[ieee 2011 ieee consumer communications and networking conference (ccnc) - las vegas, nv, usa...
TRANSCRIPT
Energy Measurements Campaign for PositioningMethods on State-of-the-Art Smartphones
Julien EberleEPFL - Lausanne
Email: [email protected]
Gian Paolo PerrucciNokia Research Center - Lausanne
Email: [email protected]
Abstract—Mobile phones have been undergoing a breathtakingevolution over the last years, starting from simple devices withonly voice services, nowadays they are referred as smartphones,powerful devices able to offer appealing services. One of theservices that has recently become popular is referred as LocationBased Services that use contextual and location information toenhance mobile user experience. As a basic assumption, this kindof services needs to be continuously aware of the phone’s location.The challenge emerges from the trade-off needed between thelocation precision and the energy consumption of the positioningmethods, the most precise being also the energy-hungriest. Thispaper presents the results of an extensive energy measurementscampaign that was conducted on different sensors and positioningmethods on a smartphone. Results will help researchers and de-velopers to design energy efficient and more accurate continuouslocation tracking algorithms to support Location Based Services.
I. INTRODUCTION AND MOTIVATIONS
Mobile phones have been undergoing a breathtaking evo-
lution over the last years, starting from simple devices with
only voice services towards the transition to smartphones
offering novel services such as mobile Internet, multimedia,
high datarate connectivity and geolocation among others. The
latter capability enables the so called Location Based Services
(LBS) that use contextual and location information to enhance
mobile user experience. One example of LBS application is
Foursquare[1] that logs visited places and allows users to
earn points and badges based on the frequency of the visits.
Other popular examples are Brightkite [2], a geolocalized
micro-blogging platform and Google Latitude [3] that can be
used to track friend’s location. All these services have one
thing in common: they all need to keep continuously track of
the position of the phone. However, smartphones are battery
driven devices and therefore have a limited amount of energy
available. For this reason in the past years there have been
several attempts to develop algorithms for continuous location
tracking that would not have a high impact on the battery life
of the devices.
The first algorithms were implemented using the single sensor
approach, as for example in the TRAC-IT project [4], a trans-
portation survey conducted in US, where GPS placed in cars
and trucks were used to track vehicles. As an other example, in
[5] and [6], Received Signal Strength (RSS) of the GSM signal
and triangulation are used for inferring the position of the
phone. Finally in [7], [8] and [9], Wi-Fi and RSS was used to
determine the position. However, each of the above mentioned
methods have limitations in terms of availability (Wi-Fi access
points are not ubiquitous, especially in rural areas), precision
(positioning using GSM does not always offer sufficient accu-
racy) or energy consumption (GPS is very energy hungry). In
recent years, algorithms developed are using a combination of
sensors available on smartphones to improve the performance
of continuous location tracking. For example Ben Abdesslem
et al.[10] for their algorithm always use the most energy
efficient positioning method when available. In practice they
turn GPS on only if the accelerometer has detected a motion
and if no Wi-Fi previously geolocated are sensed. Similarly
in RAPS (Rate-Adaptive Positioning System), a framework
developed by Paek et al. [11], the accelerometer, Cell-ID/RSS
(Received Signal Strength), user activity history and bluetooth
communications are used to decide whether to turn GPS on
or not. The CompAcc, algorithm developed by Constandache
et al.[12], uses the dead reckoning method. Starting from a
known location given by the Assisted GPS, the compass and
accelerometer evaluate the displacement and allow computing
the current location. The Assisted GPS is used from time to
time for recalibration.
One of the biggest challenges when developing this kind of
algorithms is that they can only be evaluated through field
testing. Although, when a comparison with other algorithms
is needed, it is not trivial to reproduce the same conditions of
the field testing used for the algorithm taken as a reference.
Therefore a fair comparison is not possible because a common
metric is missing. Battery life is often used for comparing dif-
ferent algorithms but the drawback is that often different phone
models are used, making the comparison unfair. Moreover, the
capacity of the batteries degrades with the time and therefore,
even if the same models are used, the battery capacity may be
altered.
Another challenge for researchers is the lack of knowledge
on the energy consumption for each of the smartphones’
sensors involved in the algorithms. In fact, to the best of our
knowledge, there is no prior work in the literature investigating
this issue in an exhaustive manner.
In this paper we present the results of an extensive energy
measurements campaign on several positioning methods and
sensors available on state-of-the-art smartphones. We believe
that by knowing the detailed energy consumption of the single
sensors, the development could be faster, the algorithms more
energy efficient and most importantly, the performance of
The 8th Annual IEEE Consumer Communications and Networking Conference - Special Session on Location Aware Technologiesand Applications on Smartphones
978-1-4244-8790-5/11/$26.00 ©2011 IEEE 937
TABLE IDESCRIPTION OF ALL THE EXPERIMENTS
Mode Environment Duration Duty cycleOff On
Offline Indoor 10 h - -
Online 2G Indoor 10 h - -
Online 3G Indoor 10 h - -
Net.Based(2G - 3G)
Indoor 2 min 10 s 10 s
Net.BasedContinuous
Indoor 1 min - -
Assisted GPS(2G - 3G)
Outdoor, not moving 1h 10 min 5 min
GPS with fix Outdoor, walking 15 min - -
GPS with fix1min interval
Outdoor, walking 20 min 1 min 5 min
GPS with fix10min inter-val
Outdoor, not moving 1 h 10 min 10 min
GPS with fix1h interval
Outdoor, not moving 3 h 1 h 10 min
GPS with fix10s interval
Outdoor, walking 15 min 10 s 5 min
Cell-ID Outdoor, driving 20 min 1 min -
WLAN Outdoor, walking 20 min 10 s -
Accelero-meter
Indoor, not moving 15 min 1 min 1 min
different algorithms can be compared with a common metric.
In the first part of the paper (Sections II and III), we present
the software used in our test-bed for the measurements and
the methodology for the experiments. Then in Sections IV and
V we summarize the results and discuss the most interesting
findings.
II. MEASUREMENTS CAMPAIGN
In this section we describe the hardware, software and the
methodology used for the measurements campaign.
A. Hardware
The smartphone used for the measurements is the Nokia
N95 8Gb. As most of the smartphones on the market, it
features the most common sensors and positioning capabilities
(GPS, WLAN, accelerometer, Bluetooth, etc.).
B. Software
For measuring the power levels we used a software devel-
oped by Nokia, namely the Nokia Energy Profiler (NEP)[13].
The NEP is able to log current, voltage, power, CPU usage
and network traffic, among others, directly on the phone. The
complete test-bed was built with Python for Symbian (PyS60)
[14] as it offered APIs that allow accessing the sensors and
positioning methods and can control the NEP. To avoid user
interaction which could lead to noise in the measurements
(backlight on, for example), a Python script was used for
switching on and off the sensors as well as starting the NEP.
The data logged was first kept in the memory and, at the end
0
100
200
300
400
500
600
0 60 120 180 240 300 360 420 480
Pow
er [
mW
]
Time [s]
Power FixCPU
0102030405060708090100
CPU
Loa
d %
Fig. 1. CPU and Power measured as the GPS was on. The fix indicate whenit returned a position. The GPS was turned off after 300 seconds, but thepower dropped only after 330 seconds.
of the measurement, written in a file to avoid extra energy
consumption due to writing process on the file system.
C. Methodology
All the tests were done using the above mentioned test-bed.
To limit the noise in the measurements, all the parts not needed
for the experiment, such as Bluetooth and the screen are
turned off before measuring. The positioning methods tested
are: GPS, Assisted GPS and the Network based positioning
method, later on referred as the NET. In addition some
sensors that can be used in such algorithms were also tested:
accelerometer, WLAN scanning, Cell-ID scanning. Finally, to
compute the overhead of each positioning method or sensor,
the standby power level was also measured. Table I describes
the conditions in which each test was made. In this table, the
”Duration” refers to the total duration of the experiment, the
duty cycle is divided in two parts: the time during which the
sensor is turned off and the time during which the sensor is
on. All the tests were repeated several times and the results
presented are obtained by averaging all the measurements.
III. EXPERIMENTS DESCRIPTION
A. Positioning methods
1) GPS: The GPS (Global Positioning System) is a passive
location sensor. Nowadays the chip is so small that it can easily
fit in most of the newest smartphones. The two main issues
related to GPS are the availability of the signal and the Time
To Fix (i.e. the time from starting the GPS to the first position
returned by the system). The GPS receiver, hereafter GPS,
needs some information to calculate its position. Basically it
needs to know the timetable of all satellites (almanac), the
current trajectories of visible satellites (ephemeris) and the
time with a very high precision. The Time To Fix or TTF
depends on the state of the GPS (cold/warm/hot). T. Clark
gives a detailed explanation on his website [15]:
• Cold: the GPS has absolutely no information.
• Warm: the GPS knows already the almanac but still needs
to download the ephemeris for the satellites it needs to
listen to.
• Hot: the GPS has already all the data and can directly
connect to the satellites.
938
1 62500
data�DOWN data�UP Power Fix
0.6
0.8
1
1.2
1.4
1.6
1000
1500
2000
2500
Pow
er�[W
]
ata�
Rate
�[bit/
s]
0
0.2
0.4
0
500
0 5 10 15 20 25 30 35 40 45 50 55 60
Da
Time�[s]
Fig. 2. A-GPS using network assistance (data UP and DOWN) in the first10 seconds on the 2G network, to allow the GPS to get quickly a fix. Thepower stays then at the same level as in the GPS only experiment.
The theoretical values for the TTF are typically a few seconds
for a hot start, around 3 minutes for a warm start and more
than 13 minutes for a cold start.
In our experiment, the GPS was first started before the
experiment, to get a fix. This way, the measurements are done
in the same conditions during the whole experiment (i.e. with
a hot start).
As one can see on the Figure 1, representing one minute
interval scanning, the time off between two scans in the power
plot is only 30 seconds. The GPS was turned off by the
software after 300 seconds, but the power level only gets down
after 330 seconds. Some other experiments were done and we
found that shutting down the GPS chip on the N95 8G needs
30 seconds1. This graph shows also clearly that the GPS needs
more CPU when it has a fix than for the search phase, but this
has almost no visible impact on the power consumption.
2) A-GPS: Most of the GPS-enabled smartphones also have
an Assisted GPS or A-GPS. The purpose is to use the other
communication channels like UMTS or GPRS2 to download
the satellites data directly from the network and therefore
reduce the time to get the first fix. Most of the time the
GPS receives enough information, like the ephemeris and the
almanac, to make a hot start in a few seconds.
This test was done on the roof of a building to ensure a clear
sky and the phone was connected to the 3G. The assisted GPS
was run every 15 minutes during 1h, but the network assistance
was only used before the first fix of each interval. As soon as
the GPS is synchronized and a location is computed, it doesn’t
need the assistance anymore for the subsequent fixes. At each
new start, after 10 minutes of inactivity, the assistance was
used in average 50% of the time.
This test was repeated a second time with the network
forced to use 2G and the trace of the power levels is shown in
Figure 2. The difference between the two experiments is that
the communication time is a bit longer when using 2G due
to the slower data rate compared to 3G, but this has a quite
1This is not the case for all smartphones. For example, the GPS chip onthe Nokia 5800 does not need any extra time to shut down.
2GPRS/EDGE are defined in the 2.5G protocol and UMTS in the 3G one.But for simplicity reason and as there are only two options to force the networkin the Nokia N95 8G, we will refer to 2G and 3G networks.
2500300035004000
1500
2000
bit/
s]
W]
Power Data�UP data�DOWN Fix
05001000150020002500
0
500
1000
0 10 20 30 40 50 60 70 80
Data
�rate
�[b
Pow
er�[m
W
Time�[s]
Fig. 3. Power used for getting the location with the NET positioning methodon the 3G network. In the 10 seconds interval, it gets twice the position.
low impact on the time to get the first fix (1 second more in
average).
3) Network based: The 2G and 3G networks define a set
of base services and among them, the Location Service. The
Nokia N95 8Gb proposes an implementation of an UE-assisted
(User Equipment) positioning method, accessible through the
GPS API interface. It behaves as if it was a GPS, able to get its
location as long as the phone is in a place covered by a cellular
network, but with a lower accuracy. It highly depends on the
density of the cellular Base Stations, and it can vary from 100
meters in a city to several kilometers in the countryside. This
method is called the Network based positioning method, but,
for simplicity reason, it will be referred in the following as
NET.
During this test, the phone was indoor, not moving and with
the network forced to 2G. Each time the positioning was turned
on, it initiated the connection and, after 5 seconds, returned
the location. Then it only needed 3 seconds more for the next
value to be returned. Like for the GPS, it needs some time to
shutdown: i.e. 2 to 4 seconds. Figure 3 shows the experiment in
which the positioning was turned on and off every 10 seconds.
Using the 2G instead of 3G increased the time to get a fix
from 5 to 6 seconds and the time to shutdown is from 5 and
8 seconds.
B. Sensors
1) Accelerometer: The internal accelerometer is able to
detect the accelerations on the 3 axes (X,Y and Z). They
can be accessed by a Python API and the values are updated
at a frequency of 30 Hz. The information given by the
accelerometer can be used in a lot of different ways by location
tracking algorithms, such as for example to count the steps or
to detect user’s activity and movements.
We made two experiments: one with continuous scanning
and one with 1 minute scanning intervals because we wanted
also to know the time to shutdown, until the power level goes
back to its idle value.
As we can see in Figure 4, the accelerometer needs no ad-
ditional time to turn on and the power level drops immediately
after shutting down the sensor. Switching off the accelerometer
doesn’t cost extra energy. So the phone can afford switching
on and off very frequently.
939
200
250
300
60
80
100
mW
]
d�%
CPU Power
0
50
100
150
0
20
40
60
Pow
er�[m
CPU
�load
0 15 30 45 60 75 90 105 120 135 150 165 180 195 210 225 240
Time�[s]
Fig. 4. The accelerometer was turned on an off every minutes. The powerand CPU load dropped immediatly.
2) WLAN scanning: Each wireless access point sends bea-
cons at regular time intervals, advertising the available WLAN.
WLAN scanning can be used as an input for positioning as
in [9] and [7], by having a mapping between a WLAN and
its location. The aim of this experiment is to know how much
energy a scan costs. During the test the user was walking
across the university campus of EPFL in Lausanne. This
exposed the phone to different conditions, from places without
any WLAN to some with up to 32 different WLANs visible.
By looking at the Figure 5 it seems that the number of WLAN
has an impact on the CPU load, but less on the power level.
Despite of the other sensors, for WLAN scanning is not that
important the power level as it is the energy spent for each
scan.3) Cell-ID scanning: In the 2G/3G network each Base
Station has a unique ID and covers a certain area, also referred
as a cell. The mobile phone being connected to that base
station can query the Cell-ID and, by using a geolocalized
database, find its position. The measurement was done in a
car driving from the campus to the city. To get the cell-ID no
special scan is needed, the values are anyway logged on the
phone for handover reasons, so it is enough to read the value
from the memory which does not cost any overhead.
C. Standby mode1) Online: In all the previous measurements, the power
needed for the phone to get connected to the network is
included. As we want to compute the overhead of each sensor
and positioning method, we need to know which part is used
by the phone just to handle the network connection when in
standby. Therefore some measurements were conducted with-
out triggering any sensors and leaving the phone in standby.
The Nokia N95 8Gb, like many other modern smartphones is
able to connect to both 2G and 3G network. As they need a
different power level, the experiment was repeated for both.For this experiment the phone was indoor, not moving for
10 hours. The network mode was firstly forced to 2G and for
the second test, to 3G.2) Offline: Some measurements, such as for the GPS, were
done with the phone in ”flying mode”, without any cellular
network connection. Therefore we needed to measure the
consumption of the phone in ”flying mode” as well, so that
the net consumption for GPS could be computed.
0
5
10
15
20
25
30
35
40
0
200
400
600
800
1000
1200
200 250 300 350 400 450 500 550 600
Nbr
of W
LAN
Pow
er [m
W] a
nd C
PU %
load
Time [s]
Power CPU WLANs
Fig. 5. The CPU load of a WLAN scan depends on the number of WLAN,but the power level seemed less sensitive to that parameter.
IV. RESULTS AND DISCUSSION
In Table II, results from the measurements described in
Section III are shown. As explained on its home page the
values given by the Nokia Energy Profiler need to be adjusted3.
The values shown in Table II have been adjusted according to
the NEP guidelines.
The first column gives the average power consumption over
all measurements for each sensor. This is only the overhead,
obtained by removing the consumption measured in the tests
9, 10 or 11, depending on the case. The four next columns
give the time needed for the sensor to return the position. As
it can vary depending on how much time elapsed from the last
position returned, the column is divided into 4 parts. Typically
the Time To Fix of the GPS depends on which information is
already available and if it is still valid. The time to shutdown
is the time from the last position returned (or the time at which
the sensor was stopped) to the time at which the power level
goes again to its idle value. In fact it can happen that when the
software turns off the sensor, the hardware needs some time
to completely shut it down. The precision gives the accuracy
in case of a positioning method. And finally the battery life is
computed as if the experiment would continuously run with
a battery capacity of 4.4 Wh (as the one used by the N95
8 Gb). It is important to note that the Assisted GPS has the
same duration of the battery life as the GPS. This is due to
the assumption that A-GPS is used only to get the first fix and
then normal GPS is used. To summarize the results in Table
II:
GPS is the power-hungriest positioning method, but the
energy needed by the GPS to get a fix depends on when it was
turned off last time. Depending on the model of smartphone
used, the time needed to shutdown must be taken into account
when designing an algorithm that would turn on and off the
GPS.
A-GPS has a network assistance which helps saving time
and it has the same energy consumption as using 30 seconds of
3”For guidance, note that a Nokia Energy Profiler measured power (orcurrent) consumption of 0.05W (or 14mA) is 0.07W (or 20mA) in reality”[13].On the Forum Nokia the complete formula is given: ”Active use cases: -2%in 0.4W (100mA) region, ie. NEP mA = 0.98 x True mA. Standby accuracy:-30% in 0.04W (10 mA) region ie. NEP mA = 0.7 x True mA. This can varybetween phone models”[16]
940
TABLE IISUMMARY OF EXPERIMENTS RESULTS. THE POWER CONSUMPTION IS ADJUSTED ACCORDING TO THE NEP RECOMMANDATIONS3
.
Test Power [mW] Time (Energy) to get position [s] ([J]) Time needed toshutdown [s]
Precision [m] Battery life [hour]
last call < 10s 1min 10min 1h
Positioning methods1. GPS ∼356 3 (1.07) 30 (10.7) 40 (14.2) 200 (71.2) 30 10 11.5
2. AGPS 3G ∼973 11 (10.7) 30 10 11.5
3. AGPS 2G ∼765 12 (9.18) 30 10 11.5
4. Net 2G ∼663 3 (1.99) 6 (3.98) 6 (3.98) 6 (3.98) 6 100 - 1000 6.2
5. Net 3G ∼1150 3 (3.45) 5 (5.75) 5 (5.75) 5 (5.75) 3 100 - 1000 3.7
Sensors6. Accelerometer ∼124 NA 0 NA 28.1
7. WLAN ∼800 0.85 (0.68) 0 100 38.7
8. Cell-ID ∼0 0 0 1000 58 - 75
Standby9. Online 2G ∼60 74.8
10 .Online 3G ∼77 58.3
11. Offline ∼36 124.7
GPS to get the first fix, but only needs 11 seconds. Of course
the 2G network is a bit slower, but as it doesn’t transfer too
much data, it consumes less energy.
Net based can quickly get the position everywhere there is a
2G/3G network for a bit less energy than the A-GPS, but with a
reduced precision (up to several kilometers in the countryside).
It needs less time to get the position if the connection is
already established (<10s column).
Accelerometer can be used for very short time interval to
detects phone’s motion, without spending too much energy to
start and stop the sensor.
WLAN scanning has a high power level, but for a short
time. Then it has a very low energy consumption. By using a
database of known WLANs with their location, it become the
second more precise positioning method.
Cell-ID scanning comes for free, but with a precision not
better than the NET, if used with a database of Cell-ID and
locations.
V. CONCLUSION
This paper presents the results of an extensive measurement
campaign that was conducted aiming at measuring power
levels, speed and accuracy of several positioning methods
and sensors present on State-of-the-Art smartphones. The
measurements were done on the Nokia N95 8Gb, using the
Nokia Energy Profiler and Python scripts as test-bed. Results
summarized in Table II, can be used as a guideline for
researchers and developers to design more accurate and more
energy efficient algorithms for continuous location tracking to
enable novel Location Based Services on smartphones.
REFERENCES
[1] Foursquare, “http://www.foursquare.com.”[2] brightkite, “http://www.brightkite.com.”[3] G. Latitude, “http://www.google.com/intl/en us/latitude/intro.html.”[4] S. Barbeau, M. A. Labrador, A. Perez, P. Winters, N. Georggi,
D. Aguilar, and R. Perez, “Dynamic management of real-time locationdata on gps-enabled mobile phones,” Final Report, March 2008.
[5] A. Varshavsky, M. Chen, J. Froehlich, D. Haehnel, J. Hightower,A. Lamarca, F. Potter, T. Sohn, K. Tang, and I. Smith, “Are gsmphones the solution for localization,” in 7th IEEE Workshop on MobileComputing Systems and Applications, 2006. WMCSA’06. Proceedings.IEEE Computer Society, 2006, pp. 20–28.
[6] N. Deligiannis, S. Louvros, and S. Kotsopoulos, “Mobile positioningbased on existing signalling messages in gsm networks.”
[7] M. A. Youssef, A. Agrawala, and A. Udaya Shankar, “Wlan locationdetermination via clustering and probability distributions,” in Proc. FirstIEEE Int. Conf. Pervasive Computing and Communications (PerCom2003), 2003, pp. 143–150.
[8] O. Baala and A. Caminada, “Location precision in indoor positioningsystem,” in Proc. Innovations in Information Technology, 2006, pp. 1–5.
[9] V. Lang and C. Gu, “A locating method for wlan based location service,”in Proc. IEEE Int. Conf. e-Business Engineering ICEBE 2005, 2005, pp.427–431.
[10] F. Ben Abdesslem, A. Phillips, and T. Henderson, “Less is more: energy-efficient mobile sensing with senseless,” in MobiHeld ’09: Proceedingsof the 1st ACM workshop on Networking, systems, and applications formobile handhelds. New York, NY, USA: ACM, 2009, pp. 61–62.
[11] J. Paek, J. Kim, and R. Govindan, “Energy-efficient rate-adaptive gps-based positioning for smartphones,” in MobiSys ’10: Proceedings ofthe 8th international conference on Mobile systems, applications, andservices. New York, NY, USA: ACM, 2010, pp. 299–314.
[12] I. Constandache, R. R. Choudhury, and I. Rhee, “Towards mobile phonelocalization without war-driving,” 2010.
[13] Nokia, “Nokia energy profiler quick start,” http://www.forum.nokia.com/Library/Tools and downloads/Other/Nokia Energy Profiler/Quick start.xhtml.
[14] ——, S60 Module Reference, Nokia, 2009.[15] T. Clark, “http://gpsinformation.net/main/gpslock.htm.”[16] Forum Nokia, “http://discussion.forum.nokia.com/forum/
showthread.php?t=160912&page=6.”
941