Improving Accuracy of WiFi Positioning System by Using Geographical Information System (GIS)

Transcription

1 AU J.T. 0(): (Jul. 006) Improvig Accuracy of WiFi Positioig System by Usig Geographical Iformatio System (GIS) Tussaai Parthorratt ad Kittipha Techakittiroj Faculty of Egieerig, Assumptio Uiversity Bagkok, Thailad Abstract WiFi positioig system with kowledge of GIS ca achieve higher accuracy i terms of distace error comparig with the oe without it. Positioig algorithm expectedly gets rid of ulikely crossig area i the regio while selectig others. Miimizatio of crossig area determiatio with costrait of geographical iformatio is a key cotributio. Proper placemet of access poits i service area is aother issue of cocer yieldig the same cosequece as optimal coordiate fittig. Learig process for determiig eviromet characteristics is essetial to eable accurate calculatio o distace from measured sigal stregth. Variatio of obstructios i oe eviromet ca maily reflect icosistet profile, especially for idoor eviromet. Keywords: Wireless positioig, WiFi, GIS, costraied optimizatio. Itroductio The proliferatio of WLAN starts from beig widely kow as so-called WiFi, which is oe of the Istitute of Electrical ad Electroics Egieers or IEEE stadards for WLAN. WiFi stads for Wireless Fidelity ad it is truly a freedom to make a coectio to Iteret without hassle of old-fashioed etwork cable. WiFi or 80.b is oe of 80. specificatios family ad it is also the first domiatig stadard i the market. Nowadays ew wireless etworkig devices are adoptig the ew stadard of IEEE 80.g, which offers higher data speed, amely drivig from Mbps to 54 Mbps, while sharig commo operatig frequecy of.4 GHz. This iovative techology is evolvig techically ad practically i couple of years leadig WLAN to be a commo sight at uiversities, airports, coffee shops, offices ad orgaizatios. That service poit or access poit is ofte referred to as wireless hotspot or hotspot i short. Icetives for developig ad stadardizig WLAN are defiitely mobility ad flexibility. This pheomeo stimulates the creatio of supportive features for ew-age laptops. However, aythig has pros, it will also have cos. This issue is largely discussed ad activatig research group to cocer about. Related works ad Backgroud Recetly, directio of research i WLAN field largely varies from pure theory to real-life applicatios. Pure theory is a factory of iovative ideas activatig maufacturer to commercialize after those ideas are prove ad stadardized. For real-life applicatio, oe of the issues curretly i cocer of wireless security is wireless positioig applicatio. Vulerability of ecryptio algorithm of both WEP ad WPA is globally realized. Therefore aother developmet for wireless security turs to be positioig techique. May past researches cotributed creative ad iovative cocept of idoor positioig system with various platform ad architecture. Some of them were that of high accuracy, but were ot practical ad cost effective. Structural ivestmet or dedicated platform for merely a positioig facility is iitially a mai iterest. 38

2 Early ad sigificat cotributio to locatio-aware system was the Active Badge System (Wat 99). The badge is attached to people or objects ad it is programmed to periodically emit ifrared sigal out with uique idetifier. Sesors are placed at kow locatio idoors to pick up radiated ifrared sigal ad trasfer those data for further locatio processig. Further improvemet was doe to elimiate lie-of-sight ad limited rage of operatig distace of ifrared. Active bats, Cricket, Motiostar, ad some others are cocrete examples of developig activities i positioig field (Hightower 00). New research tred is shifted as soo as WiFi techology emerges ad appears to be a future wireless techology. RADAR, project from Microsoft Research Group iitially develops positioig system o WiFi or WLAN platform (Bahl ad Padmaabha 000). Obviously, this research approach is cost effective ad positioig facility coexists with ormal wireless etworkig architecture. I this aspect, ay poits of i-buildig area ca be poited out provided that coverage area of access poit is wide eough. Positioig techiques ca be divided ito two broad types (Ja ad Lee 003), which are locatio figerpritig ad propagatiobased computatio. For a locatio figerpritig, site survey is required to collect sigal stregth samples at every sigle equally divided area of that buildig to build up a sigal stregth database. After learig ad surveyig process is doe, real-time sigal stregth is captured ad mappig process is started. This type of positioig techique ca be visualized as a scee aalysis. For propagatio-based computatio techique, o sigal stregth database is required beforehad. Positioig computatio is doe accordig to differet propagatio models with differet techiques of triagulatio (Hightower ad Borriello 00). Triagulatio is a basic geometric cocept for fidig crossig poit o basis of distace or agle. Each user s positio ca be calculated at ay istat. Ad the accuracy of user s positio is quatified by Euclidea distace as ( ss K Eq. ss ) + ( ss ss ) + + ( ss ss ) Half-wavelegth dipole atea is the most basic kid of atea you ca get (Ao. 006a). This is basically two bits of wire of a special legth, ad is the type of atea most commoly foud built-i to those PCMCIA wireless cards. Its radiatio patter looks like a fat doughut; you get o sigal whe you look through the "hole" (i.e. dow the wire). Therefore it is radiatig circularly i horizotal directio ad this is the reaso that we assume distace from access poit i the patter of circle. Sigal Stregth Preprocessig Relatioship of physical positio ad received sigal stregth at differet distaces away from access poits or hotspots is a key criterio for computig user s positio i WLAN. Idoor eviromet iduces a multipath fadig ad it heavily deteriorates sigal stregth i various patters. Sigal Stregth Collectio The test site for our experimet is o 5 th floor of Q-buildig, Huamark campus, Assumptio Uiversity of Thailad as show o the followig figure. Scale for the floor pla is :50 ad ote that origi poit ca be take arbitrarily. Therefore it is take at the left corer of Q-buildig itself i this case. Fig.. Q5 floor pla with markers ad access poits locatios 39

3 Predetermied locatio of access poits amed DSP-Q5 ad NAL-Q5 is also icluded i the figure together with markers alog the corridor. Those fiftee markers are poits of iterest i our experimet to be used for learig ad testig purposes. They are.5 meters apart from oe aother. Multipath Fadig Miimizatio: Sigal stregth is collected i four directios, amely, North, Earth, West ad South to miimize multipath fadig effect. Issue of sigal stregth samplig also directly reflects multipath effect i terms of directio sesitivity ad agle of receptio. Higher resolutio of samplig, higher accuracy ca be achieved i priciple. Limitatio of software beig used i this experimet will be discussed shortly i Experimet Methodology sessio. Experimet Methodology: Equipmets required for capturig sigal stregth, raw data processig ad positio determiatio are listed i categories of Hardware ad Software accordigly.. Hardware: - Access Poit or Hotspot: Two boxes of ay brad ad model of access poits are required to fuctio as sigal trasmitter. - Wireless Network Iterface Card (Wireless NIC) or WiFi card: This ca be cosidered as wireless LAN sigal receiver i this applicatio. Wireless NIC is available i various iterface types ad it ca operate o both laptop or desktop computer ad Persoal Digital Assistat (PDA) correspodigly. Latest laptop computer ad PDA are equipped with iteral Wireless NIC. - PDA or laptop computer: Desktop computer is ot suitable due to obvious reaso of mobility from poit to poit aroud the floor.. Software: - NetStumbler (Ao. 006b): This freeware program is oe of the most widely used wireless etwork auditig tools allowig detectio of 80.b, 80.a, ad 80.g etworks. It offers two versios for both Widows ad WidowsCE platform, which are Netstumbler ad Miistumbler correspodigly. Sigal stregth samples are saved ito NS file ad ca be further exported ito Microsoft EXCEL for calculatio, but Miistumbler does ot have that feature. - EXCEL: Sigal stregth average is calculated for each particular poit o the floor pla. - MATLAB: Characterizatio of wireless propagatio model ad miimizatio of error distace. Fig.. Processig steps for miistumbler Positio Computatio Coceptual Overview Coceptual overview for determiatio of WLAN user positio is preseted i the figure as follows. RSS Radius Fig. 3. Positio determiatio overview Starter for this task is to request for a floor pla of the buildig uder iterest. The 40

4 try to locate access poit s positio accordigly ad ote that origi poit of x-y coordiate is arbitrarily chose. Awareess of ay measuremet with exact referece to the correspodig floor pla has to be take care of. Distace error is calculated with respect to the exact locatio o the floor pla i had. Now sigal stregth measuremet is doe as aforemetioed istructios. After that, sigal stregth collectio from the access poits, NAL-Q5 ad DSP-Q5 are iterpreted with two propagatio models which are freespace ad wall atteuatio factor model. Sigal stregth distace relatioship directly affects the accuracy of positioig system ad reveals a surroudig atteuatio characteristic. Radio Propagatio models For a radio chael, there exists a reflectio, refractio, diffractio ad scatterig of radio waves that ifluece propagatig paths. Arrival of radio wave from differet path traveled causes multipath fadig pheomeo. Trasmitted sigal from direct ad idirect propagatio path are combied either costructively or destructively causig variatio of received sigal stregth at the receivig statio. The situatio is eve more severe for idoors commuicatio. I-buildig eviromet has differet buildig layout, costructio material, ad partitio placemet resultig i challegig ad upredictable ature of received sigal. Free-space Model: It is primarily used for the worst case cosideratio. At least this model proves ew cocept of positioig system with GIS ad uveils a tred for further improvemet. Free-space model or Friis trasmissio model is described as follows. P = P G G R T T R λ 4πd Eq. Where P R ad P T are receivig ad trasmittig power i uit of watts respectively. G R ad G T are receivig ad trasmittig atea gai. λ is a wavelegth ad d is a distace betwee trasmitter ad receiver. The expoet value of represets a path loss expoet. This ca be iterpreted as the rate of sigal decremet relative to distace. Factory values of operatig atea gai at differet distaces are required for calculatio but they are rarely foud i attached mauals. As a result, it might be icoveiet to use the above equatio. Equivaletly, free-space model is characterized i terms of referece distace as follows after few steps of mathematical maipulatio. d P R [ dbm] = P0 [ dbm] 0log. Eq.3 d 0 Sigal stregth distace relatioship of free-space model is a foudatio as the worst case to cosult with. Maximum error distace from other models that exceeds the oe calculated from free-space model is defiitely ivalid. Some segmet i the floor uder experimet is applicable to this equatio, for example, corridor, ad hallway ad so o. This model is ot uiversally appropriate for idoors, i particular, i-buildig eviromet. Wall Atteuatio Factor Model: This model represets a commo ad reliable ibuildig characteristic. Atteuatio due to a iterveig wall or partitio is described as follows. d PR [ dbm] = P dbm WAF d 0[ ] 0log. Eq.4 0 Where is a path loss expoet ad WAF is a Wall Atteuatio Factor. Overall, WAF is a umber of atteuatio from iterveig walls ad partitios. Pricipally, differet types of material possess differet electrical ad physical characteristics. Ibuildig obstructio varies from place to place ad cosidered to be specific. These values ca be obtaied by experimets. Learig Process Practically, positio determiatio based o free-space model is straightforward ad does ot require learig process as uiversal 4

5 relatioship is implied i every eviromet cosidered. Path loss expoet for free space is fixed at the value of as show i Eq.. This statically represets urealistic eviromet of free space regardless of the surroudig obstructios. Obviously, applyig this freespace model yields higher error distace. To achieve more accurate result ad to represet more realistic idoor eviromet, we adopt WAF model. I this case, values of WAF ad have to be computed from experimet before positioig service. This is cosidered to be a learig process of eviromet by which cliets are residig i. Oce the measuremet of sigal stregth at markig poits is doe, liear regressio is applied to those data sets resultig i WAF ad parameters. I my experimet, eight poits or odd-umbered poits o our test area are used for learig process. Note that a lie with rectagle marker represets experimetal result, while triagle marker represets calculated result due to liear regressio. Fig. 5. Sigal stregth at DSP-Q5 access poit Theoretically, the loger the rage from a trasmitter, the lower sigal stregth a receiver ca perceive. Shorter rage from trasmitter ca yields lower sigal stregth oly i the case of presece of a large barrier betwee trasmitter ad receiver. Ad that is applicable for the case of NAL-Q5 access poit. Physical realizatio of floor pla helps predictig received sigal stregth i correspodig poits of field. Fig. 4. Lie of sight for NAL access poit Cosideratio of straight lie joiig betwee these two access poits to every sigle poit of iterest o the floor pla ca at least differetiate amout of sigal atteuatio due to variety of obstructig objects. It is very obvious that sigal stregths received at DSP- Q5 access poit are commoly blocked with the same obstacle ad reflects i cosistet atteuatio patter as show i the followig figure. O the cotrary, sigal atteuatio tred for the case of NAL-Q5 access poit is icosistet. Fig. 6. Sigal stregth at NAL-Q5 access poit Regio divisio is essetial for the case of NAL-Q5 access poit as received sigal stregth does ot udergo commo obstructio as the case of DSP-Q5 access poit. Piecewise liear cocept appears to be applicable for this icosistet characteristics. Essetially, Geographical Iformatio is ow used to separate regio uder cosideratio. First regio costitutes poit up to poit 6, while secod regio costitutes poit 7 up to 4

6 poit 5. Liear regressio is the applied to each regio accordigly yieldig piecewise liear from poit to poit 6 at a corer ad from poit 7 to poit 5 at a corer. System Simulatio Sigal stregth collectio ca ow be traslated ito trasmitter-receiver distace with the aid of two radio propagatio models. Trasmitter-Receiver distace is the separatio distace of access poit from user. This value of distace ca be visualized as a radius of atea radiatio circle with the ceter at that access poit of sigal receptio. From access poit of view, user locatio ca be at ay poits o that atea radiatio circle. The followig figure illustrates received sigal stregth from NAL-Q5 access poit. Mathematics expressio represetig those two circular equatios with slight maipulatio is as follows. ( x x ) ( x x M ( x x Eq.5 ) ) + ( y y + ( y y + ( y y ) r ) r ) r = 0 = 0 = 0 Where x ad y are x-y coordiate of DSP-Q5 access poit, while x ad y are x-y coordiate of NAL-Q5 access poit. r ad r ca be calculated from eq.4 i terms of d. Trasfer all variables to the left had side ad leave right had side with zero. Defie i a form of matrix as show below. f( x, = f M f Eq.6 Fig. 7. Atea radiatio circle I the same maer, received sigal stregth from DSP-Q5 is traslated ito radius of atea radiatio circle with the ceter at the access poit itself. We ow have two circular equatios with kow ceter poit ad radius value. Access poit s positios are iitially assiged before layig coordiate system oto the floor pla of iterest. Simultaeous equatio solvig is required to be doe successfully to fid those two crossig poits. Ulikely crossig poits output from equatios solvig will be elimiated with the aid of Geographical Iformatio System later. Fig. 8. Crossig of two radiatio circles (x (x = M (x Eq.7 - x - x - x ) + ( y y ) r ) + ( y y ) r ) + ( y y ) r = f K y Eq.8 + f + + f ) represets the error matrix due to the icalculable effect such as multipath fadig, reflectio, refractio ad atteuatio for radio chael. mi with costraits of 0 x y.5 Eq.9 Hallway boudaries for both horizotal ad vertical dimesio are take to be costrait for miimizatio. 43

7 Costraits of x ad y coordiate are based o Geographical Iformatio of floor pla i our test site. It scopes dow ad elimiates ulikely itersectio area resultig from positioig algorithm i optimizatio stage. The followig figure displays a floor pla of our test site alog with specific x ad y dimesio of corridor sectio. It is obviously see that for corridor sectio, x ad y rages from 0 to 9.5 meters ad from 9.5 to.5 meters respectively. This yields the expressio show i Eq.9. Error distace decreases from 8.95 meters to 5.05 meters after applyig miimizatio with costrait of geographical iformatio of our floor pla. This result reveals the goodess of geographical iformatio i improvig idoor positioig system accuracy. Coclusio Fig. 9. Floor pla with corridor dimesio Results ad Discussio The costrait for miimizatio with the kowledge of geographical iformatio decreases error distace i compariso with ucostraied miimizatio as expected. This result emphasizes that with otio of ulikely locatio for that user, the error distace i determiig that user s positio ca be achieved with better performace. Table. Crossig poits ad error distace for the case of with ad without costrait optimizatio I this paper, WiFi positioig system with kowledge of Geographical Iformatio is coducted to improve accuracy of idoor positioig system. Reliability ad accuracy of the system maily depeds o umbers of learig poits, optimizatio of crossig poit fidig algorithm ad sufficiet sigal stregth samples to represet each poit with least error. Applyig the cocept of referece positio implicitly mitigate multipath fadig effect by relyig o a poit i the field. Issue of referece poit ca be divided ito static ad dyamic type. The oe used i this experimet is that of static type as the referece is fixed at positio of 50 cm. Refereces Ao. 006a. What kids of ateas are there? < html>. Ao. 006b. NetStumbler < etstumbler.com> Bahl, P.; ad Padmaabha, V RADAR: A I-buildig RF based User Loatio ad Trakig System. IEEE Ifocom. March:

8 Hightower, J.; ad Borriello, G. 00. Locatio Systems for Ubiquitous Computig. IEEE Commuic. Mag. August: Hightower, J.; ad Borriello, G. 00. Locatio Sesig Techiques. IEEE Commuic. Mag. August: Techical Report Ja, R.H.; ad Lee, Y.R A Idoor Geolocatio System for Wireless LANs. Proc. 003 It. Cof. Parallel Process. Workshops (ICPPW 03). Wat, R.; Hopper, A.; Falao, V.; Gibbos, J. 99. The Active Badge Locatio System. AM Tras. Ifo. Syst..40():