熊本大学学術リポジトリ. Kumamoto University Repositor

Transcription

1 熊本大学学術リポジトリ Kumamoto Unversty Repostor Ttle Wreless LAN Based Indoor Poston and Its Smulaton Author(s) Ktasuka, Teruak; Nakansh, Tsune CtatonIEEE Pacfc RIM Conference on Comm Computers, and Sgnal Processng - Vol. Issue date 2003 Type URL Journal Artcle Rght

2 Wreless LAN Based Indoor Postonng System WPS and Its Smulaton Teruak Ktasuka, Tsuneo Nakansh and Akra Fukuda Graduate School of Informaton Scence and Electrcal Engneerng, Kyushu Unversty 6- Kasuga-Koen, Kasuga-sh, Fukuoka , JAPAN System LSI Research Center, Kyushu Unversty Emal: {ktasuka, tun, Abstract The wreless LAN(W-F) nfrastructure s wdely used and many locaton-aware systems and servces are researched. In ths paper, wreless LAN based ndoor postonng system WPS s proposed. WPS uses wreless LAN technology to measure the locaton of each moble termnal. Moble termnals equp only wreless LAN devce to communcate and measure ts locaton, so t can be made wthout any addtonal devces for locaton sensng. Exstng wreless LAN based locaton systems measure the sgnal strength by only access ponts. On WPS, each moble termnal also measures the sgnal strength of neghborng termnals. Thereby WPS can acheve more precse locaton estmaton than exstng systems, where there are many moble termnals. We smulate the case that dsntance measurement has probablstc error. The result shows mprovement of accuracy n WPS. I. INTRODUCTION Locaton sensng technology s very mportant for an nfrastracture of an ubqutous computng envronment. Outdoor locaton sensng technology such as GPS s already developed and wdely used. For ndoor locaton sensng, there are a few technologes, but they are not wdely used currently. If locaton sensng technology whch uses the wreless LAN nfrastructure s developed, t wll be used wdely and can be a popular way. As ndoor locaton sensng, there are techonologes based on ultrasonc sensor and GPS pseudolte. Both techonologes have to need hgh nstallaton cost and mantenance cost. A system based on ultrasonc sensor[] have to nstall many sensors on a celng and connect them to a server. A system of GPS pseudolte[2] also have to nstall pseudo-satellte. In ths paper, we show the desgn of proposed postonng system WPS and the result of ts smulaton. WPS uses the sgnal strength between moble termnals and access ponts to calculate the locaton of moble termnals. Moble termnals also measure the sgnal strength of neghborng moble termnals. II. DESIGN OF WIPS A. Features WPS uses the wreless LAN nfrastructure and provdes the locaton nformaton to moble users. There were a smlar research n Mcrosoft Research[3] and Ekahau[4]. The concept of WPS have been presented n [5] already. The major features of WPS are the followngs G B G2 A C D Gn G3 Reference host Moble host Connectons between reference host and A Connecton between moble host and A Fg.. Basc desgn of WPS. A moble termanl A measure the sgnal strength of not only access ponts G, G2 and G3, but also moble termnals B, C and D. When a densty of moble users.e. WLAN termnals ncreases, accuracy of locaton nformaton becomes hgh. In the stuaton of sparse densty of locaton reference ponts such as access ponts, more precse locaton nformaton can be provded than usual system. Frst feature s matched to a rendevzous n scrouged place such as a hall of an exhbton. Second feature prevents settng cost of many access ponts from rsng. The sgnal strength of WLAN rado s used for calculaton of the locaton. However the sgnal strength does not reflect the dstance between termnals drectly, because the sgnal s faded by not only dstance but also mult-path and other causes. To acheve these features, WPS measures the sgnal strength by not only access ponts but also moble termnals. Each moble termnal measures the sgnal stregth of access pont and neghborng moble termnals. Then WPS determnes moble termnals locaton by more observatons than [3]. In Fg. as an example, a moble termnal A measures the sgnal strength of each access pont G, G2 and G3 and also measures t of each moble termnal B, C and D. On a latest work such as [3], only access ponts measure the sgnal strength of a moble termanl. Generally n WPS, /03/$7.00 c 2003 IEEE 272

3 many observed sgnal strength data are used to determne the poston of moble termnals. The number of observatons of the sgnal strength are ncreased n o(n 2 ), when a number of termnals and access ponts s denoted by n. For these observatons, WPS can acheves the above two features. B. Algorthm In WPS, a locaton server gathers the sgnal strength from each moble termnal and access pont. The server calculates the locaton of each moble termnal and notfes each moble termnal. Each access pont knows ts own absolute locaton and notfes the locaton server wth the sgnal strength. The process of locaton calculaton on a locaton server s descrbed below. Bascally the steepest descent method s used. ) Gather the lst of sgnal strength of each par of termnals and access ponts. 2) Determne the ntal postons of moble termnals. 3) Iterate the modfcaton of the postons of moble termnals, untl convergence. 4) Notfy the locaton to each moble termnals. In the Step ), each moble termnal and access pont sends the packet to the server. The packet contans the sgnal strength lst of neghborng moble termnals and access ponts. The neghbor means that the termnal s n a rado communcaton range of another termnal. Each access pont adds the locaton of tself nto the packet. In the Step 2), the server calculates an ntal postons of moble termnals by neghborng access ponts and neghborng ntalzed termnals. For each moble termnal denoted by A, a server selects the neghborng moble termnals N A and access ponts G A. Already ntalzed moble termnals are selected from N and named N A. N A, N A and G A s a set of hosts moble termnals or access ponts. The poston of termnal X or access ponts X s denoted by p X. The ntal poston of termnal A s determned by p A = ΣX GA p X +Σ X N A px n(g A )+n(n A ). n(g A ) and n(n A ) are the number of hosts n G A, N A respectvely. Intalzaton s done cyclcally untl all ntal poston of moble termnals are determned. In the Step 3), moble termnal postons are refned by the steepest descent method to mnmze the expresson: Σ n p p j d,j, =0 Σn j=+ when total number of hosts both moble termnals and access ponts are n, p s the poston of -th host, and d,j s the measured dstance between -th and j-th host. Where the poston p (k) k-th teraton, p (k+) shows the poston of -th host n s calculated by p (k+) = p (k) + α (), when α s a sutable small value α (α =0.05 s used n later smulaton), () s defned below. () =Σ n j=0u,j f(, j), u,j s an unt vector from p (k) j to p (k). f(, j) s the dfference between the measured dstance d,j and the current estmated dstance between -th and j-th hosts. u,j and f(, j) are defned by followng equatons. u,j = p(k) p (k) j, l,j = p (k) p (k) j, l,j d,j l,j f and j are neghbors, 0 f and j are not neghbors f(, j) = and l,j >d max, d max l,j f and j are not neghbors and l,j d max. d max s the typcal rado communcaton range. The teraton s fnshed when the maxmum value of () s smaller than a sutable value γ (γ =0.0 s used n later smulaton). III. SIMULATION MODEL AND RESULTS A. Smulaton Model To confrm the precson of locaton nformaton provded by WPS, we smulate WPS n 200m square plan. A rado communcaton range of each hosts s. 4, 5 or 9 access pont s located n the plan and the number of moble termnal are ncreased from 5 to 50. The 9 access ponts case s the case of enough access ponts, and the case of 4 and 5 access ponts are the sparse access ponts case. In ths smulaton, our major am s to confrm the advantage of our dstance measurement between moble termnals. Then, we compare our model wth the case that only access ponts measure the dstance to moble termnals. We also take care that the estmated dstance has probablstc error. However, we don t take care of a movement of each moble termnal, a communcaton delay, and a method of dstance estmaton from a sgnal strength. Fg. 2 shows the locaton of access ponts, through 9. In the case of 4 access ponts, the access ponts numbered to 4 are used. The access ponts numbered to 5 are used n the 5 access ponts case. Moble termnals are put n the plan n a random manner and they don t move. The cover rato of access ponts of each case s dscussed here. In the case of 9 access ponts, 95% area of 200m square plan can reach 3 or more access ponts drectly, but 4% area reaches only 2 access ponts and under % area reaches only access ponts. In the case of 5 access ponts, only 24% area reaches 3 access ponts, 53% area reaches 2 access ponts, and 23% area reaches only access pont. In the case of 4 access ponts, no area reach three access ponts drectly and 2% area does not reach any access ponts. 74% and 24% area reach respectvely only one and two access ponts. We smulate under two assumpton for each number of access ponts. Frst assumpton s measured dstance between hosts has probablstc error. Second assumpton s measured dstance has no error. Second assumpton s not sutable for real envronment, f the dstance s measured by the rado sgnal strength. However, we use the second assumpton to 273

4 6 200m 2 00 I-R4-E I-R5-E I-R9-E A-R4-E A-R5-E A-R9-E m average of error dstance [m] Fg. 2. Smulaton envronment show the avalablty on whch the densty of access ponts s very low. In the frst assumpton, the measured dstance s the sum of real dstance and probablstc error. The probablstc error s chosen under the normal dstrbuton and s n proporton to the real dstance. The proportonal error means that, f a real dstance between hosts s, we assume the measured dstance usng sgnal strength s selected n normal dstrbuton,.e. between 80m and n 95.5% case. If the dstance s half, the error s also half. In the related works[3], [4], they determne the estmated dstance from the sgnal strength n ther experments. But we do not n the smulaton. B. Smulaton Results Fg. 3 shows the result of the assumpton that the measured dstance contans 20% error n normal dstrbuton. X-axs s the number of moble termnals and Y-axs s the average error of estmated locaton. The average error s defned as an average of dfference between real poston and estmated poston of each moble termnals. On the access pont based method, the average error rate s shown as I-R4-E, I-R5-E and I-R9-E. Each lne shows the case of 4, 5 and 9 access ponts respectvely. The average errors are 42.2m, 6.8m, 6.5m n the case of 4, 5, 9 access ponts respectvely. The averages are almost the same when number of moble termnal s ncreased. The result of proposed method s shown as A-R4-E, A- R5-E and A-R9-E n Fg. 3. Each A-Rn-E s the result of n access ponts case. On the proposed method, the average error decreases as the number of moble termnal ncreases. In the case of 5 moble termnal, the average errors n 4, 5 and 9 access ponts are respectvely 8.5m(44% of access pont based method), 0.9m(65%) and 5.5m(86%). When there number of termnals Fg. 3. Smulaton results n the cases that dstance mesurement has probablstc error. I- means the result of access ponts based method, A- means proposed method and -Rn- shows the case of n access ponts. are 50 moble termnals, the average errors are 3.4m(8%), 3.m(9%) and 2.8m(42%) respectvely n 4, 5 and 9 access ponts cases. Fg. 4 shows the result of the case assumed that the measured dstance s equal to real dstance between each par of hosts. The lnes, I-R4-E0 and I-R5-E0, show the access pont based method, and the lnes A-R4-E0 and A-R5-E0 show the result of the proposed method. *-R4-* and *-R5-* cases are respectvely the case of 4 and 5 access ponts. The result of 9 access ponts case s not shown n Fg. 4, snce the error s almost zero n both access pont based method and proposed method. Access pont based method has no mprovement along wth ncreasng of moble termnal, as same as Fg. 3. Proposed method mproves the average of error along wth the ncrement of number of moble termnals. In the case of 4 access ponts, the average of error access pont based method s 42.0m. On the proposed method, the average of error are 3.8m(33% of 42.0m) and 0.3m(0.7%) respectvely n 5 and 50 moble termnals. In 5 access ponts case, the average of error access pont based method s 2.m. On the proposed method, the average of error are 2.9m(24% of 2.m) and 0.m(.0%) respectvely n 5 and 50 moble termnals. Comparng Fg. 3 and Fg. 4, we fnd that the error of dstance measurement between each par of hosts has a great nfluence on the accuracy of locaton estmaton. However n both Fgs., the number of moble termnals has an mportant 274

5 average of error dstance [m] 00 0 I-R4-E0 I-R5-E0 A-R4-E0 A-R5-E0 teraton count number of termnals Fg. 5. A-R4-E0 A-R5-E0 A-R9-E0 A-R4-E A-R5-E A-R9-E Number of teraton untl convergence number of termnals Fg. 4. Smulaton results n the cases of error-free dstance measurement. I- means the result of access ponts based method, A- means proposed method and -Rn- means the case of n access ponts. effect on proposed method. The effect s greater on the sparse access ponts case such as 4 or 5 access ponts case than 9 access ponts cases. Fnally, Fg. 5 shows the number of teratons for each case. Operaton of each teraton s descrbed n Sec. II. The parameters of teraton are α = 0.05 and γ = 0.0. We confrm that the number of teratons s not ncreased when the number of moble hosts s ncreased. Especally n the case of the sparse access pont such as 4 and 5 access ponts case, when the number of moble termnals s ncreased, teraton s decreased. Complexty of a teraton s o(n 2 ), where n s the number of hosts, both access ponts and moble termals. Then the complexty of a estmaton of all moble termnals s o(n 2 ) or below. ACKNOWLEDGMENT Ths research has been partally supported by JPSP Grant-n-Ad for Scentfc Research (B) (KAKENHI ), MEXT Grant-n- Ad for Young Scentsts (B) (KAKENHI ) and NTT. REFERENCES [] Harter, A., Hopper, A., Steggles, P., Ward, A. and Webster, P.: The Anatomy of a Context-Aware Applcaton, Proc. of ACM/IEEE MOBI- COM 99, pp , Aug [2] Gobb, H.S.: GPS Pseudoltes: Theory, desgn, and applcatons, Ph.D. Thess, Stanford Unversty, Sep [3] Bahl, P. and Padmanabhan, V. N.: RADAR: An In-Buldng RF-Based User Locaton and Trackng System, Proc. IEEE INFOCOM 2000, pp , [4] Ekahau, Inc.: Ekahau Postonng Engne 2.0, [5] Ktasuka, T., Nakansh, T., Fukuda A.: Locaton Estmaton System usng Wreless Ad-Hoc Network, Proc. the 5th Internatonal Symposum on Wreless Personal Multmeda Communcatons (WPMC 2002), pp , Oct IV. CONCLUSION We proposed ndoor postonng system WPS, and descrbe ts features. WPS can acheve better accuracy, where many moble termnals exst, and also n the place of sparse access ponts. Moble termnals can help each other to fnd ts locaton. Smulaton result s shown. Smulaton result ensure the features of WPS. When the number of moble termnals s ncreased, the precson s also ncreased. The accuracy of estmated locaton s mproved n all cases. Complexty of proposed algorthm s not greater than o(n 2 ) on the smulated cases, where n s the number of hosts, both access ponts and moble termals. 275