ArticlePDF Available

Adaptive GPS Duty Cycling and Radio Ranging for Energy-efficient Localization

Abstract and Figures

This paper addresses the tradeoff between energy consumption and localization performance in a mobile sensor network application. It focuses on fusing GPS loca- tion with more energy-efficient location sensors to bound position estimate uncertainty in order to prolong node lifetime. We consider an animal monitoring application and use empirical GPS and radio contact data from a large-scale deployment to model animal mobility, GPS and radio performance. These models are used to explore duty cycling strategies for maintaining position uncertainty within specified bounds. We then explore the benefits of using short-range radio contact logging alongside GPS as an energy-inexpensive means of lowering uncertainty while the GPS is off. Results show that GPS combined with radio-contact logging is effective in extending node lifetime while meeting application- specific positioning criteria.
Content may be subject to copyright.
Adaptive GPS Duty Cycling and Radio Ranging for
Energy-efficient Localization
Raja Jurdak
CSIRO ICT Centre
Pullenvale QLD 4069 Australia
rjurdak@ieee.org
Peter Corke
Built Environment and
Engineering
Queensland University of
Technology, Brisbane, Australia
peter.corke@qut.edu.au
Dhinesh Dharman
CSIRO ICT Centre
Pullenvale QLD 4069 Australia
dhinesh.dharman@gmail.com
Guillaume Salagnac
INSA
Lyon, France
guillaume.salagnac@insa-
lyon.fr
Abstract
This paper addresses the tradeoff between energy con-
sumption and localization performance in a mobile sensor
network application. The focus is on augmenting GPS loca-
tion with more energy-efficient location sensors to bound po-
sition estimate uncertainty in order to prolong node lifetime.
We use empirical GPS and radio contact data from a large-
scale animal tracking deployment to model node mobility,
GPS and radio performance. These models are used to ex-
plore duty cycling strategies for maintaining position uncer-
tainty within specified bounds. We then explore the benefits
of using short-range radio contact logging alongside GPS as
an energy-inexpensive means of lowering uncertainty while
the GPS is off, and we propose a versatile contact logging
strategy that relies on RSSI ranging and GPS lock back-offs
for reducing the node energy consumption relative to GPS
duty cycling. Results show that our strategy can cut the
node energy consumption by half while meeting application-
specific positioning criteria.
1 Introduction
This paper addresses the tradeoff between energy con-
sumption and localization performance for outdoor mobile
sensor network applications. Mobile network proposals of-
ten assume that GPS can provide absolute location informa-
tion [1], without considering the operational constraints and
high energy profile of this technology, which causes rapid
depletion of the node batteries.
c
2010 Association for Computing Machinery. ACM acknowledges that this contribu-
tion was authored or co-authored by an employee, contractor or affiliate of the national
government of Australia (CSIRO). As such, the Government retains a nonexclusive,
royalty-free right to publish or reproduce this article, or to allow others to do so, for
Government purposes only.
SenSys’10, November 3–5, 2010, Zurich, Switzerland.
Copyright c
2010 ACM 978-1-4503-0344-6/10/08 ...$10.00
The obvious solution is to duty cycle the GPS module to
prolong node lifetime. This comes at the cost of increased
position uncertainty. Whenever the GPS is powered down,
the uncertainty in the mobile node’s position increases with
time. Fortunately, most applications can tolerate a certain
amount of localization uncertainty. A key challenge is that,
even if the node is not moving, its uncertainty will continue
to increase until the GPS is again turned on.
This paper proposes the use of more efficient sensor
streams to infer node locations and to reduce position uncer-
tainty while the GPS is off. In particular, we focus on radio
ranging among neighbouring nodes. Nodes send radio mes-
sages to share their position and estimated uncertainty with
spatially proximal neighbours, which enables sharing of GPS
load among the nodes and an overall reduction in the use of
GPS at each node. The challenge in using multiple sensors
is to keep the position uncertainty within application-specific
bounds while determining which sensor to sample and when
to sample it.
To address this challenge, this paper proposes a strat-
egy for managing this position uncertainty/energy tradeoff
in mobile sensor networks through the use of multiple sensor
streams with different energy cost and accuracy characteris-
tics. Our initial analysis reveals that the simple combination
of GPS and contact logging reduces the power consumption
of the GPS module but incurs additional power consumption
for increased radio activity. We then focus on modeling the
evolution of position uncertainty with time and the energy
implications of different GPS duty cycling strategies and ra-
dio contact logging. Our models use empirical data from
long term animal tracking experiments [2] where the GPS
is always on, which provides ground truth position data, to
build models of node mobility and GPS performance.
This paper considers two classes of positioning applica-
tions: (1) applications that can tolerate a fixed uncertainty in
node positions, which are representative of location-based
mobile phone applications, vehicle tracking applications,
inria-00524296, version 1 - 7 Oct 2010
Author manuscript, published in "The 8th ACM Conference on Embedded Networked Sensor Systems (SenSys 2010) (2010)"
xk
xk+1
xk+1
uncertainty
AAU
Figure 1. The growing uncertainty (dashed circle) and
the maximum uncertainty bound (solid circle).
and animal tracking applications; and (2) applications where
the tolerance for position uncertainty varies based on given
landmarks, which are representative of virtual fencing [2]
and collaborative gaming applications. While we propose a
strategy for class 2 applications that exploits their variable
uncertainty to minimise energy consumption, the core fo-
cus of the paper remains around the class 1 applications that
have wider applicability. We devise an event-based contact
logging strategy for class 1 applications that triggers beacon
transmission whenever a node detects a significant change in
its position state, resulting in significant energy savings.
In sum, the contributions of this paper are three-fold:
Proposal and evaluation of new GPS duty cycling algo-
rithms that are verified with detailed simulations based
on a large empirical dataset. The algorithms use a prob-
abilistic model of the GPS lock time and empirical node
mobility models from a cattle tracking experiment, to
establish guidelines for how GPS duty cycling affects
energy efficiency and position accuracy in mobile ap-
plications.
Novel methods for combining radio ranging with GPS
position measurements. We initially explore the perfor-
mance of periodic radio beaconing with a fixed trans-
mit power versus a less constrained approach with
RSSI-based ranging and event-driven contact beacon-
ing. Simulation results confirm that the RSSI- and
event- based approaches can reduce energy consump-
tion by nearly half, with a small reduction in a position
accuracy.
Proposal and evaluation of an adaptive GPS duty cy-
cling algorithm for responding to extreme cases where
energy or accuracy have to be compromised.
2 Cooperative localization
The notation that we will use in this paper is illustrated
in Figure 1. At time kwe obtain a GPS measurement xk
and then turn off the GPS. At the next measurement xk+1,
the mobile node may be either within or outside the user-set
maximum uncertainty bound, indicated by the larger circle.
Whenever xk+1falls outside the larger circle, we denote this
as an error, since the system would have failed to deliver a
Figure 2. Cumulative distribution of observed animal
speed. 95% of measurement are below 1m/s, 100% are
below 3.5m/s
position estimate within the user-set uncertainty bound. A
primary performance metric is the ratio of errors to the total
number of GPS measurements (denoted by the error rate σ),
which is a reliability measure of our algorithm. It is impor-
tant to note that we are not attempting to estimate the exact
node position while the GPS is off. Our aim is rather to esti-
mate the bounded vicinity of the node, and to ensure that the
next GPS measurement occurs within a defined radius of the
previous measurement.
Our approach requires a means to estimate the uncertainty
bound (shown as a dashed circle) at any given time. Since we
do not know how fast nodes are moving or in what direction,
the simplest model is to assume a speed and grow the esti-
mated uncertainty linearly with time. When it approaches
the maximum allowable uncertainty bound, taking into ac-
count the GPS’s lock time, we turn on the GPS. If the node’s
speed exceeds our assumption, it could end up at the point
marked in red when the GPS obtains a fix.
2.1 User Policy
Our approach relies on user policy inputs to identify rele-
vant performance bounds. The operator of the system spec-
ifies the user policy, which can include: (1) the acceptable
error in position for mobile entities; (2) the desired battery
lifetime of the mobile devices; and (3) a preference for er-
ror or lifetime. The error comprises two components: (a) the
absolute acceptable uncertainty (AAU) which is the maxi-
mum uncertainty in Figure 1 within which we expect to be
when the GPS next turns on; and (b) the error tolerance (σ),
which is upper bound on the error rate σ. Together, these er-
ror metrics set constraints on the time that the system can use
low-end positioning sensors before the uncertainty demands
switching on the energy-hungry GPS sensor. The desired
battery lifetime sets a target for average power consumption
for the mobile device, so that each device can gauge whether
it is on track to meeting its target lifetime.
2.2 Node Mobility
We consider three attributes for modelling node mobility:
(1) the distribution of past speed data of the mobile nodes,
as shown in Figure 2 for cows; (2) the current speed; and (3)
the clustering behaviour of the mobile entities.
The distribution of past speed data is useful for estimating
the assumed speed of the mobile device, as well as the like-
inria-00524296, version 1 - 7 Oct 2010
lihood of variation from that average speed. However, the
speed at successive samples is not an independent random
variable and is strongly persistent, so we additionally use the
current speed for future speed predictions.
Clustering is another important aspect of the node mo-
bility. The clustering behavior builds on the time series of
separation distances between every pair of mobile entities.
Separation distance information can be obtained using any
combination of received signal strength indicator (RSSI) or
the difference in exchanged GPS coordinates. The cluster-
ing behavior can be described by statistics on the separation
distance between the entities and also the duration for which
the entities are in contact. These statistics enable determin-
ing expected node density and expected contact durations
among mobile nodes, which in turn guides the decision on
setting GPS duty cycle and radio beaconing metrics.
2.3 Energy Model
Our approach tracks node energy consumption by record-
ing the duty cycles of the major node components [3]: GPS
module, radio, MCU, and other components.
GPS duty cycling is a key factor in our approach. We
simply compute the GPS module energy consumption at any
time tduring the deployment as PON
gps ×DCgps ×t, where
DCgps is the GPS duty cycle, and PON
gps is the power consump-
tion of the GPS module in active mode. The GPS module is
completely powered off when not in use for a current draw
of zero. We track the radio energy consumption as the sum
of the transmission, reception, and listening energy values
respectively [3].
The MCU power consumption is heavily dependent on
the GPS and beaconing strategies, as the MCU must power
on when either the GPS module or radio are active. In
general, the MCU duty cycle and energy consumed are ex-
pressed as:
DCMCU =DCg ps +DCradio overla p (1)
EMCU =DCMCU ×PMCU ×t
where DCradio is the duty cycle of the radio, and overl a p de-
notes any overlap between the on-times for the GPS and ra-
dio. Sections 3 and 4 elaborate on MCU energy consumption
further for specific duty cycling strategies. The remaining
node components, including the regulators and audio board
for emitting audio cues to the cattle in the virtual fencing
application, have a constant energy.
2.4 Performance Tradeoffs
The user policy in our approach targets both position ac-
curacy and energy efficiency, which are dependent variables
that are often in tension: setting a low error tolerance re-
duces the achievable lifetime of mobile devices for a given
battery, because it limits opportunities for GPS duty cycling.
In the simplest terms, the system configuration would ideally
(1) maximize lifetime; and (2) minimize error rate, within
the user set AAU value. Obviously, achieving both goals si-
multaneously is not possible; however, our approach aims at
identifying the operating point that ensures that both metrics
are performing within user-specific bounds. Whenever either
of these conditions are violated, the system is not delivering
on one of the user-set policies. In section 4.3, we propose
Figure 3. A cow with our smart collar
a best-effort strategy for delivering on energy and accuracy
within user-set bounds. However, if a situation arises that
fails to meet targets for both metrics, then the system has
reached a critical point. The third user policy metric from
section 2.1 specifies the user preference for either favouring
energy efficiency or position accuracy at such critical points.
While in general the preference can be specified as weights
for favouring constraints (1) and (2), the rest of this paper
uses a boolean value that discretely favours either of the con-
straints at critical points in the deployment.
It is important to distinguish between the measured error
rate, which is visible to nodes during deployment, and the ac-
tual error rate, which can only be tracked when ground truth
data is available, such as for offline analysis. In Sections 3
and 4, we track the actual error rate as we conduct offline
analysis of empirical data for more insight into system de-
sign. Section 4.3 uses measured error rate in developing an
online strategy for managing performance.
3 GPS duty cycling
This section applies our framework to GPS duty cycling
independently of any other sensor stream. The aim is to ex-
plore different GPS duty cycling algorithms and their effect
on error rate and energy efficiency.
3.1 Motivating Application
To ground the discussion, we consider an outdoor location
monitoring application for tracking cattle using smart col-
lars, as shown in Figure 3 that contain wireless sensor nodes
and GPS modules. The goal is to track cow positions, and
in some cases, enforce exclusion zones within the paddock,
in effect a virtual fence [4, 2]. Virtual fencing is useful for
managing cattle in vast grazing lands, such as in Australia,
where farms can reach the sizes of small countries, and it
is not economically feasible to install physical fences in the
whole area. The reader should note, however, that our ap-
proach is independent of the virtual fencing aspect and gen-
erally applicable to all mobile node tracking scenarios.
The target node lifetime in this application is 3 months,
which is the interval at which the animals are brought in for
health checks, treatment and sorting. Achieving this lifetime
is a challenge because the GPS module on each node has a
large current draw 1. Our cattle monitoring application and
1Smart phones that include GPS modules involve similar con-
siderations, albeit with more regular opportunities for recharging
inria-00524296, version 1 - 7 Oct 2010
Component Power (mW)
MCU 18
Radio (TX) 79.2
Radio (RX) 29.7
GPS 165
Switch-mode Regulator 6
Linear Regulator 3.3
Audio 3
Table 1. Active mode power consumption for collar com-
ponents
cow collar setup are fully described in Wark et al. [2]. We
limit the discussion here to a high level description of the
collar setup and the energy consumption of each of its com-
ponents.
The collars incorporate a Fleck node, a GPS module,
and an audio board for generating audio cues to indicate
to animals that they are crossing a virtual fenceline. The
Fleck node itself comprises an Atmel-1281 MCU and Atmel
RF212 radio. Four D-Cell batteries in series provide power
to all the collar components via several switch-mode regula-
tors. The weight of these batteries yields a collar weight that
approaches the upper limit set by an animal ethics committee
in Australia, so larger batteries are not an option.
The power consumption of these components is shown in
Table 1. The sleep mode power consumption for the radio,
MCU, GPS, and audio is several orders of magnitude lower
than the active mode current. The power consumption of
the GPS board is the largest by far. To put things in per-
spective, Figure 4(a) illustrates a scenario with the GPS con-
stantly active. Using our collars with a 10% radio duty cycle,
based on a 6ms check interval and 60ms sleep interval, and
an always-on GPS board, the GPS accounts for 88% of the
power consumption of 209 mW and limits the nodes’ life-
time to 19 days. Figure 4(b) sets the GPS duty cycle at 5%,
while maintaining the same setup for all of the collar com-
ponents. In this scenario, the GPS accounts for about a third
of the overall node power consumption of 25mW, which ex-
tends the nodes’ lifetime by a factor of 7.5. This confirms
the importance of duty cycling the GPS on the mobile nodes.
In both scenarios, the MCU remains on only as required to
service to the GPS or radio, which couples its duty cycle to
these components’ duty cycles.
3.2 GPS lock time
The GPS module does not output a valid position estimate
immediately after being turned on. It enters a state where the
internal filter is initializing. This is referred to as acquir-
ing lock and indicated by a flag in the output data packet.
Once it has lock, the GPS reports horizontal dilution of pre-
cision (HDOP), which is an estimate of error in the horizon-
tal plane. This value tends to start high, perhaps several tens
of metres, and falls over consecutive samples to a few me-
tres. The whole process of obtaining lock and an estimate
with acceptable HDOP is referred to as obtaining a GPS po-
sition lock. We conducted an experiment over three days in
which the GPS was turned off for a random amount of time,
batteries.
(a) Always on GPS (b) GPS with duty cycle of
5%
Figure 4. Impact of GPS duty cycling: Reducing GPS
duty cycle from 100% to 5% reduces the node power con-
sumption by 7.5 times. The radio has a 10% duty cycle
in performing low power listening for both cases.
Figure 5. Observed GPS lock time as a function of off
time. The vertical scale has been limited to allow the lin-
ear characteristic to be visible.
tO. We trackied the lock time tL, that time that it takes the
GPS chip to achieve HDOP less than 10m. The data is shown
as a scatter plot in Figure 5 and we observe a linear cluster
of points that is represented by
tL=0.0245tO+0.0051
with a superimposed Poisson random process. In some
cases, it took more than 2 minutes to obtain a lock.
3.3 Design Considerations
While GPS provides the most reliable absolute location
estimate outdoors, its high power consumption demands
duty cycling in order to achieve the target lifetime. When-
ever the GPS is active, the localization uncertainty is equal
to the GPS error Ugps, which depends on the number of satel-
lites currently visible.
Whenever the GPS is powered off at time tk, and until it
next acquires lock, the position uncertainty grows progres-
sively according to
U(t)=s×[ttk]
where sis the assumed speed. Since we require U<AAU
inria-00524296, version 1 - 7 Oct 2010
and the GPS has a finite lock time we need to start the lock
process when t>Tmax where
Tmax =AAU Ugps(tk)
stL(2)
is the GPS off time. The value of scan be specific to each
mobile node, and can be a constant, or vary as a function of
time and space. The lock time, tL, has a weak dependence
on the length of time that the GPS has been off, which we
can control. It also depends on factors beyond our control,
such as the time varying satellite configuration and the de-
ployment environment. For example, some satellites will be
obscured in a deep valley or under dense foliage. In practice,
a “best estimate” of tLis used based on data such as shown
in Figure 5 or updated in real-time on the node based on the
measured time taken to acquire lock. Nevertheless, long lock
times will occur with low probability, which means that the
estimated uncertainty will occasionally exceed AAU.
loop {t}
choose speed()
if GPS == LOCK then
position GPS position
UUgps
lastlocktime = now()
GPS = OFF
else
UU+ (t * Sc)
if GPS == OFF and U+(tL*s)>AAU then
GPS = ON
end if
end if
end loop
Figure 6. GPS duty cycling strategy
Figure 6 illustrates the algorithm for duty cycling the GPS
module. The algorithm uses the choose speed() function to
select the speed model. At each time step the algorithm
checks if the GPS has lock and if so it obtains the position,
sets the uncertainty to Ugps, then powers off the GPS. Other-
wise, it grows the uncertainty according to the current speed
Sc. Once the uncertainty approaches AAU, it turns on the
GPS and the process repeats.
3.4 Bounding the uncertainty
Most positioning applications require a fixed uncertainty
bound. For a specific class of applications, such as virtual
fencing, the AAU can dynamically vary during the deploy-
ment according to application state. For instance, in the vir-
tual fence application, the AAU can vary as a function of the
node’s position and is the distance of the node from the vir-
tual fence line. If the node is 1000 meters away from the
fence, we simply require that the next time the GPS makes a
measurement, it will not have crossed the fence line. If we
assume the node moves at 1m/s (which is true 95% of the
time), then a node can keep its GPS off for more than 985
seconds, with a conservative estimate of 15 seconds for GPS
lock time. A node that was 100m away from the fence at
the last lock can keep its GPS powered off for 85 seconds.
(a) Error behaviour for a static AAU
(b) Error behaviour for a dynamic AAU
Figure 7. Effect of AAU on error behaviour
Comparatively, a fixed AAU of 50m would result in a GPS
off time of only 35 seconds. The sizeable increase in GPS off
times relative to a fixed AAU can greatly reduces the average
power consumption for dynamic AAU.
Figure 7(a) illustrates the effect of AAU on the growth
of the estimated uncertainty. The actual displacement in this
plot indicates the ground truth displacement of the mobile
node since the last GPS lock. During deployment, the actual
displacement is not visible to the nodes, as errors are only
discovered when the node acquires the next GPS lock. The
purpose in tracking the actual displacement here is to deter-
mine how well our uncertainty estimates model the nodes’
mobility.
We choose an assumed speed sof 0.4 m/sec for the nodes,
which is the average node speed in the cow dataset, and the
estimated uncertainty grows linearly while the GPS module
is powered off. When it approaches AAU, the node activates
the GPS and sets its uncertainty to Ugps when a fix is ob-
tained, resulting in periodic duty cycling of the GPS. A total
of 9 GPS locks occurred in this time window.
In Figure 7(b) the AAU is varied according to the distance
between the last observed position of the mobile node and
the virtual fence line. While the user sets an initial AAU
of 50m, the AAU rises to about 90m at the first GPS lock,
reflecting how far the cow is from the virtual fence. The
AAU increases to further to 125m and 150m at the next two
GPS locks. The higher AAU allows the node to keep its
GPS module powered off for longer reducing the number of
GPS lock attempts to 4, with the actual displacement never
exceeding the uncertainty.
3.4.1 Speed models
The assumed speed is a critical parameter. If it is too low,
then the AAU bound will be violated if the animal moves
quickly. If it is too high, the GPS off times will be very short
which is costly in terms of power. Choosing the right speed
inria-00524296, version 1 - 7 Oct 2010
Static Dynamic Probabilistic
Sc=sif (s(t)>s) i=t-lastlocktime
Sc=s(t) if (i==0)
else Sc=s(t)
Sc=sif (Sc>s)
P=t22
else
P=t11
else
Sc=P×Sc+s(1P)
(a) (b) (c)
Figure 8. The choose speed() function for each speed
model
requires a speed model, which is highly dependent on the
application. The data in Figure 2 indicates an average cow
speed of only 0.4 m/sec, and 95% of the time the cows move
at 1m/sec or less, with an absolute maximum of 3.5m/sec.
We explore three models for the assumed speed s, which
are summarised in Figure 8:
1. Static: based on a constant assumed speed
2. Dynamic: based on setting the assumed speed as the
last observed speed of the mobile node, read from the
GPS speed value from the GPS chip at last lock
3. Probabilistic: based on setting the assumed speed on the
basis of the last observed speed and a state model of the
mobile node
The static and dynamic speed models are application-
independent. The static speed model assumes a low vari-
ance from an average speed and uses this estimate for the
entire deployment. The dynamic speed model assumes a
high correlation between the most recent speed measure-
ment and the current speed estimate, i.e. persistence. The
probabilistic speed model also relies on persistence, but as-
signs application-specific probabilities for decaying speed
estimates. For instance, previous work has modeled animal
mobility as a 3-state Markov process [5]. Unlike some other
mobility models in the literature for mobile phone or car
applications [6], the animals do not follow specific routes,
rather they spend a lot of time in random foraging behaviour.
The probabilistic model in this paper is a simplification of
the Markov model of cow speed in [5] where the animal has
a slow- and a fast-moving state, states 1 and 2 respectively.
State transitions from state ito state jare referred to as tij.
Using data from a 2-day experiment we determine the prob-
abilities that at each time step the animal will transition out
of that state. The assumed speed is set to the last observed
speed, as for the dynamic model, but at each subsequent time
step it moves toward a constant assumed speed using a first-
order filter. The time constant is a function of the transition
probability of the initial state. The reasoning behind this is
that as the time since the last speed observation grows, its
significance decreases until at some point, and without any
other data we revert to a constant assumed speed.
Figure 9 illustrates the effect of each speed model on un-
certainty estimation for a constant AAU. All three models
use the same sof 0.4 m/sec. The uncertainty in the constant
(a) Error behaviour for the static speed model
(b) Error behaviour for the dynamic speed model
(c) Error behaviour for the probabilistic speed model
Figure 9. Effect of speed model on errror behaviour
speed model (Figure 9(a)) grows regularly according to s,
resulting in 6 GPS locks and an error rate of 7.5%. For the
dynamic speed model in Figure 9(b), the uncertainty growth
rate follows the changes in s. For instance, at 680 seconds,
the node acquires lock and observes that the current speed
is higher than s, and as a result, it increase sto better model
the actual motion. At the following GPS lock, the same node
observes a drop in the measured speed, and reverts to sagain
for its speed estimate. The dynamic model improves the er-
ror rate to 3% while increasing the GPS locks to 7. Finally,
the probabilistic model results in 7 lock attempts with an er-
ror rate of 3.8%.
3.5 Performance Evaluation
To evaluate the impact of our GPS duty cycling strategy
on position accuracy and energy efficiency, we have imple-
mented the duty cycling strategy in Figure 6 in a Python-
based simulator. We use a dataset which contains GPS
data from 30 cow collar nodes collected continuously (once
per second) over 2 days (100% duty cycle) from a herd at
the Belmont Research Station in central Queensland. This
dataset of cow positions represents the ground truth for all
the analysis and results in this paper. Each GPS module
is represented is represented by a Collar class, with a sim-
ple state machine that includes the probabilistic lock time
model of Figure 5 and the preceding off time. When the
GPS is turned off, the simulator evolves the uncertainty es-
timate according to the speed model. The simulator keeps
inria-00524296, version 1 - 7 Oct 2010
the GPS module off for Tmax seconds then turns it on to start
the locking process. When the simulated GPS obtains lock,
the true position is used to determine if the AAU constraint
has been violated and to update the error rate σ. The sim-
ulator also updates the average power consumption for the
GPS, radio, MCU and regulators based on the energy model
in Section 2.3, which enables the computation of the node’s
overall power consumption.
We initially fix the AAU to 50m and explore the impact
of changing the assumed speed for each of the three speed
models. The results are shown in Figure 10. As expected
the GPS power consumption dominates the node power con-
sumption in all case. The GPS power consumption increases
with assumed speed, since the faster growth in estimated un-
certainty triggers more frequent GPS fixes. Because the GPS
and radio timers fire completely independently, the overlap
parameter for computing the MCU duty cycle is zero. The
MCU power consumption increases with increasing GPS ac-
tivity, as the MCU has to remain in active state for interacting
with the GPS module. The radio power consumption, based
on a low power listening duty cycle of 5% with 6ms check in-
terval and 128ms sleep interval, is a small contributor to the
overall node power consumption, since it only sends short
data packets every minute back to the base station.
The GPS power consumption of the probabilistic speed
model has the lowest dependence on the assumed speed, as
it uses the most recent observed speed estimate for most of
the time. For low assumed speeds, the dynamic speed model
has higher power consumption than the static speed model,
because it occasionally uses a much higher assumed speed
when it happens that the speed observed over the previous in-
terval is high. As the assumed speed increases, both the static
and dynamic speed models exhibit similar behaviour, as
there are fewer instances where the dynamic model switches
to current speed values. The dynamic speed model performs
best in terms of accuracy, with its error rate σdecreasing
steadily with increasing assumed speed, from about 5% at
an assumed speed of 0.2m/sec to less than 1% for assumed
speeds above 1.2 m/s. The probabilistic and static speed
models have similar power consumption at low speeds, with
a slight advantage in error rate for the probabilistic model.
At higher speeds, the static model achieves lower error rate at
the cost of higher power consumption, since it always main-
tains a conservative assumed speed, while the probabilistic
model can often use a lower assumed speed.
Figure 11 shows the effect of enabling dynamic AAU in
a virtual fencing application according to distance from the
fence. The variance of results here, indicated by error bars,
is noticeably higher than for static AAU, because of the dis-
parity among nodes’ distance to the fence during the deploy-
ment. The dynamic AAU reduces the GPS power consump-
tion by more than half for low assumed speeds, and more
than 70% for higher assumed speeds. For the lowest assumed
speed of 0.2 m/s, the GPS power consumption is reduced to
about the same level as the MCU power consumption and the
regulators. The overall node power consumption increases
more linearly with assumed speed for dynamic AAU. At the
highest assumed speed of 1.4 m/s, the overall node power
consumption is about 40% of the static AAU case. The re-
Figure 10. Power consumption and error rate for the 3
speed models (fixed AAU). S: static, D: dynamic, and P:
probabilistic.
duced power consumption of dynamic AAU comes at the
cost of a higher error rate. This stems from the longer off
times for the GPS modules, which increases the expected
lock times and the instances where the GPS takes a very long
time to obtain a fix. While dynamic AAU can yield signifi-
cant energy savings at the cost of slightly higher error rates,
dynamically changing AAU according to the distance from
the fence is a specific feature of the virtual fencing appli-
cation and similar tracking applications with known bound-
aries. The remainder of the paper focuses on static AAU,
which is more representative of a wider class of GPS-enabled
mobile applications.
Figure 11. Power consumption and error rate for the 3
speed models (dynamic AAU). S: static, D: dynamic, and
P: probabilistic.
In summary, GPS duty cycling clearly reduces the node
inria-00524296, version 1 - 7 Oct 2010
energy consumption at the cost of positioning error rates
ranging from 0% to 15%. The dynamic speed model pro-
vides the best balance between error rate and power con-
sumption, which is why we adopt this model for the remain-
der of the paper. Although the dynamic AAU yields sizeable
benefits for the virtual fencing application, the remainder of
this paper adopts the static AAU model that is more repre-
sentative of general mobile sensor network applications. The
next sections investigate how the use of contact logging can
further improve the energy/accuracy tradeoff.
4 Coupling GPS and contact logging
Mobile nodes can use non-GPS relative localization sig-
nals to reduce position uncertainty while the GPS module is
powered off. In many positioning applications, several mo-
bile nodes may cluster together making short range RF con-
tact logging attractive. Consider that node A has powered off
its GPS and is growing its estimated uncertainty as a func-
tion of time. Receiving a beacon from node B only implies
that the two nodes are within a certain distance ˆr, which can
be conservatively estimated as the contact radius R, since the
distance between the nodes must be ˆrR.
The beacon from B also includes its last measured posi-
tion xB
kand its current uncertainty estimate UB. If UB+ˆr<
UAthen node B is nearby and has a lower uncertainty than
node A. In this case node A lowers its uncertainty estimate
UA:=UB+ˆr(3)
Lowering the estimated uncertainty enables node A to keep
its GPS off for longer, which reduces its energy consump-
tion. If all nodes run this algorithm, the energy-expensive
GPS position lock is shared across the nodes. The fairness
of the algorithm stems from the fact that when node A re-
lies on node B for its position estimate, UA>UB. If the two
nodes use the same assumed speed to grow their uncertainty,
then node A will decide to turn on its GPS before node B,
which allows B to rely on A for its position estimate in the
next cycle. Using a variable speed model can result in excep-
tions to this trend, but contact logging will result in long-term
sharing of GPS load if the assumed speed follows a random
distribution.
If a node’s estimated uncertainty approaches AAU, it
turns the GPS on. If the GPS is acquiring lock when UA
is updated and if UA<AAU s×tlock then node A should
turn off its GPS , even without acquiring lock.
Figure 13 illustrates the impact of contact logging on GPS
duty cycling for two nodes that are sending contact bea-
cons every second. Without contact logging (Figure 13(a)),
each node independently tracks its uncertainty estimate and
acquires a GPS position fix whenever its uncertainty ap-
proaches AAU, resulting in 7 fixes for node 1 and 6 fixes
for node 2. Using contact logging, the nodes can reduce
their GPS fixes to 5 and 4 locks in the same time window.
Consider the plot at 480 seconds. The uncertainty for node
2 approaches AAU, so it powers on its GPS and obtains a
position fix. Node 1, which is in the vicinity of node 2, can
rely on the latter’s recent GPS fix to reduce its uncertainty to
the 10m contact radius. This example illustrates the fairness
feature in our contact logging strategy, since nodes that re-
cently acquire lock will have a smaller uncertainty estimate
than neighbours, forcing another node to turn on its GPS in
the next round.
Our contact logging strategy depends on the settings of
two variables: (1) the contact beacon period; and (2) the con-
tact radius. The beacon period is the time between beacons
at each node. It involves a tradeoff: short contact periods
increase the likelihood of useful contact but incur a higher
energy cost.
The contact radius specifies the value of R, which must
balance the need to cover larger areas with a higher Rand
to reduce uncertainty with a lower R. Section 4.1 analyti-
cally explores this balance for the choice of R, followed by
simulations based on empirical data to determine the operat-
ing points for beacon period and contact radius that balance
these tradeoffs. Section 4.2 then investigates the impact of
beacon period on the energy/accuracy tradeoff.
4.1 Contact Radius
Figure 12. Relative distances between pairs of cows .
4.1.1 Analysis
Instead of conservatively estimated ˆr=R, we could
choose ˆr<R/2, since R/2 is the expected value of the inter-
node distances if they are uniformly distributed. The choice
of ˆrcan also rely on: offline statistics on inter-node distances
from the 2-day experiment, shown in Figure 12; or RSSI val-
ues, as discussed in Section 4.1.3. For the moment, we focus
on determining the features and dependencies of favourable
contact radii.
In contact logging, the relative localization uncertainty
between any two nodes that are in contact is ˆr, and it de-
pends on the radio propagation factors and on the applica-
tion requirements. It also relates to the value of AAU. Intu-
itively, ˆrhas to be less than AAU by at least s×tlock. In
other words, detecting that two mobile entities are in contact
within a large distance is not useful if ˆr>AAU (s×tlock)
is small.
Let P(ˆr)be the probability that the distance between two
nodes is ˆr. From a single mobile node’s perspective in
an N-node network, the probability that at least one other
node is closer than ˆris P(contact)=1(1P(ˆr))(N1).
The probability that a neighboring node has a lower uncer-
tainty is a function of the local node’s uncertainty: when
the local node’s uncertainty approaches AAU, it is highly
likely that a neighboring node’s uncertainty is lower, since
inria-00524296, version 1 - 7 Oct 2010
(a) Error behaviour for GPS duty cycling
(b) Error behaviour for GPS duty cycling coupled with contact log-
ging
Figure 13. Impact of contact logging on GPS duty cycles
the neighboring node runs the same GPS duty cycling al-
gorithm and the expected value of U=AAU/2. There-
fore, the probability that a neighboring node has a lower
uncertainty P(lower)than the local node is simply equal to
U(t)/AAU . The likelihood that a neighbour with a lower
uncertainty estimate will come into a node’s radio range is
simply Puse =P(cont act)P(lower). The selection of the best
contact radius should maximize Puse :
Puse =[1(1P(ˆr))(N1)]×U(t)
AAU
where U(t)=ˆr+(ttcont act )×s
s.t.U(t)<AAU
(4)
and where tcontact is the time of contact with a neighbor. The
product of (ttcontact )×squantifies the growth of local un-
certainty since the initial contact and indicates an additional
dependence of Puse on s.Puse is a linear function of time rep-
resented by segment starting at tcontact and ending at AAUˆr
s.
The segment length is shorter for larger ˆr, but Pcontact is also
higher for larger ˆr, which suggests an optimal point exists.
The best value for ˆrshould maximize the area below the
trapezoid bounded by the Puse and the time axis:
ZAAUˆr
s
0
Puse dt (5)
We analyze the data from the field experiment in sec-
tion 3.5 to first determine the probability density function of
the distances between each pair of cows in the 30 cow data
set. Figure 12 summarises the results. The inter-node dis-
tance is concentrated in the range 0-20m and suggests that
setting the contact radius at or below 20m can be highly ben-
eficial for cooperative localization, as P(ˆr<20)is nearly
73%. Using the virtual fencing dataset, we consider contact
radius values at 5m, 10m, 20m and 30m with respective P(ˆr)
of 0.26, 0.52, 0.73 and 0.81 and a speed of 0.4m/sec. Fig-
ure 14 compares the area under the Puse curve using equa-
tion 5 as Ngrows from 1 to 30 nodes. This shows a high de-
pendence of optimal ˆron the number of nodes in the deploy-
ment, particularly that the optimal ˆrshrinks with increasing
number of nodes. The expected ˆrthat maximizes Puse for
our 30 cow deployment is 5m.
Figure 14. Effect of number of nodes and contact radius
on Puse
4.1.2 Contact Radius Selection
We now explore the effect of the contact radius setting on
performance. The main mechanism for setting the contact
radius is to change radio receiver sensitivity and/or change
radio transmit power. Given that these setting will not pro-
vide a fixed or uniform signal transmission, the value of R
conservatively reflects the maximum contact radius.
To accurately model the tradeoffs of contact logging, we
extended the Python simulator with a Message class. Each
instance of the Message class represents a contact beacon
which enables communication between instances of the Col-
lar class. A transmitted contact beacon is registered as re-
ceived if the distance between the sender and receiver is less
than the contact radius. The time slot for sending contact
beacons and the order in which nodes send and receive con-
tact beacons are randomized to reflect reality. Each beacon
contains the following information: node ID, node position,
and node uncertainty.
Figure 15 summarizes the impact of contact radius on
node energy consumption and error rate for the entire virtual
fencing dataset. The contact beacon period is set to 1 sec-
ond. The overlap parameter here for the DCMCU is equal to
DCgps, as the frequent radio beacon transmissions and chan-
nel checking imply that the radio nearly always on when the
GPS module is on. The optimal contact radius is 5m, which
is in line with the analysis in Figure 14. Although contact
logging reduced GPS duty cycle by up to 25%, the over-
all node power consumption is only reduced by about 15%,
as contact logging increases the radio activity and associated
MCU activity. In addition, contact logging incurs an increase
in error rates over plain GPS duty cycling as nodes rely more
on neighbours’ estimates and exceed AAU more often. The
radio power consumption notably increases for larger con-
tact radii, which increases the range of node transmissions
and causes more overhearing. The conclusion from these re-
sults is that simple contact logging with a fixed radius and a
inria-00524296, version 1 - 7 Oct 2010
1 second beacon period does not yield worthwhile benefits
over simple GPS duty cycling.
Figure 15. Effect of contact radius (static AAU)
4.1.3 RSSI-based Ranging
As mentioned above, estimating the distance ˆrbetween
two nodes as the radio range Ris conservative. The use of
received signal strength indicator (RSSI) can relax this esti-
mate. While RSSI is a poor estimator of absolute range in
mobile or non-line-of-sight environments due to multi-path
signal propagation and interference, it still highly correlates
with range between two nodes [7]. Here, we use RSSI to
bound the range between two nodes as follows. Line-of-sight
RSSI measurements with varying distance are collected prior
to deployment in an open field. Let the line-of-sight RSSI
value of two nodes that are ˆrmeters apart be X. If during
a deployment, a node measures an RSSI of Xfrom a neigh-
bour, then the neighbour is at most ˆrmeters away.
Figure 16 shows 2400 line-of-sight RSSI measurements
we collected for the Atmel RF212 radio that is used in the
collar nodes. The measurements were collected by fixing a
mobile node in a field near our office and moving another
node within a 70m radius around the fixed node. Moving the
transmitter in different directions around the receiver was to
reduce any biases arising from the antenna radiation pattern
not being fully omni-directional. Both nodes remained at
an elevation of 1m from the ground. Each point in the fig-
ure corresponds to a single RSSI measurement. The solid
line represents the maximum distance at which each discrete
RSSI value was recorded by the RF212 radio in our measure-
ments. For the purpose of radio ranging, we can use this line
as a less conservative estimate of ˆrfor any RSSI X.
The last bar plot in Figure 15 shows the power consump-
tion and error rate for RSSI-based contact logging. It clearly
achieves the lowest power consumption relative to all con-
tact radii and GPS duty cycling. In terms of error rate, RSSI-
based contact logging causes an increase of 3% in error rate
over GPS duty cycle yet remains well within the 5% target
set for our application.
4.2 Beaconing Period
The baseline contact logging protocol is periodic, where
each node has an internal timer for triggering transmission
Figure 16. Line-of-Sight RSSI measurements for RF212
Radio. The bounding line represents the maximum pos-
sible line-of-sight distance for each observed RSSI value.
Figure 17. Periodic and event-based beacons
of contact beacons. Whenever neighbours hear contact bea-
cons, they determine whether to rely on their neighbour’s
position estimate if that reduces their uncertainty. The con-
tact period itself is dependent on many independent fac-
tors, including the clustering patterns of nodes, their rela-
tive speeds, terrain, and AAU. This suggests the need for
adapting the contact beacon period to these factors, which is
a non-trivial optimisation problem with highly dynamic and
unpredictable inputs.
Before attempting to solve this problem, we first ex-
plore the impact of the contact beacon period on the accu-
racy/energy tradeoff in the first part of this section. Next, we
propose an event-driven contact beacon transmission, based
on local position state changes.
4.2.1 Periodic Beaconing
To investigate the effect of beacon period on the index,
Figure 17 shows the power consumption and error rate for
contact beacon periods of 1, 5, 10, 20, and 50 seconds. In-
creasing the contact beacon period from 1 to 5 seconds re-
duces both power consumption and error rate. While a longer
beacon period causes GPS power consumption to increase, it
also shrinks the radio power consumption. For beacon peri-
ods of 1 second, we observed that the GPS power consump-
tion remains relatively high. The reason is that, on many
occasions, a node powers its GPS ON to get lock, only to
inria-00524296, version 1 - 7 Oct 2010
turn it OFF, before getting lock, due to reception of useful
message from neighbour. This continuous switching of the
GPS module prevents it from lower its consumption further
with short beacon periods.
4.2.2 Event-based Beaconing
An additional effect with periodic beaconing is the im-
plicit synchronization of several neighbours of a node that
obtains lock. When the node obtains lock, its beacon in-
form its neighbours of this event, and all neighbours set their
uncertainties accordingly. If the neighbours happen to be
within the same ˆrfrom the node, and with similar assumed
speeds s, then they are likely to turn on their GPS modules at
nearly the same time.
To counter this effect and to reduce radio power consump-
tion further, we propose an alternative to periodic contact
beaconing in the form of event-based beacons. In this ap-
proach, a node send its contact beacon whenever it locally
detects one of the following position state changes: (1) its
position is updated upon obtaining GPS lock; (2) its position
is updated upon receiving a useful beacon from one of its
neighbours; or (3) it has powered on its GPS module and is
in the process of obtaining GPS lock. A beacon sent in state
(3) serves as a GPS lock backoff beacon, and its purpose is
to inform neighbouring nodes that it will imminently obtain
lock, thereby avoiding the implicit synchronization issue.
Figure 18. Event-triggered beaconing example: nodes
only send beacons when they are getting lock, they get
lock, or they log a useful contact with a neighbor
To support event-based beaconing, nodes use a modified
beacon format that includes the expected time to obtain lock
(based on history of lock times and off times).This beacon
is known as the GPS lock backoff beacon. Figure 18 illus-
trates how event-based beaconing works. At time t1, node
A obtains a GPS lock. As a result, it sets it uncertainty
to Ugps(t1)and broadcasts a beacon declaring its id, uncer-
tainty, and position. Node B receives this beacon at time t2,
and, after determining that contact with node A reduces its
own uncertainty, sets its uncertainty according to Equation 3.
Nodes A and B then proceed to grow their uncertainty re-
gion according to their local speed models. At time t3, node
B hears a beacon from node C with a low uncertainty, and
updates its local uncertainty region. Subsequently, it broad-
casts a beacon announcing its updated uncertainty, enabling
node A to reduce its uncertainty as well at time t4. By time
t5, the uncertainty at node A grows close to AAU, so A de-
cides to power on its GPS module to avoid exceeding AAU.
It also broadcasts a GPS lock backoff beacon informing all
its neighbours of this action to avoid multiple neighbours try-
ing to simultaneously get GPS lock. Node B hears this bea-
con, and refrains from powering on its GPS module to get
lock. At time t6, node A succeeds in getting lock, reducing
its uncertainty to Ugps(t6), and broadcasts a beacon inform-
ing neighbours of this change. Finally, B hears node A’s bea-
con at time t7and updates its uncertainty as well. During all
these exchanges, four radio packets are transmitted and re-
ceived, node A’s GPS obtained 2 locks, while node B’s GPS
module remained in off mode.
The last bar plot of of Figure 17 shows the results for
event-based beaconing. With a slight increase in error rate,
it nearly halves the power consumption relative to all of the
beacon periods we consider. The energy saving from event-
based beaconing stem from 3 factors: (1) the reduced radio
beaconing enables the radio to sleep more often; and (2) the
reduced overlaps in GPS lock attempts significantly reduce
the GPS power consumption; (3) this implicitly enables the
MCU to sleep more often as well, as most radio beacons are
sent when the GPS is on.
Finally, we consider the effect of combining RSSI-based
contact logging and event-based beaconing on error rate and
power consumption. Figure 19 summarises the results of the
various mechanisms in the paper, including GPS duty cy-
cling, GPS duty cycling with contact logging (contact radius
of 5m and beacon period of 5 seconds), RSSI-based contact
logging with 5 second beacon period, event-based beaconing
with a 5 m contact radius, and event- and RSSI- based con-
tact logging. The latter yields the lowest power consumption
at nearly 30mW; however, it causes the error rate to cross the
error tolerance due to RSSI spreads. This means that event-
based beaconing with a 5 m contact radius remains the most
energy-efficient approach that meets the 5% error tolerance
for the virtual fencing application.
4.3 Managing Performance
We now outline an adaptive duty cycling [8] online algo-
rithm that we have implemented on the Fleck nodes and that
delivers both required position performance and lifetime. As
already discussed, each node evolves its uncertainty estimate
while its GPS is powered off and uses opportunistic contact
logging to reduce that uncertainty.
Whenever a node receives a GPS position lock, it deter-
mines if the fix falls more than the distance AAU from the
previous fix. The node maintains a circular list of the out-
come of this test for the last Msamples. When the error rate
in this list exceeds σ, the node is not meeting the position
performance criteria so the assumed speed is incremented
and the GPS duty cycle will increase. On the other hand, if
the error rate is below a different threshold the node is over-
delivering in terms of positioning performance and using a
higher duty cycle than it needs, in which case the assumed
speed is decremented. The logic also takes into account the
current energy state E(t)and the desired energy state Ere-
quired to meet the lifetime objective. The algorithm is shown
in figure 20. To resolve any conflicts between the energy and
accuracy constraints, the user preference input sets the prior-
inria-00524296, version 1 - 7 Oct 2010
Figure 19. Comparison of configurations. Combination
of RSSI-based and event-based beaconing has the lowest
power consumption, but exceeds error tolerance.
ity of increasing speed to lower the error rate or decreasing
speed to increase the lifetime.
loop {every GPS fix}
d(t)= [x(t)x(t0)]2+[y(t)y(t0)]2)
if d(t)>AAU then
list.insert(1)
else
list.insert(0)
end if
if list.sum
M>σthen
if user pref == accuracy OR E(t)>Ethen
s=s+δ+
end if
end if
if list.sum
M<σthen
s=sδ
end if
end loop
Figure 20. Adaptive Duty Cycling Strategy
We implemented this strategy in our simulator and set a
power consumption of 44 mW for achieving 3 months life-
time with the 4 D-cell batteries. The initial battery energy
was scaled down to match the 2-day duration of the data set.
The error tolerance is 5%. We set δ+to 3-sand δto 0.1 to
implement fast response to errors and slow energy recovery.
For M=1, Figure 21 shows the impact of the user prefer-
ence metric on the residual energy and the error rate. The
algorithm delivers the desired lifetime when the energy pref-
erence is set, and approaches the error tolerance when the
accuracy preference is set. Note that the measured error rate
indicates the node’s online measurements, while the real er-
ror rate is the ground truth value.
5 Related Work
The first instance of animal tracking by networked em-
bedded systems is the ZebraNet project [9], which tracks in-
dividual position records for animals every of a few minutes.
However, ZebraNet collars include a solar panel, making the
energy management problem quite a tractable one. Position-
ing is done by GPS only, and the nodes flood their informa-
tion by flooding in order to facilitate data acquisition by the
mobile sink. The Electronic Shepherd project [10] also uses
GPS localization for tracking herds of sheep in the moun-
tains. “Real-time” is obtained thanks to a GPRS modems in
the collar, even though position is recorded and sent back to
base only a few times per day in order to prolong battery life.
Unlike our work, nodes are asymmetric: only a few animals
in the herd wear collars including GPS and GPRS modules.
All the others wear a small ear-tag including only a micro-
controller and a radio-chip.
The work in [11] considers energy efficient localization
for base nodes on the path of the mobile nodes, and the en-
ergy implications of each approach. While the speed mod-
els in [11] are similar to this paper, their work is purely
simulation-based and does not consider the benefits of con-
tact logging or online adaptation according to energy and po-
sition changes. The work in [12] addresses the tradeoff be-
tween localization accuracy and energy efficiency. It consid-
ers static, dynamic, and dead reckoning mobility models and
studies their effect on accuracy and energy through ns2 sim-
ulations. Our work addresses the same tradeoff but uses em-
pirical GPS data to conduct analysis on three different speed
models and a realistic energy model of the GPS-enabled
nodes that includes the previously unreported stochastic lock
time model. We also consider the effect of dynamically
changing the required localization performance based on the
mobile node’s distance from an exclusion zone. Further we
explore contact logging as a complement to GPS for achiev-
ing a better balance between accuracy and efficiency. In a
similar fashion, Pattern et al. [13] provide a tuning knob
to obtain various energy-accuracy tradeoffs by the careful
choice of duty cycle and activation radius.
The works in [14] and [15] address this tradeoff as well
by considering improvements to the RIP method [16] as
the baseline localization method. A key difference between
You’s work and ours is that we support a variable AAU. An-
other distinction is their use of mote-based RSSI localization
schemes in indoor environments, which require a deployed
network infrastructure, consume less power, and require a
calibration phase. Finally, they do not consider contact log-
ging for reducing localization uncertainty.
Recent work [17] [18] has also explored collaborative lo-
calization that enables nodes to detect neighboring signals
for position refinement. Chan et al. [17] use a cluster-based
approach, where nearby nodes can establish clusters through
IEEE 802.15.4 radios as a neighbour detection sensor that
enhances WiFi localization. Our approach is similar in its
reliance on contact signals, but it differs in that the contact
signal is sent and received by the same radio, eliminating the
need for multi-radio nodes. Liu et al. [18] also use a form of
contact logging that establishes anchor nodes that know their
location, pseudo-anchor nodes that estimate their location
inria-00524296, version 1 - 7 Oct 2010
(a) User preference set to accuracy (b) User preference set to energy
Figure 21. Effect of user preference
based on contact with anchor nodes, and free nodes that rely
on anchor or pseudo-anchor nodes for their location. This
resembles the multi-hop contact logging in our virtual fenc-
ing application. However, their work focuses on algorithms
to bound the uncertainty propagation in iterative localization
techniques through the use of static anchor nodes, while we
focus on bounding the localization uncertainty of mobile an-
chors and neighbouring nodes that rely on them within an
energy constraint. In addition, nodes dynamically become
anchors in our model based on their current GPS state.
More recently, research has explored energy-aware local-
ization for mobile phones equipped with GPS. For example
Constandache et. al [19] propose an average error metric for
GPS duty cycling. Using proximity mechanisms like Wi-Fi
or GSM yields a lower average error for the same energy
budget as plain GPS duty cycling. One of their mobility
models deduces the current location of the node from recent
history when the user is driving on a straight road. Simi-
larly, the work in [20] stores historical information of loca-
tions where GPS does not work, and of average speed as a
function of time and place. Our algorithm stores no neigh-
bour history, where each node tracks only its own uncertainty
through contact logging with neighbours. This minimises
local storage requirements for individual nodes and training
requirements for establishing historical patterns.
The use of accelerometers has also been proposed as a
low power indicator of movement to supplement GPS duty
cycling [20][21]. The inclusion of inertial sensor readings,
such as magnetometers or accelerometers, is an attractive
extension to our approach, which we discuss in the next
section. Two recent works [20][22] consider Bluetooth as
a radio ranging technology, since it is pervasive in mobile
phones. Bluetooth is relatively energy-hungry for neigh-
bour discovery and frequent beaconing, as it relies on a syn-
chronous connection-oriented master slave topology. Our
work uses lower power IEEE 802.15.4 for radio ranging that
are use asynchronous broadcast channels and a flat topology.
This enables a single beacon to be detected by all neighbours
within radio range, providing a more scalable approach.
6 Conclusion and Perspectives
This paper has proposed and validated short-range radio
contact logging as an effective complement to GPS duty
cycling for balancing the positioning accuracy and energy-
efficiency tradeoff. Given a user policy that comprises a
target lifetime, acceptable uncertainty, and error tolerance,
we have established a strategy for tailoring the contact ra-
dius and beacon period for any mobile sensor network ap-
plication. For RSSI-independent contact logging, the strat-
egy considers the distribution of relative distances between
the mobile nodes to analytically determine the optimal con-
tact radius. We have confirmed that this contact radius per-
forms best through simulations based on our empirical data
set. We then proceeded to identify the most suitable beacon
period for our dataset, based on which we proposed event-
driven beaconing that reactively sends beacons only when
nodes detect state changes.
While the event-driven beaconing strategy yields the low-
est power consumption in our scenario, the optimal strategy
is highly dependent on the mobility pattern of the nodes,
their clustering characteristics, and user policy. For instance,
an application with highly mobile nodes may benefit from a
more proactive beaconing strategy that transmits beacons at
short periodic intervals. Similarly, while RSSI-based contact
logging outperformed all statically set contact radii, it may
not suit radios with no RSSI or high multi-path deployments.
Assisted GPS is a mechanism that uses cellular phone
base stations to provide ephemeris data (satellite date) to
mobile nodes in order to reduce their GPS lock time. In
WSN deployments with both static and mobile nodes, the
static nodes can supply the ephemeris data to the mobile
nodes, which in turn seed their GPS receivers on startup with
this data to achieve a lock time of 4 seconds (time-to-first-
fix). A challenge in using assisted GPS is that the receiver’s
ephemeris data must be erased every time. If a node ever
needs to switch back to using standard satellite based GPS,
the GPS receiver will take longer than normal to obtain a fix.
The static nodes that serve as a routing backbone for the
inria-00524296, version 1 - 7 Oct 2010
mobile nodes could also serve as local positioning anchors
outside the scope of assisted GPS. The area of relative po-
sitioning through powerful local beacons is indeed a well-
investigated area. While our current virtual fencing deploy-
ments do include static nodes that can be used for this pur-
pose, we are moving towards more sparse static node deploy-
ments for high scalability in monitoring mobile cattle. The
GPS duty cycling and radio contact logging strategies here
represent a step towards more sparse deployments.
Another possible enabler of more sparse deployments is
the use of inertial sensors, such as accelerometers and mag-
netometers, to refine position estimates while the GPS is
powered off. While this is a proven strategy in other GPS
tracking scenarios, its utility for low power devices is an
open question. The accelerometer in particular involves an
inherent tradeoff between its accuracy for predicting speed
and its sampling rate. It can only provide precise estimates of
speed for high sample rates, which incurs high energy cost.
The use of accelerometers as boolean indicators of motion is
certainly an option for input to the algorithm. Magnetome-
ters can certainly provide accurate heading data for low en-
ergy cost, thereby limiting the uncertainty growth within a
angular cone rather than a circle. This comes at a cost of in-
creasingly computational complexity for growing the irregu-
lar uncertainty region as the mobile nodes change directions.
Some concepts from this paper, such as dynamically
changing the AAU according to distance from the fence, are
specific to the cattle monitoring application. However, our
overall strategy is applicable to other mobile tracking appli-
cations, such as smart phone social networking applications.
Consider groups of friends with smart phones that regularly
meet and use location-based services that rely on GPS. If
all smart phones keep their GPS modules on to run the ser-
vices, they will most likely deplete their batteries quickly.
Alternatively, the use of contact logging to share the GPS
load among the smart phones can prolong the lifetime of all
devices. Applying our strategy to this application would re-
quire mobility models, clustering patterns, and application-
specific user policies to determine the best configuration.
Acknowledgements
The authors would like to thank Chris Crossman, Greg
Bishop-Hurley, Philip Valencia, Brano Kusy, and Alban
Cotillon for their valuable inputs in realising this work.
7 References
[1] M. Ilyas and R. C. Dorf, editors. The handbook of ad
hoc wireless networks. CRC Press, Inc., Boca Raton,
FL, USA, 2003.
[2] T. Wark et al. The design and evaluation of a mobile
sensor/actuator network for autonomous animal con-
trol. In IPSN ’07, 2007.
[3] R. Jurdak, A. G. Ruzzelli, and G.M. P. O’Hare. Adap-
tive radio modes in sensor networks: How deep to
sleep? In IEEE (SECON), June 2008.
[4] Z. Butler et al. From robots to animals: Virtual fences
for controlling cattle. Int. J. Rob. Res., 25(5-6), 2006.
[5] Y. Guo et al. Using accelerometer, high sample rate gps
and magnetometer data to develop a cattle movement
and behaviour model. Ecological Modelling, 220(17),
2009.
[6] G. Lin, G. Noubir, and R. Rajaraman. Mobility models
for ad hoc network simulation. In INFOCOM, 2004.
[7] K. Srinivasan and P. Levis. Rssi is under appreciated.
In EmNets, 2006.
[8] R. Jurdak, P. Baldi, and C. Videira Lopes. Adaptive
low power listening for wireless sensor networks. IEEE
Tran. on Mobile Computing, 6(8):988–1004, 2007.
[9] P. Zhang and C. Sadler et al. Habitat Monitoring
with ZebraNet: Design and Experiences, chapter Wire-
less Sensor Networks: A Systems Perspective. Artech
House, 2005.
[10] Thorstensen B. and T. et al. Syversen. Electronic Shep-
herd – a low-cost, low-bandwidth, wireless network
system. In MobiSys, 2004.
[11] J. Winter, Y. Xu, and W.C. Lee. Prediction-based strate-
gies for energy saving in object tracking sensor net-
works. In IEEE ICMDM, January 2004.
[12] S. Tilak et al. Dynamic localization control for mobile
sensor networks. In IPCCC, 2005.
[13] S. Pattern, S. Poduri, and B. Krishnamachari. Energy-
quality tradeoffs in sensor track- ing: Selective activa-
tion with noisy measurements. In IPSN, 2003.
[14] C.-W. You. Enabling Energy-Efficient Localization
Services on Sensor Network Positioning Systems. PhD
thesis, January 2008.
[15] You et al. Impact of sensor-enhanced mobility predic-
tion on the design of energy-efficient localization. Ad
Hoc Netw., 6(8), 2008.
[16] B. Kusy et al. Radio interferometric tracking of mobile
wireless nodes. In MobiSys, 2007.
[17] L. Chan et al. Collaborative localization: Enhancing
wifi-based position estimation with neighborhood links
in clusters. In Pervasive, 2006.
[18] J. Liu, Y. Zhang, and F. Zhao. Robust distributed
node localization with error management. In MobiHoc,
2006.
[19] I. Constandache, S. Gaonkar, and et al. Enloc: Energy-
efficient localization for mobile phones. In INFOCOM,
pages 2716–2720, 2009.
[20] J. Paek, J. Kim, and R. Govindan. Energy-efficient rate-
adaptive gps-based positioning for smartphones. In
MobiSys, 2010.
[21] M. B. Kjaergaard et al. Entracked: energy-efficient ro-
bust position tracking for mobile devices. In MobiSys,
2009.
[22] K. Lin, A. Kansal, and et al. Energy-accuracy trade-
off for continuous mobile device location. In MobiSys,
2010.
inria-00524296, version 1 - 7 Oct 2010

Supplementary resource (1)

... One possible explanation for this observation could be related to the difficulty of satellite acquisition while a bird is flying, as has been noted for moving animals in various mammal studies 29,31,32 , perhaps because of changes in the position and orientation of the GPS transmitter. Our findings also confirmed that longer duty cycles (of 2 h compared with those of 1 h) produced lower FLR, probably associated with the fact that more intense duty cycles increase the transmitter energy consumption and consequently reduce the device usage time 33 . In fact, Silva et al. 25 suggested that FLR due to poor GDOP (when Geometric Dilution of Precision limits the transmitter to contact with enough satellites to produce a fix) increased when the birds moved. ...
Article
Full-text available
In recent decades, global positioning system (GPS) location data and satellite telemetry systems for data transmission have become fundamental in the study of basic ecological traits in wildlife biology. Evaluating GPS location errors is essential in assessing detailed information about the behaviour of an animal species such as migration, habitat selection, species distribution or foraging strategy. While many studies of the influence of environmental and technical factors on the fix errors of solar-powered GPS transmitters have been published, few studies have focussed on the performance of GPS systems in relation to a species' biological traits. Here, we evaluate the possible effects of the biological traits of a large raptor on the frequency of lost fixes-the fix-loss rate (FLR). We analysed 95,686 records obtained from 20 Bearded Vultures Gypaetus barbatus tracked with 17 solar-powered satellite transmitters in the Pyrenees (Spain, France and Andorra), between 2006 and 2019 to evaluate the influence of biological, technical, and environmental factors on the fix-loss rate of transmitters. We show that combined effects of technical factors and the biological traits of birds explained 23% of the deviance observed. As expected, the transmitter usage time significantly increased errors in the fix-loss rate, although the flight activity of birds revealed an unexpected trade-off: the greater the proportion of fixes recorded from perched birds, the lower the FLR. This finding seems related with the fact that territorial and breeding birds spend significantly more time flying than non-territorial individuals. The fix success rate is apparently due to the interactions between a complex of factors. Non-territorial adults and subadults, males, and breeding individuals showed a significantly lower FLR than juveniles-immatures females, territorial birds or non-breeding individuals. Animal telemetry tracking studies should include error analyses before reaching any ecological conclusions or hypotheses about spatial distribution.
... SensLoc [17] uses algorithms to detect the user context, such as frequently visited places, and movement detection to adapt the GPS duty cycle. Jurdak et al. [13] propose to use short-range radio contact-logging to bound the position uncertainty when duty-cycling the GPS. SmartDC [6] provides adaptive GPS duty-cycling using prediction mechanisms for regularities in user mobility. ...
Article
Full-text available
Energy-efficient location tracking with battery-powered devices using energy harvesting necessitates duty-cycling of GPS to prolong the system lifetime. We propose an energy- and mobility-aware scheduling framework that adapts to real-world dynamics to achieve optimal long-term tracking performance. To forecast energy, the framework uses an exponentially weighted moving average filter to compute a virtual energy budget for the remainder of the forecast period. The virtual energy budget is then used as input for our proposed information-based GPS sampling approach, which estimates the current tracking error through dead-reckoning and schedules a new GPS sample when the error exceeds a given threshold. In order to improve the long-term tracking performance, the threshold is adapted based on the current energy and movement trends to balance the expected information gain from a new GPS sample with its energy cost. We evaluate our approach on empirical traces from wild flying foxes and compare it to strategies that sample GPS using fixed and adaptive duty cycles and by using dead-reckoning with a fixed threshold. Our analysis shows that the proposed information-based GPS sampling strategy reduces the mean tracking error compared to existing methods and approaches the performance of the optimal offline sampling strategy.
... In addition, GPS is power-hungry, which can drain the battery of the energy-limited mobile devices quickly [22]. To mitigate the highenergy requirement of the GPS, several techniques are proposed [12,30,38,55] that use mobile sensors with duty-cycling of the GPS to trade-off localization accuracy with energy-efficiency. To address the shortcomings of GPS-based techniques, WiFi-based outdoor localization systems have been proposed, e.g. ...
Conference Paper
Full-text available
Recent years have witnessed fast growth in outdoor location-based services. While GPS is considered a ubiquitous localization system, it is not supported by low-end phones, requires direct line of sight to the satellites, and can drain the phone battery quickly. In this paper, we propose DeepLoc: a deep learning-based outdoor localization system that obtains GPS-like localization accuracy without its limitations. In particular, DeepLoc leverages the ubiquitous cellular signals received from the different cell towers heard by the mobile device as hints to localize it. To do that, crowd-sensed geo-tagged received signal strength information coming from different cell towers is used to train a deep model that is used to infer the user's position. As part of DeepLoc design, we introduce modules to address a number of practical challenges including scaling the data collection to large areas, handling the inherent noise in the cellular signal and geo-tagged data, as well as providing enough data that is required for deep learning models with low-overhead. We implemented DeepLoc on different Android devices. Evaluation results in realistic urban and rural environments show that DeepLoc can achieve a median localization accuracy within 18.8m in urban areas and within 15.7m in rural areas. This accuracy outperforms the state-of-the-art cellular-based systems by more than 470% and comes with 330% savings in power compared to the GPS. This highlights the promise of DeepLoc as a ubiquitous accurate and low-overhead localization system.
Chapter
Dementia is one of the primary causes of dependency and disability among older people worldwide. In order to diminish the risk of wandering associated with dementia, portable GPS trackers are used by the patients. But most of these devices are plagued by either short-range or low-effective battery life not greater than 10 hours, thus becoming highly restrictive. At the same time, though techniques to make these devices more energy efficient are available in literature, many of them fail to provide sufficient accuracy required for continuous real-time location tracking of the patient. Therefore, in this paper, we propose a novel, small, and energy-efficient location tracker called Nemo that can be used as a wearable by effectively combining the novel LoRa protocol of communication, geofencing, and Adaptive GPS Duty Cycling strategies. The whole system consists of a LoRa-enabled GPS tracker as end node, a LoRaWAN gateway to relay the messages between the tracker and a central server, and also an Android/web application to receive real-time alerts. Further, the maximum range of communication and energy efficiency of the module at various environments is verified through comprehensive real-world experiments.
Article
The development of novel Information and Communication Technology (ICT) based solutions becomes essential to meet the ever increasing rate of global urbanization in order to satiate the constraint in resources. The popular ‘smart city paradigm is characterized by ubiquitous cyber provisions for the monitoring and control of such city's critical infrastructures, encompassing healthcare, environment, transportation and utilities among others. In order to manage the numerous services keeping their Quality of Service (QoS) demands upright, it is imperative to employ context aware computing as well as fog computing simultaneously. This paper investigates the feasibility of energy minimization at the fog layer through intelligent sleep and wake-up cycles of the fog nodes which are context-aware. It proposes a virtual machine management approach for effectively allocating service requests with a minimal number of active fog nodes using a genetic algorithm (GA); and thereafter, a reinforcement learning (RL) approach is incorporated to optimize the period of fog nodes’ duty cycle. Simulations are carried out using MATLAB and the results demonstrate that the proposed scheme improves energy consumption of the fog layer by approximately 11–21% when compared to existing context sharing based algorithms.
Article
Integrated GPS receivers have become a basic module in today’s mobile devices. While serving as the cornerstone for location based services, GPS modules have a serious battery drain problem due to high computation load. This paper aims to reveal the impact of key software parameters on GPS energy consumption, by establishing an energy model for a standard GPS receiver architecture as found in both academic and industrial designs. In particular, our measurements show that the receiver’s energy consumption is in large part linear with the number of tracked satellites as well as the length of the raw GPS signal. This leads to an energy efficient design of selective tracking algorithm with large satellite Geometric Dilution of Precision (GDOP) weight, well spatial distribution, and large signal intensity. Real experimental results on three typical scenarios show that our selective tracking algorithm provides better tradeoff between positioning accuracy and energy consumption than standard GPS receivers.
Chapter
"Intelligent Transportation systems" is what everyone wants to know about, and about which very little is available as know-how. ITS technologies and monitoring systems are quite popular and reasonably well deployed in developed countries, particularly the roadways and airways. ITS holds a greater promise than ever before, as both availability of niche technologies and demand for more efficient transportation systems have increased multi-fold in recent years. Of late, there are huge railway projects all over the world that spans through several techniques, such as light / heavy rails, monorails etc. Apart from the social benefits that can be envisaged, these projects are genuine examples of public-private partnerships along with global business operations. Many of these projects demonstrate a classy trend of moving towards automation of operations of very large scales. Few agent architectures are discussed in brief in this chapter.
Article
We consider the problem of data collection from a network of energy harvesting sensors, applied to tracking mobile assets in rural environments. Our application constraints favor a fair and energy-aware solution, with heavily duty-cycled sensor nodes communicating with powered base stations. We study a novel scheduling optimization problem for energy harvesting mobile sensor network, that maximizes the amount of collected data under the constraints of radio link quality and energy harvesting efficiency, while ensuring a fair data reception. We show that the problem is NP-complete and propose a heuristic algorithm to approximate the optimal scheduling solution in polynomial time. Moreover, our algorithm is flexible in handling progressive energy harvesting events, such as with solar panels, or opportunistic and bursty events, such as with Wireless Power Transfer. We use empirical link quality data, solar energy, and WPT efficiency to evaluate the proposed algorithm in extensive simulations and compare its performance to state-of-the-art. We show that our algorithm achieves high data reception rates, under different fairness and node lifetime constraints. IEEE
Chapter
“Intelligent Transportation systems” is what everyone wants to know about, and about which very little is available as know-how. ITS technologies and monitoring systems are quite popular and reasonably well deployed in developed countries, particularly the roadways and airways. ITS holds a greater promise than ever before, as both availability of niche technologies and demand for more efficient transportation systems have increased multi-fold in recent years. Of late, there are huge railway projects all over the world that spans through several techniques, such as light/heavy rails, monorails etc. Apart from the social benefits that can be envisaged, these projects are genuine examples of public-private partnerships along with global business operations. Many of these projects demonstrate a classy trend of moving towards automation of operations of very large scales. Few agent architectures are discussed in brief in this chapter.
Conference Paper
Full-text available
This paper investigates a mobile, wireless sensor/actuator network application for use in the cattle breeding industry. Our goal is to prevent fighting between bulls in on-farm breeding paddocks by autonomously applying appropriate stimuli when one bull approaches another bull. This is an important application because fighting between high-value animals such as bulls during breeding seasons causes significant financial loss to producers. Furthermore, there are significant challenges in this type of application because it requires dynamic animal state estimation, real-time actuation and efficient mobile wireless transmissions. We designed and implemented an animal state estimation algorithm based on a state-machine mechanism for each animal. Autonomous actuation is performed based on the estimated states of an animal relative to other animals. A simple, yet effective, wireless communication model has been proposed and implemented to achieve high delivery rates in mobile environments. We evaluated the performance of our design by both simulations and field experiments, which demonstrated the effectiveness of our autonomous animal control system.
Conference Paper
Full-text available
Mobile applications often need location data, to update locally relevant information and adapt the device context. While most smartphones do include a GPS receiver, it's frequent use is restricted due to high battery drain. We design and prototype an adaptive location service for mobile devices, a-Loc, that helps reduce this battery drain. Our design is based on the observation that the required location accuracy varies with location, and hence lower energy and lower accuracy localization methods, such as those based on WiFi and cell-tower triangulation, can sometimes be used. Our method automatically determines the dynamic accuracy requirement for mobile search-based applications. As the user moves, both the accuracy requirements and the location sensor errors change. A-Loc continually tunes the energy expenditure to meet the changing accuracy requirements using the available sensors. A Bayesian estimation framework is used to model user location and sensor errors. Experiments are performed with Android G1 and AT&T Tilt phones, on paths that include outdoor and indoor locations, using war-driving data from Google and Microsoft. The experiments show that a-Loc not only provides significant energy savings, but also improves the accuracy achieved, because it uses multiple sensors.
Conference Paper
Full-text available
An important feature of a modern mobile device is that it can position itself. Not only for use on the device but also for remote applications that require tracking of the device. To be useful, such position tracking has to be energy-efficient to avoid having a major impact on the battery life of the mobile device. Furthermore, tracking has to robustly de- liver position updates when faced with changing conditions such as delays due to positioning and communication, and changing positioning accuracy. This work proposes EnTracked — a system that, based on the estimation and prediction of system conditions and mo- bility, schedules position updates to both minimize energy consumption and optimize robustness. The realized system tracks pedestrian targets equipped with GPS-enabled de- vices. The system is configurable to realize different trade- offs between energy consumption and robustness. We provide extensive experimental results by profiling how devices consume power, by emulation on collected data and by validation in several real-world deployments. Results from this profiling show how a device consumes power while tracking its position. Results from the emulation indicate that the system can estimate and predict system conditions and mobility. Furthermore they provide evidence for that the system can lower the energy consumption considerably and remain robust when faced with changing system con- ditions. By validation in several real-world deployments we provide evidence that the real system works as predicted by the emulation.
Conference Paper
Full-text available
Many emerging smartphone applications require position information to provide location-based or context-aware services. In these applications, GPS is often preferred over its alternatives such as GSM/WiFi based positioning systems because it is known to be more accurate. However, GPS is extremely power hungry. Hence a common approach is to periodically duty-cycle GPS. However, GPS duty-cycling trades-off positioning accuracy for lower energy. A key requirement for such applications, then, is a positioning system that provides accurate position information while spending minimal energy. In this paper, we present RAPS, rate-adaptive positioning system for smartphone applications. It is based on the observation that GPS is generally less accurate in urban areas, so it suffices to turn on GPS only as often as necessary to achieve this accuracy. RAPS uses a collection of techniques to cleverly determine when to turn on GPS. It uses the location-time history of the user to estimate user velocity and adaptively turn on GPS only if the estimated uncertainty in position exceeds the accuracy threshold. It also efficiently estimates user movement using a duty-cycled accelerometer, and utilizes Bluetooth communication to reduce position uncertainty among neighboring devices. Finally, it employs celltower-RSS blacklisting to detect GPS unavailability (e.g., indoors) and avoid turning on GPS in these cases. We evaluate RAPS through real-world experiments using a prototype implementation on a modern smartphone and show that it can increase phone lifetimes by more than a factor of 3.8 over an approach where GPS is always on.
Conference Paper
Full-text available
Location knowledge of nodes in a network is essential for many tasks such as routing, cooperative sensing, or service delivery in ad hoc, mobile, or sensor networks. This paper introduces a novel iterative method ILS for node localiza- tion starting with a relatively small number of anchor nodes in a large network. At each iteration, nodes are localized using a least-squares based algorithm. The computation is lightweight, fast, and any-time. To prevent error from propagating and accumulating during the iteration, the er- ror control mechanism of the algorithm uses an error registry to select nodes that participate in the localization, based on their relative contribution to the localization accuracy. Sim- ulation results have shown that the active selection strategy signican tly mitigates the eect of error propagation. The algorithm has been tested on a network of Berkeley Mica2 motes with ultrasound TOA ranging devices. We have com- pared the algorithm with more global methods such as MDS- MAP and SDP-based algorithm both in simulation and on real hardware. The iterative localization achieves compara- ble location accuracy in both cases, compared to the more global methods, and has the advantage of being fully decen- tralized.
Conference Paper
Full-text available
Energy-efficient performance is a central challenge in sensor network deployments, and the radio is a major contributor to overall energy node consumption. Current energy- efficient MAC protocols for sensor networks use a fixed low power radio mode for putting the radio to sleep. Fixed low power modes involve an inherent tradeoff: deep sleep modes have low current draw and high energy cost and latency for switching the radio to active mode, while light sleep modes have quick and inexpensive switching to active mode with a higher current draw. This paper proposes adaptive radio low power sleep modes based on current traffic conditions in the network, as an enhancement to our recent RFID impulse low power wake-up mechanism. The paper also introduces a comprehensive node energy model, that includes energy components for radio switching, transmission, reception, listening, and sleeping, as well as the often disregarded micro-controller energy component to evaluate energy performance for both MicaZ and TelosB platforms, which use different MCU's. We then use the model for comparing the energy-related performance of RFIDImpulse enhanced with adaptive low power modes with BMAC and IEEE 802.15.4 for the two node platforms under varying data rates. The comparative analysis confirms that RFIDImpulse with adaptive low power modes provides up to 20 times lower energy consumption than IEEE 802.15.4 in low traffic scenario. The evaluation also yields the optimal settings of low power modes on the basis of data rates for each node platform, and it provides guidelines for the selection of appropriate MAC protocol, low power mode, and node platform for a given set of traffic requirements of a sensor network application.
Article
We consider the problem of monitoring and controlling the position of herd animals, and view animals as networked agents with natural mobility but not strictly controllable. By exploiting knowledge of individual and herd behavior we would like to apply a vast body of theory in robotics and motion planning to achieving the constrained motion of a herd. In this paper we describe the concept of a virtual fence which applies a stimulus to an animal as a function of its pose with respect to the fenceline. Multiple fence lines can define a region, and the fences can be static or dynamic. The fence algorithm is implemented by a small position-aware computer device worn by the animal, which we refer to as a Smart Collar. We describe a herd-animal simulator, the Smart Collar hardware and algorithms for tracking and controlling animals as well as the results of on-farm experiments with up to ten Smart Collars.
Article
There is a general belief in the Wireless Sensor Net- work (WSN) community that the received signal strength indicator (RSSI) is a bad estimator of the link quality. This belief is due to the existence of many asymmetry linksinolderradiossuchasCC1000andTR1000.Newer radios that are based on IEEE 802.15.4 standard such as CC2420 implement another parameter called link qual- ity indicator (LQI) which is believed to be a better in- dicator than RSSI. There is so far no extensive evalua- tionofCC2420tovalidatethisclaim.Wehaveconducted such an evaluation and our preliminary results indicate that RSSI for a given link has very small variation over time for a link. Our results also indicate that when the RSSI is above the sensitivity threshold (about -87 dBm), the packet reception rate (PRR) is atleast 85%. Around this sensitivity threshold, however, the PRR is not corre- lated possibly due to variations in local phenomena such as noise. LQI, on the other hand, varies over a wider range over time for a given link. However, the mean LQI computed over many packets has a better correlation with PRR.
Article
Energy efficiency and positional accuracy are often contradictive goals. We propose to decrease power consumption without sacrificing significant accuracy by developing an energy-aware localization that adapts the sampling rate to target’s mobility level. In this paper, an energy-aware adaptive localization system based on signal strength fingerprinting is designed, implemented, and evaluated. Promising to satisfy an application’s requirements on positional accuracy, our system tries to adapt its sampling rate to reduce its energy consumption. The contribution of this paper is fourfold. (1) We have developed a model to predict the positional error of a real working positioning engine under different mobility levels of mobile targets, estimation error from the positioning engine, processing and networking delay in the location infrastructure, and sampling rate of location information. (2) In a real test environment, our energy-saving method solves the mobility estimation error problem by utilizing additional sensors on mobile targets. The result is that we can improve the prediction accuracy by 56.34% on average, comparing to algorithms without utilizing additional sensors. (3) We further enhance our sensor-enhanced mobility prediction algorithm by detecting the target’s moving foot step and then estimate the target’s velocity. This method can improve the mobility prediction accuracy by 49.81% on an average, comparing to previous sensor-enhanced mobility prediction algorithm. (4) We implemented our energy-saving methods inside a working localization infrastructure and conducted performance evaluation in a real office environment. Our performance results show as much as 68.92% reduction in power consumption.
Conference Paper
This paper reports a new novel low-cost, wireless communication network system, called the "Electronic Shepherd" (ES). The system is innovative in the way that it supports flock behavior, meaning that a flock leader monitors the state of the other elements in the flock using low-cost radio communication equipment. The paper addresses both details of the terminal devices and communication protocols, as well as testing of the system in a real environment. The ES system was originally made to address special needs for sheep and reindeer farmers who were seeking a system to keep track of their animals during the grazing season. The system, including GPS receivers, UHF radio communication transceivers and GPRS modems, contributes a new approach for low-cost networking and service implementation, not only for the purpose of animal tracking, but also for other applications where objects are to be monitored at a low cost.