A Noble Fuzzy-based Mobile Tracking Scheme

Transcription

1 JOUNA OF NETWOKS, VO., NO., JUNE 006 A Noble Fuzzy-based Moble Trackng Scheme Soo-hang Km ETI, onverged Access Network esearch Team, aejeon, Korea sckm@etr.re.kr Jong-han ee and Yeon-Seung Shn Kunsan Natonal Unv., ept. of omputer Informaton Scence, Korea chan000@kunsan.ac.kr ETI, onverged Access Network esearch Team, aejeon, Korea shnys@etr.re.kr Abstract In ths study, we propose a novel moble trackng method based on Mult-rtera ecson Makng (MM, n whch uncertan parameters such as Plot Sgnal Strength (, the dstance between the moble and the base staton, the movng drecton, and the prevous locaton are used n the decson process usng the aggregaton functon n fuzzy set theory. In the mcrocell- or pcocell-based system the frequent movements of the moble brng about excessve traffcs nto the networks. A moble locaton estmaton mechansm can facltate both effcent resource allocaton and better Qualty of Servce (QoS provsonng through handoff optmzaton. Through numercal results, we show that our proposed moble trackng method provdes a better performance than the conventonal method usng the receved sgnal strength. Index Terms trackng, locaton management I. INTOUTION There wll be a strong need for the moble termnal trackng n the next generaton moble communcaton systems [-5]. The locaton of a Moble Termnal must be found out, e.g., n wreless emergency calls already n the near future. It s of great mportance to the effcency of next generaton moble communcaton systems to know the exact poston of the movng moble user n order to reduce the number of pagng messages and cell handover messages. Handover effcency wll be an mportant aspect n next generaton moble communcaton systems because t affects drectly to the swtchng road and QoS, partcularly n combned mcrocell of pcocell networks. The frequent movements of a moble termnal or host brng about excessve traffcs nto the network and may degrade the QoS severely. If ts locaton can be estmated, network resources may be more effectvely allocated and better QoS can be provsoned wth the combnaton of handoff optmzaton. It wll have vable roles n the communcaton networks of next generaton. Global Postonng System (GPS was ntally developed for mltary purposes but t s also utlzed for cvl applcatons such as local traffc nformaton servces and geo-locaton based applcatons. However ncorporatng GPS recevers nto handsets rases questons of cost, sze and power consumpton [6]. Other methods for locaton estmaton are based on rado sgnal propagaton such as sgnposts, dead reckonng, crcular or hyperbolc trlateraton systems, etc. Many methods and systems have been proposed based on rado sgnal strength measurement of a moble object's transmtter by a set of base statons [7-9]. ecently, adaptve schemes based on the use of cellular systems and on fuzzy logc [0], hdden Markov models [-] and pattern recognton methods [] have been used to estmate the poston of mobles. The system studed n [7] estmates moble locaton usng nformaton on contours but t does not provde a realstc search procedure. In ef. [8] the estmaton s based on the sgnal strength receved at a mult-beam antenna of a base staton n the mult-path envronment, and the angle of ts arrval (AOA. AOA s measured under the assumpton that the sgnal s n lne of sght (OS, but OS sgnal may not be receved n the mcrocell where reflectons and dffractons occur due to dense buldng envronment. In ths stuaton AOA of the strongest reflected sgnal s utlzed for estmaton, and therefore the locaton estmated dffers greatly from real one. Tme of arrval (TOA of a sgnal from a moble to neghborng base statons are used n [], but ths scheme has two problems. Frst, an accurate synchronzaton s essental between all sendng endponts and all recevng ones n the system. An error of µs n synchronzaton results to 00 m error n locaton. Secondly ths scheme s not sutable for the mcrocellular envronment because t also assumes OS envronment. Tme dfference of arrval (TOA of sgnals from two base statons s consdered n [5]. TOA scheme and TOA scheme have been studed for IS-95B where PN code of MA system can be used for the locaton estmaton. Enhanced Observed Tme fference (E-OT s a TOA postonng method based on OT feature already exstng n GSM. The moble measures arrval tme of sgnals from three or more cell stes n a network. In ths method the poston of moble s determned by trlateraton [6]. E-OT, whch reles upon the vsblty of at least three cell stes 006 AAEMY PUBISHE

2 JOUNA OF NETWOKS, VO., NO., JUNE 006 to calculate t, s not a good soluton for rural areas where cell-ste separaton s large. The above-mentoned schemes such as AOA, TOA and TOA have problems as follows. - These schemes assume that the cellular system conssts of OS areas. They get good results only under ths assumpton. - The mcrocellular system such as IMT-000 has NOS areas whch are affected by specfc reflectons and dffractons. In ths stuaton these schemes have great errors n estmaton. - In the mcrocellular envronment the ponts of the same average sgnal strength form not a crcular contour but a dstorted one. These schemes gnore the fact that the propagaton rule s affected by many parameters. - They rely only on the nformaton related to rado sgnal such as sgnal strength. Ther accuraces are affected by short-term fadng, shadowng or dffracton. In ths study, to enhance estmaton accuracy, we propose a scheme based on Mult-rtera ecson Makng (MM whch consders multple parameters: the sgnal strength, the dstance between the base staton and moble, the movng drecton, and the prevous locaton. Ths process s based on three step locaton estmatons whch can determne the moble poston by gradually reducng the area of the moble poston [7]. Usng MM, the estmator frst estmates the locatng sector n the sector estmaton step, then estmates the locatng zone n the zone estmaton step, and then fnally estmates the locatng block n the block estmate step. Ths paper s organzed as follows. Secton II descrbes locaton defnton for our system. Usng the concepts descrbed n Secton II, our estmaton scheme s consdered n Secton III. Smulaton envronment and results are shown n Secton IV n order to compare our schemes wth other schemes. Fnally concludng remarks are gven n Secton V. II. OATION MOE The locaton model for our study s shown n Fg.. The locaton of a MT wthn a cell can be defned by dvdng each cell nto sectors, zones and blocks and relatng these to the sgnal level receved by t at that pont. It s done automatcally n three phases of sector defnton, zone defnton and block defnton. Then the locaton defnton block s constructed wth these results. They are performed at the system ntalzaton before executng the locaton estmaton. The sector defnton phase dvdes a cell nto sectors, and assgns a sector number to blocks belongng to each sector. The zone defnton phase dvdes each sector nto zones, and assgns a zone number to blocks belongng to each zone. The block defnton phase assgns a block number to each block. In order to ndcate the locaton of each block wthn a cell, -dmensonal vector (dstance d, angle a s assgned to each block. After the completon of ths phase each block has a set of block nformaton. The locaton estmator presents the poston of the Moble Termnal (MT usng a block number. In our paper we need the drecton nformaton to dentfy the relatve locaton of a block from the BS so we use the polar coordnates converted from the rectangular coordnates. sector 5 sector 6 zone n sector zone n- y axs zone 0 zone ± ± (x,y (x,y (x,y (x,y (x,y block n sector sector Fgure. Sectors, Zones and Blocks. x axs sector That s, n order to ndcate the locaton of each block wthn a cell, we use the vector data whch s obtaned by convertng the rectangular coordnate of the block to the polar coordnate wth the orgn of the BS. Based on the above locaton defnton method, each block s gven hs locaton nformaton whch conssts of one center pont (x, y and four of area pont (x j, y j as shown n Fg.. By comparng the locaton nformaton of each block and the poston nformaton of the MT, the poston of the MT wthn a cell s estmated. Usng MM and the block object whch s constructed as descrbed above, the proposed scheme estmates the optmal block at whch the MT s located. Ths scheme s mplemented as an estmator at the BS. The estmator s started wth a tmer, and the estmaton s performed sequentally n three steps: sector estmaton, zone estmaton, and fnally block estmaton. A. Sector efnton The sector defnton phase dvdes a cell nto sx sectors and assgns a sector number to blocks belongng to each respectve sector as shown n Fg.. All blocks belongng to the same sector have the same sector number. In order to dentfy whch blocks belong to whch sector, we utlze the drecton nformaton of a vector whch shows the drecton of a block from the orgn of the base staton. The sector defnton procedure s summarzed as follows. vde a cell nto sx sectors of a sze. Assgn an angle for each sector at the nterval of / respectvely. ompare the drecton nformaton of a block wth the angle of each sector, and then determne whch sector t belongs to. Assgn the correspondng sector number to each block. 006 AAEMY PUBISHE

3 JOUNA OF NETWOKS, VO., NO., JUNE 006 B. Zone efnton The zone defnton phase dvdes each sector nto zones, and assgns a zone number to blocks belongng to each respectve zone as shown n Fg.. Ths defnton phase can be descrbed wth two dfferent algorthms dependng on OS model and NOS model. Each cell s dvded nto n zones, wth each zone classfed by threshold as shown n Fg.. Z_A_ Z_A_ Z_A_n Z_OP n BS Zone Zone Zone n stance 0 n BS BS Fgure. Zone defnton wth. The followng algorthm summarzes the zone defnton procedure for OS model; - Select each threshold by consderng. - In order to map the sgnal strength onto the drecton nformaton, determne the dstance functon for each threshold wth (. p A ( d = k k log( d + u( d d = k k log( d + υ( ( p B ( d In ( ndcates the dstance between two base statons, and d the dstance between the base staton A and the moble. The k s proportonal to the transmsson power of staton and k has an offset value dependng on the rado propagaton envronment. Two random sgnals u (d, υ (d represent power dstrbutons of sgnals receved at a dstance d from staton A and from staton B respectvely. They have..d (dentcal ndependent dstrbuton wth Gaussan dstrbuton of N ( μ( d, σ. In our study the sgnal attenuaton n NOS envronment s estmated based on the dstance n OS model, and ths result s reflected n the block object nformaton. hanges n OS and NOS envronment are depcted by k. a. lassfy zones usng the dstance functon. b. Assgn a same zone number and a threshold to all blocks that belong to a same zone. But above algorthm for OS s not suffcent n the envronment where blocks have a buldng or a hll. The boundary lne for dentfyng each zone of the same sgnal strength s severely dstorted due to shadowng and dffracton. Further refnement wth NOS data s requred as shown below. a. Identfy zones by above OS algorthm. b. From blocks belongng to the same zone, select blocks whch need NOS offset. As shown n Fg., each cell s dvded nto blocks and a poston s defned for each block. urng ths cell plannng the blocks needng NOS offset are selected. Ths nformaton s stored n the data base. c. Assgn NOS offset, k, of ( to the blocks selected. d. Fnd the dfference between k and threshold whch s determned by OS model. e. Accordng to (, a new zone s defned to whch the correspondng block wll belong. Here, the dfference represents the dfference between the poston of the zone determned accordng to OS model and that of when k s assgned to the correspondng block.. Block efnton In the block defnton phase a block number s assgned to each block as shown n Fg.. Start y A x : Passed block : ross pont : oad Fgure. Identfyng a block usng a vector. 0 B End In order to ndcate the locaton of each block wthn a cell, we use the vector data obtaned by convertng the rectangular coordnate of a block to the polar coordnate wth the orgn of the base staton. Each vector has the nformaton on a dstance and an angle. If the moble moves along the path (from A to B, t ndcates the locaton of the block that the moble passes through wthn the cell. In order to represent the locaton of a block relatve to the base staton, we assgn a - dmensonal vector (d, a to each block. The set of block nformaton s called the block object. The block object contans the followng nformaton: the sector number, the zone number, the block number, the vector data (d, a, the maxmum and the mnmum values of the average for the OS block, the compensated value for the NOS block, and a bt for ndcatng node or edge. 006 AAEMY PUBISHE

4 JOUNA OF NETWOKS, VO., NO., JUNE 006 III. MOBIE TAKING BASE ON FMM A. Fuzzy mult-crtera decson parameters In our paper, the receved sgnal strength, the dstance between the MTs and the BS, the prevous locaton, and the movng drecton are consdered as decson parameters. The receved sgnal strength from the present base staton and ts neghborng base statons has been used n many schemes, but t has very rregular profles due to the effects of rado envronments. The dstance, whch s measured based on sgnal receved from the present base staton, s consdered because t can explan the block allocaton plan; however, t may also be naccurate due to the effect of mult-path fadng, etc. It s not suffcent by tself. We consder the prevous locaton. It s normally expected that the estmated locaton should be near the prevous one. Therefore, f the estmated locaton s too far from the prevous one, the estmaton may be regarded as naccurate. We also consder the movng drecton. Usually the MT s most lkely to move forward, less lkely to move rghtward or leftward, and least lkely to move backward more than one block. The low-speed MT (.e., a pedestran has a smaller movng radus, that s, the dstance that a MT s expected to move from hs current locaton s short and a more complex movng pattern, whle the hgh-speed MT (.e., a motor vehcle has a larger radus and a smpler pattern. In moble trackng usng MM, the ecson functon s defned by combnng the degree of satsfacton for multple evaluaton parameters, and the decson s made on the bass of hs functon. The evaluaton parameter can be seen as a proposton. A compound proposton s formed from multple evaluaton parameters wth a connectve operator, and the total evaluaton s performed by totalng the values for the multple parameters wth connectve operators. In ths method errors n the evaluaton parameters mpose mlder changes on the total evaluaton value than n bnary logcs. Ths method can also consder multple naccurate and nsuffcent evaluaton parameters smultaneously and can compensate for them. Ths results n the optmal decson. In our study the measure of the rato was used for ndcatng the evaluaton parameter and a weght was mposed accordng to the degree of mportance of each evaluaton parameter. B. Membershp functon The membershp functon wth a trapezodal shape s used for determnng the membershp degree of the MT because t provdes a more versatle degree between the upper and the lower lmts than the membershp functon wth a step-lke shape. et us defne the membershp functons for the s from neghborng BSs. The membershp functon of, μ (, s gven by (. s the sgnal strength receved from the BS, s s the lower lmt, and s s the upper lmt. values are used for formng the membershp functon wth a trapezodal shape and for determnng the membershp degree of estmated sgnal strength. 0, < s s μ = (, s s ( s s, > s Now we defne the membershp functon of the dstance. The membershp functon of the dstance μ ( s gven by (, where s the dstance between the BS and the MT [8-9]. d s the upper lmt, and d s the lower lmt., < d d μ = (, d d ( d d 0, > d The membershp functon of the prevous locaton of the MT μ ( s gven by (, where s a scalar value defned by the estmated current locaton, E,, E s also a scalar value respectvely and a trapezodal fuzzy number for the prevous locaton (a vector value of a moble [8-9]. E,, E s used for determnng the membershp degree of estmated current locaton. In other words, f a current locaton estmated s too far away from a prevous locaton (that s, f the membershp degree s too small, t s lkely that we have ncorrect locaton estmated. μ 0; E, E E ( =, E, E E 0, < E E E E E E E > E The membershp functon of the movng drecton μ ( s gven by (5. s a scalar value defned by the movng drecton of the MT, H,, H means the prevous drectons and used for determnng the membershp degree of estmated movng drecton. The movng drecton ( s defned by comparng the vector nformaton of prevous block and that of the estmated block. μ 0, H, H H ( =, H, H H 0, H H H < H H H H > H. ocaton estmaton Most of the MM approaches face the decson problem n two consecutve steps: aggregatng all the judgments wth respect to all the crtera and per decson ( (5 006 AAEMY PUBISHE

5 JOUNA OF NETWOKS, VO., NO., JUNE alternatve and rankng the alternatves accordng to the aggregated crteron. Also our approach uses ths twosteps decomposton [0-]. Each decson problem nvolves n alternatves and m lngustc attrbutes correspondng to m crtera. Thus, decson data can be organzed n a m n matrx. The decson matrx for alternatves s gven by (6: μ ( μ ( μ = μ ( μ ( n μ ( μ ( μ ( μ ( n μ ( μ ( μ ( μ ( n μ ( μ ( μ ( μ ( The weghtng vector for evaluaton crtera can be gven by usng lngustc termnology wth fuzzy set theory []. It s a fnte set of ordered symbols to represent the weghts of the crtera usng the followng lnear orderng: very hgh hgh medum low very low. Weghtng vector W s represented as (7. nm (6 W = ( w, w, w, w (7 The fuzzfcaton procedure leads to a performance n m matrx μ [0,] where each element μ nm expresses how much the n-th alternatve satsfes the m-th crteron. Therefore, each low of the performance matrx s a fuzzy set μ m expressng the satsfacton of the m-th crteron n the unverse of the avalable alternatves [-]. By multplyng the weghtng vector by the decson matrx, the performance matrx s gven by (8: μ ( w μ( w μ( w μ( w μ( w μ( w μ( w μ( w = μ μ ( w μ( w μ( w μ( w (8 μ( n w μ( n w μ( n w μ( nm w Gven the decson matrx and the weghtng vector, the decsonmakng objectve for the general fuzzy MM problem s to rank all alternatves by gvng each of them an overall preference ratng wth respect to all crtera [-]. GMV (Generalzed Mean Value s used for rankng the alternatves accordng to the aggregated crteron. The GMV for alternatves s represented as (9. where ( + ( A + B + A B m( μ n = (9 [( + ( A + B ] A = μ w, ( n ( n w = μ, = μ ( w B = μ w nm ( n, respectvely., and Sector estmaton based on mult crtera parameters The decson parameters consdered n the Sector Estmaton step are the sgnal strength, the dstance and the prevous locaton. The MT s estmated to be located at the sector neghborng to the BS whose total membershp degree s the largest. The sector estmaton s performed as follows. Procedure : Membershp degrees are obtaned usng the membershp functon for the sgnal strength, the dstance and the prevous locaton. Procedure : Membershp degrees obtaned n Procedure for the BS neghborng to the present staton are totalzed usng the fuzzy connectve operator as shown n (0. μ = μ μ ( μ ( (0 ( We obtan ( by mposng the weght on μ. The reason for weghtng s that the parameters used may dffer n ther mportance. ωμ = μ ( W + μ ( W + μ ( W ( where W s the weght for the receved sgnal strength, W for the dstance, and W for the locaton. Also W + W + W s, W s 0.5, and W and W are 0., 0., respectvely. Procedure : Blocks wth the sector number estmated are selected from all the blocks wthn the cell for the next step of the estmaton. Selecton s done by examnng sector number n the block object nformaton. Zone estmaton based on mult crtera parameters The decson parameters consdered n the Zone Estmaton step are the sgnal strength, the dstance and the movng drecton. From the blocks selected n the sector estmaton step, ths step estmates the zone of blocks at one of whch the MT locates usng the followng algorthm. Procedure : Membershp degrees are obtaned usng the membershp functon for the sgnal strength, the dstance and the movng drecton. Procedure : Membershp degrees obtaned n Procedure s totalzed usng the fuzzy connectve operator as shown n (. μ = μ μ ( μ ( ( ( We obtan ( by mposng the weght on μ. ωμ = μ ( W + μ ( W + μ ( W ( where W, W, and W are assumed to be 0.6, 0., 0., respectvely. Procedure : Blocks whch belong to the zone estmated above are selected for the next step. It s done by examnng the zone number of the blocks selected n the sector estmaton. Block estmaton based on mult crtera parameters The decson parameters to be consdered n the Block Estmaton step are the sgnal strength, the dstance and the movng drecton. From the blocks selected n the zone estmaton step, ths step uses the followng 006 AAEMY PUBISHE

6 6 JOUNA OF NETWOKS, VO., NO., JUNE 006 algorthm to estmate the block n whch the MT may be located. Procedure : Membershp degrees are obtaned usng the membershp functon for the sgnal strength, the dstance and the movng drecton. Procedure : Membershp degrees obtaned n Procedure are totalzed usng the fuzzy connectve operator as shown n (. μ = μ μ ( μ ( ( ( We obtan (5 by mposng the weght on μ. ωμ = μ ( W + μ ( W + μ ( W (5 where W, W, and W are assumed to be 0.6, 0., 0., respectvely. Procedure : The selecton s done by examnng the block number of the blocks selected n the zone estmaton. IV. PEFOMANE ANAYSIS A. Smulaton Model The movng pattern s descrbed by the changes n movng drecton and velocty. In our study we assume that low speed mobles, pedestrans, occupy 60% of the total populaton n the cell and hgh-speed mobles, vehcles, 0%. One half of the pedestrans are assumed to be stll and another half movng. Also the prvate owned cars occupy 60% of the total vehcle, the tax 0% and the publc transportaton 0%. Vehcles move forward, leftward/rghtward and turn-back. The movng velocty s assumed to have a unform dstrbuton. The walkng speed of pedestrans s 0~5km/h, the speed of prvate cars and taxs 0~00km/hr, and buses 0~70km/h. The speed s assumed to be constant durng walkng or drvng. Fgure shows the road used n our smulaton to consder traffc envronments. The black crcle ndcates the branch of the road, and the shaded areas are blocks that the road passes through. Each block s a square and ts sde s assumed to have the length of 0m. The tme needed for a hgh speed moble to pass through a block s calculated from BT = r / υ where r s the length of the road segment crossng at each block and v the moble speed. As shown n Fgure, BT s dependent on r. We can consder four dfferent values - r, n m (crossng dagonally, n m (/ crossng, n m (/ crossng and n m (/ crossng - accordng to whch porton of a block each road segment crosses through. In order to reflect more realstc nformaton nto our smulaton, t s assumed that the sgnal strength s sampled every 0.5sec, 0.sec, 0.sec, 0.sec and 0.05sec for the speed of 0km/h, 0km/h, 50km/h, 70km/h and 00km/h, respectvely. If BT s too small, we cannot obtan enough samples to calculate the average sgnal strength. We consder the followng smulaton parameters regardng the receved sgnal strength. The value of k, whch ndcates the changes n OS/NOS envronments, s n the range of 0 through 50. The mean sgnal attenuaton by the path-loss s proportonal to.5 tmes the propagaton dstance, and the shadowng has a log-normal dstrbuton wth a standard devaton of σ = 6dB. A value of receved sgnal strength less than 6 db s regarded as an error, whch s therefore excluded from the calculaton. B. Smulaton esults The effect of the block sze on the estmaton performance of MM and the exstng schemes s shown n Fg.. As the block sze becomes smaller, the accuraces of the three schemes decrease. The accuraces of AOA and TOA decrease rapdly. On the other hand, the performance of MM s least affected by block sze because t addtonally utlzes the prevous locaton and dstance between the moble and the base staton for estmaton. Estmaton rate AOA TO A MM The block sze[m] Fgure. The estmaton rate and the block sze. The estmaton rate or accuracy of our proposed scheme dependng on the moble speed s shown n Fg. 5. The accuracy of AOA and TOA becomes lower rapdly snce the sgnal measurement error would be large as the moble speed ncreases. The performance of MM s least affected by the moble speed because the nformaton such as movng drecton and prevous locaton are consdered n MM and, therefore, errors durng the sgnal evaluaton step decrease. Estmaton rate AOA TOA MM The speed[km/h ] Fgure 5. The estmaton rate and the moble termnal s speed. 006 AAEMY PUBISHE

7 JOUNA OF NETWOKS, VO., NO., JUNE The estmaton results of three schemes for the stuaton where the hgh speed moble moves along a straght or curved sector boundary area s shown n Fg. 6. In ths fgure the horzontal and vertcal axes represent the relatve locaton of the area observed and the path generated n ths smulaton. esults are shown for Area Parttonng (AP, Vrtual Area (VA [5, 7] and MM from left to rght n ths fgure respectvely. As can be seen AP sometmes selects faulty locatons far away from the generated path. That s because naccurate results n sector estmaton stage are escalated nto zone and block estmaton. VA has better accuraces for curved path. In our understandng t may be attrbuted to the fact that the average value of plot sgnal strengths sampled by hgh speed moble passng through two sectors falls nto the range of values of the sector boundary area. The performance of MM s less affected durng a left turn or rght turn. A left or rght turn causes abrupt sgnal dstorton, but ther effects on estmaton can be compensated for by usng nformaton on prevous locaton and dstance to base staton. Especally, t can reduce the possblty of selectng a faulty locaton far away from present locaton by usng nformaton on prevous locaton. (a Area Parttonng Scheme. (b Vrtual Area Scheme. We compare our scheme, MM wth VA [7], E- OT and TOA n Fg. 7 []. In ths fgure the moble mantans ts y poston at,000m and traverse x axs from x = 0m to x =,000m. The y axs means how the drms vares wth x poston of moble. stance oot Mean Square (rms stands for the root-mean-square value of the dstances from the true locaton pont of the poston fxes n a collecton of measurements. In order to get the estmated values for comparson we take an average of 0 values for each moble poston. We assume NOS envronment and the sgnal level of moble may change abruptly due to shadowng. It shows that the performance of MM s least affected by abrupt change of sgnal level. MM has the most accurate result. Ths may well be attrbuted to the fact that t mposes less weght on the receved sgnal strength n NOS area and, nstead, greater weghts on other parameters such as the dstance between moble and base staton, prevous locaton, and movng drecton are consdered as decson parameters. stance root mean squre, y=km MM VA E-OT TOA Movng dstance of moble termnal [km] Fgure 7. omparson of the estmaton accuracy. V. ONUSIONS In ths study, we proposed a MM-based moble trackng method for estmatng more accurately the moble locaton by consderng multple parameters, such as the sgnal strength, the dstance between the base staton and moble, the movng drecton, and prevous locaton. We have demonstrated that our scheme ncreases the estmaton accuracy when the moble moves along a boundary area. The effect of weght factor varatons on the estmaton performance of our scheme and the determnaton of the optmal weght should be the subject of a future study. Also further researches are requred on ther mplementaton and applcatons to the handoff and channel allocaton strateges. EFEENES (c MM scheme. Fgure 6. The estmaton results on the move. [] B. Z. Theodore, G. V. Konstantnos, P. T. hrstos, A. Z. Nkolaos and A. N. Nkos, "Global oamng n Next- Generaton Networks," IEEE ommuncatons Magazne, pp. 5-5, February 00. []. Govann, F. oberto,. Pergorgo,. Salvatore, E. Marcello, P.. Smon and. Ada, "End-User Servces n 006 AAEMY PUBISHE

8 8 JOUNA OF NETWOKS, VO., NO., JUNE 006 Premum IP Networks," IEEE omm. Magazne, pp. 5-60, January 00. [] M. Vctor,. A. u, G. arlos, I. M. Jose, B. hrstophe, M. Erc and. Marco, "An IP-Based QoS Archtecture for G Operator Scenaros," IEEE Wreless omm., pp. 5-6, June 00. [] S. Y. Hu and K. H. Yeung, "hallenges n the Mgraton to G Moble Systems," IEEE omm. Magazne, pp. 5-59, ecember 00. [5] J.. ee, B. Y. yu and J. H. Ahn, "Estmatng the Poston of Mobles by Mult-rtera ecson Makng," ETI Journal, Vol., Num., pp. -7, Aug. 00. [6] G. M. juknc and. E. chton, Geolocaton and Asssted GPS, IEEE omputer, Vol., No., pp.- 5, Feb. 00. [7] W. G. Fgel, N. H. Shepherd and W. F. Trammell, "Vehcle locaton by a sgnal attenuaton method," IEEE Trans. Veh. Technol., vol. VT-8, pp. 0-09, Nov [8] G.. Ott, "Vehcle locaton n cellular moble rado systems," IEEE Trans. Veh. Tech., Vol. VT-6, pp. -6, Feb [9] M. Hatta and T. Nagatsu, "Moble ocaton Usng Sgnal Strength Measurements n a ellular System, IEEE Transactons on Vehcular Technology, vol. VT9, pp5-5, May 980. [0] H.. Song, "Automatc Vehcle ocaton n ellular ommuncaton Systems," IEEE Transactons on Vehcular Technology, vol., pp90-908, Nov. 99. [] O. Kennemann, "Pattern ecognton by Hdden Markov Models for Supportng Handover ecsons n the GSM system," Proc. 6th Nordc Semnar g. Moble ado omm., Stockholm,Sweden, pp.95-0, 99. [] T. Nypan and O. Hallngstad, ellular Postonng by atabase omparson and Hdden Markov Models., PW00, pp. 77-8, Oct. 00 [] O. Kennemann, "ontnuous ocaton of Movng GSM Moble Statons by Pattern ecognton Technques, n Proc. 5th Int. Symp. Personal, Indoor, Moble, ado omm., denhaag, Holland, pp.60-6, 99. [] H. Staras and S. N. Honkman, "The accuracy of vehcle locaton by trlateraton n a dense urban envronment," IEEE Trans. Veh. Tech., vol VT-6, pp. 8-, Feb. 97. [5] T. S. appaport, J. H. eed and B.. Woerner, "Poston ocaton Usng Wreless ommuncatons on Hghways of the Future," IEEE ommuncatons Magazne, pp. -, Oct. 00. [6] Y. A. Sprto, On the Accuracy of ellular Moble Staton ocaton Estmaton, IEEE Trans. Veh. Technol., vol. 50, no., pp , 00. [7] J.. ee and Y. S. Mun, "Moble ocaton Estmaton Scheme," SK telecommuncatons evew, Vol. 9, No. 6, pp , ec [8] M.. Austne, "recton based handoff algorthms for urban mcrocells," VT99, Vol., pp. 0-05, 996. [9]. H. Yeh and H. eng, An algorthm for fuzzy multcrtera decson makng, IEEE IIPS '97, Vol., pp , 997. [0]. J., J. Y. Kuo, and H. W. T., Fuzzy decson makng through relatonshps analyss between crtera, Fuzzy Systems Symposum, Soft omputng n Intellgent Systems and Informaton Processng, pp. 96 0, 996. [] Z. H. J. and S. H. J., Applcaton of fuzzy logc to engneerng desgn and confguraton problems-a survey, IEEE Internatonal onference on Fuzzy Systems, Vol., pp. 0-5, 996. []. Naso and B. Turchano, "A Fuzzy Mult-rtera Algorthm for ynamc outng n FMS," IEEE ISM 998, Vol., pp. 57-6, Oct []. H. Yeh and H. eng, "An Algorthm for Fuzzy Mult- rtera ecson Makng," IEEE IIPS 997, pp , 997. [] S.. Km, J.. ee, Y. S. Shn and K.. ho, Moble Trackng usng Fuzzy Mult-rtera ecson Makng, Internatonal onference on Moble Ad-hoc and Sensor Networks 005 (MSN 05, ecture Notes n omputer Scence (NS, vol. 79/005, pp , ec Soo-hang Km was born on February 6, 96 n Samcheok, Korea. He receved hs Ph.. n computer and communcaton engneerng from hungbuk Natonal Unversty, hungju, Korea n 006. Before that, he receved a M.S. n computer scence from hungnam Natonal Unversty, aejeon, Korea and a B.S. n computer scence from Hongk Unversty, Seoul, Korea, n 995 and 986, respectvely. He s a prncpal engneer n Moble Access esearch Group at Electroncs and Telecommuncatons esearch Insttute (ETI, Korea. Hs current research nterests are n the areas of rado resource management, moble access networks of next generaton moble communcaton system and All-IP. He s actve n Next Generaton Moble ommuncaton (NGM Forum of Korea. Jong-han ee was born on September 5, 967 n Taean, Korea. He has receved an M.S. and Ph.. n computer scence and engneerng from Soongsl Unversty, Seoul, Korea, n 996 and 000, respectvely. He s a professor at the ept. of Informaton Scence, Kunsan Natonal Unversty. He was a senor member of the engneerng staff n the Flexble Access Networks ab. of the Electroncs and Telecommuncatons esearch Insttute (ETI. He has publshed more than 0 papers n Internatonal journals, and he has 5 patents. Hs current research nterests are n the area of resource management and handover for the next generaton moble multmeda networks. Yeon-Seung Shn receved hs M.S n statstcs from Korea Unversty, Seoul, Korea n 987. Before that, he receved a B.S. n statstcs from Kangwon Natonal Unversty, huncheon, Korea, n 98. He s a team leader n Moble Access esearch Group at Electroncs and Telecommuncatons esearch Insttute (ETI, Korea. Hs current research nterests are n the areas of moble access systems and S. 006 AAEMY PUBISHE