You are on page 1of 18

Handover location accuracy for travel time estimation in GSM and UMTS

David Gundlegrd and Johan M Karlsson

Linkping University Post Print

N.B.: When citing this work, cite the original article.

Original Publication: David Gundlegrd 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: Linkping University Electronic Press http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-16517

Handover Location Accuracy for Travel Time Estimation in GSM and UMTS

David Gundlegrd Johan M Karlsson Department of Science and Technology, Linkping University, SE-601 74 Norrkping, Sweden

Abstract
In this paper, field m easurem ents from the GSM and UMTS netw orks are analysed in a road traffic inform ation context. The m easurem ents ind icate a p otentially large im provem ent using UMTS signalling d ata com pared to GSM regard ing hand over location accuracy. These im provem ents can be used to generate real-tim e traffic inform ation w ith higher quality and extend the geographic usage area for cellular based travel tim e estim ation system s. The r esults confirm previous reports ind icating that he technology has a large p otential in GSM and they also show that the potential m ight be even larger u sing UMTS. Assum ing that non vehicle term inals can be filtered out, that veh icles are tracked to the correct route and that hand overs can be pred icted co rrectly, a conclusion from the experim ents is that the hand ov er location accuracy in both GSM and UMTS w ill be sufficient to estim ate useful travel tim es, also in urban environm ents. If there is a scalable w ay of d oing this filtering, tracking and pred iction in urban environm ents is not clear to the authors t od ay.

57

1 Introduction
The behaviour 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 pared to the GSM netw ork, w ill affect the quality of the generated traffic inform ation. Efforts to estim ate traffic inform ation from cellular netw orks started in m id 1990s, w hen the research p roject CAPITAL w as initiated [4]. The projects follow ing CAPITAL have taken m any d ifferent approaches to extract inform ation from the cellular netw orks, and several pilot projects are currently active in, am ong other countries, USA, Israel, Sw ed en and the N etherland s. A general travel tim e estim ation algorithm can inclu d e the fo llow ing basic steps: location d ata collection, id entification of vehicle located term inals, m ap m atching, possible route d eterm ination and travel tim e calculation. In ad d ition to these basic steps the algorithm should be able to compensate for anom alies like term inals in a bus w ith a separate lane, term inals on trains, term inals on m otorcycles or bicycles etc. It is very d ifficult to filter out bicycles and m otorcycles, w hich m eans that they w ill bias the travel tim e estim ation. H ow ever, recently there have been efforts in trying to filter out public transports using e.g. tim e-tables [5]. With know led ge of the sections w ith separate lanes other ad -hoc filtering m ethod s could m ost likely im prove accuracy further. Map matching in this context is analysed in e.g. [6]. The route d eterm ination problem is d iscu ssed in e.g. [7, 8]. A survey and a detailed d escription about the d ifferences in location d ata for generating road traffic inform ation from the GSM and UMTS netw orks can be found in [9]. Recent extensive surveys of the area can be found in [10, 11, 12, 13]. Although the technology of using cellular netw orks to estim ate road traffic inform ation has been subject for analysis for quite som e tim e, it is still far from being m ature. It is not clear w hat to expect from these system s in term s of accuracy, availability and coverage. The potential of the system is although quite clear, it is possible to retrieve a lot of traffic d ata in a cost efficient w ay, i.e. by using existing signalling d ata w ithout the need to invest in sensor infrastructure. Evolving com m unication system s w ill change com m unication patterns, and both the d esign of the com m u nication system and the com m unication pattern of the users w ill affect the potential of estim ating road traffic inform ation from cellular netw orks. In ord er to realise a safer and m ore efficient transportation system , it is likely that vehicles need to com m unicate m ore frequently w ith each other and w ith som e kind of traffic control centre.

58

This traffic generates cellular netw ork signalling d ata that can be very useful in ord er to estim ate road traffic inform ation in a cost efficient w ay. Interes tingly, w hen vehicles com m unicate m ore, these system s w ill generate better traffic inform ation. Better traffic inform ation w ill probably generate m ore com m unication, w hich once again w ill generate better traffic info rm ation, and this can be seen as a positive spiral of traffic inform ation collection. This paper focus on com paring hand over location inform ation betw een GSM and UMTS networks in tw o d ifferent environm ents, a sparse urban e nvironm ent w ith slow traffic and a suburban highw ay environm ent w ith hig her speed lim it. The hand over location d ata can be an integr al part w hen estim ating road traffic inform ation using passive m onitoring of cellular system s. In ord er to estim ate a travel tim e or average speed after the correct route has been id entified , tw o w ell d efined locations are need ed . The tim e betw een these locations are then used as travel tim e for the segm ent or for calculation of average segm ent speed . The locations can be d efined by any kind of event that can be d etected through the cellular netw ork and relates to a certain ge ographic position, e.g. a certain signal strength pattern or a han d over. H ence, the hand over location accuracy estim ation can be used to assess the potential of both GSM and UMTS in estim ating road traffic inform ation. It can be seen as a low er bound on potential accuracy, since d ed icat ed processing of raw d ata, i.e. m easurem ent reports, m ight generate m ore accurate locations. On the other hand , d ue to the fact that hand over locations are id e ntified w ith a large num ber test d rives, it can also be seen as an upper bound on hand over location accuracy. H and over location accuracy should not be m ixed up w ith travel tim e accuracy, since a num ber of factors except hand over location a ccuracy affect the final travel tim e estim ation, e.g. tracking the term inals to the correct route and filtering ou t non-vehicle term inals.

2 Measurement objectives
N um erous sensor types are available to m easure road traffic state inform ation such as speed , d ensity and flow . Stationary sensors, e.g. ind uctive loops and IR sensors, m easure vehicle traffic param eters in a given location. Floating sensors are located in vehicles and m easure the param eters for a given vehicle at d ifferent locations. The vehicles that are equipped w ith sensors are often referred to as probe vehicles or floating cars. License plate m atching technologies m easure the travel tim e betw een vid eo cam era locations. Different types of sensors have d ifferent advantages and d raw backs. Which sensor that should be used is d epend ent on traffic cond itions, road netw ork structure and financial aspects, but also the m ain application of d ata is relevant to this

59

choice. Tw o types of applications can be d istinguished , incid ent d etection and travel tim e estim ation. The perform ance of d ifferent system s for traffic inform ation d epend s on a num ber of factors related to the m easurem ent proced ure and the num ber of sensors, but also on how the perform ance m etrics are d efined . Id eally w e w ould w ant to know our ow n travel tim e given that w e start our travel at a certain tim e in the near future. H ow ever, that is a quite challenging task, and the aim is currently to report the historical travel tim e d ating e.g. five m inutes back. Travel tim es flu ctuate d ue to ind ivid ual d riving patterns and it is not obvious if w e w ant to know the low est, average or highest travel tim e in t he reporting interval. Speed fluctuations d ue to d river behaviour d ecreases w hen the road gets m ore congested and the incid ent or slow d ow n d etection is another im portant application. A com m on perform 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 perform ance. For stationary sensors the tim e to d etect an incid ent d epend s on the tim e it takes for the incid ent to prop agate to the sensor. In this case the sensor spacing is the crucial factor. For floating sensors the tim e to d etect an incid ent d epend s on the tim e for the sensors to propagate to the incid ent. This tim e is highly correlated to the num ber of vehicles w ith sensors (penetration) and the reporting interval of the sensors. Using signalling d ata from cellular netw orks gives a potentially high num ber of floating sensors w ith a very short reporting interval. The issue in these kind s of system s is the relatively low location accuracy and the rest of this paper investigates this issue further. The best results are m ost lik ely obtained w hen fusing inform ation from several sensors w ith d ifferent characteristics.

3 Location data in GSM and UMTS


The type of location d ata available for a term inal in GSM and UMTS d epend s on the state of the term inal. The state, on the other hand , d epend s on how the term inal is u sed , i.e. u sed for surfing the w eb, m aking telephone calls etc. In both GSM and UMTS, term inals used for real-tim e services generate the m ost d etailed location d ata and w e w ill focus on this kind of d ata throughout the paper. Term inals used for real-tim e services continuously send m easurem ent r eports about the rad io environm ent in ord er to assist the netw ork in the han d over d ecision. These m easurem ent reports together w ith hand over points can be used to track the route of the vehicle and calculate travel tim es. The m ea surem ent reports in GSM are sent every 480 m s and they contain the signal

60

quality of the serving base station, signal strength of surround ing base st ations and a tim ing advance (TA) value that gives a rough estim ation of the d istance to the serving base station. The m easurem ent reports in UMTS are m ore flexible, the netw ork can d eterm ine the reporting strategy d ynam ically. A large d ifference com pared to GSM is the use of event triggered m easurem ent reports, i.e. the m obile term inal m akes m easurem ents and send s m ea surem ent reports w hen an event has occurred . The event triggered reports can be replaced or com bined w ith period ic reports. In UMTS the period ic m easu rem ent report interval is configurable betw een 0.25 and 64 second s, d epend ing on rad io environm ent and the state of the m obile term inal [12]. The frequency of event triggered reports are depend ent of the frequency of actual events, e.g. a new rad io link ad d ition to the active set, but also on the operator configurable param eters tim e-to-trigger, hysteresis and and offset value. More d etailed inform ation on UMTS m easurem ent reports can be found in e.g. [14, 15, 16]. Signal strength and quality of serving base station(s) are sim ilar to the ones in GSM. The m axim um num ber of su rround ing base stations that can be m easured is increased from 6 in GSM to 32 in UMTS. A TA value is not calculated in UMTS (WCDMA) netw orks since it is not a TDMA based system , but other tim e alignm ent m easurem ents are available, e.g. round trip tim e and tim e d ifference betw een base stations [15]. An im portant d ifference betw een GSM and UMTS is the possibility to use soft hand over in UMTS. This m eans that a term inal can be connected to several base stations simultaneously 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 surem ent reports containing rad io param eters and hand over points can be used . When it com es to calculating travel tim es it is very im portant to have tw o accurate estim ations of the vehicles p osition in ord er to m ake a good estim ation of the travel tim e betw een those points. The hand over points in GSM are a good cand id ate to estim ate those positions. H ow ever, 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 purpose of this paper is to verify how efficiently these soft hand over points can be used to tie the vehicle to a certain position in the road netw ork.

4 Location experiments
The aim of the location experim ents is to com pare the location accuracy of hand over points in GSM and UMTS. The first experim ents w ere carried out on a 900 m eters long street segm ent in a sparse urban environm ent. The segm ent w as d riven fifteen tim es back and forth w ith test equipm ent for GSM

61

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 sim ultaneou sly, both w ith ongoing telephone calls. For com parison another test run w as m ad e in a d ifferent environm ent. The second test w as perform ed in a subu rban highw ay environm ent w ith less com plex GSM cellular structure. Since hand overs are analysed , w e assu m e a passive m onitoring approach for travel tim e calculations, i.e. existing signalling traffic from the cellular netw orks are used to estim ate travel tim es. In the sparse urban environm ent eight hand over zones w ere d etected for UMTS (see Fig. 1). In six of the hand over zones, a specific rad io link w as ad d ed , rem oved or both in every test run. In hand over zone 7, a soft hand over occurred in tw elve of the fifteen test runs and in hand over zone 8, a soft hand over w as com pleted in nine test runs. Durin g the fifteen test runs the UMTS term inal w as connected to four d ifferent base stations. The m ain changes betw een the d ifferent test runs w ere the num 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 occurred at all in hand over zone 7 and 8. For the GSM test runs, four hand over zones could be d istinguished (see Fig. 2). During the fifteen test runs, the GSM term inal w as connected to nine 5 4

4 6 1

3 7 3 2 8 1 2

Figure 1. Handover zones in UMTS


*

Figure 2. Handover zones in GSM.

TEMS = Telecommunication Management System

62

d ifferent cells. In the four hand over zones a hand over occurred in every round , how ever, no specific hand over, i.e. betw een the sam e tw o cells, o ccurred in every test run. Figure 1 and 2 ind icate quite large d ifferences in the behaviour of the GSM and UMTS netw orks regard ing hand overs. In the suburban highw ay env ironm 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 using seven test runs on a 1.5 km road segm ent. N ext, a m ore d etailed analysis is m ad e regard ing the differences in term s of hand over location accuracy and how it can affect the estim ation of travel tim es.

5 Handover location accuracy


Several approaches are possible w hen d eterm ining the location of a hand over in UMTS. Since soft hand over is used , rad io links are ad d ed , rem oved or r eplaced in the active set. The typical scenario in the experim 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 rm ally the new ly ad d ed rad io link becom es stronger som ew here w ithin the hand over zone and is eventually the only one rem aining , but this is not necessarily the case. Straightforw ard m ethod s to d eterm ine the hand over location are to use the first rad io link ad d ition or rem oval, the last rad io link ad d ition or rem oval or the point w hen a new rad io link is the strongest. Using the rad io link ad d itions possibly generates m ore hand over points, but the accur acy m ight be better using the point in tim e w hen a new rad io link is stronger. In practice, im plem entation d etails w ill settle w hich d ata or com bination of d ata that w ill be used . An insight of the experim ents is that using all rad io link ad d itions or points w here a new rad io link is stronger rend ers a lot of am biguity w ithin hand over zones, and therefore the first or last rad io link ad d ition in a hand over zone w ere used in the location accuracy calculations below . In GSM hard hand overs are used , w hich m eans that there is no d oubt about w here a hand over is located . Of course, several stages of the hand over could be d efined as the actual hand over point, but that is of less im portance as long as the sam e d efinition is used for both calibration and estim ation.
Sparse Urban Environment

The percentage of times a specific hand over is com pleted in the sam e hand over zone, d riving the sam e route several tim es, is in this paper referred to as consistency. The consistency of hand over points w ill affect the travel tim e estim ations. If a pred icted hand over d oes not occur, it w ill feed the system

63

w ith no d ata if the hand over sequence cannot be id entified or corrupt d ata if the vehicle is tracked to the w rong route. The consistency of the d ifferent hand over points for GSM and UMTS are show n in Figure 3.

Handover consistency in UMTS and GSM


Consistency percentage
100% 80% 60% 40% 20% 0% UMTS UMTS UMTS UMTS UMTS UMTS UMTS UMTS GSM 1 2 3 4 5 6 7 8 1 GSM 2 GSM 3 GSM 4

Handover zone

Figure 3. Sparse urban environm ent. H and over consistency in UMTS and GSM.

Com pared to the UMTS case, the GSM hand overs w ere m uch m ore scattered . Several hand overs occurred outsid 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 ccurred in these zones in several of the test runs. The GSM inconsistency on the other hand , w as d ue to hand over betw een d ifferent cells in the sam e hand over point. A reason for this can be that the UMTS netw ork has m ore free capacity than the GSM netw ork, and there is a possibility that the inco nsistency of the GSM hand overs are d ue to cell capacity lim itations. H ow ever, the test runs w ere d riven at tw o d ifferent days and tim es and it is not likely that the netw ork w as congested in both occasions. Since the results w ere sim ilar for GSM in both test runs, a congested netw ork d oes not seem to be a good explanation. A m ore probable explanation m ight be the d ifferent cell stru ctures of GSM and UMTS in the test area. Du e to relatively few su bscribers in the UMTS netw ork, a hierarchical cell structure ha s not yet been d eveloped , as often is the case for the GSM netw ork. A hierarchical cell structure m ight explain 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.

64

The hand over location accuracy is assessed using d ata from the 15 test runs to calculate an average hand over point and m easure the d eviation from this point in the d ifferent test runs. The d eviation is m easured from the ave rage point of all hand overs w ithin a hand over zone. It sh ould be noted that since the GSM hand overs w ere very inconsistent, hand overs between d iffe rent cells w ill be grouped to the sam e average point, w hich w ill affect the GSM hand over location accuracy m easurem ents negatively. The hand overs in UMTS are d efined as the first or last rad io link ad d ition d epend ing on han d over zone. Missing hand overs in a zone are ignored in the accuracy m ea surem ents, i.e. the consistency and accuracy d iagram s should be assessed and analysed ind epend ently. All hand over positions are m apped to the oned im ensional space of the traversed road segm ent. The positioning error of the GPS receiver is includ ed in all m easurem ents.

Handover Location Error


100,0 80,0

Meters

60,0 40,0 20,0 0,0

Mean error Standard deviation Max error

3 SM G

TS

TS

TS

TS

TS

TS

TS

TS

SM

SM

U M

U M

U M

U M

U M

U M

U M

U M

Handover zone

Figure 4. Sparse urban environm ent. Mean and m ax hand over location error (in d eviation from average hand over point) together w ith the stand ard d eviation of the m easurem ents.

Figure 4 show s that the accuracy is quite 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 accuracy 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 um error. The large m axim um error of hand over zone 5 in UMTS w as d ue to an ou tlier in one of the round s. The UMTS accuracy seem s in general m uch better than for GSM, w hich, as m entioned previously, partly could be explained by d ifferent cell structure and traffic pattern in the area. It should be noted that it is possible that the soft hand over technique used in UMTS m ight rend er m uch

65

SM

m ore stable hand over points com pared to GSM, but it is not clear from the experim ents w hat w ill happen if the traffic pattern and cell structure of the UMTS netw ork changes. In the m easurem ents the tru e value of the average hand over point for the d ata set is used to calculate the location error. This value w ill not be know n in a travel tim e estim ation system , but instead est im ated using test runs or coverage m aps.
Suburban Highway Environment

The previous hypothesis that the low accuracy and consistency of the GSM hand overs is d ue to a com plex hierarchical cell structure is evaluated 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 structure and the road is a highw ay w ith speed lim it 90 km / h. In this environm ent it is also possible to evaluate how the v ehicle speed affects the hand over accuracy. It is possible that a higher speed gives a larger hand over zone d ue to larger movem ent betw een m eas urem ent reports. The results of the test runs are show n in Figure 5 below .

Handover Location Error


160,0 140,0 120,0 100,0 80,0 60,0 40,0 20,0 0,0 GSM1 GSM2 GSM3 UMTS1 UMTS2 UMTS3 Handover Zone

Meters

Average error Standard deviation Max error

Figure 5. Suburban highw ay environm ent. Mean and m ax hand over location error (in d eviation from average hand over point) together w ith the stand ard d eviation of the m easurem ents.

The consistency for all evaluated hand over zones is 100% for both GSM and UMTS. This im plies that the low consistency for GSM in the urban enviro nm ent can be d ue to the cell structure. The location accuracy of the UMTS hand overs is in the sam e ord er as the best ones from the urban environm ent, i.e. very good . On the other hand , the GSM m easurem ents have a large variation 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 results w e can

66

expect that the vehicle speed d oes not have a d ram atic effect on the accuracy, the hand over consistency can be very affected by cell structure and the UMTS hand overs seem to have m uch better location accuracy. Com paring the location error of all 68 hand overs perform ed in GSM and all the 99 hand overs p erform ed in UMTS, it is possible to d raw better conclusions regard ing a general d ifference in hand over location accuracy. The ave rage location error of all hand overs in GSM is 23.1 m and 6.3 m in UMTS. Consid ering the null hypothesis that there is no d ifference in the average hand over location error value betw een GSM and UMTS, it can be d iscard ed w ith a tw o-sid ed p -value 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 com pared to GSM, consid ering the sam e netw ork cond itions. It is also quite lik ely that this statem ent is valid for other netw ork cond itions.

6 Potential Travel Time Accuracy


The purpose of calculating the han d over location error, as d escribed above, is to evaluate the potential of a travel tim e estim ation system based on m onito ring active cell phones. By using tw o hand over locations, a travel tim e can be calculated for a vehicle passing betw een them . The accur acy and the consistency of the hand over location w ill be a very im portant factor for the travel tim e accuracy. It is im portant to m ention that the travel tim es are calculated und er the assum ption that all vehicles are tracked to the correct road segm ent and that it is id entified as a vehicle, hence the term potential travel tim e accuracy is used . In a real system , these tasks are typically very challenging, esp ecially in an urban environm ent. This problem is d iscussed in d etail in e.g. [6, 7]. Figure 6 show s an exam ple of estim ated and actual hand over locations d uring a test run. D E is the d istance betw een the estim ated hand over locations. These are typically calibrated using coverage m aps or test runs. In this case D E is the d istance betw een the average hand over locations, since they w ere estim ated based on all of the test runs. D i is the d istance betw een the actual hand over points d uring test run i, w here i 1..15 . The actual travel tim e d u ring test run i (Ti) is sim ply the tim e to travel th e d istance D E, w hereas the estim ated travel tim e d uring test run i (TEi) is the tim e to travel the d istance D i.

67

DE Di
Estimated location of handover Actual location of handover

Figure 6. Overview of estim ated and actual hand over locations for travel tim e est im ations.

Figure 7 and 8 show s the actual travel tim e (T i) and the estim ated travel tim e (TEi) together w ith actual m ean speed and estim ated m ean speed for a street segm ent traversed in the test runs. The actual and estim ated travel tim es and m ean speed s are calculated using the TEMS logfile betw een hand over zones 3 and 4 in both the GSM and the UMTS netw ork. In test run tw o for GSM no estim ation is show n since the hand over w as blocked . In fact, also hand over blocking events could be u sed as location d ata, but using them as input to travel tim e estim ation can be risky d ue to increased possibility of extrem e outliers. Also test run tw o in UMTS lacks travel tim e estim ation, in this case the reason w as equipm ent d isconnection just b efore hand over zone 3.

Travel Time and Mean speed - GSM North Direction


Seconds and km/h
90 80 70 60 50 40 30 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Estimated Mean speed (km/h) Mean speed (km/h) Travel Time (s) Estimated Travel Time (s)

Round

Figure 7. Travel tim es and m ean speed estim ated w ith GSM hand over zones 3 and 4.

68

Travel Time and Mean speed - UMTS North Direction

Seconds and km/h

90 80 70 60 50 40 30 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Estimated Travel Time (s) Travel Time (s) Estimated Mean speed (km/h) Mean speed (km/h)

Round
Figure 8. Travel tim es and m ean speed 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 over 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 rT TEi ror, since the travel tim e error is d efined as the absolute d ifference i . As can be seen in Figu re 6 above, the estim ated UMTS travel tim es follow the actual travel tim es extrem ely w ell. H ow ever, as can be seen in Figure 5, also the estim ated GSM travel tim es follow the actual travel tim es in a good w ay. Four m ain factors affecting travel tim e accuracy are hand over location accu racy, hand over consistency, num ber of available probes in a tim e interval and road segm ent length. H and over location accuracy and consistency are d iscu ssed above, but also the num ber of available pr obes and the road segm ent length are im portant for travel tim e accuracy, interestingly they are also correlated . Increasing the road segm ent length w ill m ake the hand over location error less significant and yield better travel tim e accuracy. More vehicle p robes w ill give a better accuracy d ue to the possibility of averaging the tra vel tim e value betw een several probes in a travel tim e reporting interval. H ow ever, increasing the road segm ent length w ill also d ecrease the num ber of ongoing telephone calls or d ata sessions that com pletes the w hole segm ent, w hich gives less probes 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 seful in for exam ple urban environm ents. The road segm ent lengths (d istance betw een average hand over points) of the experim ents w ere 750 and 785 m eters for GSM and UMTS, respectively.

69

The experim ents ind icate that it is possible to m ake good travel tim e estim ations from cellular netw orks for this segm ent length, even sh orter segm ent lengths are possible d epend ing on rad io environm ent, num ber of probes and accuracy requirem ents. Another factor that w ill affect the travel tim e accuracy is the average vehicle speed 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 quite d rastically. This problem w ill typically be com m on w hen d eploying this kind of system s in urban environm ents.

7 UMTS Characteristics and Potential Benefit


In the com ing years there w ill be m any countries 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 ple, the nu m ber of new UMTS term inals has passed the num ber of new GSM term inals. H ence, using also UMTS term inals for travel tim e estim ation w ill increase the num ber of floating sensors in the system , w hich is an im portant factor w hen it com es to both travel tim e accuracy and tim e to incid ent d etection. Increasing the num ber of vehicle sam ples in a travel tim e reporting interval is very im portant for accuracy, especially since the system prod u ces relatively noisy m easurem ents. The higher d ata rate and shorter d elay together w ith d ynam ic m easur em ent reporting in UMTS m akes the netw ork to react m uch faster to changes in the rad io environm ent, w hich affects the location accuracy of netw ork events. This, in com bination w ith the soft hand over principle that m akes a rad io link ad d ition or rem oval w ithout large d elay, m ight be the reason to the m uch better UMTS location accuracy in the evaluated tests. Generally speaking there is also a better synchronization betw een base stations and m obile term inals in UMTS and (eventually) a m ore d ense rad io netw ork w hich gives a potentially higher location resolution. This is ind epend ent of w hether hand overs or som ething else is u sed to d eterm ine specific locations of the m obile term inal on the road . The higher location accuracy in the UMTS netw ork can be used to m ake the travel tim e accuracy better or m aintaining the relative accuracy w hile m aking the travel time segm ents shorter. Shorter travel tim e segm ents are necessary in e.g. urban environm ents and are also useful w hen d etecting incid ents. The field tests show that the potential accuracy of the UMTS hand overs is m uch better than for GSM. H ow ever, it should also be noted that there is m uch m ore inform ation available in UMTS, lead ing to potential am biguity and real-tim e d ata processing problem s. As an exam ple there w ere 243 rad io

70

link ad d itions in UMTS in the seven test runs of the high w ay environm ent w hereas the sam e test runs generated only 28 hand overs in GSM.

8 Conclusions
Since the behaviour of the cellular netw orks is im plem entation and enviro nm ental d epend ent, it is d ifficult to m ake general conclusions from lim ited tests. H ow ever, valuable insights w ere m ad e about the behaviour of the d ifferent netw orks w hen it com es to estim ating travel tim es based on hand over signalling d ata. The experim ents ind icate relatively good hand over location accuracy for both GSM and UMTS. The hand over location accuracy and consistency w as m uch better in UMTS than in GSM. A com plex cell structure m ight be an e xplanation to the inconsistency in the GSM hand overs, this can also explain som e of the problem s w ith system d eploym ent in urban environm ents. M ore calibration w ork is therefore need ed in the parts of the netw ork that has a com plex cell structure in ord er to pred ict hand overs correctly. These parts of the net w ork can easily be id entified using the operators cell plan. Assum ing that non vehicle term inals can be filtered out, vehicles are tracked to the co rrect route and that hand overs can be pred icted correctly, a conclusion from the experim ents is that the hand over location accuracy w ill be su fficient to estim ate useful travel tim es, also in urban environm ents. If there is a scalable w ay of d oing this pred iction and filtering in urban environm ents is not clear to the authors tod ay. As soon as there are enough probes, using GSM signalling d ata to estim ate road traffic inform ation on highw ays seem s quite w ell established and prom ising. To d o the sam e thing in urban environm ents is a m ore challenging task. Com plem enting GSM system s w ith UMTS signalling d ata w ill im prove travel tim e estim ation accuracy, and has the potential to extend coverage for the system s also to urban areas. After all, there is a lot m ore signalling d ata from vehicles available in u rban environm ents, and it w ould be a w aste not to use it. H and over algorithm s are not d esigned to m ake hand over d ecisions in sp ecific geographic positions, and are hence not optim ised to d eterm ine a location. This im plies that separate analysis of m easurem ent reports can increase the accuracy of location events. H ow ever, this is trad ed against m ore d ata processing and the accuracy of hand over locations can in this case give good input about the potential of the technology.

71

References
[1] Bolla, R., and Davoli, F.: Road Traffic Estim ation from Location Tracking Data in the Mobile Cellular Netw ork. Proceed ings of WCN C, Chicago, Septem ber 2000, pp. 1107-1112 [2] Ygnace, J-L and Drane, C, Cellular Telecom m unication and Tran sportation Convergence, Proceed ings of ITSC, Oakland , August 2001, pp. 16-22 [3] Karhum ki, T, The Utilisation of GSM -N etw ork in Travel Tim e Mon itoring, ITS Workshop on Road Monito ring, Im peria, 2002 [4] University of Maryland Transportation Stud ies Center. Final Evalu ation Report for the CAPITAL-ITS Operational Test and Dem onstration Program . University of Maryland , College Park, 1997 [5] Schollm eyer, R, Wiltschko, T, Classificati on of Public Transport Veh icles u sing Cellular Mobile Rad io Data, Proceed ings of ITS Europe, Aalborg, 2007 [6] H ellinga, B, Liping, F, Takad a, H , Obtaining Traveller Inform ation via Mobile Phone Location Referencing Challenges and Opportun ities, Proceed ings of the Transportation Factor, 2003 [7] Mangold , S, Kyriazakos, S, Applying Pattern Recognition Tec hniques based on H id d en Markov Mod els for Vehicular Positioning Location in Cellular N etw orks, Proceed ings of VTC 99, Am ste rd am , 1999 [8] H ellebran d t, M, Mathar, R, Location Tracking of Mobiles in Cellular Rad io N etw orks, IEEE Transactions on Vehicular Technology, Vol. 48, N o. 5, Septem ber 1999, pp. 1558-1562 [9] Gund legrd , D and Karlsson, J M, Generating Road Traffic Info rm ation from cellular N etw orks N ew Possibilities in UMTS, Pr oceed ings of ITS-T, Chengd u, 2006 [10] H ellinga, B, Izad panah, P, An Opportunity Assessm ent of Wireless Moitoring of N etw ork-Wid e Road Traffic Cond itions, Final report prepared for Ministry of Transportation of Ontario, 2007 [11] Subbarao, W.V. et. al, Travel Tim e Estim ation using Cell Phones for H ighw ays and Road w ays, Final report prepared for Departm ent of Transportation, 2007 [12] Virginia Transportation Research Council, Probe -based Traffic Monitoring State-of-the-Practice Report, N CH RP 70-01, 2005 [13] Qiu, Z, Cheng, P, State of the Art and Practice: Cellular Probe Tec hnology Applied in Advanced Traveler Inform ation System , Proceed ings of TRB, Washington, 2007

72

[14] H iltunen, K, Binu cci, N , Bergstrm , J, C om parison Betw een the Period ic and Event-Triggered Intra-Frequency H and over Measurem ent Reporting in WCDMA, Proceed ings of WCN C, 2000 [15] 3GPP, Requirem ents for support of radio resource m anagem ent, TS 25.133, v. 7.2.0, Decem ber 2005 [16] 3GPP, Rad io Resource Control (RRC); Protocol Specification TS 25.331, v. 8.0.0, September 2007

73