[ieee 2011 ieee international conference on pervasive computing and communications workshops (percom...

2
COAL: Context Aware Localization for High Energy Efficiency in Wireless Networks Yanliang Liu Dept. of Computer Science and Engineering, The University of Texas at Arlington Arlington, TX 76019, Email: [email protected] Abstract—Localization is one of the principal enabling wireless services now penetrating into every aspect of our life. Unfortu- nately, present localization schemes are often inaccurate (based on signal-distance conversion), or energy inefficient (such as GPS), or labor intensive (requiring previous field signal data collection). In this paper, a novel localization scheme named COAL, for COntext Aware Localization, is put forward to aim at both energy efficiency and accuracy. The core idea is to exploit users’ context information such as an ongoing event to facilitate the localization scheme. With the facilitation of these context in- formation, localization frequency can be significantly reduced to conserve energy while preserving high degree of accuracy. Besides that, COAL is complementary to existing location schemes using any wireless signal and can be implemented as an enhancement to them. Keywords-Localization, energy efficiency, user context I. I NTRODUCTION As wireless network is penetrating into every bit of modern life, associated data services are becoming increasingly pop- ular. Localization is one of the key enabling technologies for wireless data service, where the location of a mobile user can be determined with desirable accuracy. When it comes to localization, a popular and straightfor- ward approach is GPS. However, the high energy cost often prohibits it to sustain over long time on mobile devices, and it is not applicable in indoor environment either. Another way is triangulation based on wireless signals such as WiFi, by calculating the distance between the mobile node and land- marks based on received signal strength. Unfortunately, due to inherent channel fluctuation, this estimation will produce coarse and unstable localization results. More new localization schemes have employed fingerprinting to increase accuracy [1]. In these schemes, the location of the mobile device are determined based on its received signal pattern compared to prior collected field signal data. Evidently these schemes require significant efforts in collecting training data a prior and high localization frequency to maintain accuracy. Among existing proposed localization approaches, a subset of them have specifically focused on high accuracy with constraint on energy consumption such as EEMSS [2] and CenceMe [3]. In this paper we propose a novel localization scheme termed COAL (for COntext Aware Localization), which also targets at both energy efficiency and accuracy. Our key idea is to leverage users’ context information such as an ongoing event to facilitate the localization scheme. Note- worthily, our scheme is complementary to existing location schemes using any wireless signal and can be implemented as an enhancement to them. II. COAL: CONTEXT AWARE LOCALIZATION SCHEME Below, we formally present our assumptions and ap- proaches. A. Landmarks and Signal Strength Assuming that signals from multiple WiFi access points (APs) received by a mobile terminal is denoted by an AP vector and corresponding signal strength vector AP rec =< AP rec1 , AP rec2 , ..., AP recn > S rec =<S rec1 ,S rec2 , ..., S recn > where S reci is the received signal strength for AP reci . With the AP list and signal strength vector, the mobile device can be localized by using methods already proposed in the literature, for example triangulation or fingerprinting proximity matching [1]. However, for a known environment such as a campus, this information can be used to simplify and enhance the localization process as well as achieve classroom- level accuracy without complicated calculation. B. Context Enhancement for Reduced Localization Frequency As discussed above, localization can be performed given the list of access points and proper localization schemes. However, to avoid outdated location information, conventional localiza- tion schemes have to periodically refresh the access point list, either perform mobile based computation or communicate with localization server for facilitation. These periodic operations can potentially be energy hungry and drain precious battery energy. In our scheme, we employ user context information, such as the schedule of a seminar, to reduce the frequency of these periodic operations and hence conserve energy. Since during the seminar, the user is highly likely to stay at the same location and localization can be performed at very low frequency; when the seminar is about to end, the moblie device can increase localization frequency in order to capture potential user movement. 1) Overall Operation: Firstly, a user’s location will be determined based on received WiFi signal by the grouped AP list as discussed before, and then we enhance the localization result (either in a classroom, hallway, or outside a building) by context information. If the determined location, together with current time, concur an event such as a class in the system, we will reduce the localization frequency (including rescanning WiFi signals) based on the starting and ending time of the event. Fourth Annual PhD Forum on Pervasive Computing and Communications 978-1-4244-9529-0/11/$26.00 ©2011 IEEE 401

Upload: vuongbao

Post on 26-Feb-2017

213 views

Category:

Documents


1 download

TRANSCRIPT

COAL: Context Aware Localization for High Energy Efficiency in Wireless Networks

Yanliang LiuDept. of Computer Science and Engineering, The University of Texas at Arlington

Arlington, TX 76019, Email: [email protected]

Abstract—Localization is one of the principal enabling wirelessservices now penetrating into every aspect of our life. Unfortu-nately, present localization schemes are often inaccurate (basedon signal-distance conversion), or energy inefficient (such asGPS), or labor intensive (requiring previous field signal datacollection). In this paper, a novel localization scheme namedCOAL, for COntext Aware Localization, is put forward to aim atboth energy efficiency and accuracy. The core idea is to exploitusers’ context information such as an ongoing event to facilitatethe localization scheme. With the facilitation of these context in-formation, localization frequency can be significantly reduced toconserve energy while preserving high degree of accuracy. Besidesthat, COAL is complementary to existing location schemes usingany wireless signal and can be implemented as an enhancementto them.

Keywords-Localization, energy efficiency, user context

I. INTRODUCTION

As wireless network is penetrating into every bit of modernlife, associated data services are becoming increasingly pop-ular. Localization is one of the key enabling technologies forwireless data service, where the location of a mobile user canbe determined with desirable accuracy.

When it comes to localization, a popular and straightfor-ward approach is GPS. However, the high energy cost oftenprohibits it to sustain over long time on mobile devices, andit is not applicable in indoor environment either. Another wayis triangulation based on wireless signals such as WiFi, bycalculating the distance between the mobile node and land-marks based on received signal strength. Unfortunately, dueto inherent channel fluctuation, this estimation will producecoarse and unstable localization results. More new localizationschemes have employed fingerprinting to increase accuracy[1]. In these schemes, the location of the mobile device aredetermined based on its received signal pattern comparedto prior collected field signal data. Evidently these schemesrequire significant efforts in collecting training data a priorand high localization frequency to maintain accuracy.

Among existing proposed localization approaches, a subsetof them have specifically focused on high accuracy withconstraint on energy consumption such as EEMSS [2] andCenceMe [3]. In this paper we propose a novel localizationscheme termed COAL (for COntext Aware Localization),which also targets at both energy efficiency and accuracy.Our key idea is to leverage users’ context information such asan ongoing event to facilitate the localization scheme. Note-worthily, our scheme is complementary to existing locationschemes using any wireless signal and can be implemented asan enhancement to them.

II. COAL: CONTEXT AWARE LOCALIZATION SCHEME

Below, we formally present our assumptions and ap-proaches.

A. Landmarks and Signal Strength

Assuming that signals from multiple WiFi access points(APs) received by a mobile terminal is denoted by an APvector and corresponding signal strength vector

APrec =< APrec1, APrec2, ..., APrecn >

Srec =< Srec1, Srec2, ..., Srecn >

where Sreci is the received signal strength for APreci.With the AP list and signal strength vector, the mobile

device can be localized by using methods already proposedin the literature, for example triangulation or fingerprintingproximity matching [1]. However, for a known environmentsuch as a campus, this information can be used to simplify andenhance the localization process as well as achieve classroom-level accuracy without complicated calculation.

B. Context Enhancement for Reduced Localization Frequency

As discussed above, localization can be performed given thelist of access points and proper localization schemes. However,to avoid outdated location information, conventional localiza-tion schemes have to periodically refresh the access point list,either perform mobile based computation or communicate withlocalization server for facilitation. These periodic operationscan potentially be energy hungry and drain precious batteryenergy.

In our scheme, we employ user context information, suchas the schedule of a seminar, to reduce the frequency ofthese periodic operations and hence conserve energy. Sinceduring the seminar, the user is highly likely to stay at thesame location and localization can be performed at very lowfrequency; when the seminar is about to end, the mobliedevice can increase localization frequency in order to capturepotential user movement.

1) Overall Operation: Firstly, a user’s location will bedetermined based on received WiFi signal by the grouped APlist as discussed before, and then we enhance the localizationresult (either in a classroom, hallway, or outside a building) bycontext information. If the determined location, together withcurrent time, concur an event such as a class in the system, wewill reduce the localization frequency (including rescanningWiFi signals) based on the starting and ending time of theevent.

Fourth Annual PhD Forum on Pervasive Computing and Communications

978-1-4244-9529-0/11/$26.00 ©2011 IEEE 401

2) Context Information Gathering: While numerous re-search results on context modeling and event prediction,particularly in pervasive computing domain, we employ a morestraightforward approach in order to focus on the localizationpart itself. The context information we utilize is gatheredfrom the web and further analyzed at the backend server.These information include public class schedule informationavailable through the university, public personal calender suchas Google/Outlook calendar, or even a seminar announcementavailable on a mailing list.

3) Localization Frequency Determination: Once the con-text information is obtained and matched to the user’s currentlocation, it can be used to determine required localizationfrequency. For example, when the context of a user is a class,the backend server will consider the starting and ending timeof the class to reduce localization frequency. If the class willlast an extended period of time, for example 1 hour, the user islikely to remain at the same location during this time. A muchlower localization frequency can be used here, for example,every T1 = 30 minutes, during this time period. As the classis approaching its end, the frequency can be increased in orderto timely detect the user’s movement.

C. Individual Localization Schedule

Although context information such as class schedule can bereliable sources to predict upcoming location changes whenan event ends, it is also risky to use it solely since ending ofan event is not fixed in real life. Our solution is to increasethe localization frequency of each user toward the end of theevent. Basically, localization task will be assigned to individualmobile node in a group gathering alternatively: if a fewusers detect a location change, indicating possible end of thegathering, other users will be alerted by the backend serverand the localization frequency then will be increased for all.

D. Context Prediction

In addition to the events, the schedule of which are availablefrom public calendars, ad-hoc group or events that are notin public calendars can happen at a specific location for anextended period of time as well. These can also be consideredas context information to enhance our localization scheme.

The approach we employ to detect a temporary groupgathering is based on the similarity of behavior for a numberof users at a fixed location. Basically, if multiple users sharethe same APstr and similar range Sstr for a period of time,we can predict with high confidence that they are in the sameplace and an event is in place, therefore localization frequencycan be reduced. Then, based on some context information suchas location and time, we can try to find potential events atthe same place and time from Google search results. From thesearch results, simple data mining technique will be employedto extract event information that has high probability, forexample the structural characteristic of the web pages.< table >

< tr >< td > 11/30/2010 < /td >

< td > University Center < /td >< /tr >

< /table >For structural information, we can extract the useful infor-

mation by decoding the HTML tags. For non-structural sched-ule information, we are considering more advanced algorithmsto detect the semantic relationship and then extract them out.

If there is an event taking place in the same time and loca-tion, we can reduce the localization frequency by the same wayas last section; otherwise, we can treat it as a temporary groupmeeting, assigning different value to localization frequency.

III. EXPERIMENT RESULTS

As shown below, COAL conserves 90% energy comparedto normal scheme and outperforms the EEMSS and CenceMe.

Fig. 1. Localization Times and Battery Life Comparison

IV. CONCLUSION AND FUTURE WORK

In this paper, we propose COAL, a context aware localiza-tion scheme for mobile networks to aim at energy efficiencyand high accuracy. By leveraging users’ context such as classesor events or potential users’ behavior, COAL can significantlyreduce localization frequency and hence conserve energy formobile devices as well as maintain high accuracy. Addition-ally, we design schemes that can quickly detect transient periodwhen user locations change. Both real life experiments andsimulations on synthetic data are performed on HP iPQA smartphones. We plan to conduct a medium scale test on campus.

V. ACKNOWLEGEMENT

I thank my advisor, Dr. Yonghe Liu for his insightful advicein preparing this paper.

REFERENCES

[1] A. Taheri, A. Singh, and A. Emmanuel, “Location fingerprinting oninfrastructure 802.11wireless local area networks (wlans) using locus,”in Proceedings. 29th Annual IEEE International Conference on LocalComputer Networks (LCN 2004), 2004, pp. 676–683.

[2] Y. Wang, J. Lin, and M. Annavaram, “A framework of energy efficientmobile sensing for automatic user state recognition,” in Proceedings ofthe 7th international conference on Mobile systems, applications, andservices. MobiSys ’09, 2010, pp. 179–192.

[3] E. Miluzzo, N. D. Lane, K. Fodor, R. P. H. Lu, and M. M. S.B. E. Z. T. Campbell, “Sensing meets mobile social networks: Thedesign,implementation and evaluation of the cenceme application,” inProceedings of the 6th ACM conference on Embedded network sensorsystems. SenSys ’08, 2008, pp. 337–350.

402