Energy Modeling of Multihop Energy Neutral WSN Using Wake-up
Receiver
Fayc¸al Ait Aoudia, Matthieu Gautier and Olivier Berder
University of Rennes 1, IRISA, INRIA, Rennes, France
1 STAGE OF THE RESEARCH
Wireless sensor networks (WSN) constitute a key
technology for many applications that require virtual-
physical interactions, such as military surveillance,
environmental monitoring, home automation and
health monitoring. For many long-term applications,
a severe bottleneck is the limited amount of energy
provided by individual batteries commonly used to
power WSN. Energy harvesting forms a promising
technology to tackle this problem. Moreover, the
most energy consuming task in a wireless node is usu-
ally communication. Recent improvements in wake-
up receivers (WURx) hardware make them a privi-
leged direction to reduce the consumption of radio on
wireless nodes.
Using both technologies can enable WSN to
achieve long term sustainability while maintaining
quality of service requirements fulfilled. However,
this requires careful design of both power managers
(PMs) and MAC protocols to efficiently use these two
emerging technologies together. The aim of our work
is to design efficient PMs and MAC protocols for
multihop energy harvesting wireless sensor networks
(EH-WSN) in which each node is equipped with a
WURx. In this paper, we present the energetic model
and two PMs on which we are currently working on.
2 OUTLINE OF OBJECTIVES
Wireless nodes are commonly powered with individ-
ual batteries. For long-term applications, the limited
amount of energy available in batteries becomes a se-
vere bottleneck and the challenge of maximizing the
lifetime of battery-powered WSN has received con-
siderable attention. Two promising technologies to
tackle the problem of energy management in WSN
are energy harvesting and WURx. Indeed, enabling
powering nodes thanks to energy harvested in their
environment instead of batteries is a privileged way
to achieve long term sustainability. Moreover, recent
improvements in the hardware of WURx make them a
promising technology to reduce the power consump-
tion of communication in WSN, which is usually the
most consuming task.
The goal of our work is to jointly design en-
ergy managementsystems and MAC protocols to effi-
ciently use these emerging technologies in the context
of multihop WSN. The rest of this section motivated
the use of each of this technology.
2.1 Energy Harvesting Sensor Networks
With advances in energy harvesting techniques, it is
now possible to build sensor networks in which each
node is entirely powered by energyharvested in its en-
vironment (Le et al., 2013b; Kansal et al., 2007; Vig-
orito et al., 2007). In EH-WSN, each node has at least
one energy harvester and one or more energy storage
devices. The storage devices are used to buffer en-
ergy and thus enable the node to survive periods dur-
ing which harvested energy is not enough to power it.
In those platforms, the challenge is not to maximize
the lifetime as in battery-powered WSN, but to ensure
long-term sustainability.
Sustainability is obtained when over long periods
harvested energy is greater or equal to consumed en-
ergy, an operating mode called Energy Neutral Oper-
ation (ENO) (Kansal et al., 2007). To achieve the best
energy efficiency while satisfying the ENO condition,
all the harvestedenergy should be used to increase the
nodes performance. The harvested energy then equals
the consumed energy over long periods. This operat-
ing mode is called ENO-Max (Vigorito et al., 2007).
2.2 Wake-up Receivers
Usually, the energy consumed for communication is
dominant over all the other nodes activities, such as
sensing or computing. To successfully achieve com-
munication, when a node transmits a packet, the des-
tination nodes should be listening the medium to re-
ceive data from the sender. This task is particularly
challenging in the context of WSN because the duty
3
Ait Aoudia F., Gautier M. and Berder O..
Energy Modeling of Multihop Energy Neutral WSN Using Wake-up Receiver.
Copyright
c
 2015 SCITEPRESS (Science and Technology Publications, Lda.)

cycling schemes used to improve energy efficiency
make this synchronization not trivial. Multiple ren-
dezvous schemes were proposed, and they can be
classify into two types (Huang et al., 2013):
Synchronous Schemes which require the nodes to
be synchronized such that the wake up time of
each node is known a priori. These schemes re-
quire synchronization between nodes, which con-
sumes a lot of energy. In the context of EH-WSN,
each node may have its own duty cycle, chosen to
be the most appropriate to achieve sustainability.
As the dutycycle changesfrequently,especially in
highly dynamical environments, the need of fre-
quent synchronization among nodes often makes
these schemes unpractical. Moreover, the nodes
wake up even when there is no data to transmit or
receive, which results in idle listening.
Asynchronous Schemes which use preamble sam-
pling. Each node chooses its active schedule inde-
pendently of other nodes and wakes up for short
durations to check if there is a transmission on the
channel. When a node wants to transmit a packet,
it must first transmit a preamble long enough to
make sure all the potential receivers detect it and
get the data frame. In those schemes, preamble
signaling consumes a lot of energy and nodes still
need to follow a duty cycle to avoid deafness.
Recent improvements in WURx hardware make them
a promising solution for WSN (Magno and Benini,
2014; Marinkovic and Popovici, 2011). The basic
idea of WURx is to allow a node to be waking up
from deep sleep by another node. When a node wants
to transmit a packet, it first transmits a wake up bea-
con. The receiver WURx hardware detects this bea-
con and wakes up the node. Moreover, many WURx
propose addressing capabilities, which allow a node
to insert in the wake up beacon the unique identifier
of the receiver. In this way, the WURx only wakes up
the node if it reads its own address in the wake up bea-
con. This feature avoids a node from waking up all its
neighbors when it wants to transmit data to only one
of them.
3 RESEARCH PROBLEM
Considering multihop WSNs, nodes have two func-
tions. First, as sensors, they must perform sensing
operations to generate new data and send them to the
sink. Second, they must be relays for other nodes
to permit them to forward their packets. We refer to
these tasks as the sensing task and the relay task re-
spectively. As packets aggregation can significantly
reduce energy consumption, the packets sending can
be uncoupled from these two tasks and seen as a third
one.
The available energy for each node must be effi-
ciently allocated between these tasks. Note that, in
the context of a node equipped with WURx, perform-
ing a relay task must follow the reception of a wake
up beacon, i.e. an interrupt from the WURx. We call
the reception of a wake up beacon a relay request. In
the context of EH-WSN, energy allocation is guided
by the following constraints:
Sustainability: each node must operate under ENO-
Max condition
Quality of Service: the network must fulfill some
Quality of Service (QoS) requirements that de-
pend on the application. A QoS metric can oper-
ate at node scale (e.g. minimum sensing rate) or at
network scale (e.g. minimum end-to-end delay).
Two classical examples illustrate how the QoS re-
quirements strongly depend on the application. In an
event detection application, a node only sends data
when it detects an event (e.g. fire detection). Most of
the time, a node does not send anything, but when an
event is detected, it must be transmitted to the sink in
the shortest delay possible. Therefore, an important
QoS metric is the end-to-end delay that should be as
small as possible. A good strategy is to send a packet
as soon as it is ready. Moreover, energy allocation
should favor the relay task in order for the packets to
reach the sink in the shortest delays.
In monitoring applications, QoS requirements
might be a minimum sensing rate for each node. If
the end-to-end delay is not an important constraint,
for example because the data are not going to be pro-
cessed immediately anyway, a good strategy for the
energy allocation policy is to favor the sensing task in
order to maximize the sensing rate.
These two examples illustrate how the design
of an efficient PM depends on the application and
the QoS requirements. We aim to design efficient
power management schemes and MAC protocols
to efficiently tackle the energy allocation problem.
Our work targets EH-WSN in which each node is
equipped with a WURx device capable of addressing
functionalities.
4 STATE OF THE ART
Research in EH-WSN is a hot area and much atten-
tion is drawn to several topics such as harvester tech-
nologies, storage devices, power management, MAC
and routing protocols and others. Similarly, many ef-
SENSORNETS2015-DoctoralConsortium
4

forts are devoted to WURx hardware. Yet, combin-
ing energy harvesting technologies and WURx is just
emerging and, to the best of our knowledge, there has
been only little research in that direction.
Magno et al. proposed a power device (Magno
et al., 2012) featuring multi-source harvesting capa-
bilities, multiple energy storage devices, fuel cells and
radio triggercapabilities. This device is also equipped
with a TI MSP430 microcontroller. The idea is to let
the device executing the energy management policy
programs. It is also responsible for waking up the
node when a wake up signal arises and offers energy
monitoring features to the node. Therefore, in con-
trast to our work, the aim of the authors is not to pro-
pose energy management policies, but a device to ex-
ecute them.
In (Magno et al., 2013), the authors introduced a
power device to be connected to an existing node. The
device they proposed is equipped with two harvesters,
one battery and one WURx. One of the harvester is
use to supply the battery and power the node, while
the other is a piezeoelectric MEMS harvester used to
supply the WURx in order to achieve “zero power”
WURx.
Le et al. proposed in (Le et al., 2013a) an adapta-
tion of the TICER MAC protocol for EH-WSN node
equipped with WURx. Each node uses a PM to dy-
namically adapt the duty cycle to achieve ENO-Max.
The authors show that by using a WURx, idle lis-
tening is drastically reduced and thus more energy is
available for increasing the QoS. The main limitations
of this work are that only one-hop network is con-
sidered and the lack of collisions avoidance scheme.
Also, the addressing capabilities of the WURx are not
used.
5 METHODOLOGY
In this section, we present the energy model we are
currently working on and two PMs exploiting this
model.
5.1 Energetic Model
We suppose that the time is divided into equal length
cycles of duration T. Energy conservation over any
cycle k leads to the following equation:
P
H
(k)T − P
L
T −
1
η
E
C
(k) = ∆E
S
(k) (1)
where, P
H
(k) is the average harvested power, P
L
is the
leakage power and is assumed to be constant. E
C
(k)
is the energy consumed by the node, and ∆E
S
(k) is the
Figure 1: Global software architecture of our system. The
relay requests predictor is only needed for the Predictive Al-
locator PM (see section 5.2.2), and is therefore dashed. The
PM sets the sensing task execution rate and allow/forbid the
execution of the relay task.
variation of stored energy. The factor η ∈ [0, 1] takes
into account a reduced energy efficiency due to power
conversion.
ENO-Max condition is achieved when consumed
energy equals harvested energy, i.e. ∆E
S
(k) = 0.
Therefore, ensuring ENO-Max operating state means
maintaining the stored energy at a constant value de-
noted E
ref
S
.
In practice, the PM is executed at the beginning of
every cycle. We denote the residual energy E
res
(k) =
E
S
(k) − E
ref
S
, where E
S
(k) is the amount of stored en-
ergy. Moreover, we have to predict the production
of the energy source over the cycle. We denote the
predicted average harvested power over the cycle k
c
P
H
(k). We can now rewrite (1) as follows:
c
P
H
(k)T + E
res
(k) − P
L
T −
1
η
E
C
(k) = 0 (2)
The PM has two levers for controlling the node
power consumption: setting the sensing task rate and
allowing/forbidding the execution of the relay task
when a relay request arises. If the relay task is not al-
lowed to be executed, then the relay request is simply
ignored. Figure 1 shows the global software architec-
ture of our system.
Let us now model the node activity. Two separate
cases are defined: the packets are forwarded as soon
as they are ready (immediate transmission), and the
packet are aggregated and sent later.
5.1.1 Immediate Transmission
In the first case, we assume that the packets are for-
warded as soon as they are ready. Then we have:
E
C
(k) = N
S
(k)(E
S
+ E
T
)+N
R
(k)(E
R
+ E
T
)
+ P
S
T
Sleep
(k)
(3)
where N
S
(k) and N
R
(k) are the number of sensing
tasks and relay tasks respectively executed over the
cycle. E
S
, E
R
and E
T
refer to constant energy costs for
performing a sensing task, a relay task and a packet
transmission respectively. We thus can rewrite (2) as
EnergyModelingofMultihopEnergyNeutralWSNUsingWake-upReceiver
5

follows:
c
P
H
(k)T+E
res
(k) − P
L
T −
1
η

N
S
(k)(E
S
+ E
T
)
+ N
R
(k)(E
R
+ E
T
) + P
S
T
Sleep
(k)

= 0
(4)
where P
S
is the node power consumption when it is
in sleep state and T
Sleep
(k) the time spent in sleep
state. As we assume low duty cycles, we approximate
T
Sleep
(k) by T for all k.
5.1.2 Aggregation
We assume in this section that we use packets aggre-
gation. We propose to postpone the transmission of
all the packets generated or received during the cycle
k − 1 to the next cycle k. To reduce collisions, we
send them at a random time. We denote E
TA
(k) given
in (5) the cost of sending all the packets which have
been ready to be sent during the cycle k− 1.
E
TA
(k) =

N
S
(k− 1) + N
R
(k− 1)

E
T
(5)
We thus have:
E
C
(k) = N
S
(k)E
S
+ N
R
(k)E
R
+ E
TA
(k) + P
S
T (6)
and we can rewrite (2) as follows:
c
P
H
(k)T+E
res
(k) − P
L
T −
1
η

N
S
(k)E
S
+ N
R
(k)E
R
+ E
TA
(k) + P
S
T

= 0
(7)
5.2 Possible Designs of Power Managers
We propose in this section two designs of PMs on
which we are currently working on. These PM share a
common global design and are divided into two com-
ponents. The first component is called Cyclic PM
(CPM) and is executed at the beginning of every cy-
cle. The second component, called WURx interrupt
handler (WIH), is executed after each WURx inter-
rupt. The aim of the latter component is to decide
whether a relay task should be performed or not in re-
sponse to a relay request. This design is illustrated in
Figure 2.
We assume that a QoS constraint is the maximum
duration, denoted T
max
S
, separating two sensing tasks.
In the rest of this section, only the equations for the
packets aggregationcase are shown, but the equations
describing the immediate transmission case can be
obtained similarly.
5.2.1 Reactive Allocator PM
The idea of this PM is to always try to maximize the
sensing rate. When a relay request arises, the WIH
Figure 2: Illustration of the behaviour of the PMs. The CPM
is executed at the beginning of the cycle. Then, after every
WURx interrupt, the WIH is executed and decides whether
the relay task is executed or not. On this illustration, the
relay task is executed after the first WURx interrupt, but
not after the second. Finally, the sensing tasks are executed
periodically according to the rate set by the PM.
allows a relay task to be executed only if it estimates
that it will not increase the sensing period over T
max
S
.
At the beginning of each cycle, the CPM computes
the greatest number of sensing tasks achievable over
the cycle by allocating all the available energy to this
task, i.e. N
R
(k) = 0. From (7) we have:
f
N
0
S
(k) =
1
E
S

η(
c
P
H
(k)T+E
res
(k) − P
L
T)
− P
S
T − E
TA
(k)

(8)
and the initial sensing rate is thus T
0
S
(k) =
T
f
N
0
S
(k)
.
After the i
th
relay request which arise at the time
t
i
, the WIH estimates the greatest number of sensing
tasks achievable until the end of the cycle if a relay
task is performed. If we denote N
i
S
(k) the number of
sensing tasks performedsince the beginning of the cy-
cle, then:
f
N
i
S
(k) =
1
E
S

η(
c
P
H
(k)T + E
res
(k) − P
L
T) − P
S
T
− E
TA
(k) − iE
R
− N
i
S
(k)E
S

(9)
The WIH can then compute the expected sensing rate
e
T
i
S
(k) =
T−t
i
f
N
i
S
(k)
. If
e
T
i
S
(k) ≤ T
max
S
, then a relay task is
performed and the new sensing rate is set to
e
T
i
S
(k).
Otherwise, the relay request is ignored and the sens-
ing rate stays unchanged.
This PM favors the sensing task over the relay
task. It aims to always maximize the sensing rate by
not pre-allocating energy for the relay task. If a relay
request arise, then the node will perform a relay task
only if that does not deteriorate the sensing rate below
the fixed limit.
5.2.2 Predictive Allocator PM
This PM allocates energy budgets for each task (sens-
ing and relay) by estimating the needs of the relay
task based on past observations. Unlike the previous
SENSORNETS2015-DoctoralConsortium
6

PM, the CPM sets the sensing rate at a fixed value for
the whole cycle. Moreover, the CPM also sets a max-
imum number of relay requests that can be satisfied
over the cycle, denoted N
max
R
(k). Then, the job of the
WIH is simply to allow execution of relay tasks while
less than N
max
R
(k) of them were performed since the
beginning of the cycle.
At the beginningof each cycle, the CPM estimates
the number of relay requests that will arise based on
past observations. This predicted value is denoted
c
N
R
(k). Then, it computes the corresponding number
of sensing tasks that can be performed:
f
N
S
(k) =
1
E
S

η(
c
P
H
(k)T + E
res
(k) − P
L
T) − P
S
T
− E
TA
(k) −
c
N
R
(k)E
R

(10)
To ensure sensing QoS constraint, the PM chooses
N
S
(k) = max
n
f
N
S
(k),
T
T
max
S
o
. Then, the sensing rate
for the cycle k is set to T
S
(k) =
T
N
S
(k)
and N
max
R
(k) is
set to:
N
max
R
(k) =
1
E
R

η(
c
P
H
(k)T + E
res
(k) − P
L
T) − P
S
T
− E
TA
(k) − N
S
(k)E
S

(11)
Unlike the previous PM, this one allocates energy
for the relay task at the beginning of every cycle. The
amount of energy allocated for the relay task is based
on prediction of the future amount of relay requests
that will arise. If the number of relay requests is over-
estimated, then too much energy will be allocate for
relaying at the expense of the sensing task. There-
fore, the sensing rate will be suboptimal. Conversely,
if the number of relay requests is underestimated, then
some of them will be ignored. In that case, the sens-
ing rate is larger than what it would be if the number
of relay requests was accurately predicted. But ignor-
ing relay requests can decrease the overall network
performance.
5.2.3 Network Scale Considerations
For each node u, we assume that the routing layer
provides a set of admissible forwarders F(u). Then,
when a node wants to forward a packet, it chooses one
of the forwarders and sends a wake-up beacon, i.e.
a relay request. If the forwarder refuses the request,
then it tries another one, until one of the forwarders
accepts it. We think that this will favor routing path
containing nodes with high energy harvesting rates.
Another important consideration is how to avoid
nodes which are favored by the routing algorithm to
be flooded by relay requests and thus to always run
with low sensing rates. One possibility is to allow
nodes to refuse every relay request after running a re-
lay task for a period of time. The duration of this pe-
riod will depend on the energetic status of the node.
We believethatthis scheme will favorthe use of nodes
with good energy harvesting rate because they will
have shorter “deaf” periods.
Another possibility is to enable the nodes to
exchange their energetic status with their one-hop
neighbors. Then, when a node needs to forward a
packet, it can choose randomly among the admissi-
ble forwarders using a statistical distribution that fa-
vor nodes with a good energetic status. If a node u
is aware of some energetic status metric S(v) (e.g.
N
max
R
(k)) for all v∈ F(u), then it can choose randomly
among its admissible forwarders with the probability
of choosing the node v being
S(v)
∑
i∈F(u)
S(i)
.
Moreover, we are currently working on the design
of a MAC protocol to efficiently implement the relay
request mechanism.
6 EXPECTED OUTCOME
In the current state of our work, we did not evalu-
ate our proposal. We plan to first experiment using
the OMNeT++ simulation framework (OMNeT++,
2014), and then on real hardware using the PowWow
platform (pow, 2014) and the WURx from (Magno
and Benini, 2014).
REFERENCES
(2014). PowWow: Power Optimized Hardware
and Software FrameWork for Wireless Motes.
http://powwow.gforge.inria.fr/. [Online; accessed
December-2014].
Huang, P., Xiao, L., Soltani, S., Mutka, M., and Xi, N.
(2013). The evolution of mac protocols in wireless
sensor networks: A survey. Communications Surveys
Tutorials, IEEE, 15(1):101–120.
Kansal, A., Hsu, J., Zahed, S., and Srivastava, M. B. (2007).
Power management in energy harvesting sensor net-
works. ACM Transactions on Embedded Computing
Systems (TECS).
Le, T. N., Magno, M., Pegatoquet, A., Berder, O., Sentieys,
O., and Popovici, E. (2013a). Ultra low power asyn-
chronous mac protocol using wake-up radio for en-
ergy neutral wsn. In Proceedings of the 1st Interna-
tional Workshop on Energy Neutral Sensing Systems,
ENSSys ’13, pages 10:1–10:6, New York, NY, USA.
ACM.
EnergyModelingofMultihopEnergyNeutralWSNUsingWake-upReceiver
7

Le, T. N., Pegatoquet, A., Sentieys, O., Berder, O., and
Belleudy, C. (2013b). Duty-cycle power manager
for thermal-powered wireless sensor networks. In
Personal Indoor and Mobile Radio Communications
(PIMRC), 2013 IEEE 24th International Symposium
on, pages 1645–1649.
Magno, M. and Benini, L. (2014). An ultra low power high
sensitivity wake-up radio receiver with addressing ca-
pability.
Magno, M., Jackson, N., Mathewson, A., Benini, L., and
Popovici, E. (2013). Combination of hybrid energy
harvesters with mems piezoelectric and nano-watt ra-
dio wake up to extend lifetime of system for wireless
sensor nodes. In Architecture of Computing Systems
(ARCS), Proceedings of 2013 26th International Con-
ference on, pages 1–6.
Magno, M., Marinkovic, S., Brunelli, D., Popovici, E.,
O’Flynn, B., and Benini, L. (2012). Smart power
unit with ultra low power radio trigger capabilities for
wireless sensor networks. In Proceedings of the Con-
ference on Design, Automation and Test in Europe,
DATE ’12, pages 75–80, San Jose, CA, USA. EDA
Consortium.
Marinkovic, S. and Popovici, E. (2011). Nano-power wire-
less wake-up receiver with serial peripheral interface.
Selected Areas in Communications, IEEE Journal on,
29(8):1641–1647.
OMNeT++ (2014). OMNeT++ Network Simulation Frame-
work. http://www.omnetpp.org/. [Online; accessed
November-2014].
Vigorito, C., Ganesan, D., and Barto, A. (2007). Adaptive
control of duty cycling in energy-harvesting wireless
sensor networks. In 4th Annual IEEE Communica-
tions Society Conference on Sensor, Mesh and Ad Hoc
Communications and Networks, SECON, pages 21–
30.
SENSORNETS2015-DoctoralConsortium
8