from the external circuit. Whereas, during an intermittent discharge, no
electrons are supplied from the outer circuit and hence no real charge recovery is pos-
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:19
Cathode
Anode
Electrolyte
Separator
Load
−+
M+→M+
e−→e−→
Fig. 13: Battery schematic during discharge. Shuttle ions (M+) are oxidized at the
anode and move towards the cathode inside the cell. The electron (e−) released during
oxidation travel through the outer circuit to power the load.
sible. Therefore, the edges (Pj(k)) corresponding to charge recovery in the stochastic
battery models shown in Fig. 3 are invalid from an electrochemical perspective since
no self-replenishment of active materials takes place in the cell.
6.2. Overpotential and Rate Capacity Effect
In this section, we explain the overpotential of a battery, which is the fundamental
principle of the electrochemical phenomenon called rate capacity effect. The reduced
energy output or runtime of an intermittent discharge compared to a continuous dis-
charge observed in our experimental results in Section 5.3 is explained based on the
overpotential and rate capacity effect.
Overpotential. In the context of this paper, the overpotential describes the fact that
whenever a current is drawn from a battery, the voltage of that battery will drop de-
pending upon the magnitude of the current drawn. The equilibrium voltage of the cell
is defined as the cell voltage at chemical equilibrium of the battery at a given SoC and
temperature; it can be expressed as E0. In theory, to obtain maximum energy output
from a battery, the cell voltage ETshould follow the discharge profile of the equilib-
rium voltage E0. However, the cell voltage deviates from the equilibrium voltage as
soon as a current is drawn from the cell. This deviation is termed as overpotential η
and can be expressed as:
η=E0−ET(8)
The overpotential is caused by various kinetic limitations and can be divided into
three main parts: ohmic overpotential (ηohmic), activation overpotential (ηactivation) and
concentration overpotential (ηconcentration) [Winter and Brodd 2004]:
η=ηohmic +ηactivation +ηconcentration (9)
The ohmic overpotential is governed by Ohm’s law and arises from internal resistances
of the ion conducting electrolyte and the electron conducting construction materials of
the battery (electrodes, current collector, terminals). The activation overpotential, also
known as electron transfer overpotential, arises from kinetic hindrance during the
charge-transfer reaction and can be described by the ButlerVolmer and Tafel equa-
tions [Bard and Faulkner 1980]. The concentration overpotential is caused by limited
mass transport when diffusion arises from a gradient in concentration and can be de-
scribed by Fick’s laws. The practical relevance of the overpotentials is best understood
in a scenario where a constant current Iext flows through a load. The same current is
then flowing through every interface and along every path in the battery cell. If the
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:20 S. Narayanaswamy et al.
Fig. 14: Measured influence of the power rate on the overpotential and the voltage
relaxation of a NiMH cell.
transport of shuttle ions (M+) or electrons (e−) is hindered at a position X along this
path, the enforced external current Iext builds up a local polarization and thus an elec-
tric field. This field adds additional driving force to the ions (or electrons) at position
X. The field becomes stronger until it is just sufficient to locally create a current Ilocal
at position X that is equal to the external current Iext. The local polarization voltage
(overpotential), associated with this field, however, diminishes the externally available
voltage ETas compared to the theoretical cell voltage E0. A detailed review of different
sources of overpotentials and their underlying mechanism is explained in [Park et al.
2010], [Bernardi and Go 2011].
The total overpotential is a function of current rate, SoC, temperature, battery chem-
istry, battery design and age of the battery. The influence of power rate on the overpo-
tential is crucial to understand the experimental results in Section 5.3. Fig. 14 illus-
trates the measured effect of different power rates on the overpotential and the voltage
relaxation behavior of a NiMH battery. The cell was given a 5s pulse (discharge) with
currents of 250, 500 and 1000 mA and allowed to rest after (pause). It was found that
the higher the power rate, the higher the overpotentials. The influence of power rate on
each individual overpotential (ηohmic,ηactivation,ηconcentration) depends on many factors,
however, all overpotentials will increase with increasing power rate. Intermittent dis-
charge always has a higher peak power and therefore a higher overpotential compared
to the continuous discharge with the equivalent average power.
The different overpotentials have different time responses as shown in Fig. 14 de-
pending on the underlying processes. The ohmic overpotential is very fast and appears
instantaneously (microseconds), the activation overpotential occurs in the range of mil-
liseconds and the concentration overpotential is in the range of seconds up to hours
[Jossen 2006]. The different time responses of the overpotential have an influence on
the energy output obtained from the battery and therefore they are required to be con-
sidered while optimizing the communication data traffic of the WSN as explained in
Section 7.2.
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:21
0 1,000 2,000 3,000 4,000
0.8
0.9
1
1.1
1.2
1.3
1.4
Time [s]
Voltage [V]
Fig. 15: Measured influence of different overpotentials during an intermittent dis-
charge of a NiMH cell. As seen from the graph, the ohmic overpotential is dominating
at higher SoC values and the concentration overpotential is more prominent towards
the end of discharge
Rate capacity effect. In battery terminology, the C-rate is often used to define the
charge or discharge current of a battery. 1C corresponds to the current necessary to
charge or discharge the battery completely in one hour. A 2C rate would be the equiv-
alent current to charge or discharge a battery in half an hour and 0.5C corresponds to
two hours of charge or discharge. This definition makes it easier to compare batteries
with different capacities and quickly access their power capabilities. The rate capacity
effect descibes the decrease of usable capacity of a battery with increasing C-rates. The
relation between the discharge current and the battery output capacity is modeled by
Peukert’s law as
In
batt ·t=C(10)
where tis the time required to discharge the battery at current Ibatt,Cis the ca-
pacity of the battery in A h and nis the Peukert constant. The Peukert constant de-
pends upon the type of battery and it is directly related to the internal resistance of
the cell [Rekioua 2014]. Typically, the different overpotentials (ηohmic,ηactivation and
ηconcentration), reflect the resistance of the cell and this is the reason for batteries to
show a rate capacity effect. As explained previously, the overpotentials are a complex
function of several terms such as temperature, current rate, SoC, battery chemistry,
cell design (high power or high energy cell), age of the battery, etc. [Pop et al. 2008].
Therefore, it is difficult to model the individual contribution of different overpotentials
to the rate capacity effect. A rough estimation regarding the contribution of different
overpotentials is possible from Fig. 14, where we observe that the ohmic overpoten-
tial (ηohmic) is dominating the other overpotentials (ηactivation and ηconcentration) in this
specific SoC value.
Nevertheless, to further understand the influence of different overpotentials on the
rate capacity effect, we performed an intermittent discharge experiment with a NiMH
cell. The test pattern consists of 10 sTON time with 1000 mA discharge current followed
by 10 s of rest with 0mA. The experiment was repeated till the cell voltage reached its
cut-off value of 0.9V as specified in its datasheet. Magnifications of the voltage profile
at three different SoC values (fully charged, partially discharged and towards end of
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:22 S. Narayanaswamy et al.
discharge) are provided. It can be seen that, at high SoC values, the drop due to the
ohmic overpotential is higher compared to other overpotentials. Even though we can
observe some contribution from the activation overpotential at this SoC range, it is
not significant compared to the drop due to the ohmic overpotential. At the partially
discharged state, the rate capacity effect of the cell is mainly dominated by the ohmic
overpotential, as seen in Fig. 15. On the other hand, towards the end of discharge,
the voltage drop due to the concentration overpotential becomes prominent. This is
because the shuttle ions that are moving inside the cell from anode to cathode, as
explained in Section 6.1, are limited with their mobility. This in turn obstructs the
arrival of further shuttle ions towards the cathode and this obstruction results in a
higher overpotential as shown in Fig. 15. To summarize, for this specific NiMH cell,
at full and partially charged states, the ohmic overpotential dominates and towards
the end of the discharge, the concentration overpotential becomes more significant.
Nevertheless, the usable energy and the average voltage of the battery decrease with
increasing power rates. Therefore, in order to obtain higher energy output and thereby
a longer battery runtime, the discharge current rate of the battery has to be reduced
in order to mitigate the rate capacity effect.
Summary. In summary, underlying electrochemical processes strongly depend on the
cell chemistry, the design of the battery and external factors like current rate and tem-
perature. From an electrochemical point of view, self-replenishment of active materials
does not take place during the idle periods of an intermittent discharge. Therefore the
intermittent discharge has no benefit compared to a continuous discharge performed
with the equivalent average power.
7. MODEL AMENDMENTS AND POWER MANAGEMENT TECHNIQUES
Based on our experimental results obtained in Section 5.3 and the electrochemical
explanation of battery behavior provided in Section 6, in this section we suggest nec-
essary amendments for the stochastic battery models explained in Section 3.1. More-
over, we outline both hardware and software based power management techniques to
increase the runtime of the WSN by reducing the rate capacity effect.
7.1. Recommended Amendments to Existing Stochastic Battery Models
From our experimental evaluations in Section 5, we identify that the intermittent dis-
charge of the battery provided less energy output or runtime when compared with the
corresponding continuous discharge performed with equivalent average power. This
implies that the charge recovery does not occur in these battery chemistries during the
idle periods of an intermittent discharge as claimed by the existing literature. More-
over, as explained in Section 6, from an electrochemical perspective it is not possible
for a battery to recover charge during the idle periods of an intermittent discharge as
claimed in the literature. The self replenishment of active materials inside the bat-
tery does not happen spontaneously because the battery cell requires an electron from
the outer circuit to reduce the oxidized shuttle ions at the anode side. Since the ex-
ternal circuit, if it is not a charging device, does not supply electrons, this reaction is
infeasible in case of a real battery cell. Therefore, we suggest that the charge recovery
edges (Pj(k)) in the existing Markov chain model shown in Fig. 3 should be removed,
since the self-replenishment of active materials does not take place during an inter-
mittent discharge. With these changes, the stochastic model with recovery effect gets
transformed into a battery discharge model.
7.2. Power Management Techniques
In this section, we suggest hardware and software based approaches for power man-
agement that can be used to improve the energy output obtained from the battery by
reducing the rate capacity effect.
Hardware based power management approaches. Discharging the battery with
high peak power decreases the overall energy output, because of the rate capacity
effect. This is observed from our experimental results shown in Table II, where the
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:23
=
=
DC-DC
B
CWSN
Fig. 16: Hybrid power supply architecture for obtaining higher energy output from the
battery by minimizing the rate capacity effect [Shin et al. 2011]. The supercapacitor
Chandles the higher peak power experienced during an intermittent discharge and is
efficiently charged at a lower continuous discharge rate from the battery Busing the
DC-DC converter.
continuous discharge test with 8mW discharge power provided an energy output of
1.6W h and the corresponding intermittent discharge test with (TON/TOFF)ratio of 1:4
(40 mW[50 s] / 0mW[200 s]) provided an energy output of 1.48 W h. An energy gain of
approximately 8% was obtained for this continuous discharge test compared to the
corresponding intermittent discharge test. This is equivalent of comparing the power
supply configuration A (cells are connected in parallel) to the power supply configura-
tion B (cells are switched periodically) as shown in Fig. 6 for N= 5. Therefore, from
these experiments, it can be concluded that configuration A performs better than con-
figuration B because of the reduced peak power drawn from the cell. This observation
is a clear contradiction to the results obtained based on stochastic battery models in
[Chiasserini and Rao 2001a] where it is claimed that switching between multiple bat-
teries to power the WSN yields longer runtime than the constant parallel-connected
battery setup due to the charge recovery effect. By contrast, our experimental results
show that switching between the batteries provides reduced energy output compared
to operating them in parallel. This applies to all tests performed on alkaline cells ac-
cording to the evaluation procedure in Table I. Therefore, instead of switching between
the batteries in a multiple-battery power supply architecture, we suggest to operate
them in a constant-parallel connected fashion (configuration A) in order to obtain a
higher energy output by reducing the peak power drawn from each cell.
In certain applications a multiple-battery setup is not possible due to size and weight
constraints. In such cases, the existing power management approaches turn OFF the
WSN during inactive periods in order to reduce the power drawn from the battery.
However, for a battery this behaves like an intermittent discharge with ON periods
followed by OFF periods and as shown from our experimental results that a continu-
ous discharge with equivalent average power provides higher energy output than per-
forming an intermittent discharge. Therefore, in such cases we recommend a battery-
supercapacitor hybrid power supply architecture as shown in Fig. 16, where a super-
capacitor is accompanied with a battery to mitigate the rate capacity effect due to the
intermittent discharge as proposed in [Shin et al. 2011]. Having the advantage of high
power density compared to that of batteries, supercapacitors efficiently handle higher
peak powers as experienced during an intermittent discharge. Higher energy output or
runtime is obtained by charging the supercapacitor at a reduced average power from
the battery and optimally scheduling the wake-up/sleep time of the WSN considering
the amount of charge present in the supercapacitor ([Tanevski et al. 2013]).
Software based power management approaches. Apart from the hardware based
power management solutions, the energy output of the battery might be maximized
through appropriate scheduling of the communication data. From Table II and Fig-
ure 9a, we can see that for the same (TON/TOFF )ratio of 1:1, the test case with
a shorter TON and TOFF time (40 mW[0.5s]/0mW[0.5s]) provided an energy output
of 1.49 W h. However, the test cases 40 mW[5s]/0mW[5s] and 40 mW[50 s]/0mW[50 s],
even though with a longer rest period compared to the former test case, resulted in
a reduced energy output of 1.44 W h and 1.43 W h, respectively. As discussed in Sec-
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:24 S. Narayanaswamy et al.
0
Pout
Time [s]
Power [mW]
0
Pout
Time [s]
Tst Ton−tx/rx
Pbb−tx/rx
Tst Ton−tx/rx
Pbb−tx/rx
TOFF
(a) Shorter TON and TOFF (b) TOFF longer than TON
Fig. 17: Scheduling the wake-up/sleep time of the WSN to obtain higher energy output
considering the electrochemical properties of the battery. (a) A shorter TON and TOFF is
more favorable and (b) on the other hand if shorter TON time is not possible then a
longer TOFF time is required to compensate the concentration overpotential.
tion 6, the concentration overpotential is visible only in the time domain of seconds
and due to the shorter TON time of 0.5s, the battery has a reduced overall overpo-
tential, especially at the end of discharge. On the other hand, the test cases with
TON times of 5s and 50 s have a higher overall overpotential and therefore the en-
ergy output of these tests is lower, even though their TOFF times are larger than the
test case 40 mW[0.5s]/0mW[0.5s]. This indicates that a periodic data transfer mode,
as explained in Fig. 7a, with shorter TON time does not enter the time domain of the
concentration overpotential of the battery. On the other hand, buffering the data for
a longer time and transmitting in a burst mode as shown in Fig. 7b results in a re-
duced energy output due to longer TON time. However, increasing the value of TOFF al-
lows the cell to relax and to compensate the concentration overpotential induced by
the longer TON time (Fig. 9b and 9c). Therefore, we suggest that while optimizing the
wake-up/sleep time of the WSN, the electrochemical properties of the battery such as
overpotential and rate capacity effect should also be considered along with the start-up
losses in the transceiver. A shorter TON and TOFF time is more favorable and also saves
cost in terms of buffer size required for holding incoming data during the TOFF pe-
riod. If shorter TON times are not possible to be scheduled in an application then the
TOFF value must be considerably larger than the TON value as shown in Fig. 17, in
order to compensate for the higher concentration overpotential.
7.3. Future Work
Through our systematic experimental evaluations we identified that there is no ex-
istence of charge recovery effect in batteries. Moreover, we identify the rate capacity
effect as the dominant electrochemical phenomenon that should be considered for max-
imizing the energy output of the battery power supply used in WSN applications. Our
future work in this direction involves analyzing and characterizing the gain obtained
from the hybrid power supply architecture consisting of batteries and supercapaci-
tors outlined in the previous subsection. Moreover, developing optimal communication
data shaping methodologies, considering the electrochemical properties discussed in
this paper will be the focus of our future work.
8. CONCLUDING REMARKS
This paper provides an experimental evaluation of the battery recovery effect in the
domain of WSNs. The general architecture of a typical WSN and two operating modes
that have an impact on the energy output of the battery are analyzed in detail. In
contrast to state-of-the-art approaches, which analyze the charge recovery effect in
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:25
batteries through stochastic battery models, this paper proposes a systematic evalu-
ation procedure for experimentally verifying the existence of battery charge recovery
effect. This evaluation procedure was used to verify three different battery chemistries
(alkaline, NiMH and Li-Ion) using a high accuracy battery tester. The experimental re-
sults do not show any charge recovery effect by performing intermittent discharges on
these three battery chemistries within the operating power range of WSNs. Moreover,
the continuous discharge performed with equivalent average power of the intermittent
discharge provided higher energy output due to the reduced peak power. Therefore, the
dominant electrochemical phenomenon that governs the energy output of the battery
is the rate capacity effect and no charge recovery takes place in these investigated bat-
tery chemistries. This analysis complies with the electrochemical explanation of the
battery behavior during an intermittent discharge. Upon identifying that the rate ca-
pacity effect as the dominant electrochemical phenomenon, we outlined both hardware
and software based power management approaches in order to obtain higher energy
output from the battery by minimizing the rate capacity effect. As a part of future
work, we will characterize the gain obtained from the hybrid power supply architec-
ture consisting of a battery and a supercapacitor outlined in the previous subsection.
Moreover, developing optimal communication data shaping methodologies, considering
the electrochemical properties discussed in this paper will be the focus of our future
work.
REFERENCES
I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. 2002. Wireless sensor networks: a survey.
Computer Networks. 38, 4 (Jan 2002), 393 – 422. DOI:http://dx.doi.org/10.1016/S1389-1286(01)00302-4
G. Anastasi, M. Conti, and D. M. Francesco. 2009. Extending the Lifetime of Wireless Sensor Networks
Through Adaptive Sleep. IEEE Transactions on Industrial Informatics. 5, 3 (Aug 2009), 351–365.
DOI:http://dx.doi.org/10.1109/TII.2009.2025863
A. J. Bard and L. R. Faulkner. 1980. Electrochemical methods: fundamentals and applications. Wiley, New
york, USA.
BaSyTec Tester 2013. Datasheet of BaSyTec CTS Battery Tester. (Mar 2013). http://www.basytec.de
L. Benini, D. Bruni, A. Macii, E. Macii, and M. Poncino. 2003. Discharge current steering for
battery lifetime optimization. IEEE Transactions on Computers 52, 8 (Aug 2003), 985–995.
DOI:http://dx.doi.org/10.1109/TC.2003.1223633
L. Benini, G. Castelli, A. Macii, E. Macii, M. Poncino, and R. Scarsi. 2001b. Extending Lifetime of Portable
Systems by Battery Scheduling. In Proceedings of the Conference on Design, Automation and Test in
Europe. IEEE, 197–203. DOI:http://dx.doi.org/10.1109/DATE.2001.915024
L. Benini, G. Castelli, A. Macii, and R. Scarsi. 2001a. Battery-driven dynamic power management. IEEE
Design and Test of Computers. 18, 2 (Mar 2001), 53–60. DOI:http://dx.doi.org/10.1109/54.914621
D. M. Bernardi and J. Y. Go. 2011. Analysis of pulse and relaxation behavior in lithium-ion batteries. Journal
of Power Sources 196, 1 (Jan 2011), 412–427. DOI:http://dx.doi.org/10.1016/j.jpowsour.2010.06.107
U. Boker, T. A. Henzinger, and A. Radhakrishna. 2014. Battery Transition Systems. SIGPLAN Notices 49, 1
(Jan 2014), 595–606. DOI:http://dx.doi.org/10.1145/2578855.2535875
S. Castillo, N. K. Samala, K. Manwaring, B. A. Izadi, and D. Radhakrishnan. 2004. Experimental Analysis of
Batteries Under Continuous and Intermittent Operations. In Proceedings of ESA/VLSI. IEEE, 18–24.
C. K. Chau, F. Qin, S. Sayed, M. H. Wahab, and Y. Yang. 2010. Harnessing Battery Recovery Effect in Wire-
less Sensor Networks: Experiments and Analysis. IEEE Journal on Selected Areas in Communication
28, 7 (Sep 2010), 1222–1232. DOI:http://dx.doi.org/10.1109/JSAC.2010.100926
Y. Chenfu, W. Lili, and L. Ye. 2015. Energy Efficient Transmission Approach for WBAN
Based on Threshold Distance. IEEE Sensors Journal 15, 9 (Sep 2015), 5133–5141.
DOI:http://dx.doi.org/10.1109/JSEN.2015.2435814
C. F. Chiasserini and R. R. Rao. 1999a. A model for battery pulsed discharge with recovery ef-
fect. In Proceeding of Wireless Communications and Networking Conference. IEEE, 636 – 639.
DOI:http://dx.doi.org/10.1109/WCNC.1999.796721
C. F. Chiasserini and R. R. Rao. 1999b. Pulsed Battery Discharge in Communication Devices. In Proceed-
ings of fifth Annual International Conference on Mobile Computing and Networking. ACM, 88–95.
DOI:http://dx.doi.org/10.1145/313451.313488
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:26 S. Narayanaswamy et al.
C. F. Chiasserini and R. R. Rao. 2001a. Energy efficient battery management. IEEE Journal on Selected
Areas in Communications 19, 7 (Jul 2001), 1235–1245. DOI:http://dx.doi.org/10.1109/49.932692
C. F. Chiasserini and R. R. Rao. 2001b. Improving battery performance by using traffic shaping
techniques. IEEE Journal on Selected Areas in Communications 19, 7 (Jul 2001), 1385–1394.
DOI:http://dx.doi.org/10.1109/49.932705
J. Choi and H. Cha. 2010. A Processor Power Management Scheme for Handheld Systems Consider-
ing Off-Chip Contributions. IEEE Transactions on Industrial Informatics 6, 3 (Aug 2010), 255–264.
DOI:http://dx.doi.org/10.1109/TII.2010.2050330
P. Chowdhury and C. Chakrabarti. 2005. Static task-scheduling algorithms for battery-powered DVS sys-
tems. IEEE Transactions on Very Large Scale Integration (VLSI) Systems 13, 2 (Feb 2005), 226–237.
DOI:http://dx.doi.org/10.1109/TVLSI.2004.840771
S. Dasika, S. Vrudhula, K. Chopra, and R. Srinivasan. 2004. A Framework for Battery-Aware Sensor
Management. In Proceedings of the Conference on Design, Automation and Test in Europe. 962–967.
DOI:http://dx.doi.org/10.1109/DATE.2004.1269017
M. Day. 2009. Using power solutions to extend battery life in MSP430 applications. In App note Texas In-
struments. http://www.ti.com/lit/an/slyt356/slyt356.pdf
M. Dhanaraj, S. Jayashree, and C. S. R. Murthy. 2005. A Novel Battery Aware MAC Protocol
for Minimizing Energy X Latency in Wireless Sensor Networks. In High Performance Com-
puting. Lecture Notes in Computer Science, Vol. 3769. Springer Berlin Heidelberg, 312–321.
DOI:http://dx.doi.org/10.1007/11602569 34
R. W. Erickson and D. Maksimovic. 2001. Fundamentals of power electronics, 2nd edition. Springer.
T. F. Fuller, M. Doyle, and J. Newman. 1994. Relaxation Phenomena in Lithium-Ion-Insertion Cells. Journal
of The Electrochemical Society 141 (Jan 1994), 982–990. DOI:http://dx.doi.org/10.1149/1.2054868
GSP Li-Ion 2012. Datasheet of GSP062530 Li-Polymer cell. (Jun 2012). http://www.greatpowerbattery.com.
hk/rechargeable.html
M. Horton, D. Culler, K. Pister, J. Hill, R. Szewczyk, and A. Woo. 2002. MICA, The Commercializa-
tion of Microsensor Motes. Sensors Magazine 19 (Apr 2002), 40–48. http://www.sensorsmag.com/
networking-communications/mica-the-commercialization- microsensor-motes-1070
S. Jayashree, B. S. Manoj, and C. S. R. Murthy. 2004. On Using Battery State for Medium Access Control
in Ad Hoc Wireless Networks. In Proceedings of the 10th Annual International Conference on Mobile
Computing and Networking. ACM, 360–373. DOI:http://dx.doi.org/10.1145/1023720.1023755
M. R. Jongerden and B. R. Haverkort. 2009. Which battery model to use? IET Software 3, 6 (Dec 2009),
445–457. DOI:http://dx.doi.org/10.1049/iet-sen.2009.0001
M. R. Jongerden, A. Mereacre, H. Bohnenkamp, B. R. Haverkort, and J. Katoen. 2010. Computing Optimal
Schedules of Battery Usage in Embedded Systems. IEEE Transactions on Industrial Informatics 6, 3
(Aug 2010), 276–286. DOI:http://dx.doi.org/10.1109/TII.2010.2051813
A. Jossen. 2006. Fundamentals of battery dynamics. Journal of Power Sources 154, 2 (Mar 2006), 530–538.
DOI:http://dx.doi.org/10.1016/j.jpowsour.2005.10.041
H. Junyoung, H. Jiman, and C. Yookun. 2009. EARQ: Energy Aware Routing for Real-Time and Reliable
Communication in Wireless Industrial Sensor Networks. IEEE Transactions on Industrial Informatics
5, 1 (Feb 2009), 3–11. DOI:http://dx.doi.org/10.1109/TII.2008.2011052
R. M. LaFollette. 1995. Design and performance of high specific power, pulsed discharge, bipolar lead acid
batteries. In Proceedings of Tenth Annual International Battery Conference on Applications and Ad-
vances. IEEE, 43–47. DOI:http://dx.doi.org/10.1109/BCAA.1995.398511
K. Lahiri, S. Dey, D. Panigrahi, and A. Raghunathan. 2002. Battery-Driven System Design: A New Frontier
in Low Power Design. In Proceedings of the Asia and South Pacific Design Automation Conference. IEEE,
261–267. DOI:http://dx.doi.org/10.1109/ASPDAC.2002.994932
R. Min, M. Bhardwaj, S. Cho, E. Shih, A. Sinha, A. Wang, and A. Chandrakasan. 2001. Low-power wire-
less sensor networks. In Proceedings of Fourteenth International Conference on VLSI Design. 205–210.
DOI:http://dx.doi.org/10.1109/ICVD.2001.902661
L. Mo, W. Yajun, and W. Yu. 2011. Complexity of Data Collection, Aggregation, and Selection
for Wireless Sensor Networks. IEEE Transactions on Computers 60, 3 (Mar 2011), 386–399.
DOI:http://dx.doi.org/10.1109/TC.2010.50
R. F. Nelson, R. Rinehart, and S. Varley. 1997. Ultrafast pulse discharge and recharge capabilities of thin-
metal film battery technology. In Proceddings of the 11th International Pulsed Power Conference (Digest
of Technical Papers), Vol. 1. IEEE, 636–641. DOI:http://dx.doi.org/10.1109/PPC.1997.679411
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

Experimental Evaluation of Battery Recovery Effect in Wireless Sensor Nodes 00:27
P. Nuggehalli, V. Srinivasan, and R. R. Rao. 2006. Energy efficient transmission scheduling for delay con-
strained wireless networks. IEEE Transactions on Wireless Communications 5, 3 (Mar 2006), 531–539.
DOI:http://dx.doi.org/10.1109/TWC.2006.1611083
Panasonic NiMH 2000. Datasheet of HHR-75AAA/HT NiMH cells. (May 2000). http://www.actec.dk/
Panasonic/pdf/HHR75AAA.pdf
M. Park, X. Zhang, M. Chung, G. B. Less, and A. M. Sastry. 2010. A review of conduc-
tion phenomena in Li-ion batteries. Journal of Power Sources 195, 24 (Dec 2010), 7904–7929.
DOI:http://dx.doi.org/10.1016/j.jpowsour.2010.06.060
F. D. Pellegrini, D. Miorandi, S. Vitturi, and A. Zanella. 2006. On the use of wireless networks at low level
of factory automation systems. IEEE Transactions on Industrial Informatics 2, 2 (May 2006), 129–143.
DOI:http://dx.doi.org/10.1109/TII.2006.872960
L. Peng, Q. Chunming, and W. Xin. 2004. Medium access control with a dynamic duty cycle for sensor net-
works. In Proceedings of Wireless Communication and Networking Conference, Vol. 3. IEEE, 1534–1539.
DOI:http://dx.doi.org/10.1109/WCNC.2004.1311671
J. Polastre, J. Hill, and D. Culler. 2004. Versatile Low Power Media Access for Wireless Sensor Networks. In
Proceedings of the 2nd International Conference on Embedded Network Sensor Systems. ACM, 95–107.
DOI:http://dx.doi.org/10.1145/1031495.1031508
V. Pop, H. J. Bergveld, D. Danilov, P. P. L. Regtien, and P. H. L. Notten. 2008. Battery management systems:
accurate state-of-charge indication for battery powered applications (springer verlag ed.). Vol. 9. Springer
Verlag. DOI:http://dx.doi.org/10.1007/978-1- 4020-6945-1
W. Qing, Q. Qinru, and M. Pedram. 2000. An interleaved dual-battery power supply for battery-operated
electronics. In Proceedings of the Asia and South Pacific Design Automation Conference. IEEE, 387–390.
DOI:http://dx.doi.org/10.1109/ASPDAC.2000.835130
V. Raghunathan, C. Schurgers, P. Sung, and M. B. Srivastava. 2002. Energy-aware wire-
less microsensor networks. IEEE Signal Processing Magazine 19, 2 (Mar 2002), 40–50.
DOI:http://dx.doi.org/10.1109/79.985679
D. Rakhmatov and S. Vrudhula. 2003. Energy Management for Battery-powered Embedded
Systems. ACM Transactions on Embedded Computing Systems 2, 3 (Aug 2003), 277–324.
DOI:http://dx.doi.org/10.1145/860176.860179
R. Rao, S. Vrudhula, and N. Chang. 2005. Battery Optimization vs Energy Optimization: Which to
Choose and when?. In Proceedings of the International Conference on Computer-aided Design. 439–445.
DOI:http://dx.doi.org/10.1109/ICCAD.2005.1560108
Recovery Effect Results 2015. Recovery effect measurement results repository. (2015). https://bitbucket.org/
recoveryeffect/experimental-results.git
D. Rekioua. 2014. Wind Power Electric Systems: Modeling, Simulation and Control. Springer London.
DOI:http://dx.doi.org/10.1007/978-1-4471-6425- 8
P. Rong and M. Pedram. 2006. Battery-aware power management based on Markovian decision processes.
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 25, 7 (Jul 2006),
1337–1349. DOI:http://dx.doi.org/10.1109/ICCAD.2002.1167609
S. Sarkar and M. Adamou. 2003. A framework for optimal battery management for wire-
less nodes. IEEE Journal on Selected Areas in Communications 21, 2 (Feb 2003), 179–188.
DOI:http://dx.doi.org/10.1109/JSAC.2002.807335
E. Shih, S. Cho, F. S. Lee, B. H. Calhoun, and A. Chandrakasan. 2004. Design Considera-
tions for Energy-Efficient Radios in Wireless Microsensor Networks. Journal of VLSI sig-
nal processing systems for signal, image and video technology 37, 1 (May 2004), 77–94.
DOI:http://dx.doi.org/10.1023/B:VLSI.0000017004.57230.91
D. Shin, Y. Kim, J. Seo, N. Chang, Y. Wang, and M. Pedram. 2011. Battery-supercapacitor hybrid system for
high-rate pulsed load applications. In Proceedings of the Conference on Design, Automation and Test in
Europe. 1–4. DOI:http://dx.doi.org/10.1109/DATE.2011.5763295
A. Sinha and A. Chandrakasan. 2001. Dynamic power management in wireless sensor networks. IEEE
Design and Test of Computers 18, 2 (Mar 2001), 62–74. DOI:http://dx.doi.org/10.1109/54.914626
M. Tanevski, A. Boegli, and P. Farine. 2013. Power supply energy optimization for ultra
low-power wireless sensor nodes. In IEEE Symposium on Sensors Applications. 176–181.
DOI:http://dx.doi.org/10.1109/SAS.2013.6493581
Varta Alkaline 2003. Datasheet of AAA-LR03 Alkaline cells. (Jul 2003). http://www.elektronik.ropla.eu/pdf/
stock/vmb/lr03-aaa.pdf
A. G. Werner, K. Lorincz, M. Ruiz, O. Marcillo, J. Johnson, J. Lees, and M. Welsh. 2006. Deploying a wireless
sensor network on an active volcano. IEEE Journal on Internet Computing 10, 2 (Mar 2006), 18–25.
DOI:http://dx.doi.org/10.1109/MIC.2006.26
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.

00:28 S. Narayanaswamy et al.
M. Winter and R. Brodd. 2004. What are batteries, fuel cells, and supercapacitors? Chemical reviews 104
(Sep 2004), 4245–4269. DOI:http://dx.doi.org/10.1021/cr020730k
W. Ye, J. Heidemann, and D. Estrin. 2004. Medium Access Control with Coordinated Adaptive Sleep-
ing for Wireless Sensor Networks. IEEE/ACM Transactions on Networking 12, 3 (Jun 2004).
DOI:http://dx.doi.org/10.1109/TNET.2004.828953
Received July 2015; revised December 2015; accepted February 2016
ACM Transactions on Design Automation of Electronic Systems, Vol. 00, No. 00, Article 00, Pub. date: 2016.