Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 W-F Indoor Locaton Based on RSS Hyper-Planes Method Ch-Kuang Hwang and Kun-Feng Cheng Department of Electrcal Engneerng, Chung Hua Unversty No. 707, Sec., Wufu Rd., Hsnchu, 00, Tawan, R.O.C. Tel: 0-58-6867 Fax: 0-58-646 Emal: smon@chu.edu.tw Abstract Ths paper presents a wreless postonng technque based on receved sgnal strength (RSS hyper-planes for ndoor locaton system. One major problem n the ndoor wreless locaton s the serously dfferent effects of RSS between the lne of sght (LOS and non-lne of sght (NLOS. Snce shadowng and mult-path effects can result large varance n the measured RSS, and the degree of the varance wll heavly depends on the surroundng envronment. Fndng an effectve locaton estmaton technque to smooth away these effects s the man focuses n ths paper. Therefore, n the wreless propagaton envronment the RSS nformaton from multple access ponts (AP are recorded and processed can somehow provde an overlappng coverage area of nterest. Then the hyper-planes technque s appled to effectvely mgrate the NLOS and to calculate a superor locaton of the desred moble termnal (MT. Keywords: receved sgnal strength (RSS, hyper-plane, non-lne of sght (NLOS. Introducton Locaton estmaton n wreless technology has attracted great nterests n recent research, because t s an mportant tas n the applcaton of wreless communcatons. Recently, the moble locaton servce has been drven by FCC (Federal Communcaton Commttee whch mandated the wreless networs to provde the locaton for emergency calls. The Enhanced 9 demands the locaton accuracy requrement for phase as 00m for 67% of tme and 00m for 95% of tme for networ-based locaton systems []. However, the moble locaton system s only sutable for outdoor envronments, but not for ndoor envronments []. The attenton on developng an ndoor locaton technology based on the WLAN s gradually ncreasng whle the popularzaton of the WLAN s growng. The superorty n usng ths technque can utlze the extng hardware wthout affectng the orgnal networ access functon, and wll also create a popular value added. One major problem needed to be conquered n the ndoor wreless locaton s the NLOS effects on the RSS. As mentoned before, the mult-path and shadowng effects cause great dfferences n the measurng RSS, and the degree of the dfference s vared wth envronment. How to fnd an effectve locaton estmaton technque to smooth away the affecton of NLOS, and to reduce the error of locaton estmaton s the research focus n ths paper. The research of ths paper s amed at solvng the postonng problem n the automatc materal handlng system (AMHS whch already exsted n a semconductor plant. Frst, for the possble wreless propagaton envronment n a semconductor plant we establshed the model whch can descrbe the relatonshp between RSS and the dstance and set up the smulaton test envronment, and then executed the practcal ste-survey based on the establshed model. We can use the hyper-plane locaton estmaton technque to smooth away the NLOS effects. And then, a W-F ndoor locaton system s bult for the semconductor plant envronment. Ths paper s organzed as follows. Some related wors n locaton determnaton technologes are surveyed n Secton. Our research methodology and how to buld a reasonable ndoor locaton system are descrbed n Secton. The smulaton results are dscussed n Secton 4. Fnally, we present our conclusons n Secton 5.. Related Wor The free space propagaton model has been very popularly used to predct receved sgnal strength when the transmtter and receved have a clear, unobstructed lne-of-sght (LOS path between them. As wth most large-scale rado wave propagaton models, the free space model predcts that receved power may decay as a functon of the separaton dstance between transmtter-recever (T-R. The free space power receved by a recever antenna whch s separated from a radatng transmtter antenna by a dstance d, s gven by Frs free space equaton [], PG t tgrλ Pr ( d = ( (4 π d L where Pt s the transmtted power, Pr ( d s the receved power whch s a functon of the T-R separaton dstance d (meters, G t s the transmtter antenna gan, G s the recever antenna gan, L s r 7
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 the system loss factor whch s not drectly related to propagaton ( L, and λ s the wavelength n meters. Furthermore, t s clear that Eq. ( does not hold for d = 0. For ths reason, large-scale propagaton use a close-n dstance d 0 denoted as a nown receved power reference pont. The receved power Pr ( d, at any dstance d > d0, can be related to P r at d 0. The value Pr ( d0 may be predcted from Eq. (, or may be measured n the rado envronment by tang the average receved power at many ponts located at a close-n radal dstance d 0 from the transmtter. Thus, usng Eq. (, the receved power n free space at a dstance greater than d 0 s gven by PG t tgrλ PG t tgrλ d0 (4 π d L (4 π d0 L d d0 Pr ( d0 for d d0 Pr ( d = = ( = > d If P r s n dbm unt, the receved power s gven by d Pr( d[ dbm] = Pr( d0 0log ( d0 Normally, the reference dstance d 0 for the practcal system usng low-gan antenna n the -GHz regon s typcally chosen to be m n ndoor envronments. The system nsde buldngs to use RSS from off-the-shelf WLAN adapters to detect the MT locaton s RADAR [4], whch operates by measurng and recordng RSS nformaton of a number of sampled locatons durng off-lne phase and nferrng the locaton of an MT based on the record RSS fngerprntng database nformaton n the on-lne phase. The RADAR locaton technque s performed n two phases. Frst, n the off-lne phase, the locaton fngerprnts are collected by performng the survey of RSS from three access ponts (APs at nown locaton. The RSS s measured wth enough statstcs to create a database or a table of predetermned RSS values n the target area. These measurements consst of average values of the RSS (the average computed usng several samples at the nown locaton. In summary the RSS values at three APs are used as the locaton fngerprnt. Second, n the on-lne phase, the system operaton n the target area, MT reports the RSS from each AP and the system determnes the best match between the on-lne observatons and any pont n the off-lne model. The locaton of the best matchng pont s reported as the locaton estmate. The RADAR system dsregard that the effects of the floor and nstead consder the effects of obstacles (walls between the transmtter and the recever. The wall attenuaton factor (WAF model s descrbed by d nw * WAF nw < C Pr( d = Pr( d0 0nlog d (4 0 C* WAF nw C where Pr s n unts of dbm, n s the rate at whch the path loss ncreases wth dstance, C s the maxmum number of obstructons (walls up to whch the attenuaton factor maes a dfference, nw s the number of obstructons (walls between the transmtter and the recever, and WAF s the wall attenuaton factor. In general the values of n and WAF depend on the buldng layout and constructon materal, and are derved emprcally.. Research Methodology. Expermental Test Bed Ths experment regards the actvty center of Chung Hua Unversty as the ste-survey of envronment, and ths ste s a LOS envronment. We selected a pont n the ste as the transmt pont (TP. We consder that the rado wave propagaton whch sent out from the TP can present the nature dspersed and spread out to all drectons n the space. It s just le a spherod expandng constantly, so we assume that regard TP as the center of a crcle, then recevng pont (RP s on the sphere. Set up the poston of star drecton on the sphere (RP~5, as show n Fg., then as the fve drectons to execute the ste-survey of RSS at these RPs wth varous dstances. We used two laptop computers wth Lnux, PCMAIA wreless networ card based on Atheros chp wth an extendable 5dB omn-drectonal antenna for the experment. The purpose of usng the extendable antenna s amed for applyng the drectve antenna to test. Ths experment used the current IEEE 80. standard. The applcaton program developed and fnshed n language C under Lnux operatng system. The current drver program for the networ card s that t can notfy the operatng system core to wor by nterrupt. The networ drver has been desgned n the ln layer. When the networ card receves a pacet, t wll produce an nterrupt for the nterrupt handler to process ths pacet. Normally, after fnshng the nterrupt process, the pacet s sent to the upper layer (Networ layer for further process. However, the open drver code can be downloaded from the nternet, and the networ card can be set up to operate n montor mode for recevng the raw pacet. Under ths mode, the networ card would fetch every pacet through t. Therefore, the selected wreless card based on Atheros chp can be programmed n ths montor mode to retreve the RSS nformaton from the receved pacet wthout the further process such as the upper layer (Networ layer. That s, our applcaton program n the montor mode can flter and judge the correct transmsson nformaton to fetch the needed locaton 8
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 nformaton. Ths locaton system can have better estmated tmer because the locaton nformaton s obtaned n the lower layer.. Data Collecton Each tme the transmt test s conducted only on one of fve RPs. We begn wth testng the LOS envronment by placng a laptop at the fxed TP poston. The other one s placed on one of the fve RPs drectons whch have the same separaton dstance m. Then the mean and varance of the correspondng RSS are obtaned by conductng one thousand tmes of the transmsson test. After fnshng one of RPs, we then move ths laptop to the next one to test. The testng procedure performs contnuously untl fnshng the test of fve drectons. The separaton dstance has been set up as,.5, 5, 7.5, 0, 5, 0, 5, and 0m, respectvely, and the testng follows the same procedure as the one wth m dstance. Smlarly, under the NLOS envronment the same procedure s also conduced by consderng the usual case of propagaton path n the semconductor plant, such as the MT (TP s shadowed by the neghbor machne platform. We placed a metal shadowng object for the TP wth 0.5m separaton on the propagaton path. The thcness of metal shadowng objects for testng are selected to exceed the wavelength of carrer frequency.. Predcton model of RSS The mean RSS predcton model can be establshed accordng to Eqs. ( and (4, so the power receved Pr( d by a dstance d, s gven by d Pr( d = Pr( d0 0nlog (5 d0 where n s the path loss exponent ndcatng the rate at whch the path loss ncreases wth dstance, d0 s the close-n reference dstance whch s determned from measurements close to the transmtter, Pr ( d0 s the close-n reference receved power. Accordng to our collected RSS mean data, the Eq. (5 can be wrtten n the matrx form as P log.5 0 (.5 / r( d + Pr( d d d0 M = 0 n M (6 Pr( d0.0 + Pr( d0 log ( d0.0 / d0 It can be wrtten as follows y = 0n x (7 T ( - T x x x y or n = (8 0 Eq. (8 usually can apply to both of the LOS and NLOS envronments but wth dfferent n. Fg. shows the curve for RSS mean predcton model. If Eq. (6 s used to fnd out only sngle n for all dstances, the fttng error wll not be sutable for all the ranges. Therefore, we mae dfferent predcton of n for varous dstances or ranges wth the same values of ste-survey RSS. Fg. shows the curve for the RSS mean predcton model wth dfferent predctng n durng varous perods..4 Locaton Algorthm Based on Three APs In the general geometrcal trangulaton locaton researches, they assumed that the measured nose s addtve and the NLOS error s a large postve bas whch causes the measured ranges to be greater than the true ranges [5]. Under the assumpton, the MT locaton wll guarantee to le n the overlapped regon (enclosed by ponts U, V and W of the range crcles as shown n Fg. 4. But, t s noted that the ntersect on of three crcles may not be overlapped wth the real measurement results. Therefore, wth the above assumpton we have to judge whether the three crcles ntersect or not n our locaton algorthm. If two crcles ntersect as depcted n Fg. 5, then the trangle can be drawn as Fg. 6. Form Fg. 6 we can get r = Dcosθ (9 Form cosne axoms we also have D = D + R DRcosθ (0 = D + R rr where D s the dstance between AP and the estmate and D s the dstance from to the estmate. By the defnton of the radus, we have ( ( R = x x + y y ( The r can be rewrtten as by rearrangng (0 D + R D r D + R D = = R R R r The parameter q s then defned as follows. q r D + R D = = R R ( ( The range of the parameter q can determne the shape of the trangle. For example, If 0< q < D D R 9
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 If q > D > D If q < 0 D < D D Furthermore, we can judge two crcles whether ntersecton or not by the relatonshp among D, D and R, such as n case of ntersecton ( D + D R q r = R R D D + R D = (4 In case of non-ntersecton ( D + D < R, we have q = r = D R D + D R D R D (5 The estmate of the hyper-plane ntersectng wth the R s denoted as (x,y whch can be evaluated as follows. x = qx + ( q x (6 y = q y + ( q y Ths normal vector of the hyper-plane s gven by a = x x, b = y y (7 Form Eqs. (6 and (7, the hyper-plane equaton can be formed as a ( x- x + b ( y- y = 0 (8 Fg.7 shows the ntersected stuaton of three straght lnes (hyper-planes whch forms the possble area for the MT. The least square error method s appled wth weght factors to determne the pre-estmate of locaton coordnates. We begn the defnton of the cost functon related to estmated error j, j σ = j ax + by d H = (9 a + b σ j= j where the parameterσ j s the standard devaton of the measured RSS at each AP. By mplementng the mnmzaton of mn{ } x, y H, the pre-estmate ( x, y can be obtaned by the followng equatons: al+ al + al abl+ abl + abl x y ab l+ abl+ ab l bl+ bl+ bl ad l+ adl+ ad l = bd l+ bdl + bd l A W = b - W = A b where j=, j σ j a + b j= σ j l =, W (0 x =, y Because all of three hyper-planes wll not parallel together, the matrx A s nvertble by checng ts non-zero determnate ( ab ab ll + ( ab ab ll + ( ab ab ll We then hybrd all possble combnaton of LOS and NLOS cases, that s, we combne all possble pre-estmates ( x, y to estmate the fnal one. If the number of APs s desgned to be n, then there are n = all possble pre-estmated coordnates. For n = case, the ntermedate estmate s formulated as the sum wth the weghtngs related to the nverse square errors, 8 ( x, y e ( xm, ym = ( 8 = e = where e = Δd and Δ d = ( x - x + ( y - y. If there are more than three APs (n> n a target area, tae neghborng three APs as a locaton area. Therefore, many dfferent locaton areas combne to obtan the estmate. At each locaton area wll determne a MT estmate poston by Eq. (. The orgnal varances can be used as a weghtng functon to obtan the fnal estmate. And n the effectve area, that s, 0< q < or D+ D < R for case, the weghtng s wrtten as βm = ( σ Some ntermedate estmate poston may locate outsde the effectve locaton area, for the case of only one outsde the effectve area (one q < 0 or q >, the weghtng functon s reformulated as βm = ( σ 40
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 In case of more than one outsde the effectve area (two q < 0 or q >, we dscard the estmate, that s, β m = 0. ( xy, = m= β ( x, y m m= m β m m (4.5 Locaton Algorthm Based on Four APs We use four APs to predct the MT poston coordnates ( x, yz,. Because the AMHS heght s assumed to be nown, so the nown heght can mpose some lmtaton on z-coordnaton to ensure the accuracy of z-coordnaton. From Eq. (8, the hyper-plane s adjusted to nclude the z-coordnaton, and t can be rewrtten as a( x x + b( y y + c( z z = 0 (5 From Eq. (9 by addng lmtaton of the nown heght on z-coordnaton h wth the assocated gan g on the z-coordnaton, Eq. (9 can be modfed as 4 4 σ j=, j j ax + by + cz d H = + g( z- h 4 a + b + c σ j= j (6 Smlarly by evaluatng mn { H } x, y, z, Eq. (0 can be rewrtten as follows wth W = [ x y z ] T A W = b - W = A b Moreover, Eq. ( can be rewrtten as follows 6 ( x, y, z e ( xm, ym, zm 6 = e where 4 e = Δ d and = (7 = (8 ( - ( - ( - Δ d = x x + y y + z z 4. Smulaton Results Ths ndoor locaton system based on three or four APs s consdered n ths smulaton. There are three locatons for the MT to be estmated, one of the three locatons s not n the effectve area. Then, the performance or accuracy of the proposed locaton algorthm s conducted n the D plane (x-y plane, as shown n Fg. 8. Three MT and four APs postons are lsted as the followng: MT (-5,0, MT (0,0, MT (5,0, AP (-0,0, AP (-0,-0, AP(0,-0, and AP4 (0,0, respectvely. All numercal values are expressed n meters. Locaton estmate based on three and four APs are smulated n the paper. In case of three APs, there are four possble trangles, whch are denoted as T (formed by AP, AP and AP, T (formed by AP, AP and AP4, T (formed by AP, AP and AP4 and T4 (formed by AP, AP and AP4, respectvely. Based on the establshed model as shown n Fg., we randomly generate 000 RSSs for the LOS case and 000 RSSs for the NLOS case wth the assocated measured varance for each AP, and the results of estmated error are shown n the followng table. Table. Average error based on the proposed technque for three and four APs. Average error Area MT MT MT Three APs T 0.5 0.7 0. T 0.6 0.7 0. T 0.7 0.6 0.6 T4 0.7 0.4 0.6 Four APs 0.7 0.5 0.9 By observng the smulaton results, we now that the proposed ndoor locaton technque can predct the MT locaton effectvely for both of three APs and four APs cases. In the meantme, t also demonstrates the good locaton estmaton ablty for the MT outsde the effectve locaton area. 5. Conclusons In ths experment the major challenge s the random MT poston, so LOS and NLOS effects mpose dffculty on predctng the dstance based on the measurng RSS. So, t becomes very dffcult to dstngush NLOS and LOS. Therefore, wthout dentfyng NLOS and LOS cases, n ths paper we consder all possble combnaton of both cases at the desred APs to predct the MT poston. The correspondng varance and estmated errors are utlzed as the weghtng functons. Moreover, the proposed method can be mplemented wthout the assumpton that the measured range s greater than the true ranges. Our proposed method s also good for the case of the MT outsde the effectve area. Acnowledgements Ths wor s supported by NSC 95--E-6-06, and NSC 96--E-6-05. 4
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 References [] J. M. Zagam, S. A. Parl, J. J. Bussgang, K. D. Melllo, Provdng Unversal Locaton Servces Usng a Wreless E9 Locaton Networ, IEEE Communcatons Magazne, Vol. 6, pp. 66-7, Aprl. [] I. K. Aduse, K. Kyamaya, and K. Jobmann, Moble postonng technologes n cellular networs: an evaluaton of ther performance metrcs, n Proc. MILCOM 00, vol., Oct. 00, pp.9-44. [] T. S. Rappaport, Wreless Communcatons: Prncples and Practce, nd, NJ: Prentce Hall PTR, 00. [4] P. Bahl and V. N. Padmanabhan, RADAR: An In-Buldng RF-based User Locaton and Tracng System, n Proc. IEEE INFOCOM, 000, pp. 775 784. [5] C. D. Wann and H.C. Chn, Hybrd TOA /RSSI Wreless Locaton wth Unconstraned Nonlnear Optmzaton for Indoor UWB Channels, IEEE WCNC, 007, March 007, pp. 940 945. RP Fg.. Curves for mean of RSS predcton after revson U D AP V W AP AP D D RP RP5 0m.5m m TP Fg. 4. Measured range crcles and the assocated ntersected area. RP RP4 Fg.. RPs and TP confguraton U AP AP ( x, y ( x, y Fg. 5. Standard ntersecton of two crcles Fg.. The curve for RSS mean predcton 4
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 U θ D D AP r (, AP (, θ θ x y ( x, y x y R Fg. 6. The assocated trangle of the standard ntersecton of two crcles. Fg. 7. Intersectng area of the three straght lnes AP AP4 MT MT MT AP AP Fg. 8. APs and MT confguraton 4
Chung Hua Journal of Scence and Engneerng, Vol. 5, No. 4, pp. 7-4 (007 44