A Pervasive Indoor-Outdoor Positioning System

Transcription

1 70 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 A Pervasve Indoor-Outdoor Postonng System Lonel Reyero 1, Glles Delsle 1 INRS-EMT, Unversté du Québec, Montréal, Canada, H5A 1K6, lonel.reyero@telecom.com Internatonal Insttute of Telecommuncatons, Montréal, Canada, H5A 1K6 glles.delsle@telecom.com Abstract A new concept called Always Best Located (ABL) s ntroduced to acheve a pervasve ndoor-outdoor postonng system. Desgned as a combnaton of WLAN and GPS, ths hybrd postonng system s shown to yeld ubqutous coverage and best-effort accuracy n any envronment. An orgnal mappng algorthm has been conceved to enable the system to dscover the locaton of nearby WLAN access ponts on the fly, wthout requrng human nterventon or connecton to an exstng database. Followng an outlne of postonng algorthms and locaton based applcaton requrements, a comparatve study of postonng technologes based on GPS, GSM, WLAN, and Bluetooth s presented. As revealed by ths benchmark, none of the exstng technologes suffce to provde ubqutous coverage; however a combnaton GPS and WLAN s shown to provde excellent coverage and accuracy ndoors as well as outdoors. New algorthms are then presented to llustrate how these two technologes can be combned to provde a contnuous servce by swtchng from one technology to another, to accurately locate the user ndoors, and to automatcally dscover locaton of nearby WLAN access ponts. Fnally the ABL system s valdated over tme by several users and s shown to yeld 97% avalablty the equvalent of more than 3 hours of servce aday. Index Terms postonng, locaton, GPS, WLAN, GSM, ubqutous I. INTRODUCTION Wreless postonng has enabled computer systems to calculate ther own geographc poston. Ths has lead to the apparton of a wde set of applcatons of whch navgaton, fleet management, and local search are just a few examples. Wreless postonng s now seen as an enabler whch s part of most of new generaton applcatons. Numerous wreless postonng systems have been developedforallkndofcommuncatonnterface FM rado, GSM networks, WLAN networks, Bluetooth networks, RFID devces, UWB rado, and even ultrasound. Despte ths large number of systems, creatng a locaton-aware applcaton for the masses s stll a techncal challenge. Most of the exstng postonng systems operate n a confned envronment, such as nsde a buldng, and therefore they are not yet able to offer a servce n any place users may travel to, ndoors as well as outdoors. Moreover, exstng systems requre nvestment n a new nfrastructure as well as a complex calbraton phase. In our opnon, an deal postonng system should be capable of trackng the user locaton n any place he or she may travel to, ndoors as well as outdoors. Smlarly to the prncple of % avalablty of telecommuncaton networks, postonng systems should offer a servce at any place and any tme users may travel to. Moreover, as applcatons value ncreases wth respect to the accuracy of the postonng, an deal postonng system should offer an accurate servce, ndoors as well as outdoors. Fnally, an deal postonng system should use an exstng nfrastructure, and should have a calbraton phased reduced to the mnmum. Our objectve s to create such a postonng system, by followng a hybrd approach consstng n combnng exstng technologes to offer an accurate and ubqutous servce [1]. Our research methodology starts by evaluatng accuracy and coverage of postonng based on four exstng technologes GPS, GSM, WLAN, and Bluetooth n dfferent envronments of the cty of Montreal, Canada. These measurements have shown that GPS and WLAN offer the most complementary accuracy and coverage, and have therefore been selected to develop our hybrd postonng system. Ths system has been developed as the combnaton of three dfferent postonng components. The frst component s a place detecton algorthm, based on WLAN, offerng a room level accuracy nsde buldngs. GPS s used as the second component to offer an excellent accuracy and coverage n outdoors envronments. Fnally, our thrd component s a WLAN postonng algorthm whch s used n unfamlar ndoors areas and outdoors where GPS s unavalable. Sophstcated dscovery algorthms have been created so that WLAN access ponts can be located automatcally as users go ahead wth ther daly actvtes. All of these components have been separately tested before beng ntegrated n a fnal prototype, whch has been carred by three partcpants for a week, n order to valdate ts ntended ubqutous coverage. II. POSITIONING APPROACHES,ALGORITHMS AND TECHNOLOGIES A. Postonng Approaches Postonng systems are ether desgned to be networkbased or termnal-based, as llustrated on fg. 1. Network-based postonng systems are typcally runnng on a server n a telecommuncaton network and are usng measurements from several antennas to calculate users locaton, e.g., GSM base statons or WLAN access ponts. These systems are usually more accurate, as they have access to fne graned data such as

2 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER the antenna locaton, tlt, transmsson power, whch termnal-based systems have not. In addton, networkbased systems are more relable as the nfrastructure s managed by a sngle company. However, these systems may suffer from a lmted coverage, as they can only operate where the company network s deployed. Consequently they suffer from scalablty, as they requre deployment of new nfrastructure to extend the coverage. Termnal-based systems calculate the user locaton on the termnal tself by usng the avalable communcaton nterfaces to make rado measurements. Ths approach has the advantage to use the exstng nfrastructure of GSM antennas or WLAN access ponts. However, the locaton of these antennas has to be known so the user locaton can be nferred from ther detecton. Creatng a database of known antenna locaton s one of the man ssues of the termnal-based approach. It s often bult by dscoverng antennas locaton whle drvng n each and every street, whch s tme consumng and lead to scalablty ssues. Termnal-based postonng systems are also more flexble, and can easly combng several technologes such as GSM, WLAN, and Bluetooth whch are present on most hgh-end termnals. All research results n ths paper are obtaned by followng a termnal-based approach. Postonng algorthms based on proxmty consst n assocatng the user locaton to the locaton of a reference pont. For nstance, the most popular proxmty algorthm s referred to Cell-ID n GSM networks. It calculates the user locaton as the locaton of the GSM base staton to whch the user s currently connected. Accuracy of these algorthms s functon of the dstance at whch a reference pont can be detected, whch s typcally between 100 meters and 10 klometers for GSM networks. Trlateraton Trlateraton algorthms calculate the user locaton from a measure of the dstance between the user and 3 or more reference ponts. Ths dstance can be nferred from a measure of the sgnal strength or the tme dfference of arrval between sgnals comng from the three reference ponts. Accuracy of these algorthms s functon of the dstance at whch reference ponts can be detected, the accuracy of the dstance measurements, and the qualty of the algorthm. User locaton s calculated by solvng equaton system llustrated n fg.. Fgure 1. Termnal-based and network-based postonng B. Locaton types Geographc nformaton usng a broad varety of expressons such as a name, an address, an ntersecton, a descrpton, a dstance, or a coordnate, s communcated everyday. These expressons can be sorted n three man categores of locaton: absolute, relatve, and symbolc. Absolute locatons are coordnates whose orgn s the center of the Earth and whch can be used to represent any locaton on the surface of the planet. Relatve locatons are coordnates relatve to a buldng or a dstrct, whch can be used to represent locatons nto that lmted area. Symbolc locatons are names, tags, or any addresses that can be understood by users, such as Home or Work for nstance. These types of locaton are usually more meanngful n nterpersonal communcatons. r r r 1 3 x + y Fgure. Trlateraton algorthm ( x ) + ( y j) ( x k) + ( y l) Trangulaton Trangulaton dffers from trlateraton by relyng on the measurements of angles nstead of dstances to nfer the user locaton. As shown on fg. 3, the user locaton s calculated by measurng the sgnal angle of arrval from or more reference ponts. Accuracy of these algorthms depends on the dstance at whch reference ponts can be detected, as well as the qualty of the angle measurements and of the algorthm. The user locaton s calculated as the ntersecton of two lnes, as represented n fg. 3. (1) C. Postonng Algorthms Proxmty y y ( x ) sn( α ) ( x k) sn( β ) + j + l ()

3 7 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 Fgure 3. Trangulaton algorthm Centrod Postonng algorthms of type centrod calculate the user locaton as a weghted average of the locaton of several reference ponts. A weght w s calculated for each detected reference pont as a functon of the receved sgnal strength, the reference pont coverage, or the relablty of the reference pont. Each algorthm may calculate t n a dfferent way. Then, the user locaton s calculated as a weghted average of the detected reference ponts locaton (x,y ), as descrbed n fg. 4. x y Fgure 4. Centrod Algorthm w x w y Accuracy of these algorthms depends on the number of reference ponts used, the dstance at whch these reference ponts are detected, as well as the number of nformaton used to calculate the weght of each reference pont. These algorthms have the advantage to be smple and to requre mnmal computatonal power, whch make them especally sutable for moble devces. Place Detecton Place detecton algorthms sgnfcantly dffer from the other types of algorthms, as they do not calculate the user locaton as a coordnate from the detecton of reference ponts of known locaton. Place detecton algorthms determne the user locaton as a place recognzed among a lst of known places. These algorthms operate n two phases: place defnton and place recognton. Durng the place defnton phase, a place detecton algorthm creates a profle for each place to be recognzed. A place profle contans a sgnature of the place, unque nformaton that dfferentates t from any other. Ths can be create from sgnal strength measurements, or smply contans the lst of antennas dentfers detected from the place n queston. Durng the place recognton phase, rado measurements are recorded and compared to each place profle. The probablty the user s located at each of these places s calculated, and the user s consdered to be located at the place of hghest probablty. These algorthms have the advantage not to requre a large database of known reference ponts locaton. However, they may also suffer from a lack of coverage, snce they wll only operate when the user located at one of the known places. (3) D. Technologes Global Postonng System (GPS) GPS s a global postonng system developed by department of defense of the Unted States of Amerca []. A constellaton of 4 satelltes transmt sgnals contanng ther locaton as well as accurate tmng nformaton. A recever s able to calculate ts own locaton by measurng the tme dfference of arrval of the sgnals comng from at least 3 satelltes. As recevers requre a drect lne of sght to the satelltes, GPS does not operate ndoors. It also operates wth hgher postonng errors n urban centers, due to multpath caused by sgnal reflecton on large buldngs. Ths s a serous drawback for socal applcatons targetng end users, as t s ndoors and n urban centers that users are spendng most of ther tme. GPS coverage s typcally avalable less than 10% of the tme, durng a day of a regular user [3]. Therefore GPS s napproprate to develop socal applcatons requrng to a postonng servce n any place user may travel to. However, GPS global outdoor coverage has made t a success for navgaton and fleet management applcatons. GPS offers an excellent accuracy outdoors, whch s typcally nferor to 10 meters n more than 95% of the tme [4], but t can easly reach 50 meters of naccuracy n urban centers. Improvement systems such as DGPS and WAAS make use of terrestral antennas and satelltes to broadcast a correcton sgnal to the recevers. WAAS recevers typcally yeld errors lower than 3 meters n more than 95% of the trals [4]. Cellular network based postonng Wreless postonng based on cellular networks conssts n usng base statons as ponts of reference to locate the wreless termnals. There are numerous wreless postonng systems such as E-OTD (Enhanced- Observed Tme Dfference) for GSM [5], A-FLT (Advanced-Forward Lnk Trlateraton) for CDMA 000, whch are usng one or more base statons as pont of references. Ther accuracy ranges from 100 meters to 800 meters, dependng on the base staton densty. Ths relatvely low accuracy makes these systems napproprate to navgaton applcatons. However, the ubquty of cellular networks n urban centers, make t a great soluton for local search applcatons and nterpersonal communcaton. The nvestment requred to deploy the postonng systems, as well as the per request charge appled by network operators, have slowed down the adopton of locaton-aware applcatons. However, the Enhanced-911 rules adopted n the Unted States by the FCC, mposng 97% of emergency calls to be located wthn 50 to 300 meters, s helpng to accelerate wreless postonng systems adopton. WLAN based postnong WLAN access ponts too can be used as ponts of reference to locate wreless termnals. Numerous postonng systems based on WLAN have been subject to research and even commercalzaton [6], [7], [8], [9],

4 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER [10], [11]. Most of these postonng systems are usng trlateraton or fngerprntng algorthms based on sgnal strength measurements. WLAN access ponts short coverage enables these systems to be very accurate ndoors, wth mean errors between 1 and 0 meters dependng on the algorthm used. Ths excellent accuracy coupled wth WLAN nfrastructure s low cost make WLAN based postonng an excellent soluton for ndoors enterprse applcatons such as asset and staff trackng. WLAN postonng systems have also been developed for outdoor envronments [3], [1], [13], [14]. Snce the massve adopton of WLAN technology, access ponts have been deployed n enterprse, resdences, coffee shops, and publc places. In urban centers, several access ponts have overlappng coverage at any gven place, thus enablng a contnuous postonng servce n these urban areas. However, ths coverage rapdly dsappears as the populaton densty lowers, such as n rural areas. In order to use WLAN postonng outdoors, the locaton of detected access ponts has to be known. Obtanng the locaton of all access ponts n a cty s a complcated, tme consumng process, whch requres drvng n each and every street to detect and nfer access ponts locaton. TABLE I. Applcatons LOCATION-AWARE APPLICATIONS REQUIREMENTS Requred coverage Requred accuracy Adequate postonn gsystem Navgaton Outdoor ~0m GPS Fleet Management Outdoor ~10m GPS Inventory and staff trackng Interpersonal communcaton Indoor ~5m WLAN, RFID Indoor, outdoor ~5m - Local Search Indoor, outdoor < 100m GSM Gamng Indoor, Outdoor ~30m - Vrtual Tours Outdoor ~30m GPS Local nteracton Indoor ~10m - Local advertsng Indoor, outdoor ~50m - E. Applcatons A synthess of exstng locaton-aware applcatons, as well as ther requrements n terms of postonng accuracy and coverage, s presented n Table I. As t appears, most professonal applcatons can be acheved by exstng postonng systems. However, lots of end consumer applcatons cannot be bult on exstng postonng systems. They requre an ubqutous coverage, both ndoors and outdoors, whch cannot be provded wth the requred accuracy by exstng postonng systems. F. Hybrd Postonng Technologes The prncple behnd hybrd postonng s to recognze that there s no sngle postonng technology able to accurately locate users ndoors as well as outdoors. However, exstng technologes are complementary, and can be combned to meet these objectves. For nstance, GPS operates very accurately outdoors but does not ndoors, whle WLAN based postonng s accurate ndoors, but lack of coverage outdoors. A hybrd postonng system combnng these two technologes could then offer coverage both outdoors, usng GPS, and outdoors, usng WLAN. Ths approach appears deal to meet requrements of end-user applcatons requrng a postonng servce n any place users may travel to. To go even further, an deal postonng system should provde the followng specfcatons: Ubqutous coverage Outdoor accuracy Indoor accuracy Use exstng nfrastructure Mnmal confguraton Our objectve s to create a hybrd postonng system combnng exstng technology whch s gong to meet these requrements. III. EXPERIMENTAL EVALUATION OF POSITIONING TECHNOLOGIES AND ASSOCIATED ALGORITHMS As presented n the prevous secton, there s no postonng technology able to offer an accurate postonng n any place users are travelng to. However, exstng technologes seem to be complementary n term of coverage and accuracy. For nstance, GPS operates accurately outdoors but s out of coverage ndoors, whle WLAN s accurate ndoors, but lack of coverage outdoors. Therefore, t seems ntutve that combnng technologes can lead to a ubqutous postonng system. In order to valdate ths statement a coverage and accuracy survey of four exstng technologes namely GPS, GSM, WLAN, and Bluetooth have been conducted n dfferent envronments. The obtaned results have revealed the most complementary technologes, whch would be best combned together. A. Methodology Expermental setup Expermental data are recorded usng a laptop computer, equpped wth WLAN and Bluetooth adapters, and connected to a GPS and a GSM phone. Fg. 5 llustrates the expermental setup. An applcaton, developed for the experment, records wreless measurements for each technology n a separate trace fle. The applcaton records every second the GPS locaton, the number of GPS satelltes, the dentty and sgnal

5 74 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 strength of detected WLAN access ponts, the dentty and sgnal strength of detected GSM base statons, and the dentty and type of detected Bluetooth devces. Bluetooth scannng beng sgnfcantly slower, t was only completed at a rate of one scan every 15 seconds. Procedure Outdoors, the equpment s placed n a car, whch s drven as slowly as possble across the dfferent expermental envronments. Traces are recorded twce n each envronment. The frst trace s used to learn where the detected WLAN access ponts and GSM base statons are located n order to create a database of known wreless reference ponts. The second trace s used to compute the user locaton, and estmate coverage and accuracy for each technology. Coverage s calculated as the percentage of samples for whch a vald locaton could be calculated. Accuracy s calculated as the dstance measured between the estmated locaton and the one obtaned usng GPS. Indoors, the equpment s placed on a tray and s moved at walkng speed. As GPS s not avalable ndoors, t could not be used as a reference to estmate accuracy of the other technologes. Therefore, accuracy has not been measured ndoors, and a sngle trace has been recorded n every envronment. Coverage has been measured as the percentage of samples for whch a wreless reference pont has been detected. Envronments Outdoors, accuracy and coverage have been measured n 5 dstnct envronments of varyng populaton densty: downtown, resdental, suburban, countrysde, and hghways. Indoors, coverage has been measured n dfferent types of buldng ncludng campus, lbrares, shoppng centers, offces, transportaton hub. A total of 5 hours of measurements have been recorded. B. Algorthms Unlke GPS, wreless nterfaces for WLAN, GSM, and Bluetooth do not drectly compute the user locaton. These nterfaces only reveal the dentty and sgnal strength of nearby wreless beacons. Therefore, a postonng algorthm has to be developed to calculate the user locaton from the measured nformaton. An algorthm of type centrod has been developed and used for WLAN, GSM, and Bluetooth. Ths algorthm requres the locaton of the detected wreless reference ponts n order to operate. So our frst step s to develop an algorthm to create a database contanng the locaton of these reference ponts. Fgure 5. Expermental setup Buldng the database of known reference ponts The frst phase s to dscover the locaton of reference ponts namely WLAN access ponts, GSM base statons, and Bluetooth devces present n our expermental envronments. In order to do t, an automatc algorthm that calculates reference ponts locaton from a detecton trace and assocated GPS locatons has been deployed. For the sake of smplcty, the algorthm s presented for the partcular case of locatng WLAN access ponts, but t s also appled to GSM base statons and Bluetooth devces. As the car carryng the expermental equpment s drvng on a street, t s gong to detect a WLAN access ponts several tmes, as llustrated on fg. 6. Each tme ths access pont s detected, the sgnal strength s recorded along to the current GPS locaton. The access pont s then supposed to be located where t has been detected wth the strongest receved sgnal strength. So the access pont locaton s not based on a sngle GPS locaton, a mean of the GPS locatons recorded for the 10% strongest sgnal strength s used. The estmated locaton of ths access pont s then recorded n a database along to the access pont dentty, ts estmated coverage, ts maxmum measured sgnal strength, and ts total number of detecton. The estmated coverage s calculated as the dstance between the estmated access pont locaton and the locaton of the detecton the furthest away. Ths process s used to locate WLAN access ponts, GSM base statons, and Bluetooth devces usng the trace recorded n each expermental envronment.

6 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER Fgure 6. Dscovery of a WLAN access pont. Calculatng the user locaton Once the wreless reference ponts have been located, usng our frst trace, the user locaton can be calculated for each technology, usng the second trace. An algorthm of type centrod s used to calculate the user locaton, as a weghted mean of the locaton of detected wreless reference ponts, as llustrated on fg. 7. Our algorthm calculates the weght gven to a reference ponts as a functon of ts receved sgnal strength, ts coverage, and ts total number of detecton. Ths generc algorthm s detaled below for the partcular case of calculatng the user locaton usng WLAN, but t s used smlarly for GSM and Bluetooth. Fgure 7. Centrod Algorthm Let s suppose the user termnal detects M WLAN access ponts, among whch only N have known locaton. Let x and y be the lattude and longtude of access pont, belongng to [1, N] and let w be the weght assocated to t. Then the user locaton s calculated as: X N 1 w x N ; Y w y (4) 1 The weght w of an access pont s calculated as a functon of an estmaton of the dstance d separatng the access pont and the user, an estmaton of the coverage c of the access pont, and an estmaton of the access pont relablty f, as represented by the followng expresson: w d. c. (5) The varable d s an estmaton of the dstance between the access pont and the user. Ths dstance s calculated usng the expresson f p MAX p 10.σ d d0 10, (6) where p s the sgnal strength of access pont measured by the user termnal, and p MAX s the maxmum sgnal strength of access pont, as stored n the database of known access ponts. Varable d 0 s chosen to be equal to 5m and σ to be equal to. The greater s the dfference between the measured sgnal strength p and ts known maxmum p MAX, the longer s the dstance d between the user and the access pont. The hgher s the dstance d, the lower s the weght w assocated to the access pont. The varable c s an estmaton of access pont coverage, as t s stored n the database of known access ponts. Coverage s sgnfcant nformaton, snce the smaller t s, the smaller the uncertanty of the user locaton s. Therefore, the smaller the coverage of access pont s, the greater the weght w assocated to t s. Fnally, the varable f s an estmaton of the relablty of the data related to access pont stored n the database of known access ponts. The locaton, maxmum receved sgnal strength, and coverage of access ponts are not exact values. They have been emprcally measured durng the dscovery phase, and thus may be largely erroneous for some access ponts. The varable f s used to dmnsh the weght of access ponts wth such unrelable data. An access pont s consdered to have relable data f t has been detected a suffcent number of tmes durng the dscovery phase. For nstance, an access pont detected a sngle tme durng the dscovery phase must have an estmated locaton and coverage largely erroneous. The varable f followng expresson: f max ( detecton, f ) s gven the value of the max (7) where detecton s the total number of detecton of access pont durng the dscovery phase, and f max s a threshold set to 0, above whch all access ponts are judged to be equally relable. Ths expresson has the effect to lower the weght wth data nferred from less than 0 detectons durng the tranng phase. C. Results GPS As ground truth was not used durng the experment, GPS accuracy could not be determned. The fdelty of the measurements has been calculated nstead, as the mean dstance between the two traces of GPS measurements for each envronment. Results of Table II confrm that GPS s an excellent outdoor postonng system, as t offers a complete outdoor coverage as well as an excellent accuracy. Even downtown, where large buldngs are causng multpath, vald locaton could be obtaned at any tme. However, postonng was not as accurate as n other envronments, and the error was hgher than 50m n certan streets, as shownnfg.8.

7 76 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 Fgure 8. GPS postonng error n dense urban envronment Indoors, satelltes sgnals have only been detected for 7% of the measurements and have never been strong enough to compute a vald locaton. GSM Accuracy has been measured wth GSM by usng two dfferent algorthms. The frst s trval and assocate the user locaton to the locaton of the cell tower t s currently connected to. The second algorthm s the centrod algorthm presented above, whch make use of several cell towers n the user vcnty. GSM s the only technology to offer a ubqutous coverage n all consdered envronments. However, the obtaned accuracy s much lower than other technologes, wth a mean error between 150 and 900 meters. Accuracy s manly dependent on the base staton densty. Usng several nearby cell towers nstead of one yelds an ncrease n accuracy between 0% and 30%, whch does not change the scale of the results. TABLE II. ACCURACY AND COVERAGE OF WIRELESS POSITIONING BASED ON GPS, GSM, WLAN, AND BLUETOOTH. Outdoors GPS Indoor 0% - Downtown 7m Resdental 3m Suburban m Countrysde 5m Hghways 5m WLAN Results obtaned for WLAN shows that t s a good postonng technque n urban areas, as accuracy s below 50m and coverage s almost complete. Bufferng s an mportant factor n the coverage measurements. A typcal WLAN adapter s lstenng for the presence of access ponts beacon frames durng a short tme nterval on each channel, usually between 50 and 10 ms. As beacon frames are sent perodcally, usually every 100ms, there s a hgh probablty the WLAN adapter wll not be lstenng to the rght channel when the beacon frame wll be transmtted. Consequently, a WLAN adapter wll not detect all surroundng access ponts n a sngle scan. Bufferng the results of several scans s necessary to obtan a complete lst of these access ponts. Usng a sldng buffer of several scans has for consequence to ncrease the number of access ponts detected, and therefore to ncrease the coverage of the wreless postonng system. Ths bufferng technque sgnfcantly mproves the coverage measured n the expermental envronments. Bluetooth The obtaned results show there s not enough Bluetooth devces n the envronment to provde a sgnfcant coverage for a postonng system. No Bluetooth devces have been detected outdoors. Indoors, Bluetooth devces have been detected for only 31% of the scans. Furthermore, around 90% of these Bluetooth devces were moble devces that can not be used as a reference pont for postonng. Therefore, there are not enough statc Bluetooth devces n the envronment that could be used to develop a wreless postonng system. C) Toward an ubqutous postonng system Accordng to the results n Table II, t appears that none of the consdered technology s able to offer an accurate and ubqutous postonng servce by tself. However, GPS and WLAN provde complementary coverage GPS s excellent outdoors whle WLAN operates best ndoors and n urban centers and could therefore be combned to create a ubqutous postonng system. Ths s the approach of our Always Best Located postonng system. GSM Cell-ID GSM Centrod WLAN same as outdoors 19m ~5m a 177m 98% ~5m a 6m 384m 301m 7m 438m 318m 9% 45m 941m 680m 46% 77m 704m 57m 73% 155m IV. OUR APPROACH Our approach s to create a hybrd postonng system combnng GPS and WLAN to offer an accurate and contnuous postonng servce n any place user may travel to. Three dfferent postonng methods have been developed on top of these two technologes and have been ntegrated nto a sophstcated swtchng mechansm to locate the user wth the most accurate method avalable. Bluetooth 31% 0% ~m a - 0% - a accuracy has not been measured ndoors. These values are gven as examples of what has been obtaned by prevous research projects [15], [16], [17], [18]. 0% - 0% - 0% - A. The Always Best Located (ABL) prototype Always Best Connected s a popular concept n new generaton networks consstng n a moble termnal s

8 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER ablty to automatcally select the best connecton avalable to communcate. For nstance, a mult-mode termnal s able to automatcally swtch between UMTS to WLAN to offer the fastest connecton avalable. By analogy, the term Always Best Located (ABL) have been defned as the ablty to automatcally select the best postonng method avalable. Ths s the ablty we have ntegrated nto our prototype, called ABL, to automatcally swtch between three postonng methods: place detecton, GPS postonng, and WLAN postonng, to offer the best accuracy and the broadest coverage. Place detecton, whch s used by default, makes use of WLAN to recognze the place at whch the user s currently located among a lst of user defned places. Ths postonng method s very accurate, and can be used to dstngush the room n whch the user s currently located. Users are requred to create a profle for each place to be recognzed. Therefore, place detecton s only used n famlar places, such as home, work, and anywhere users are spendng a sgnfcant amount of tme. If the user s currently not located at a known place, the prototype automatcally swtches to GPS postonng, whch s used to provde excellent accuracy and coverage outdoors. If GPS coverage s unavalable, WLAN postonng s used. Ths postonng method s relyng on the centrod algorthm presented n the prevous secton to calculate the absolute user locaton. Selected as a last resort, WLAN postonng s used n any area deprved of GPS coverage and user defned places. Each of these postonng methods s summarzed n Table III, and the swtchng mechansm ntegratng them together s llustrated n fg. 9. Detaled operaton of the place detecton and WLAN postonng methods are presented n the followng secton. The fnal secton of the paper emphaszes the ABL prototype and an experment to valdate ts coverage n real stuatons. TABLE III. POSITIONING METHODS OF THE ABL PROTOTYPE Place detecton GPS postonng WLAN postonng Algorthm Hstogram model None Centrod Postonng type Symbolc Absolute Absolute Accuracy ~ room level ~ 5m ~ 30m Envronments Famlar places, e.g., home, work Outdoors Anywhere else V. INDOOR POSITIONING AND PLACE DETECTION Place detecton s the man postonng method of our prototype. It s used to dstngush the place, or room, n whch the user s locaton among a lst of user defned places. As numerous place detecton algorthms have already been subject to ntensve research [7], [8], [19], [0], t was decded to mplement some of them and compare ther accuracy n a real stuaton n order to select the most accurate algorthm. The frst part of ths secton presents the analytcal foundaton of these algorthms, and the second one presents the comparatve accuracy expermentaton. Fgure 9. Operaton of the Always Best Located prototype. A. Place detecton algorthms Algorthms presented n ths secton have already been presented [30] and only the mnmum essental to understand our paper s presented here. The three dfferent algorthms compared n our experment are varants of a generc place detecton algorthm, whch operates n two phases: place defnton, and place recognton. The frst phase, place recognton, conssts n defnng a profle for each place to be recognzed. Ths profle contans a mathematcal of the receved sgnal strength of each access pont detected at the current place. Ths model s represented by the followng expresson, f l, ( x), (8) where, f l, (x) returns the probablty an access pont s measured wth sgnal strength x at the place l. The profle of a place l s the set of models obtaned each access pont detected.

9 78 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 The second phase, place recognton, conssts n recognzng the place n whch the user s located among the lst of defned places. The algorthm compares the current WLAN measurements to the place profles to calculate the probablty the user s located n each place l, and assumes he s located n the place of hghest probablty. In order to do t, the probablty P(L O) s calculated, where L s the event correspondng to the user presence n place l, ando s the event correspondng to the current WLAN measurements. P(L O) s calculated usng the followng expresson, P O L ( L O) ( ) P( O) P. (9) As P(O) s constant for any place l, tsdsmssed.only P(O L) s calculated, whch s the probablty of obtanng the current WLAN measurements at the place l. The event O, whch s represent the current WLAN measurements, s defned as the vector <o 1,o,o 3,..o, o M > where o s the sgnal strength measured for access pont. Then,P(O L) s calculated by the followng expresson, M ' ( L) f l ( o ) P O 1, (10) whch s equvalent to calculate the product of the probablty of measurng each access pont wth sgnal strength o at place l. P(O L) s calculated for every place l, and the user s assumed to be located n the place of hghest probablty. Three varants of ths algorthm, each dfferng by the functon f l, (x) used to model the access ponts sgnal strength dstrbuton have been mplemented. The three varants respectvely use a Gaussan model, a hstogram model, and a Kernel model. The mathematcal detals can be found n [8]. B. Expermentaton Method The objectve of ths experment s to compare the accuracy of the three algorthm varants n order to select the most accurate one to be ntegrated nto our postonng prototype. Ths experence has been conducted n the offce of the Internatonal Insttute of Telecommuncatons (IIT), whch are composed of small offces, a few large meetng rooms, and 5 WLAN access ponts, whch have not been moved or added for ths experment. A total of 18 places have been defned for the experment. For each place, 400 samples of the access ponts sgnal strength have been recorded, and used to create the place model for each varant of the algorthm. Once the places have been defned, a moble termnal s moved from a place to another and ts locaton s calculated usng each of the algorthms varant. The moble termnal s recordng nearby WLAN access ponts sgnal strength, and calculates t own locaton usng the three algorthms. The place recognzed by each algorthm, s compared to the actual user locaton, and the accuracy of an algorthm s calculated as the percentage of trals leadng to successful place recognton. The moble termnal s moved across the offce space and remans mmoble at least one mnute n each of the 18 places defned. The samples recorded whle the moble termnal was movng n between two defned places are dsmssed. Results As llustrated n Table IV, the obtaned results show that the algorthm relyng on a hstogram model of the access ponts sgnal strength dstrbuton yelds the best accuracy, by recognzng the correct place n 87% of the trals. Therefore, ths algorthm has been selected to mplement the place detecton component of our prototype. The measured accuracy s lower than what was reported n exstng publcatons [7], where 95% of successful place recognton has been obtaned. However, ths dfference can be explaned by the number of access ponts present n the envronment. For nstance, [7] have used more than 30 access ponts, whle only 5 access ponts have been used n ths experment. As our postonng system s targetng end users, t s also hghly probable they wll not have to defne as many places as we dd. Indeed, users wll certanly only defne places where ther actvtes are the most ntense. Consequently, the obtaned accuracy wll rse as the number of place s reduced. TABLE IV. ACCURACY OF EACH PLACE DETECTION ALGORITHM Algorthm varant Accuracy Gaussan Model 74% Hstogram Model 87% Kernel Model 80% VI. WLAN POSITIONING WLAN postonng s used to calculate the absolute user locaton, lattude and longtude, when the user s n an area deprved of GPS coverage and of user defned places. Ths postonng method uses the sgnal strength of nearby access ponts to calculate the user locaton usng the centrod algorthm presented n secton III. The locaton of detected access ponts has to be known so ths algorthm can be used. Usually, the locaton of these access ponts has to be manually dscovered by drvng n each and every street. In order to avod ths tme ntensve phase, automatc dscovery algorthms have been desgned, enablng the moble termnal to buld ts own database of known access ponts. These dscovery algorthms are executed n parallel to the postonng operatons, so the newly dscovered access ponts are fed nto the database and can be mmedately used by postonng algorthms. Two algorthms have been desgned. The frst algorthm s relyng on GPS to nfer nearby access ponts locaton, whle the second algorthm s able to operate ndependently to dscover access ponts locaton n GPS deprved areas such as nsde buldngs.

10 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER A. Dscovery Algorthms Dscovery wth GPS Dscovery wth GPS s performed usng an ordnary algorthm, such as the one presented n secton III. Ths algorthm perodcally scans for nearby access pont. Every tme an access pont s detected ts sgnal strength s measured and assocated to the current GPS locaton. An access pont s then assumed to be located where t has been detected wth the strongest sgnal strength, as llustrated n fg. 6. Ths algorthm operates very well and enable smple dscovery of access ponts where GPS coverage s avalable. However, when GPS s out of coverage, a new algorthm has to be used. Dscovery wthout GPS Dscovery wthout GPS s performed usng a newly desgned algorthm operatng n three stages: creaton of a meshed network of proxmty lnks, calculaton of the shortest path, and calculaton of the access ponts locaton. Stage 1 conssts n creatng a meshed network of proxmty lnks, whch represent the topology of WLAN access ponts n the envronment. In ths stage, the algorthm perodcally scans for nearby access ponts. If two access ponts are detected smultaneously, or at a small tme nterval, a proxmty lnk s created between each of them. A proxmty lnk s just vrtual nformaton meanng that two WLAN access ponts are close to each other. Indeed, f two access ponts are detected smultaneously, t means ther coverage overlaps and therefore they are located at a close dstance from each other. In addton, each proxmty lnk s gven a cost whch represents an estmaton of the dstance separatng the two access ponts. Ths cost c s calculated as the sum of dstance d 1, n between the user and access pont 1, and dstance d, n between the user and access pont, as n the followng expresson, c d + d d.10 P0 P1 P0 P 10µ + 10µ d (11) P 1 and P are the measured sgnal strength of access pont 1 and. The other varables have been emprcally chosen so that do s equal to 0.3m, P 0 s equal to -0 dbm, and s equal to 3. For nstance, f two access ponts are smultaneously detected wth receved sgnal strength -60 dbm and -80 dbm, the cost of the proxmty lnk created n-between them s equal to the sum of dstance d 1, 6.5m, and dstance d, 30m, whch adds up to 36.5m. Every tme these two access ponts are detected smultaneously the cost s calculated, and only ts mnmal value s recorded. Of course, the cost s not an accurate estmaton of the dstance separatng the two access ponts, as the presence of concrete walls or metal objects may alter the receved sgnal strength even over a short dstance. Some of these access ponts have a known locaton, obtaned usng the dscovery algorthm wth GPS, whle others have an unknown locaton, as they have only been detected out of GPS coverage. These lnks connected access ponts of known locaton to access ponts of unknown locaton are systematcally created as users travel n and out of the GPS coverage. In fne, the dscovery algorthm wthout GPS has created a meshed network of access ponts coverng the area out of GPS coverage, such as llustrated n fg. 10. The followng stages of the algorthm consst n usng ths meshed network to nfer the locaton of unknown access ponts. Stage conssts n calculatng for each access pont of unknown locaton the shortest path to the access ponts of known locaton, as llustrated n fg. 11. The cost of the shortest path s also calculated, therefore provdng a noton of the dstance between the access pont of unknown locaton, and the surroundng access ponts of known locaton. Shortest paths are calculated usng Djkstra [1]. Overall, ths stage conssts n fndng for each access pont of unknown locaton the reference ponts that wll be used to nfer ts locaton. Fgure 10. Stage 1: creaton of a meshed network of access ponts. Fgure 11. Stage : Shortest path calculaton to access ponts of known locaton Stage 3 conssts n calculatng the locaton of access ponts of unknown locaton by usng the nformaton obtaned at stage. A postonng algorthm of type centrod s used to nfer the locaton of each unknown

11 80 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 access pont as the weghted average of the locaton of known access ponts lnked to t. The weght gven to each access pont of known locaton s equal to the cost of the shortest path to that access pont, as calculated at stage. Ths algorthm enables our dscovery process to automatcally nfer the locaton of nearby access ponts even n areas out of GPS coverage. The dscovered access ponts and ther estmated locaton are stored n a database, whch s later used by our WLAN postonng algorthm to locate the user when GPS s unavalable. database of access ponts located usng the dscovery algorthm relyng on GPS. When the database created by our new dscovery algorthm wthout GPS has been used, the postonng error has approxmately doubled. B. In stu experment The objectve of ths experment s to assess the qualty of our dscovery algorthms by comparng the accuracy of WLAN postonng when t s based on a database bult usng dscovery ether wth GPS or wthout GPS. Fgure 13. The expermental dstrct. TABLE V. POSITIONING ACCURACY OF GPS AND WLAN BASED ON DISCOVERY WITH AND WITHOUT GPS. Postonng error Postonng method Average 95th percentle GPS 3.7m 7.m WLAN dscovery wth GPS WLAN dscovery wthout GPS 8.5m.1m 16.4m 38.7m Fgure 1. Stage 3: Calculaton of unknown access ponts locaton. The expermental dstrct s a resdental area of medum populaton densty, of approxmately 1 km area, whch s llustrated n fg. 13. In order to do t a test envronment n the cty of Montreal has been selected. All access ponts have been dscovered usng the two dfferent algorthms, wth and wthout GPS, and have been stored the dscovered access ponts n two separate databases. For the dscovery algorthm wthout GPS, a complete meshed network of access ponts have been created by walkng n each street three tmes, so ther locaton have been nferred wth the best accuracy. As ths algorthm requres knowledge of a few access ponts locaton to nfer the locaton of the others, a set of known access ponts located on the border of the expermental dstrct has been used as nput. Once the databases have been created, a moble termnal has been moved across the expermental dstrct, and ts locaton has been calculated by GPS, and WLAN usng each of the two databases. Ground truth from dgtals map from Navteq has been used as a reference to calculated the accuracy of the three postonng technque. The obtaned results, presented n Table V, confrm GPS excellent accuracy, wth a postonng error nferor than 7m for 95% of the samples. WLAN postonng has yeld an excellent accuracy when t has been usng the C. Indoor Outdoors transtons Our prototype has then been tested n real stuatons n downtown Montréal. Transtons from ndoors and outdoors envronments has been examned to valdate the contnuty of the postonng servce. Our frst remark was the proper operaton of the dscovery algorthms. Dscovery wth GPS s relable outdoors, and just one or two passages across a buldng are enough to dscover the locaton of ndoor access ponts. As a result, WLAN postonng can be used as a relay to GPS n areas where t s unavalable, n order to offer a contnuous postonng servce. For nstance, the prototype wll automatcally swtch to WLAN postonng as a user enters a buldng, and swtch back to GPS once coverage s recovered, as llustrated n fg. 14. Outdoor to ndoor transtons occur when GPS coverage s lost as a user step nto a buldng. As soon as that happens, the prototype automatcally swtches to WLAN postonng. Ths technology transton s completed n less than a second, therefore provdng an unnterrupted servce. However, accuracy drops as soon as WLAN s used, especally n the frst seconds after the transton. Ths may occur as access ponts from nearby buldngs are stll detected when the user step n a buldng. In order to make transtons smoother, the last GPS locaton can be repeated for a few seconds to let tme to the user to walk deeper nto the buldng so that access ponts from

12 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER surroundng buldngs do not nterfere n the locaton calculaton. Fgure 14. WLAN and GPS transtons as a user walks across a buldng. Indoor to outdoor transtons happens when GPS coverage s recovered. These transtons occur a few seconds after the user step out of a buldng, as delay of approxmately 10 seconds s requred to acqure a satellte fx. Awhle, WLAN postonng s used. Fnally, a feld test has been completed n Montreal s underground cty a klometer long seres of underground tunnels connectng offces and shoppng centers n the downtown core. There, the ablty of our prototype to dscover WLAN access ponts locaton over a large dstance n order to provde a contnuous postonng servce could be valdated. For the frst part of the feld test, the user was walkng outdoors, then entered n an underground tunnel, and kept gong for a dstance of approxmately 1km. The user locaton could be calculated almost all the way durng87%ofthetmetobeexact exceptnafew tunnels of the underground cty where WLAN access ponts could not be detected. The database of access ponts used for ths experment was created by walkng outdoors on the man streets parallel to the underground cty, and then by passng twce nsde the tunnels. Ths has shown that just a few passages are necessary to create a dense database of access ponts, even over large ndoor dstances, and offer a contnuous, yet farly accurate postonng. VII. THE ABL PROTOTYPE The ABL prototype s ntegratng the three postonng methods, namely place detecton, GPS, and WLAN postonng, nto a swtchng mechansm, as prevously descrbed on fg. 9. The prototype should be able to keep track of the user locaton ndoors as well as outdoors to offer a ubqutous servce wth the best accuracy avalable. Our objectve s to valdate ths clam by conductng a coverage survey n real stuatons. In order to do t, our prototype has been carred by three partcpants for the duraton of a week, and detaled coverage statstcs have been recorded for each postonng method supported by the prototype and are presentednthssecton. A. Experment The objectve of the experment s to measure the coverage offered by the ABL prototype durng users every day actvtes. In order to do t, the prototype has been gven to 3 partcpants for the duraton of a week. The prototype had to be kept on durng the entre week to record every second the user locaton wth the three dfferent postonng methods and ther combnaton. The coverage of each postonng methods and ther combnaton nto the ABL prototype have been calculated from these measures. A total of more than 600,000 samples have been recorded n a week for each partcpant. The coverage of a postonng method s calculated as the percentage of samples contanng a vald locaton. As samples are recorded every second, the measured coverage represents the tme percentage durng whch a postonng method was avalable. Intally, the prototype had an empty database of known access ponts. Access ponts had to be dscovered as users were movng n ther daly actvtes. In addton, no place had been defned. Partcpants were asked to defne at least two places, one at home and at work. Partcpants 1 and have a smlar profle. They both lve n a house n suburban area and use ther car to commute to work n downtown Montreal. They are back home early and do not leave home afterward. Partcpant 3 lves n a small resdental buldng n a dstrct of medum densty. He walks 30 mnutes and takes the subway for 1 hour to commute to work n downtown. The ABL prototype has been developed as an applcaton for Pocket PC, usng Mcrosoft Vsual Studo 005, and.net Compact Framework. It has been deployed on HP Ipaq hw6945 termnals whch ntegrate both GPS and WLAN. B. Results Results of Table VI show the coverage measured for each postonng method and for ther combnaton nto the ABL prototype. Place detecton has yeld approxmately 90% coverage for every partcpant. As all partcpants has only defned two places, home and work, ths smply means that they spent approxmately 90% of ther tme ether at home or at work. Ths result also shows that t s crucal for a postonng system to operate properly n these two places, as t s where users wll spent the vast majorty of ther tme. GPS coverage reaches 40% for the two frst partcpants but s only 6% for the thrd. These low results confrm that GPS, alone, s not adapted to end user applcatons, as t s not avalable n place users spend most of ther tme. It s exactly what has been observed for partcpant 3, as GPS was out of coverage both home and at work, and was only avalable durng short outdoor commutes. Comparatvely, the frst two partcpants have obtaned much hgher results, only because they have obtaned a sporadc GPS coverage whle at home as they placed the termnal at proxmty of a wndow.

13 8 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER 008 WLAN postonng coverage s measured to approxmately 80%, whch confrms that access ponts are suffcently numerous to offer a vast coverage. The ABL prototype, combnng the three postonng methods, yelds approxmately 95% coverage for all partcpants. 97% coverage s even reached for partcpants 1 and. The 3% durng whch a locaton could not be calculated only represents a tme perod of 45 mnutes n a day. So the ABL prototype was able to successfully locate partcpants 1 and durng 3 hours and 15 mnutes per day. Results obtaned for partcpants 3 are slghtly lower, wth coverage of 94%, representng a perod of 1 hour and 30 mnutes a day wthout coverage. That could have been expected as partcpant 3 commutes to work by subway for approxmately 45 mnutes a day, a perod durng whch no wreless sgnals can be receved. TABLE VI. MEASURED COVERAGE FOR EACH POSITIONING METHOD AND THEIR COMBINATION. Postonng method Partcpant 1 Partcpant Partcpant 3 Place detecton 91% 93% 88% GPS postonng 37% 45% 6% WLAN postonng 91% 87% 88% ABL postonng 97% 97% 94% Results of Table VII present the role that each postonng method plays n the coverage obtaned by the ABL prototype. Place detecton represents 90% of the prototype s coverage, and s therefore a crucal component. The combnaton of GPS enables to grow the coverage of 4%. Even though ths result may seem small, t represents an hour a day of addtonal servce, coverng users outdoor trps. Fnally, WLAN postonng adds an addtonal 1% to the prototype s coverage, whch represents 15 mnutes of servce per day. TABLE VII. COVERAGE COMPOSITION OF THE ABL PROTOTYPE. Coverage composton Partcpant 1 Partcpant Partcpant 3 Total ABL 97% 97% 94% %obtanedby Place detecton %obtanedby GPS %obtanedby WLAN postonng 91% 93% 88% 4% 3% 5% % 1% 1% VIII. CONCLUSION Our research objectve was to conceve a wreless postonng system able to meet end user applcatons requrements of ubqutous coverage, outdoors and ndoors accuracy, use of exstng nfrastructure, and absence of confguraton. Our approach was to develop a hybrd postonng system combnng exstng technologes to accurately operate n any place users may travel to. An n stu coverage and accuracy survey of the four man wreless technologes, namely GPS, GSM, WLAN, and Bluetooth, has been conducted to select the most complementary technologes. The results obtaned n numerous envronments of the cty of Montreal have shown that GPS and WLAN offer complementary coverage, the frst operatng at best outdoors and the second ndoors and n urban centers. Then, a hybrd postonng system on top of these two technologes has been bult, by developng an algorthm capable of swtchng to the most accurate postonng method dependng on the crcumstances. Our frst postonng method used WLAN access ponts and exstng place detecton algorthms to recognze the room n whch the user s currently located. If no room could be recognzed, GPS was used as our second postonng method. WLAN postonng, the last postonng method, was used when GPS was unavalable, n areas such as unfamlar buldngs. Dscovery algorthms have been developed to automatcally locate WLAN access ponts as users go, so they can later be postoned. Fnally, all of these postonng methods have been ntegrated nto one postonng system, whch have been carred by 3 partcpants for the duraton of a week. Ths experment has shown that our hybrd postonng system s able to keep track of the user locaton durng 95% of the tme, or hours and 45 mnutes a day. Ths postonng system consttute another step toward the creaton of truly pervasve postonng servces, by offerng nearly ubqutous coverage, excellent accuracy outdoors usng GPS, room level accuracy ndoors, and the use of an exstng nfrastructure. New research projects could lead to mprovements of the current postonng system by allowng the combnaton of any rado nterface such GPS, GSM, WLAN, Bluetooth, or RFID, and developng a flexble framework so the postonng system could be deployed on any termnal and use the best postonng method avalable at any tme. Power consumpton could also greatly be mproved by desgnng new algorthms usng a sample nterval of a few mnutes nstead of a second. Fnally, the ablty to locate a user n any place he may travel to open new perspectves of research, especally n presence and context awareness. By relyng on an accurate poston of the user at any tme and any place, new applcatons could be develop to nfer the user actvtes, surroundng envronments, habts, and ntentons, n order to better serve hm. REFERENCES [1] L. Reyero, and G. Y. Delsle. Always Best Located, a pervasve postonng system. Proceedngs of the Second Internatonal Symposum on Wreless Pervasve Computng, pp , February 007. [] B. W. Parknson, and J. J. Splker. Global Postonng System: theory and applcatons. Amercan Insttute of Aeronautcs and Astronautcs, 793 p, [3] A. LaMarca, et al. Place Lab: Devce Postonng Usng Rado Beacons n the Wld. Lecture Notes n Computer Scence, vol. 3468, pp , 005. [4] D. L. Wlson, Davd L. Wlson's GPS Accuracy Web Page, [5] Wong, Alex. Bell Moblty and ESRI conference on locatonaware servces, January 006, unpublshed.

14 JOURNAL OF NETWORKS, VOL. 3, NO. 8, NOVEMBER [6] V. Bahl, and V. N. Padmanabhan. RADAR: An n-buldng RFbased user locaton and trackng system. Nneteenth Annual Jont Conference of the IEEE Computer and Communcatons Socetes, vol., pp , 000. [7] A. Haeberlen, et al. Practcal robust localzaton over large-scale wreless networks. Proceedngs of the 10th annual nternatonal conference on Moble computng and networkng, pp , 004. [8] T. Roos, et al. A Probablstc Approach to WLAN User Locaton Estmaton. Internatonal Journal of Wreless Informaton Networks, vol. 9, no. 3, pp , 00. [9] J. Krumm, and E. Horvtz. LOCADIO: Inferrng Moton and Locaton from W-F Sgnal Strengths. Frst Annual Internatonal Conference on Moble and Ubqutous Systems: Networkng and Servces, pp. 4-13, 004. [10] Z. Xang, et al. A wreless LAN-based ndoor postonng technology. IBM Journal of Research and Development, vol. 48, no. 5, pp , 004. [11] Ekahau Inc., Ekahau Postonng Engne, [1] J. Letchner, D. Fox and A. LaMarca, "Large-Scale Localzaton from Wreless Sgnal Strength". Proceedngs of the Natonal Conference on Artfcal Intellgence, pp. 15-0, 005. [13] Y. C. Cheng, et al. Accuracy characterzaton for metropoltanscale w-f localzaton. Proceedngs of the Internatonal Conference on Moble Systems, Applcatons, and Servces, pp , 005. [14] J. Krumm, and K. Hnckley. The NearMe Wreless Proxmty Server. Proceedngs of the Sxth Conference on Ubqutous Computng, pp , 004. [15] V. Otsason, et al. Accurate GSM Indoor Localzaton. Proceedngs the Seventh Internatonal Conference on Ubqutous Computng, pp , 005. [16] A. Varshavsky, et al. Are GSM Phones THE Soluton for Localzaton?. Proceedngs of the Seventh IEEE Workshop on Moble Computng Systems & Applcatons, pp. 34-4, 006. [17] Skyhook Wreless, W-F Postonng System, [18] M. Nlsson, J. Hallberg, and K. Synnes. Postonng wth Bluetooth. Proceedngs of the 10th Internatonal Conference on Telecommuncatons, vol., pp , 003. [19] M. Pacga, and H. Lutfyya. Herecast: An Open Infrastructure for Locaton-Based Servces Usng WF. IEEE Internatonal Conference on Wreless And Moble Computng, Networkng And Communcatons, vol. 4, pp. 1-8, 005. [0] J. Hghtower, et al. Learnng and recognzng the places we go. Proceedngs of the Seventh Internatonal Conference on Ubqutous Computng, pp , 005. [1] E. W. Djkstra. A note on two problems n connexon wth graphs, Numersche Mathematk, vol. 1, no. 1, pp , December Lonel Reyero was born n France n 198. He has receved a M.Sc. n telecommuncatons from INRS- Telecommuncatons, n Montreal, Quebec, Canada n 007 and a M.Sc. n telecommuncatons from Natonal Insttute of Appled Scences, n Lyon, France n 005. Snce 007, he s a Software Consultant for Summt-Tech Communcatons Inc., n Montréal, Quebec, Canada. Prevously, he has worked as Researcher at the Internatonal Insttute of Telecommuncatons. Mr. Reyero s a member of the Order of Engneers of the Provnce of Quebec and a member of the IEEE. He was awarded a Grant of Excellence from the Internatonal Insttute of Telecommuncatons n 006. Glles-Y. Delsle s currently Drector, Technology Integraton Centre at Technopôle Defense and Securty n Valcarter, Québec, Canada. From June 004 to March 008, he was Vce-Presdent Research at the Internatonal Insttute of Telecommuncatons n Montréal, Canada. Prevously, he was Drector and Professor at the School of Informaton Technology and Engneerng at the Unversty of Ottawa from 00 to 004 and he has been a Professor of Electrcal and Computer Engneerng at Laval Unversty, Québec, Canada, snce 1973, where he was head of the department from 1977 to From June 199 to June 1997, he was also Drector of INRS- Telecommuncatons, a research nsttute whch s a part of the Unversté du Québec. He s nvolved n research work n ntellgent antenna array, radar cross-secton measurements and analytcal predctons, moble rado-.channel propagaton modelng, personal communcatons and ndustral realzaton of telecommuncatons equpment. Dr. Delsle s a member of the Order of Engneers of the Provnce of Québec and Professonal engneers of Ontaro, Past- Presdent of the Canadan Engneerng Accredtaton Board, Fellow of the Canadan Academy of Engneerng, Past Canadan Presdent of URSI, Past Presdent of ACFAS, Fellow of the Insttute of Electrcal and Electroncs Engneers (IEEE), the Canadan Engneerng Insttute, the Canadan Academy of Engneerng and of the Insttuton of Engneerng and Technology (IET-UK). In 1986, he was awarded the J. Armand Bombarder prze of ACFAS for outstandng techncal nnovaton and hs work n technology transfer has been recognzed by a Canada Award of Excellence n 1987.The Canadan Councl of Professonal Engneers has recognzed ts contrbuton to the professon by awardng hm the prestgous Mertorous Servce Award for Professonal Servce n 004.Dr Delsle has supervsed the work of over a hundred graduate and post-graduate students over the last 30 years. He has been elected to the Canadan Hall of Fame n telecommuncatons n October 007.