An Accurate and Fast WLAN User Location Estimation Method Based on Received Signal Strength

Transcription

1 An Accuate and Fast WLAN Use Locaton Estmaton Method Based on Receved Sgnal Stength Mnghua Zhang and Shensheng Zhang Depatment of Compute Scence and Engneeng, Shangha Jao Tong Unvesty, Chna {zhangmh, Abstact. Locaton estmaton s a cucal component of locaton-awae applcatons. Postonng based on eceved sgnal stength (RSS) n weless netwoks s consdeed a pomsng and nexpensve soluton. Exstng technques only use RSS fom some fxed access ponts (APs) deployed wthn the aea of nteest to estmate use locaton. Though expements on the popetes of RSS, t s found that RSS fom fa access ponts can dstngush dffeent locatons moe easly. In ths pape, we popose to ntoduce RSS fom APs outsde of the aea to ncease locaton estmaton accuacy. We also pesent an onlne maxmum matchng method to select the most possble locatons fst, thus educng the computatonal cost ncued by usng moe RSS values and mpovng the speed of locaton estmaton. Ou new locaton estmaton method s mplemented and compaed wth elated wok n a pactcal weless netwok. Expemental esults llustate that the poposed method can gve a hghe degee of accuacy. Keywods: Use locaton estmaton; Receved sgnal stength; Weless local aea netwok; Maxmum matchng; Locaton estmaton speed. 1 Intoducton Locaton estmaton s an mpotant task n today s pevasve computng applcatons [1] that use locaton nfomaton on objects o human bengs to povde sevces wthout use nteventon. A typcal postonng system s the Global Postonng System (GPS) [2], yet t doesn t wok well n ndoo envonments o uban aeas whee hgh buldngs sheld the satellte sgnals. As the polfeaton of weless local aea netwok (WLAN) and the declnng cost of moble devces, geat attenton have been pad to the eceved sgnal stength (RSS)-based locaton estmaton technques. Such postonng systems buld on an aleady exstng, wdely avalable weless netwok nfastuctue and wok ndoos as well as outdoos. RSS-based postonng technques usually wok n two phases: ado map buldng phase and locaton detemnaton phase. Dung the fst phase, the eceved sgnal stength fom obseved access ponts (APs) togethe wth the coespondng samplng poston ae saved, esultng n a so-called ado map. Dung the locaton detemnaton phase, the eal-tme sgnal stength measuements ae compaed to data n the ado map and the neaest match s etuned as the estmated use locaton. In Y. Sh et al. (Eds.): ICCS 2007, Pat III, LNCS 4489, pp , Spnge-Velag Beln Hedelbeg 2007

2 An Accuate and Fast WLAN Use Locaton Estmaton Method 59 ndoo envonments, the popagaton of ado sgnals suffes fom sevee mult-path effects, nose and ntefeence [3]. RSS-based locaton estmaton technques need to deal wth the nosy chaactestcs of the weless channel. Those chaactestcs cause the eal-tme sgnal stength samples measued n the locaton detemnaton phase to devate sgnfcantly fom those stoed n the ado map, theeby lmtng the accuacy of such systems. Cuent RSS-based technques can be boadly dvded nto two categoes: detemnstc technques [4, 5, 6] and pobablstc technques [7, 8, 9]. Ou wok les n the fst categoy. Howeve, none of the pevous postonng systems take nto account the ole of access ponts outsde the aea of nteest. The RADAR system [4] uses the k-nnss (k-neaest neghbo n sgnal space) algothm. They collect sgnals fom thee fxed access ponts whch can be obseved n all pedetemned locatons. The Hous [8] system mpoves upon the pefomance of RADAR by usng the Jont Clusteng technque. Roos [7] adopts RSS fom 10 access ponts deployed wthn the testbed. In [10], Kaemaungs nvestgated the popetes of eceved sgnal stength epoted by IEEE b weless netwok nteface cads. Hs expemental esults ndcate that the RSS values fom the same AP at two dffeent locatons may be dffcult to dstngush fo postonng puposes when the RSS level s hgh n whch case t tends to have a lage degee of vaaton. To dstngush locatons effcently and mpove locaton estmaton accuacy, ths fndng nspes us to use RSS fom the access ponts deployed not only wthn the aea of nteest but also outsde of the aea. We also popose an onlne maxmum matchng method n whch the set of obseved APs s used to choose possble locatons fst, allevatng the buden of computaton ncued by adoptng moe RSS values. We test ou new locaton estmaton method n a pactcal weless netwok and compae wth othe wok. Expemental esults show that the method can acheve a hghe degee of accuacy. The est of the pape s constucted as follows. Secton 2 llustates the fluctuaton of the ado sgnal stength n WLAN, whch s affected by many dstubances. In secton 3, we elaboate on ou locaton estmaton methodology ncludng the onlne maxmum matchng method. We pefom expements and compae ou method wth othes n secton 4. Secton 5 concludes the pape and descbes some potental futue wok. 2 Popagaton of Rado Sgnals n WLAN The popagaton of ado sgnals s affected by many dstubances [11]. The IEEE b standad uses lcense-fee 2.4GHz band to communcate wth each othe such that othe devces usng the same band lke Bluetooth devces, mcowave ovens wll be a souce of ntefeence. Moeove, 70% pat of a human body s wate, and the 2.4GHz s the esonance fequency of wate. Thus human body s an absobe of ado sgnals. Mult-path fadng [12, 13] s anothe common phenomenon n ado sgnal popagaton. A tansmtted sgnal can each the eceve though dffeent paths, each havng ts own ampltude and phase. These dffeent components combne and poduce a dstoted veson of the tansmtted sgnal.

3 60 M. Zhang and S. Zhang Fg. 1 gves a typcal example of the nomalzed hstogam of the sgnal stength eceved fom an access pont at a fxed locaton. People walk n the aea, and doos open and close. Such changes n the envonment can explan the fluctuaton of sgnal stength shown n the fgue. Fg. 1. An example of nomalzed sgnal stength dstbuton 3 Poposed Locaton Estmaton Method 3.1 Rado Map Buldng Phase Suppose the locaton aea A has N selected samplng postons, A = {poston 1 N}. A RSS sample n ou expement s a 6-tuple: (Id, Tme, RSS, SSID, BSSID, Poston), whose elements ae explaned n detal n Table 1. Table 1. Explanaton of a RSS sample vecto Element name Explanaton Example Id Sequence numbe of detected access pont 1, 2, 3, Tme Samplng date :43:03 RSS Receved sgnal stength n dbm -76, -81, SSID Sevce Set Identfe 1x.net.sjtu BSSID MAC addess of the access pont C Poston 2-dmenson coodnates and oentaton of the (12.386, 1.54, south) samplng locaton In each samplng poston poston, seveal RSS samples ae collected n ode to epesent the fluctuaton of ado sgnal stength. Then the aveage value of sgnal stength avgss j eceved fom each obseved AP, whose MAC addess s bssd j, s computed. Wth the nfomaton the ado map s bult up. Recods n the ado map

4 An Accuate and Fast WLAN Use Locaton Estmaton Method 61 can be descbed n (1), whee L s the numbe of obseved APs n each samplng poston. = ( poston, avgss 1 N, 1 j L. j, bssd j ), (1) 3.2 Locaton Detemnaton Phase Dung the locaton detemnaton phase, suppose the eal-tme RSS samples ae epesented by (2), whee L s the numbe of access ponts detected by use. U = {( m _ avgss, m _ bssd ) 1 L} (2) Befoe seach the neaest locaton n the ado map, we use an onlne maxmum matchng method to choose a subset of ado map contanng the possble locatons fst. As each AP has ts own sgnal s coveage aea, the set of obseved APs ae not totally the same fo dffeent locatons. Based on the APs n the eal-tme RSS samples, the postons n the ado map assocated wth a maxmum ntesecton of the same APs fom the sub set P. It can be descbed fomally as: P' = { poston AP m = { U AP = { AP = { xy xy n. m _ bssd 1 n L'},. bssd. bssd ( poston )( AP 1 N ( poston xy xy m. poston = poston. poston = poston AP > AP A P')}, 1 y L}, ' m AP 1 y L}. Then the smple and effcent Eucldean dstance s selected as matchng metc between eal-tme RSS samples and RSS data saved n the ado map. The poston n P, whch has a mnmum Eucldean dstance n sgnal space s consdeed as the estmated use locaton. Eucldean dstance s computed by (4): L' mn( D j = ( U. m _ avgss j. avgss) ), = 1 (4) U. m _ bssd = j. bssd,. poston P'. j 2 ) (3) 4 Expemental Results and Analyss 4.1 Expemental Testbed We pefom ou expements n the second floo of the thd buldng of the Electonc Infomaton and Electcal Engneeng School buldngs. The layout of the floo s

5 62 M. Zhang and S. Zhang shown n Fg. 2. The aea has a dmenson of 100 metes by 20 metes. It ncludes a long hallway, a meetng oom, lots of offces and labs. To measue the RSS fom dffeent APs, we use an ASUS pocket PC wth an embedded weless Ethenet cad to captue the sgnal. We use WRAPI [14], a thd paty lbay fo etevng the sgnal stength. Thee ae altogethe 30 access ponts that can be detected. They ae deployed not only n the second floo, but also n othe floos o n neghbong buldngs. Fo buldng up the ado map, we take the samplng locatons along the hallway on a gd wth cells of 5h2 metes (the hallway wdth s 2 metes). Fo each locaton, 10 samples ae collected fo each oentaton. Fg. 2. The layout of the second floo of the thd buldng of the Electonc Infomaton and Electcal Engneeng School buldngs 4.2 Expemental Results In ths secton, we evaluate the pefomance of ou poposed method. An ndependent test set collected on dffeent tme and day fom that fo the ado map buldng phase s used. We conduct expements and compae wth othe methods. In the fst expement, we only use the RSS values fom access ponts deployed wthn the testbed. The method s just lke the 1-NNSS method n the RADAR system, so we call t the RADAR method n Fg. 3. In the second expement, we only adopt the RSS data fom thee stongest APs n each samplng poston, whch s called as the 3- stongest AP method. Ths method s smla to the technque n [8] on the selecton of RSS data. In the last expement, we use RSS data fom all obseved APs. Expemental esults ae llustated n Fg. 3. Locaton estmaton eo s defned as the Eucldean dstance between the eal poston and ts estmated poston. The esults of ou poposed method ae qute postve, fo t s moe accuate than the othe two methods. Fg. 4 gves the cumulatve dstbuton of eo dstance n the thee expements. Usng the poposed method, ove 73 pecent postons ae located wthn 5 metes, whch s hghe than the othe two methods. Ths demonstates that moe RSS data can mpove the locaton estmaton accuacy.

6 An Accuate and Fast WLAN Use Locaton Estmaton Method 63 Fg. 3. Compason of locaton estmaton eo n the thee expements Fg. 4. Eo dstance CDF of the thee expements Snce we use RSS fom moe access ponts to estmate use locaton, t s possble that the calculatng tme wll ncease. Theefoe we also pefom an expement to measue the locatng tme. The pesonal compute s confgued wth 512MB of RAM, Intel(R) Pentum(R) M pocesso 1.5GHz. The thee methods ae mplemented n MATLAB and Table 2 lsts the CPU tme each method uses. In geneal, the moe access ponts sgnal stength nfomaton s used, the moe locatng tme s needed. Although the numbe of access ponts used n ou method s much moe than the RADAR method, yet the dffeence of CPU tme needed s vey small (0.13 seconds). Howeve the aveage eo of the RADAR method s about two tmes that of ou poposed method. Theefoe t s a toleable compomse fo the mpovement of locaton estmaton accuacy. In the Table 2, the temendous ncease of CPU tme spent on the 3-stongest AP method s due to the sot opeaton to choose the fst thee stongest access ponts. On the locaton estmaton accuacy, t s vey close to the RADAR method.

7 64 M. Zhang and S. Zhang Table 2. Compae locatng tme of the thee expements Compason Items Poposed Method RADAR 3-Stongest AP Method Numbe of access ponts matched Aveage eo (m) CPU tme (s) Concluson and Futue Wok Moble use s cuent locaton s one of the most mpotant and fequently-used contexts n pevasve computng envonments. Postonng based on eceved sgnal stength s consdeed a pomsng and nexpensve soluton. Howeve, eceved sgnal stength especally n ndoo envonments s ntefeed by many dstubances. Ths pape poposes a new use locaton estmaton method to mpove accuacy, whch uses RSS data fom access ponts deployed not only wthn the aea of nteest but also outsde of the aea. RSS data fom the fa APs ae used to bette dstngush locatons. Moeove an onlne maxmum matchng method s pesented to quckly get the most possble locatons n ode to mpove the speed of locaton estmaton. Expemental esults show that, wth the 2 5 metes samplng gd, ou method can estmate use locaton to wthn 5 metes wth ove 73% accuacy. In the same testbed, the poposed method s bette than the methods used n othe wok on accuacy. It s also emakable that the samplng gd n the poposed method s bgge than that n the elated wok, whch means fewe calbaton effots ae needed. Futhemoe, though anothe expement the ncease of locatng tme by usng moe access ponts s neglgble elatve to the mpovement of locaton estmaton accuacy. In ths pape, the poposed method belongs to the detemnstc technque. But the dea of usng moe RSS data fom fa APs and the onlne maxmum matchng method ae not lmted to t. They can also be used n the pobablstc technque. In the futue, we would lke to do moe extensve expements, studyng the effect of paametes such as the sze of samplng gd, the numbe of access ponts and ty to pove ou dea n a fomal way. Acknowledgments. Ths eseach s suppoted by Shangha Commsson of Scence and Technology Gant (05SN07114, 03DZ19320). The authos would lke to thank Habn Me fo hs valuable advce. Thanks also to anonymous evewes fo the pespcacous comments. Refeences 1. Hazas, M., Scott, J., Kumm, J.: Locaton-Awae Computng Comes of Age. IEEE Compute, Vol. 37, No. 2, (2004) Enge, P., Msa, P.: Specal Issue on GPS: The Global Postonng System. Poceedngs of the IEEE, Vol. 87, No. 1, Januay (1999) Hashem, H.: The Indoo Rado Popagaton Channel. Poceedngs of the IEEE, Vol. 81, No. 7, (1993)

8 An Accuate and Fast WLAN Use Locaton Estmaton Method Bahl, P., Padmanabhan, V. N.: RADAR: An In-buldng RF-based Locaton and Tackng System. In: Poceedngs of the IEEE INFOCOM 2000, Tel-Avv Isael, Mach (2000) Ahmad, U., Gavlov, A., Sungyoung Lee, Young-Koo Lee: Modula Multlaye Pecepton fo WLAN Based Localzaton. In: Poceedngs of 2006 Intenatonal Jont Confeence on Neual Netwoks, Vancouve BC, Canada, July (2006) Smalagc, A., Kogan, D.: Locaton Sensng and Pvacy n a Context-Awae Computng Envonment. IEEE Weless Communcatons, Vol. 9, No. 5, Octobe (2002) Roos, T., Myllymak, P., T, H., Mskangas, P., Sevanen, J.: A Pobablstc Appoach to WLAN Use Locaton Estmaton. Intenatonal Jounal of Weless Infomaton Netwoks, Vol. 9, No. 3, July (2002) Youssef, M., Agawala, A., Shanka, A. U.: WLAN Locaton Detemnaton va Clusteng and Pobablty Dstbutons. In: Poceedngs of IEEE Intenatonal Confeence on Pevasve Computng and Communcatons. IEEE Compute Socety (2003) Casto, P., Chu, P., Kemenek, T., Muntz, R.: A Pobablstc Locaton Sevce fo Weless Netwok Envonments. In: Poceedngs of the Intenatonal Confeence on Ubqutous Computng. Spnge LNCS 2201, Beln Hedelbeg (2001) Kaemaungs, K., Kshnamuthy, P.: Popetes of Indoo Receved Sgnal Stength fo WLAN Locaton Fngepntng. In: Poceedngs of the Fst Annual Intenatonal Confeence on Moble and Ubqutous Systems - Netwokng and Sevces. IEEE Compute Socety (2004) Ladd, A. M, et al.: Robotcs-Based Locaton Sensng usng Weless Ethenet. In: Poceedngs of the 8th Annual Intenatonal Confeence on Moble Computng and Netwokng. ACM Pess (2002) Stallngs, W.: Weless Communcatons and Netwoks. Fst Edton. Pentce Hall (2002). 13. Youssef, M., Agawala, A.: Small-Scale Compensaton fo WLAN Locaton Detemnaton Systems. In: Poceedngs of the IEEE WCNC 2003, Vol. 3. Mach (2003) WRAPI. Webste: