Handover location accuracy for travel time
estimation in GSM and UMTS
David Gundlegård and Johan M Karlsson
Linköping University Post Print
N.B.: When citing this work, cite the original article.
Original Publication:
David Gundlegård and Johan M Karlsson, Handover location accuracy for travel time estimation in GSM and UMTS, 2009, IET Intelligent Transport Systems, (3), 1, 87-94.
http://dx.doi.org/10.1049/iet-its:20070067
Copyright: Institution of Engineering and Technology (IET)
http://www.theiet.org/
Postprint available at: Linköping University Electronic Press
Handover Location Accuracy for Travel Time
Estimation in GSM and UMTS
David Gundlegård
Johan M Karlsson
Department of Science and Technology,
Linköping University,
SE-601 74 Norrköping,
Sweden
Abstract
In this p ap er, field m easu rem ents from the GSM and UMTS netw orks are an-alysed in a road traffic inform ation context. The m easu rem ents ind icate a p o-tentially large im p rovem ent u sing UMTS signalling d ata com p ared to GSM regard ing hand over location accu racy. These im p rovem ents can be u sed to generate real-tim e traffic inform ation w ith higher qu ality and extend the ge o-grap hic u sage area for cellu lar based travel tim e estim ation system s. The r e-su lts confirm p reviou s rep orts ind icating that he technology has a large p o-tential in GSM and they also show that the poo-tential m ight be even larger u s-ing UMTS. Assu m s-ing that non vehicle term inals can be filtered ou t, that veh i-cles are tracked to the correct rou te and that hand overs can be p red icted co rrectly, a conclu sion from the exp erim ents is that the hand ov er location accu -racy in both GSM and UMTS w ill be su fficient to estim ate u sefu l travel tim es, also in u rban environm ents. If there is a scalable w ay of d oing this filtering, tracking and p red iction in u rban environm ents is not clear to the au thors t o-d ay.
1 Introduction
The behaviou r of the GSM netw ork w hen it com es to generating road traffic inform ation has been analysed for m ore than a d ecad e, e.g in [1-3]. H ow ever, the UMTS netw ork has not been a target for the sam e analysis, w hich m eans that it is not clear how the d ifferent characteristics of the UMTS netw ork, com p ared to the GSM netw ork, w ill affect the qu ality of the generated traffic inform ation. Efforts to estim ate traffic inform ation from cellu lar netw orks started in mid 1990’s, when the research p roject CAPITAL was initiated [4]. The p rojects follow ing CAPITAL have taken m any d ifferent ap p roaches to extract inform ation from the cellu lar netw orks, and several p ilot projects are cu rrently active in, am ong other cou ntries, USA, Israel, Sw ed en and the N etherland s. A general travel tim e estim ation algorithm can inclu d e the fo l-low ing basic step s: location d ata collection, id entification of vehicle located term inals, m ap m atching, p ossible rou te d eterm ination and travel tim e calcu -lation.
In ad d ition to th ese basic step s the algorithm shou ld be able to com p ensate for anom alies like term inals in a bu s w ith a sep arate lane, term inals on trains, term inals on m otorcycles or bicycles etc. It is very d ifficu lt to filter ou t bicy-cles and m otorcybicy-cles, w hich m eans that they w ill bias the travel tim e estim a-tion. H ow ever, recently there have been efforts in trying to filter ou t p u blic transp orts u sing e.g. tim e-tables [5]. With know led ge of the sections w ith sep arate lanes other ad -hoc filtering m ethod s cou ld m ost likely im p rove ac-cu racy fu rther. Map m atching in this context is analysed in e.g. [6]. The rou te d eterm ination problem is d iscu ssed in e.g. [7, 8]. A su rvey and a d etailed d e-scrip tion abou t the d ifferences in location d ata for generating road traffic in-form ation from the GSM and UMTS netw orks can be fou nd in [9]. Recent e x-tensive su rveys of the area can be fou nd in [10, 11, 12, 13].
Althou gh the technology of u sing cellu lar netw orks to estim ate road traffic inform ation has been su bject for analysis for qu ite som e tim e, it is still far from being m atu re. It is not clear w hat to exp ect from these system s in term s of accu racy, availability and coverage. The potential of the system is althou gh qu ite clear, it is p ossible to retrieve a lot of traffic d ata in a cost efficient w ay, i.e. by u sing existing signalling d ata w ithou t the need to invest in sensor i n-frastru ctu re. Evolving com m u nication system s w ill change com m u nication p atterns, and both the d esign of the com m u nication system and the com m u -nication p attern of the u sers w ill affect the p otential of estim ating road traffic inform ation from cellu lar netw orks. In ord er to realise a safer and m ore effi-cient transp ortation system , it is likely that vehicles need to com m u nicate m ore frequ ently w ith each other and w ith som e kind of traffic control centre.
This traffic generates cellu lar netw ork signalling d ata that can be very u sefu l in ord er to estim ate road traffic inform ation in a cost efficient w ay. Interes t-ingly, w hen vehicles com m u nicate m ore, these system s w ill ge nerate better traffic inform ation. Better traffic inform ation w ill p robably generate m ore com m u nication, w hich once again w ill generate better traffic info rm ation, and this can be seen as a p ositive sp iral of traffic inform ation collection.
This p ap er focu s on com p aring hand over location inform ation betw een GSM and UMTS netw orks in tw o d ifferent environm ents, a sp arse u rban e n-vironm ent w ith slow traffic and a su bu rban highw ay enn-vironm ent w ith hig h-er sp eed lim it. The hand ovh-er location d ata can be an integr al p art w hen esti-m ating road traffic inforesti-m ation u sing p assive esti-m onitoring of cellu lar systeesti-m s. In ord er to estim ate a travel tim e or average sp eed after the correct rou te has been id entified , tw o w ell d efined locations are need ed . The tim e betw een these locations are then u sed as travel tim e for the segm ent or for calcu lation of average segm ent sp eed . The locations can be d efined by any kind of event that can be d etected throu gh the cellu lar netw ork and relates to a certain ge o-grap hic p osition, e.g. a certain signal strength p attern or a han d over. H ence, the hand over location accu racy estim ation can be u sed to assess the p otential of both GSM and UMTS in estim ating road traffic inform ation. It can be seen as a low er bou nd on p otential accu racy, since d ed icat ed p rocessing of raw d ata, i.e. m easu rem ent rep orts, m ight generate m ore accu rate locations. On the other hand , d u e to the fact that hand over locations are id e ntified w ith a large nu m ber test d rives, it can also be seen as an u p p er bou nd on hand over location accu racy. H and over location accu racy shou ld not be m ixed u p w ith travel tim e accu racy, since a nu m ber of factors excep t hand over location a c-cu racy affect the final travel tim e estim ation, e.g. tracking the term inals to the correct rou te and filtering ou t non-vehicle term inals.
2 Measurement objectives
N u m erou s sensor typ es are available to m easu re road traffic state inform ation su ch as sp eed , d ensity and flow . Stationary sensors, e.g. ind u ctive loop s and IR sensors, m easu re vehicle traffic p aram eters in a given location. Floating sensors are located in vehicles and m easu re the p aram eters for a given vehicle at d ifferent locations. The vehicles that are equ ip p ed w ith sensors are often referred to as p robe vehicles or floating cars. License p late m atching technol-ogies m easu re the travel tim e betw een vid eo cam era locations. Different typ es of sensors have d ifferent ad vantages and d raw backs. Which sensor that shou ld be u sed is d ep end ent on traffic cond itions, road netw ork stru ctu re and financial asp ects, bu t also the m ain ap p lication of d ata is relevant to this
choice. Tw o typ es of ap p lications can be d istingu ished , incid ent d etection and travel tim e estim ation.
The p erform ance of d ifferent system s for traffic inform ation d ep end s on a nu m ber of factors related to the m easu rem ent p roced u re and the nu m ber of sensors, bu t also on how the p erform ance m etrics are d efined . Id eally w e w ou ld w ant to know ou r ow n travel tim e given that w e start ou r travel at a certain tim e in the near fu tu re. H ow ever, that is a qu ite challenging task, and the aim is cu rrently to rep ort the historical travel tim e d ating e.g. five m inu tes back. Travel tim es flu ctu ate d u e to ind ivid u al d riving p atterns and it is not obviou s if w e w ant to know the low est, average or highest travel tim e in t he rep orting interval. Speed flu ctu ations d u e to d river behaviou r d ecreases w hen the road gets m ore congested and the incid ent or slow d ow n d etection is another im p ortant ap p lication. A com m on p erform ance m etric is the tim e for a system to d etect an incid en t. The d efinition of an incid ent varies, e.g. 40% slow d ow n, and affects the p erform ance. For stationary sensors the tim e to d etect an incid ent d ep end s on the tim e it takes for the incid ent to p rop a-gate to the sensor. In this case the sensor sp acing is the cru cial factor. For floating sensors the tim e to d etect an incid ent d ep end s on the tim e for the sensors to p rop agate to the incid ent. This tim e is highly correlated to the nu m ber of vehicles w ith sensors (p enetration) and the rep orting interval of the sensors. Using signalling d ata from cellu lar netw orks gives a p otentially high nu m ber of floating sensors w ith a very short reporting interval. The is-su e in these kind s of system s is the relatively low location accu racy and the rest of this p ap er investigates this issu e fu rther. The best resu lts are m ost lik e-ly obtained w hen fu sing inform ation from several sensors w ith d ifferent characteristics.
3 Location data in GSM and UMTS
The typ e of location d ata available for a term inal in GSM and UMTS d ep end s on the state of the term inal. The state, on the other hand , d ep end s on how the term inal is u sed , i.e. u sed for su rfing the w eb, m aking telep hone calls etc. In both GSM and UMTS, term inals u sed for real-tim e services generate the m ost d etailed location d ata and w e w ill focu s on this kind of d ata throu ghou t the p ap er.
Term inals u sed for real-tim e services continu ou sly send m easu rem ent r ep orts abou t the rad io environm ent in ord er to assist the netw ork in the han d -over d ecision. These m easu rem ent rep orts together w ith hand -over p oints can be u sed to track the rou te of the vehicle and calcu late travel tim es. The m ea s-u rem ent rep orts in GSM are sent every 480 m s and they contain the signal
qu ality of the serving base station, signal strength of su rrou nd ing base st a-tions and a tim ing ad vance (TA) valu e that gives a rou gh estim ation of the d istance to the serving base station. The m easu rem ent rep orts in UMTS are m ore flexible, the netw ork can d eterm ine the rep orting strategy d ynam ically. A large d ifference com p ared to GSM is the u se of event triggered m easu r e-m ent rep orts, i.e. the e-m obile tere-m inal e-m akes e-m easu ree-m ents and send s e-m ea s-u rem ent reports w hen an event has occs-u rred . The event triggered rep orts can be rep laced or com bined w ith p eriod ic rep orts.
In UMTS the p eriod ic m easu rem ent rep ort interval is configu rable betw een 0.25 and 64 second s, d ep end ing on rad io environm ent and the state of the m obile term inal [12]. The frequ ency of event triggered rep orts are d ep end ent of the frequ ency of actu al events, e.g. a new rad io link ad d ition to the active set, bu t also on the operator configu rable p aram eters tim e -to-trigger, hystere-sis and and offset valu e. More d etailed inform ation on UMTS m easu rem ent rep orts can be fou nd in e.g. [14, 15, 16]. Signal strength and qu ality of serving base station(s) are sim ilar to the ones in GSM. The m axim u m nu m ber of su r-rou nd ing base stations that can be m easu red is increased from 6 in GSM to 32 in UMTS. A TA valu e is not calcu lated in UMTS (WCDMA) netw orks since it is not a TDMA based system , bu t other tim e alignm ent m easu rem ents are available, e.g. rou nd trip tim e and tim e d ifference betw een base stations [15].
An im p ortant d ifference betw een GSM and UMTS is the p ossibility to u se soft hand over in UMTS. This m eans that a term inal can be connected to sev-eral base stations sim u ltaneou sly in UMTS, w hereas in GSM the term inal is only connected to one base station at the tim e. To track a vehicle, both m ea s-u rem ent rep orts containing rad io p aram eters and hand over p oints can be u sed . When it com es to calcu lating travel tim es it is very im p ortant to have tw o accu rate estim ations of the vehicle’s p osition in ord er to m ake a good estim ation of the travel tim e betw een those p oints. The hand over p oints in GSM are a good cand id ate to estim ate those p ositions. H ow eve r, in UMTS the term inal w ill not change from one base station to another, instead rad io links w ill be ad d ed to and rem oved from the term inals active set and the p u rp ose of this p ap er is to verify how efficiently these soft hand over p oints can be u sed to tie the vehicle to a certain p osition in the road netw ork.
4 Location experiments
The aim of the location exp erim ents is to com p are the location accu racy of hand over p oints in GSM and UMTS. The first exp erim ents w ere carried ou t on a 900 m eters long street segm ent in a “sp arse” u rban environm ent. The segm ent w as d riven fifteen tim es back and forth w ith test equ ip m ent for GSM
and UMTS (Ericsson TEMS*
Investigation 7.0) and a GPS receiver. Signalling d ata w as collected from a GSM term inal and a UMTS term inal s im u ltaneou s-ly, both w ith ongoing telep hone calls. For com p arison another test ru n w as m ad e in a d ifferent environm ent. The second test w as p erform ed in a su bu r-ban highw ay environm ent w ith less com p lex GSM cellu lar stru ctu re. Since hand overs are analysed , w e assu m e a p assive m onitoring ap p roach for travel tim e calcu lations, i.e. existing signalling traffic from the cellu lar netw orks are u sed to estim ate travel tim es.
In the sp arse u rban environm ent eight hand over zones w ere d etected for UMTS (see Fig. 1). In six of the hand over zones, a sp ecific rad io link w as ad d -ed , rem ov-ed or both in every test ru n. In hand over zone 7, a soft hand over occu rred in tw elve of the fifteen test ru ns and in hand over zone 8, a soft hand over w as com p leted in nine test ru ns. Du rin g the fifteen test ru ns the UMTS term inal w as connected to fou r d ifferent base stations. The m ain changes betw een the d ifferent test ru ns w ere the nu m ber of tim es a rad io link w as ad d ed w ithin a hand over zone and w hether a hand over occu rred at all in hand over zone 7 and 8.
For the GSM test ru ns, fou r hand over zones cou ld be d istingu ished (see Fig. 2). Du ring the fifteen test ru ns, the GSM term inal w as connected to nine
* TEMS = Telecommunication Management System
Figure 2. Handover zones in GSM.
1
2 3 4
Figure 1. Handover zones in UMTS
2 4 5 6 7 8 3 1
d ifferent cells. In the fou r hand over zones a hand over occu rred in every rou nd , how ever, no sp ecific hand over, i.e. betw een the sam e tw o cells, o c-cu rred in every test ru n.
Figu re 1 and 2 ind icate qu ite large d ifferences in the behaviou r of the GSM and UMTS netw orks regard ing hand overs. In the su bu rban highw ay env i-ronm ent several UMTS and GSM hand over zones w ere d etected . Three of these hand over zones for both GSM and UMTS w ere analysed in d etail u sing seven test ru ns on a 1.5 km road segm ent. N ext, a m ore d etailed analysis is m ad e regard ing the d ifferences in term s of hand over location accu racy and how it can affect the estim ation of travel tim es.
5 Handover location accuracy
Several ap p roaches are p ossible w hen d eterm ining the location of a hand over in UMTS. Since soft hand over is u sed , rad io links are ad d ed , rem oved or r e-p laced in the active set. The tye-p ical scenario in the exe-p erim ents w as that a rad io link w as ad d ed and rem oved several tim es in a hand over zone. N o r-m ally the new ly ad d ed rad io link becor-m es stronger sor-m ew here w ithin the hand over zone and is eventu ally the only one rem aining , bu t this is not nec-essarily the case. Straightforw ard m ethod s to d eterm ine the hand over loca-tion are to u se the first rad io link ad d iloca-tion or rem oval, the last rad io link ad d i-tion or rem oval or the p oint w hen a new rad io link is the strongest. Using the rad io link ad d itions p ossibly generates m ore hand over points, bu t the accu r a-cy m ight be better u sing the p oint in tim e w hen a new rad io link is stronger. In p ractice, im p lem entation d etails w ill settle w hich d ata or com bination of d ata that w ill be u sed . An insight of the exp erim ents is that u sing all rad io link ad d itions or p oints w here a new rad io link is stronger rend ers a lot of am bigu ity w ithin hand over zones, and therefore the first or last rad io link ad d ition in a hand over zone w ere u sed in the location accu racy calcu lations below .
In GSM hard hand overs are u sed , w hich m eans that there is no d ou bt abou t w here a hand over is located . Of cou rse, several stages of the hand over cou ld be d efined as the actu al hand over p oint, bu t that is of less im p ortance as long as the sam e d efinition is u sed for both calibration and estim ation.
Sparse Urban Environment
The p ercentage of tim es a sp ecific hand over is com p leted in the sam e hand o-ver zone, d riving the sam e rou te seo-veral tim es, is in this p ap er referred to as consistency. The consistency of hand over p oints w ill affect the travel tim e estim ations. If a p red icted hand over d oes not occu r, it w ill feed the system
w ith no d ata if the hand over sequ ence cannot be id entified or corru p t d ata if the vehicle is tracked to the w rong rou te. The consistency of the d ifferent hand over p oints for GSM and UMTS are show n in Figu re 3.
Figu re 3. Sp arse u rban environm ent. H and over consistency in UMTS and GSM.
Com p ared to the UMTS case, the GSM hand overs w ere m u ch m ore scattered . Several hand overs occu rred ou tsid e the hand over zones and the hand overs w ithin the zones w ere betw een d ifferent cells. The reason that tw o of the UMTS zones d id not have 100% consistency w as that no soft hand over o c-cu rred in these zones in several of the t est ru ns. The GSM inconsistency on the other hand , w as d u e to hand over betw een d ifferent cells in the sam e hand over p oint. A reason for this can be that the UMTS netw ork has m ore free cap acity than the GSM netw ork, and there is a p ossibility that the inco n-sistency of the GSM hand overs are d u e to cell cap acity lim itations. H ow ever, the test ru ns w ere d riven at tw o d ifferent d ays and tim es and it is not likely that the netw ork w as congested in both occasions. Since the resu lts w ere sim i-lar for GSM in both test ru ns, a congested netw ork d oes not seem to be a good exp lanation. A m ore p robable exp lanation m ight be the d ifferent cell stru c-tu res of GSM and UMTS in the test area. Du e to relatively few su bscribers in the UMTS netw ork, a hierarchical cell stru ctu re ha s not yet been d evelop ed , as often is the case for the GSM netw ork. A hierarchical cell stru ctu re m ight exp lain som e of the scattered hand overs and the fact that several d ifferent hand overs w ere d etected in the sam e hand over zone.
Handover consistency in UMTS and GSM
0% 20% 40% 60% 80% 100% UMTS 1 UMTS 2 UMTS 3 UMTS 4 UMTS 5 UMTS 6 UMTS 7 UMTS 8 GSM 1 GSM 2 GSM 3 GSM 4 Handover zone C o n s is te n c y p e rc e n ta g e
The hand over location accu racy is assessed u sing d ata from the 15 test ru ns to calcu late an average hand over p oint and m easu re the d eviation from this p oint in the d ifferent test ru ns. The d eviation is m easu red from the ave r-age p oint of all hand overs w ithin a hand over zone. It sh ou ld be noted that since the GSM hand overs w ere very inconsistent, hand overs betw een d iffe r-ent cells w ill be grou p ed to the sam e average p oint, w hich w ill affect the GSM hand over location accu racy m easu rem ents negatively. The hand overs in UMTS are d efined as the first or last rad io link ad d ition d ep end ing on han d -over zone. Missing hand -overs in a zone are ignored in the accu racy m ea s-u rem ents, i.e. the consistency and accs-u racy d iagram s shos-u ld be assessed and analysed ind ep end ently. All hand over p ositions are m ap p ed to the one-d im ensional sp ace of the traverseone-d roaone-d segm ent. The p ositioning error of the GPS receiver is inclu d ed in all m easu rem ents.
Figu re 4. Sp arse u rban environm ent. Mean and m ax hand over location error (in d e-viation from average hand over p oint) together w ith the stand ard d ee-viation of the m easu rem ents.
Figu re 4 show s that the accu racy is qu ite good for a m ajority of the hand over zones in both GSM and UMTS. The m ean error is below 20 m eters for m ore or less all of the UMTS hand over zones, an d below 40 m eters for the GSM zones. H ow ever, a large variation in accu racy betw een d ifferent hand over zones can also be seen, as w ell as a large d ifference betw een m ean and m axim u m error. The large m axim u m error of hand over zone 5 in UMTS w as d u e to an ou tlier in one of the rou nd s. The UMTS accu racy seem s in general m u ch better than for GSM, w hich, as m entioned p reviou sly, partly cou ld be exp lained by d if-ferent cell stru ctu re and traffic p attern in the area. It shou ld be noted that it is p ossible that the soft hand over techniqu e u sed in UMTS m ight rend er m u ch
Handover Location Error
0,0 20,0 40,0 60,0 80,0 100,0 UM TS 1 UM TS 2 UM TS 3 UM TS 4 UM TS 5 UM TS 6 UM TS 7 UM TS 8 GSM 1 GSM 2 GSM 3 GSM 4 Handover zone M e te rs Mean error Standard deviation Max error
m ore stable hand over p oints com p ared to GSM, bu t it is not clear from the exp erim ents w hat w ill hap p en if the traffic p attern and cell stru ctu re of the UMTS netw ork changes. In the m easu rem ents the tru e valu e of the average hand over p oint for the d ata set is u sed to calcu late the location error. This valu e w ill not be know n in a travel tim e estim ation system , bu t instead est i-m ated u sing test ru ns or coverage i-m ap s.
Suburban Highway Environment
The p reviou s hyp othesis that the low accu racy and consistency of the GSM hand overs is d u e to a com p lex hierarchical cell stru ctu re is evalu ated w ith a second test in a d ifferent road and rad io environm ent. This environm ent has m ost likely a flat GSM cell stru ctu re and the road is a highw ay w ith sp eed lim it 90 km / h. In this environm ent it is also p ossible to evalu ate how the v e-hicle sp eed affects the hand over accu racy. It is p ossible that a higher sp eed gives a larger hand over zone d u e to larger m ovem ent betw een m eas u rem ent rep orts. The resu lts of the test ru ns are show n in Figu re 5 below .
Figu re 5. Su bu rban highw ay environm ent. Mean and m ax hand over location error (in d eviation from average hand over p oint) together w ith the stand ard d eviation of the m easu rem ents.
The consistency for all evalu ated hand over zones is 100% for both GSM and UMTS. This im p lies that the low consistency for GSM in the urban enviro n-m ent can be d u e to the cell stru ctu re. The location accu racy of the UMTS hand overs is in the sam e ord er as the best ones from the u rban environm ent, i.e. very good . On the other hand , the GSM m easu rem ents have a large varia-tion w here one hand over zone has an average error of 20 m eters w here as another one has an average error of alm ost 80 m eters. From the resu lts w e can
Handover Location Error
0,0 20,0 40,0 60,0 80,0 100,0 120,0 140,0 160,0
GSM1 GSM2 GSM3 UMTS1 UMTS2 UMTS3
Handover Zone M e te rs Average error Standard deviation Max error
exp ect that the vehicle sp eed d oes not have a d ram atic effect on the accu racy, the hand over consistency can be very affected by cell stru ctu re and the UMTS hand overs seem to have m u ch better location accu racy.
Com p aring the location error of all 68 hand overs p erform ed in GSM and all the 99 hand overs p erform ed in UMTS, it is p ossible to d raw better conclu -sions regard ing a general d ifference in hand over location accu racy. The ave r-age location error of all hand overs in GSM is 23.1 m and 6.3 m in UMTS. Con-sid ering the nu ll hyp othesis that there is no d ifference in the average hand o-ver location error valu e betw een GSM and UMTS, it can be d iscard ed w ith a tw o-sid ed p -valu e of less than 10-6. This m eans that it is very likely that the long term average valu e of hand over location error is low er in UMTS co m -p ared to GSM, consid ering the sam e netw ork cond itions. It is also qu ite lik e-ly that this statem ent is valid for other netw ork cond itions.
6 Potential Travel Time Accuracy
The p u rp ose of calcu lating the han d over location error, as d escribed above, is to evalu ate the p otential of a travel tim e estim ation system based on m onito r-ing active cell p hones. By u sr-ing tw o hand over locations, a travel tim e can be calcu lated for a vehicle p assing betw een them . The accu r acy and the con-sistency of the hand over location w ill be a very im p ortant factor for the travel tim e accu racy. It is im p ortant to m ention that the travel tim es are calcu lated u nd er the assu m p tion that all vehicles are tracked to the correct road segm ent and that it is id entified as a vehicle, hence the term p otential travel tim e accu -racy is u sed . In a real system , these tasks are typ ically very challenging, esp e-cially in an u rban environm ent. This p roblem is d iscu ssed in d etail in e.g. [6, 7].
Figu re 6 show s an exam p le of estim ated and actu al hand over locations d u ring a test ru n. DE is the d istance betw een the estim ated hand over loca-tions. These are typ ically calibrated u sing coverage m ap s or test ru ns. In this case DE is the d istance betw een the average ha nd over locations, since they w ere estim ated based on all of the test ru ns. Di is the d istance betw een the actu al hand over p oints d u ring test ru n i, w here i1..15. The actu al travel tim e d u ring test ru n i (Ti) is sim p ly the tim e to travel th e d istance DE, w hereas the estim ated travel tim e d u ring test ru n i (TEi) is the tim e to travel the d is-tance Di.
Figu re 6. Overview of estim ated and actu al hand over locations for travel tim e est i-m ations.
Figu re 7 and 8 show s the actu al travel tim e (Ti) and the estim ated travel tim e (TEi) together w ith actu al m ean sp eed and estim ated m ean sp eed for a street segm ent traversed in the test ru ns. The actu al and estim ated travel tim es and m ean sp eed s are calcu lated u sing the TEMS logfile betw een hand over zones 3 and 4 in both the GSM and the UMTS netw ork.
In test ru n tw o for GSM no estim ation is show n since the hand over w as blocked . In fact, also hand over blocking events cou ld be u sed as location d ata, bu t u sing them as inp u t to travel tim e estim ation can be risky d u e to in-creased p ossibility of extrem e outliers. Also test ru n tw o in UMTS lacks travel tim e estim ation, in this case the reason w as equ ip m ent d isconnection ju st b e-fore hand over zone 3.
Figu re 7. Travel tim es and m ean sp eed estim ated w ith GSM hand over zones 3 and 4. Travel Time and Mean speed - GSM North Direction
30 40 50 60 70 80 90 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Round S e c o n d s a n d k m /h
Estimated Travel Time (s) Travel Time (s)
Estimated Mean speed (km/h) Mean speed (km/h) DE Di Estimated location of handover Actual location of handover
Figu re 8. Travel tim es and m ean sp eed estim ated w ith UMTS hand over zones 3 and 4.
Interestingly there seem s to be a correlation betw een the first and second hand over location, i.e. w hen the hand over to a certain cell is late, the ha nd o-ver from this cell to the next cell is also late. If there is su ch a correlation, the travel tim e error w ill be sm aller than ind icated by the hand over location e r-ror, since the travel tim e error is d efined as the absolu te d ifference Ti TEi .
As can be seen in Figu re 6 above, the estim ated UMTS travel tim es follow the actu al travel tim es extrem ely w ell. H ow ever, as can be seen in Figu re 5, also the estim ated GSM travel tim es follow the actu al travel tim es in a good w ay. Fou r m ain factors affecting travel tim e accu racy are hand over location accu racy, hand over consistency, nu m ber of available p robes in a tim e interval and road segm ent length. H and over location accu racy and consistency are d iscu ssed above, bu t also the nu m ber of available p r obes and the road seg-m ent length are iseg-m p ortant for travel tiseg-m e accu racy, interestingly they are also correlated . Increasing the road segm ent length w ill m ake the hand over loca-tion error less significant and yield better travel tim e accu racy. More vehicle p robes w ill give a better accu racy d u e to the p ossibility of averaging the tra v-el tim e valu e betw een several probes in a travv-el tim e rep orting interval. H ow ever, increasing the road segm ent length w ill also d ecrease the nu m ber of ongoing telep hone calls or d ata sessions that com p letes the w hole segm ent, w hich gives less p robes for averaging. Shorter road segm ents for travel tim es also lead to m ore d etailed traffic inform ation, w hich m ight be u sefu l in for exam p le u rban environm ents.
The road segm ent lengths (d istance betw een average hand over p oints) of the exp erim ents w ere 750 and 785 m eters for GSM and UMTS, resp ectively.
Travel Time and M ean speed - UM TS North Direction
30 40 50 60 70 80 90 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Round Se c o n d s a n d k m /h
Estimated Travel Time (s)
Travel Time (s)
Estimated Mean speed (km/h)
The exp erim ents ind icate that it is p ossible to m ake good travel tim e estim a-tions from cellu lar netw orks for this segm ent length, even sh orter segm ent lengths are p ossible d ep end ing on rad io environm ent, nu m ber of probes and accu racy requ irem ents. Another factor that w ill affect the travel tim e accu racy is the average vehicle sp eed in the hand over zones. If a vehicle is travelling slow ly in a hand over zone, a sm all location error can affect the travel tim e qu ite d rastically. This p roblem w ill typ ically be com m on w hen d ep loying this kind of system s in u rban environm ents.
7 UMTS Characteristics and Potential Benefit
In the com ing years there w ill be m any cou ntries w ith a com bination of UMTS and GSM term inals. Most likely the share of UMTS term inals w ill grow . In Sw ed en, for exam p le, the nu m ber of new UMTS term inals has p assed the nu m ber of new GSM term inals. H ence, u sing also UMTS term inals for travel tim e estim ation w ill increase the nu m ber of floating sensors in the system , w hich is an im p ortant factor w hen it com es to both travel tim e accu -racy and tim e to incid ent d etection. Increasing the nu m ber of vehicle sam p les in a travel tim e rep orting interval is very im p ortant for accu racy, esp ecially since the system p rod u ces relatively noisy m easu rem ents.
The higher d ata rate and shorter d elay together w ith d ynam ic m easu r e-m ent rep orting in UMTS e-m akes the netw ork to react e-m u ch faster to changes in the rad io environm ent, w hich affects the location accu racy of netw ork events. This, in com bination w ith the soft hand over p rincip le that m akes a rad io link ad d ition or rem oval w ithou t large d elay, m ight be the reason to the m u ch better UMTS location accu racy in the evalu ated tests. Generally sp ea k-ing there is also a better synchronization betw een base stations and m obile term inals in UMTS and (eventu ally) a m ore d ense rad io netw ork w hich gives a p otentially higher location resolu tion. This is ind ep end ent of w hether hand overs or som ething else is u sed to d eterm ine sp ecific locations of the m obile term inal on the road .
The higher location accu racy in the UMTS netw ork can be u sed to m ake the travel tim e accu racy better or m aintaining the relative accu racy w hile m aking the travel tim e segm ents shorter. Shorter travel tim e segm ents are necessary in e.g. u rban environm ents and are also u sefu l w hen d etecting in-cid ents.
The field tests show that the p otential accu racy of the UMTS hand overs is m u ch better than for GSM. H ow ever, it shou ld also be noted that there is m u ch m ore inform ation available in UMTS, lead ing to potential am bigu ity and real-tim e d ata p rocessing p roblem s. As an exam p le there w ere 243 rad io
link ad d itions in UMTS in the seven test ru ns of the high w ay environm ent w hereas the sam e test ru ns generated only 28 hand overs in GSM.
8 Conclusions
Since the behaviou r of the cellu lar netw orks is im p lem entation and enviro n-m ental d ep end ent, it is d ifficu lt to n-m ake general conclu sions fron-m lin-m ited tests. H ow ever, valu able insights w ere m ad e abou t the behaviou r of the d if-ferent netw orks w hen it com es to estim ating travel tim es based on hand over signalling d ata.
The exp erim ents ind icate relatively good hand over location accu racy for both GSM and UMTS. The hand ov er location accu racy and consistency w as m u ch better in UMTS than in GSM. A com p lex cell stru ctu re m ight be an e x-p lanation to the inconsistency in the GSM hand overs, this can also exx-p lain som e of the p roblem s w ith system d ep loym ent in u rban environm ents. M ore calibration w ork is therefore need ed in the p arts of the netw ork that has a com p lex cell stru ctu re in ord er to p red ict hand overs correctly. These p arts of the netw ork can easily be id entified u sing the op erator’s cell p lan. Assu m ing that non vehicle term inals can be filtered ou t, vehicles are tracked to the co r-rect rou te and that hand overs can be p red icted corr-rectly, a conclu sion from the exp erim ents is that the hand over location accu racy w ill be su fficient to estim ate u sefu l travel tim es, also in u rban environm ents. If there is a scalable w ay of d oing this p red iction and filtering in u rban environm ents is not clear to the au thors tod ay.
As soon as there are enou gh p robes, u sing GSM signalling d ata to estim ate road traffic inform ation on highw ays seem s qu ite w ell established and p rom -ising. To d o the sam e thing in u rban environm ents is a m ore challenging task. Com p lem enting GSM system s w ith UMTS signalling d ata w ill im p rove travel tim e estim ation accu racy, and has the p otential to extend coverage for the system s also to u rban areas. After all, there is a lot m ore signalling d ata from vehicles available in u rban environm ents, and it w ou ld be a w aste not to u se it.
H and over algorithm s are not d esigned to m ake hand over d ecisions in sp e-cific geograp hic p ositions, and are hence not optim ised to d eterm ine a loca-tion. This im p lies that sep arate analysis of m easu rem ent rep orts can increase the accu racy of location events. H ow ever, this is trad ed against m ore d ata p rocessing and the accu racy of hand over locations can in this case give good inp u t abou t the p otential of the technology.
References
[1] Bolla, R., and Davoli, F.: “Road Traffic Estim ation from Location Tracking Data in the Mobile Cellular Network”. Proceedings of WCN C, Chicago, Sep tem ber 2000, p p . 1107-1112
[2] Ygnace, J-L and Drane, C, “Cellu lar Telecom m u nication and Tran s-portation Convergence,” Proceedings of ITSC, Oakland, August 2001, p p . 16-22
[3] Karhumäki, T, “The Utilisation of GSM-Network in Travel Time Moni-toring,” ITS Workshop on Road Monitoring, Imperia, 2002
[4] University of Maryland Transp ortation Stu d ies Center. Final Evalu ation Rep ort for the CAPITAL-ITS Op erational Test and Dem onstration Program . University of Maryland , College Park, 1997
[5] Schollmeyer, R, Wiltschko, T, “Classification of Public Transport Vehi-cles u sing Cellu lar Mobile Rad io Data, Proceed ings of ITS Eu rop e, Aalborg, 2007
[6] Hellinga, B, Liping, F, Takada, H, “Obtaining Traveller Information via Mobile Phone Location Referencing – Challenges and Op p ortu n ities, Proceed ings of the Transp ortation Factor, 2003
[7] Mangold, S, Kyriazakos, S, “Applying Pattern Recognition Techniques based on H id d en Markov Mod els for Vehicu lar Positioning Location in Cellular Networks”, Proceedings of VTC ’99, Amsterdam, 1999 [8] H ellebran d t, M, Mathar, R, “Location Tracking of Mobiles in Cellu lar
Radio Networks”, IEEE Transactions on Vehicular Technology, Vol. 48, N o. 5, Sep tem ber 1999, p p . 1558-1562
[9] Gundlegård, D and Karlsson, J M, “Generating Road Traffic Infor-m ation froInfor-m cellu lar N etw orks – N ew Possibilities in UMTS”, Pr o-ceed ings of ITS-T, Chengd u , 2006
[10] Hellinga, B, Izad panah, P, “An Opportunity Assessment of Wireless Moitoring of N etw ork-Wid e Road Traffic Cond itions,” Final rep ort p rep ared for Ministry of Transp ortation of Ont ario, 2007
[11] Subbarao, W.V. et. al, “Travel Time Estimation using Cell Phones for Highways and Roadw ays”, Final report prepared for Department of Transp ortation, 2007
[12] Virginia Transportation Research Council, “Probe-based Traffic Moni-toring State-of-the-Practice Rep ort,” N CH RP 70-01, 2005
[13] Qiu, Z, Cheng, P, “State of the Art and Practice: Cellular Probe Technology Ap p lied in Ad vanced Traveler Inform ation System , Procee d -ings of TRB, Washington, 2007
[14] Hiltunen, K, Binucci, N, Bergström, J, “Comparison Between the Peri-od ic and Event-Triggered Intra-Frequ ency H and over Measu rem ent Reporting in WCDMA”, Proceedings of WCNC, 2000
[15] 3GPP, “Requirem ents for support of radio resource management“, TS 25.133, v. 7.2.0, Decem ber 2005
[16] 3GPP, “Radio Resource Control (RRC); Protocol Specification“ TS 25.331, v. 8.0.0, Sep tem ber 2007
