Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 do:10.21311/001.39.11.21 RFID Postonng Based on Vehcle Postonng Subsystem Wehan Wang* Beng Yuanda Internatonal Proect Management Consultng Co. Ltd., Beng 100088, Chna *Correspondng author (Emal:19126516@qq.com) Ll Chen USTB, Beng 100000, Chna Abstract Wth the rapd development of moble computng and embedded system, people has been ncreasngly nterested on locaton awareness, and has been more and more concerned about vehcle postonng servces system. However, rado frequency dentfcaton technology (RFID) s characterzed of non-contact, NLOS, large transmsson range and hgh cost performance etc. Consequently, the paper manly studed the RFID postonng algorthm based on vehcle postonng subsystem. Frst of all, the paper analyzed LANDMARC and VIRE algorthm. It proposed trangular grd layout specfc to VIRE system algorthms, and compared the postonng performance of rectangular grd layout. Under the crcumstance of no other data processng means appled, the postonng precson of trangular grd layout s hgher than that of rectangular grd layout. In addton, the lnear nterpolaton used specfc to VIRE algorthm s not applcable to the defects at electromagnetc envronment. The paper adopted BP neural network model to solve the sgnal strength value of vrtual tag, and compared t wth Newton nsertng method. The results ndcate that the postonng precson of neural network- VIRE s 14.9% and 10.2% hgher than that of lnear nterpolaton-vire and newton nterpolaton-vire respectvely. Key words: Vehcle Postonng, RFID LARNDMARC, VIRE, BP Neural Network 1. FOREWORD In recent years, wreless communcaton system has been growng at an amazng speed. Wreless technology has entered communcaton, ndustral, medcal, consumer applcatons, logstcs transportaton and publc securty and other felds, ncludng ndoor and outdoor postonng(yu, Zheng and We, 2015). Its development s closely related to people's daly lfe. At present, self-organzed wreless sensor network can provde nformaton servces such as trackng and navgaton. And the development of wreless network enables the users to extensvely obtan wreless nformaton access thus to put forward hgher requrements on accurate postonng of wreless network. However, the exstng wreless postonng technologes nclude GPS (Han, Km and Park, 2009), Zgbee(Hong, Huang and Chen, 2013), Bluetooth(Jang, 201; X, Jang and Lao, 2014) and UWB(Lu, Lu, Xu and 2012) etc.. However, the locatng process of all the wreless postonng technologes s sgnfcantly restrcted. For example, the sgnal of GPS s weak n concentrated downtown and places wth many buldngs. And the updatng of postonng nformaton s long and cannot satsfy quck postonng requrements. Furthermore, the cost of GPS recevng apparatus s hgh. It s applcable to areas wth better economc condtons, and s not sutable for large-area promoton. Rado frequency dentfcaton technology (RFID), an automatc dentfcaton technology emerged n 1990s, can automatcally dentfy the obect va radofrequency sgnal (Saab, Mhanna and Salba, 2009), and acqure relevant nformaton of the obect wth dfferent types of sgnals. Furthermore, RFID tag has realzed contactless operaton, and t s characterzed of small sze and large capacty, dspense wth vsble lght source, hgh penetrablty, as well as strong ant-polluton capacty and durablty. In addton, the NLOS dentfcaton and tag of RFID tags can work n harsh envronments, support fast readng and wrtng of multple obects, and realze real-tme locaton and long-term trackng management(song, L and Tang, 201; Wang, N and L, 2014)). It can be extensvely appled n outdoor postonng due to the low cost and hgh postonng accuracy (Saab and Nakad, 2011). Consequently, the paper manly studed the RFID postonng applcaton based on vehcle postonng subsystem, and focused on dscussng the mpact of dfferent postonng algorthms on poston accuracy (Song, L and Zhang, 2016). Frst of all, t ntroduced the postonng method of RFID technology system, focused on analyzng the LANDMARC and VIRE algorthm based on recevng sgnal strength, and analyzed the mpact of algorthm parameters on postonng accuracy. Secondly, lnear nterpolaton s used specfc to VIRE system as an mprovement. The dstance between the tags to reader s appled as the nput 166
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 nformaton, and set the receved sgnal strength as output nformaton. It connects the two through the ntroducton of BP neural network. Fnally, t compared the smulaton results and exstng RFID postonng algorthm, and analyzed the model algorthm accuracy after mprovement and optmzaton. 2. RSSI-BASED RFID POSITIONING ALGORITHM AND ITS OPTIMIZATION RFID system s manly composed of the tag, RFID reader, antenna and a wreless network system. Accordng to dfferent measurement technques, the sgnal strength, tme for recevng the sgnal, tme dfference of arrval and sgnal recepton angle provded by dfferent e-tags to the RFID reader can be obtaned. Therefore, RFID ndoor postonng method s manly dvded to the followng four categores: Receved Sgnal Strength Indcator (RSSI), Tme of Arrval (TOA), Tme Dfference of Arrval (TDOA) and Angle of Arrval (AOA). The paper focused on the analyss of RSSI-based RFID postonng algorthm. 2.1. LANDMARC Systematc Algorthm It s assumed that there are U readers, H reference tags and M postonng-pendng tags n the envronment to be postoned. Then the process for LANDMARC algorthm to derve the locaton of postonng-pendng tag s as followng. And the LANDMARC system structure s as shown n Fg. 1 below. Fgure 1. LANDMARC system structure schematc dagram 1: The sgnal feld strength vector of reference tag RT receved by each reader s defned as S ( 1, 2,..., R SR SR S RU ). Where: S s the RSSI value of RT on the reader Ru u. The sgnal feld strength vector of postonng-pendng tag LT receved by each reader s defned as S ( 1, 2,..., L SL SL S LU ). Where: S s the Lu RSSI value of LT on the reader u. The feld strength dstance between RT and LT s defned as shown n Formula 1 below: Where: u lower sgnal feld strength dfference between U u 1 E, (1, M ), (1, H ) ( S S ) u u Lu Ru s the feld strength component of Eucldean dstance on the reader u. The smaller RT and 2 (1) E, the LT on readers wll be, namely the closer between reference tag and postonng-pendng tag. 2. Calculate the feld strength Eucldean dstance between LT and reference tag RT, and form the vector 1 2 E (,,..., H E E E ). Select the smallest K element(s) among E, and form the set 1 2 EP (,,..., k EP EP EP ) accordng to ther sze. It s dentfed that k reference tag(s) correspondng to EP s the nearest neghbor reference tag of RT. 3. Dfferent weghtngs of each reference tag are gven at postonng process accordng to the sze of EP. And the estmate coordnates of LT calculated accordng to emprcal formula s as shown n the formula 2 as shown below: 167
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 k LT LT RT RT 1 ( x, y ) w ( x, y ) 1 2 ( EP ) w, (1, k) k 1 2 ( EP ) 1 (2) Where: w s the weghted value of the adacent reference tag. ( x, y ) refers to the known physcal locaton of LT of the adacent reference tag. And for the reference tag close to RT, the postonng nformaton s endowed wth hgher weght n the postonng process. 2.2. VIRE Systematc Algorthm and Its Optmzaton VIRE algorthm s to further ntroduce vrtual reference tag based on LANDMARC algorthm and on the premse of no addtonal reference tag. More accurate locaton of trackng obect can be obtaned through excludng small probablty to further mprove the postonng accuracy of LANDMARC algorthm. Where, the ntroduced vrtual tag structure schematc dagram s as shown n Fgure 2 below: RT RT Fgure 2. VIRE system structure schematc dagram Snce the algorthm ntroduced vrtual reference tag, the algorthm manly consdered the sgnal strength value on vrtual reference tag. Common VIRE system s manly obtaned through lnear nterpolaton. Consequently, place n-1 vrtual tags between two adacent actual reference tags. Therefore, each vrtual physcal grd contans (n+1)2-4 vrtual reference tags and actual reference tags. Where, the sgnal strength value on horzontal lne and vertcal lne can be calculated wth formula 3 and 4 respectvely. Sk ( Ta n, b ) Sk ( Ta, b ) p Sk ( Ta n, b ) ( n 1 p) Sk ( Ta, b ) Sk ( Tp, b ) Sk ( Ta, b ) p n 1 n 1 (3) Sk ( Ta, b n ) Sk ( Ta, b ) q Sk ( Ta, b n ) ( n 1 q) Sk ( Ta, b ) Sk ( Ta, q ) Sk ( Ta, b ) q n 1 n 1 Where: Sk ( T, ) s the sgnal strength from the vrtual tag to the k reader on the coordnate system (, ) ; a [ ] n, b [ ] n, 0 p n 1 and p refer to the modulo operaton of to n ; 0 q n 1 and q refer to the modulo operaton of to n. Therefore, n a vrtual network grd, snce the sgnal strength of actual reference tag can be obtaned through a RFID reader and that of vrtual reference tag can be derved va sgnal strength nterpolaton or other fttng model, we can mark the areas wth smlar sgnal strength of tags to be measured, and carry out ntersecton operaton thus to obtan the coordnate value of the tags to be measured. Through the above VIRE postonng algorthm subsystem, the key pont of the algorthm s how to obtan the sgnal strength value of the vrtual tag. Tradtonal VIRE algorthms, however, adopt nterpolaton algorthm. Snce the sgnal strength s not presentng lnear relaton to dstance, tradtonal VIRE algorthms have bgger errors. Consequently, the paper optmzed the VIRE algorthm through establshng a BP neural network model. It set the dstance between the tag and RFID reader as nput nformaton and that between actual tags recevng (4) 168
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 to sgnal strength as output nformaton to establsh a neural network model. In addton, t analyzed the postonng precson based on newton nterpolaton-vire, lnear nterpolaton-vire and neural network-vire algorthm. Where, model solvng flow block dagram s as shown n Fg. 3 below: Intalz aton Calculate the sgnal strength value The sgnal strength of vrtual tag can be solved by utlzng lnear Mark the areas wth smlar strength of the tags Weghted qualtatve method of LANDMARC Fgure 3. Solvng flow block dagram of VIRE system model 3. ANALYSIS OF MODEL SIMULATION RESULTS The paper manly carred out smulaton comparson and analyss of the factors affectng ndoor postonng accuracy, ncludng the mpact of the number of vrtual reference tags on postonng accuracy, and that of the layout of dfferent reference tags on postonng accuracy. In addton, t performed smulaton comparson of LANDMARC system, lnear nterpolaton-vire system, newton nterpolaton-vire system and neural network-vire system, and evaluated the advantages and dsadvantages of each algorthm. The settngs of smulaton experment are as followng: Set the coordnates of the four card readers as (0, 0), (0, 8), (8, 0) and (8, 8), namely a 8*8m postonng area; there are a total of 16 reference tags. The dstance between reference tag and car reader s 1m and that between reference tag s 2m. The path loss exponent can be set as 2 snce the dstance between reference tags s 2m. In the paper, set the nearest adacent tag k=4, and selected 20 tags to be measured. The experment dstrbuton dagram s as shown n Fg. 4: Fgure 4. Smulaton test dstrbuton dagram 3.1. Smulaton Comparson of Reference Tags Layout It can be seen from RSSI-based RFID postonng algorthm that dfferent reference tags layouts ntroduced wll defntely affect the postonng. The common layout s n rectangle. However, rectangle layout wll 169
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 calculate the coordnates of the tags to be measured drectly takng the trangle formed by the most adacent tags accordng to the rectangle dstrbuton. Therefore, the coordnate error of the tags to be measured s relatvely bg. In order to reduce the postonng error arsng from reference tag layout, the reference tag can be set accordng to form of trangular layout. Set the lnear nterpolaton -VIRE postonng algorthm as an example. The cumulatve probablty of postonng error of dfferent reference label layout s as shown n Fg. 5 below. Fgure 5. Cumulatve probablty curve of dfferent layout It can be seen from the above smulaton results as shown n Fg. 5 that whle not addng reference tag and changng the rectangle layout to trangle layout, the cumulatve probablty curve of trangle layout wll be lower than that of rectangle layout. We can see from the Fg. that n case of an error of 0.4m, the error cumulatve probablty of trangular layout s only about 78%, whle that of rectangle layout s up to 93%. Therefore, the postonng accuracy of reference tag n trangular layout wll be hgher than that n rectangular layout. 3.2. Smulaton Comparson of Dfferent Vrtual Tag Number In the general cases, the more vrtual tags and the hgher dstrbuton densty, the hgher postonng accuracy wll be. But the vrtual tags shall not be added blndly only because no cost s needed. Generally, f the number of vrtual tags s oversze, t cannot mprove the postonng error, but nstead, t wll ncrease computng tme of system postonng. The paper set neural network-vire postonng algorthm as an example, and analyzed the nfluence of dfferent vrtual tags on postonng accuracy. Where, the smulaton results are as shown n Fg. 6: Fgure 6. Impact of dfferent vrtual tags on postonng error It can be seen from Fg. 6 that when N*N (sum total of vrtual tags) s 4*4, the postonng error s 0.17m; when N*N s 5*5, the postonng error s 0.173m; when N*N s10*10, the postonng error s 0.169m; when N*N s 15*15, the postonng error s 0.168m; when N*N s 25*25, the postonng error s 0.166m; and when N*N s 30*30, the postonng error s 0.166m. We can see that when the number of vrtual tags s 25*25, postonng error bascally remans around 0.166m wthout change. Therefore, when the number of vrtual tags has reached to certan value, t wll not mprove the postonng accuracy anymore. So f t shows no obvous mprovement of postonng accuracy, the number of vrtual tags should be reduced to ncrease postonng calculatng speed of the system and mprove the postonng accuracy. 170
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 3.3 Smulaton Comparson of Dfferent Model Algorthms Two ndoor postonng systematc algorthms were presented n Chapter 3: LANDMARC systematc algorthm and VIRE systematc algorthm. And n Chapter 4, an ndoor postonng method by utlzng Newton's dfference was proposed based on the two systematc algorthms. Expermental smulaton comparson of the mpacts of the three algorthms on postonng accuracy was ntroduced as below. Where, the ntal parameters of the models are: path loss exponent n=2, threshold value th=2.2, the most adacent tag number k=6 and vrtual tags number N=5. 3.3.1 Comparson of LANDMARC Algorthm and Lnear Interpolaton-VIRE Algorthm The experment conducted smulaton error comparson of the 20 ponts to be measured by adoptng LANDMARC systematc algorthm and VIRE systematc algorthm as shown n Fg. 7. We can see from the Fg. that except that the errors of 13# and 18# are close, all the other ponts, both the ones at center poston, 2#, 3#, 4#, 6#, 7#, 8#, 9#, 10#, 11#, 12#, 15#, 16# and 19#, and the ones at edge poston, 1#, 5#, 17# and 20#, the postonng error of VIRE systematc algorthm wll be much lower than that of LANDMARC systematc algorthm. The experment has verfed that by ntroducng the lnear nterpolaton of VIRE algorthm and utlzng adacent dagrams, the mpossble postons of the ponts to be measured can be elmnated. In addton, two weghtng factors are ntroduced to reduce the error whch has sgnfcantly mproved the postonng precson. The errors of LANDMARC algorthm and lnear nterpolaton-vire algorthm postonng system are 1.1158 and 0.3554 respectvely. Fgure 7. Comparson of the postonng error of LANDMARC algorthm and lnear nterpolaton-vire algorthm 3.3.2 Comparson of Lnear Interpolaton-VIRE Algorthm and Newton Interpolaton-VIRE Algorthm Although VIRE system postonng algorthm has ntroduced vrtual reference tag and nserted vrtual tags va lnear nterpolaton, the postonng accuracy has been sgnfcantly mproved comparng wth LANDMARC systematc algorthm. Fgure 8. Comparson of the postonng error of newton nterpolaton-vire and lnear nterpolaton -VIRE algorthm 171
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 However, due to the complex and vared ndoor envronment, as well as the multpath effect, vrtual tag cannot solve the sgnal strength value totally accordng to lnear nterpolaton whch wll cause bg error on postonng. Therefore, we ntroduced newton nterpolaton to obtan the sgnal strength of vrtual tag thus to reduce the error of vrtual tag sgnal strength. We carred out experment smulaton for VIRE systematc algorthm and newton nterpolaton systematc algorthm as shown n Fg. 8: We can see from the Fg. 8 that except that both the ones at center poston, 2#, 3#, 4#, 6#, 7#, 8#, 9#, 10#, 11#, 12#, 15#, 16# and 19#, and the ones at edge poston, 1#, 5#, 13#, 17# and 18#, the postonng error of newton nterpolaton-vire algorthm wll be much lower than that of VIRE systematc algorthm. The postonng errors of newton nterpolaton-vire algorthm and lnear nterpolaton -VIRE algorthm are 0.3087 and 0.3554 respectvely. 3.3.3 Comparson of Newton Interpolaton-VIRE Algorthm and Neural Network-VIRE Algorthm Through the analyss of newton nterpolaton-vire algorthm above, we can see that n case of multple outer dameter effects and dsturbance factors exstng n the postonng envronment, the sgnal strength of vrtual tag shall be derved by utlzng nonlnear nterpolaton algorthm. However, newton nsertng method s manly as smooth connecton specfc to newton nsertng method, not takng overfttng of the data nto consderaton. Consequently, the paper proposed to obtan the vrtual tag by settng BP neural network as the core. The sgnal strength value of vrtual tag was obtaned through BP neural network. And postonng smulaton was carred out. The postonng error dstrbuton and postonng dstrbuton results of newton nterpolaton algorthm are as shown n Fg. 9 and 10. Fgure 9. Comparson of the neural network-vire and newton nterpolaton-vire algorthm Fgure 10. Postonng dstrbuton results of neural network-vire system 172
Rev. Téc. Ing. Unv. Zula. Vol. 39, Nº 11, 166-173, 2016 It can be seen from Fg. 9 that for all postonng ponts, the postonng error obtaned through neural network predcton algorthm s lower than that of newton nsertng algorthm, partcularly on 14# and 20# tag. The error postonng result has reduced about 0.5. Consequently, the sgnal energy value of vrtual tag obtaned through neural network s more n lne wth the actual envronment status. The postonng system error of neural network-vire algorthm and newton nterpolaton-vire algorthm are 0.3087 and 0.2064 respectvely. 4. CONCLUSIONS We can get the followng conclusons based on the above model smulaton results: (1) Snce the rectangular reference layout has the defect as error amplfcaton, we proposed a trangle reference tag layout and verfed ts superorty. Through comparng the error probablty cumulatve fgure of rectangular layout, t shows that the postonng accuracy of trangular layout s hgher than that of rectangular layout. (2) The smulaton experment llustrates that the more vrtual tags ntroduced n VIRE algorthm do not means hgher postonng accuracy. Instead, after vrtual tags reachng a certan number, t wll not mpose mpact on postonng accuracy mprovement. (3) We conducted error comparson smulaton of LANDMARC algorthm, lnear nterpolaton-vire algorthm, newton nterpolaton-vire algorthm and neural network-vire algorthm. The experment results show that the postonng accuracy of lnear nterpolaton-vire algorthm s hgher than that of LANDMARC algorthm, and that of newton nterpolaton-vire s hgher than that of VIRE systematc algorthm. Comparng wth all of the above algorthms, neural network-vire algorthm presents hgher postonng accuracy, and s more capable to perform hgh accuracy postonng at harsh envronment. REFERENCES Dexn Yu, Zheng K, We Z. (2015) Vehcle Postonng Method Based on RFID n Vehcular Ad-Hoc Networks. Journal of Donghua Unversty(Englsh Edton), 32(5), pp.800-806. Han S, Km J, Park C H, et al. (2009) Optmal Detecton Range of RFID Tag for RFID-based Postonng System Usng the k-nn Algorthm, Sensors, 9(6), pp.4543-4558. Hong B W, Huang Y J, Chen C Y, et al. (2013) Fuzzy neural network based RFID postonng and navgaton method for moble robots, Research Journal of Appled Scences Engneerng & Technology, 6(7), pp.1233-1239. Jang H. (2012) Desgn and Implementaton of a RFID Based Postonng Method for Indoor Vehcle Montorng System, Computer Scence, 39(2), pp.29-33. Lu C, Xao-Chun L U, Jn-Song X U. (2012) A desgn of urban vehcle poston system based on GPS-RFID. Journal of Tme & Frequency, 35(3), pp.169-174. Saab S S, Mhanna W, Salba S. (2009) Conceptualsaton study for usng RFID as a stand-alone vehcle postonng system, Internatonal Journal of Rado Frequency Identfcaton Technology & Applcatons, 2(1/2), pp.27-45. Song X, L X, Tang W, et al. (2016) A fuson strategy for relable vehcle postonng utlzng RFID and n-vehcle sensors. Informaton Fuson, 31(C), pp.76-86. Saab S S, Nakad Z S. (2011) A Standalone RFID Indoor Postonng System Usng Passve Tags, Industral Electroncs IEEE Transactons on, 58(5), pp.1961-1970. Song X, L X, Zhang W, et al. (2016) RFID Applcaton for Vehcle Fuson Postonng n Completely GPS-dened Envronments, Engneerng Letters, 24(1), pp.19-23. Wang J, N D, L K. (2014) RFID-based vehcle postonng and ts applcatons n connected vehcles, Sensors, 14(3), pp.4225-38. X L, Jang M Y, Lao C Z. (2014) Indoor Postonng System Based on RFID, Appled Mechancs & Materals, 543-547, pp.2627-2631. 173