RESEARCH Open Access A novel vehicular location prediction ... · RESEARCH Open Access A novel vehicular location prediction based on mobility patterns for routing in urban VANET

Xue et al. EURASIP Journal on Wireless Communications and Networking 2012, 2012:222http://jwcn.eurasipjournals.com/content/2012/1/222

RESEARCH Open Access

A novel vehicular location prediction based onmobility patterns for routing in urban VANETGuangtao Xue, Yuan Luo*, Jiadi Yu and Minglu Li

Abstract

Location information is crucial for most applications and protocol designs in high-speed vehicular ad-hoc networks(VANETs). In traditional approaches, this is obtained by object tracking techniques that keep tracking the objectsand publish the information to the users. In highly dynamic environments, however, these approaches are notefficient as the target objects in VANETs are typically vehicles that present high mobility. Their locations keepchanging in a large range so that the tracking and information publication algorithms have to be frequentlyinvoked to obtain the instant locations of the objects. To deal with this problem, we propose a novel approachbased on the observation that in high-speed VANET environment, the target objects are strictly constrained by theroad network. Their mobilities are well patterned and many patterns can clearly be identified. These patterns cansmartly be leveraged so that a large amount of control overhead can be saved. Towards this end, in this article weadopt Variable-order Markov model to abstract Vehicular Mobility Pattern (VMP) from the real trace data in Shanghai.We leverage VMP for predicting the possible trajectories of moving vehicles which help to keep the timelyeffectiveness of the evolutional location information. To reveal the benefits of VMP, we propose a Prediction-basedSoft Routing Protocol (PSR), taking VMP as an advantage. The experimental results show that PSR significantlyoutperforms existing solutions in terms of control packet overhead, packet delivery ratio, packet delivery delay. Incertain scenarios, the control packet overhead can be saved by up to 90% compared with DSR, and 75% comparedwith WSR.

Keywords: Vehicular ad-hoc network, Prediction models, Vehicular mobility pattern, Routing protocols

IntroductionLocation information is crucial for most applicationsand protocol designs in high-speed vehicular ad-hoc net-works (VANETs), ranging from information exchangingto in-network storage. In traditional approaches, locationinformation can be obtained through localization techni-ques. With certain object tracking and information pub-lication mechanisms, the locations of mobile object arealso available for users. Localizations and object trackingare extensively studied topics and many useful algo-rithms have been proposed. Recently, real traffic traceand maps and even traffic pattern have been introducedto assist routing in vehicular networks [1-3].In highly dynamic environments such as the VANETs,

however, these approaches are not efficient due to thehigh mobility of objects. In VANETs, objects are typic-ally vehicles that present the mobility of hundreds of

* Correspondence: [email protected] Jiao Tong University, Shanghai, China

© 2012 Xue et al.; licensee Springer. This is anAttribution License (http://creativecommons.orin any medium, provided the original work is p

kilometers per hour. Therefore, the locations of the ve-hicle objects keep changing dramatically in a large scale.This nature demands the localization techniques to befrequently invoked and the location information to becontinuously updated, incurring a large amount of com-munication and control overhead. Recall that the com-munication capacity of wireless networks is constrainedby the wireless medium [4]. As the network scales up,the demand for control packet exchange increases whilethe network capacity decreases, leading the problem tobecome more serious. In other words, these traditionalapproaches are not scalable in large-scale VANETs.In this article, we propose a novel approach, which is

based on our observation that vehicles’ urban environ-ments are well behaved and can accurately be predicted.More specifically, VANETs in an urban environment isstructured based on the traffic transportation networksuch as the roads, bridges, and tunnels. Vehicles have tostrictly follow the road and travel along single direction

Open Access article distributed under the terms of the Creative Commonsg/licenses/by/2.0), which permits unrestricted use, distribution, and reproductionroperly cited.

mailto:[email protected]://creativecommons.org/licenses/by/2.0

Xue et al. EURASIP Journal on Wireless Communications and Networking 2012, 2012:222 Page 2 of 14http://jwcn.eurasipjournals.com/content/2012/1/222

of each road segment. When the speeds of vehicles areavailable (it can be obtained through speedometer onvehicles directly), the locations of the vehicles in the fu-ture short period of time can be calculated by a simpleequation. Moreover, vehicles in urban area often haveclear destinations and the desired transport routes arelimited. When the destinations are predicted accordingto the source of the vehicles, the present location, andthe moving directions, the locations of the vehicles in arelatively long time can also accurately be predicted in alarge degree. As such for each vehicle we can obtain itslocation in a proactive manner rather than the trad-itional reactive manner, and a large amount of controloverhead can be saved.To validate this idea, we firstly extract Vehicular Mo-

bility Pattern (VMP) by employing the Variable-orderMarkov (VOM) models [5] from real trace data collectedfrom over 4,000 taxis over several months in Shanghai.We find that because of the intrinsic nature of roads,such as single and dual carriageway, free way and the in-dividual driving habit, there exist large amounts of re-usable mobility patterns in the traffic trace, whichaccounts for around 40% of the whole traces, i.e., VMPtypically includes fixed route or vehicle’s favorite pathsgiven the starting place and the destination of vehicles.To see the benefits of VMP, we propose then aPrediction-based Soft Routing Protocol (PSR) in whichthe traffic trace and the real digital road map are utilizedto assist packet routing. In PSR, the disseminated stateinformation carries vehicle’s current state and the pre-dictive states, and the state information is only requestedand updated when the last predictive state informationis not consistent with the vehicle’s current state, whichsignificantly save the control packet overhead. Finally,extensive experimental results show that VMP exhibitsquit high accuracy, and offers significant enhancementto routing design in cutting control overhead. In PSR,the control traffic overhead increases linearly with thenumber of nodes in the network, regardless of networksize or mobility.The rest of the article is organized as follows. In the fol-

lowing section, we present the network model and theVOM scheme which is used to generate VMP. We discussthe design of PSR in Section “PSR design”, followed by theperformance evaluations in Section “Performance evalu-ation”. Section “Related study” gives a review of relatedworks. We conclude the study in Section “Conclusions” aswell as the possible future work directions.

VMPLet T = r1, r2, . . ., rn denote a vehicle node trajectory se-quence, where ri depict the node’s ith passing road, r1,r2, . . . rn ∈ R = {R

1, R2, . . . Rm}, R is the set of all roads,and m = |R| is the cardinality of R. A sequence segment

rik is denoted as ri

k = riri+1. . .ri+k-1, where k is the lengthof the sequence segment and ri

0 = ε.

Definition 1The term VMP is a trajectory segment ri

k with highprobability, that is f (ri

k) = Pr (ri+k-1|rik–1) ≧ σ, where σ ∈

[0, 1] is a predefined threshold.

VMP in the real tracePrevious study [6] shows people’s regularity of move-ment and repetition of journeys to the same place. Ouranalysis on the traffic trace also shows that people havea high degree of regularity in their movement despitethe complex driving behavior. For example, consider thecondition of roads. Freeways normally have limitedaccesses and outcomes. The vehicles’ speed and direc-tion are relatively stable and we can easily know vehicles’future trajectories based on their current position andvelocity information until they reach the end of freeways.Or if a road only has one connected road on some endwhich is meanwhile the popular path, we can estimatethat most vehicles will turn that way, with a very fewexceptions making U-turns to the prior road. Also, thepaths to some hot spots are relatively fixed.Figure 1 displays some VMP mined in accordance to

the road condition. South Chongqing Road is a bidirec-tional freeway, on which vehicle nodes are characterizedwith high speed and run all the way along the freewayuntil they arrive at the outcomes. VMP in Figure 1ashows pairs of bidirectional segments in accordance toabove analysis. Another example, the path from urbanarea to Shanghai Pudong International Airport is a high-way, which is preferred by most drivers to go to the air-port. Therefore, it forms VMP as shown in Figure 1b.We can also take the behavior of individual vehicle nodes

into account. Admittedly, there is no apparent regulation tofollow due to diverse individual habits. Yet, we still uncoversome hidden patterns. Since people are prone to repeatingthe same journey to the same place [6], we are able to minethe potential VMP from their historical statistics.We randomly choose a set of real traces of one taxi

with period of 6 months to generate VMP, and mark thepatterns correspondingly on the map to get a straightfor-ward view as shown in Figure 2. From this distributionof VMP on the map of Shanghai urban area, we find thatVMP occupies a great proportion of roads.

VMP generationThe VMP mining problem presents interesting stochas-tic chains of finite order which means transition prob-abilities depend on a finite suffix of the past and the setof the lengths of all suffix is bounded. More specifically,for a vehicle node in the current road rc, its possible pat-terns can be rc–k

krc→ rc+1 (1 ≤ k ≤ K), where K is the

Figure 1 VMP on the digital road map of Shanghai. (a) VMP around South Chongqing Rd. on the digital map. (b) VMP from urban area toShanghai Pudong International Airport.


maximal number of the proceeding roads of rc and is apredefined value. Clearly, K is the upper bound on themaximal Markov order. Among all these possible pat-terns, the ones whose probability is above the value σwill be the final VMP.The tool Markov chain has widely been used for pre-

dicting the future location of an object. In a Markovchain however, each random variable in a sequence witha Markov property depends on a fixed number of ran-dom variables. Consequently, the number of possiblepatterns would be very large: patterns which incur over-whelming complexity to check all the possible patterns.

We reduce the cost by pruning unnecessary patterns.First, since our patterns are not with the same length,VOM model is more adaptive in our problem whichenables the state space reduced significantly. Second, al-though there are totally |R| roads, the patterns possiblywith high frequency are obviously the ones whose con-secutive sequence segments are connected roads. Third,the value K is generally a small number as shown laterin Section “Performance evaluation” so that the value isset to 5 in our simulation.We adapt an effective VOM model [3], which is very

popular in the area of lossless compression and is also

Figure 2 Distribution of VMP on the map of Shanghai urban area.


used widely in sequence prediction for estimating theprobability and mining VMP. The algorithm is asfollows.

Step 1 Incrementally parsing procedure. We sequentiallyparse r1

n into ‘phrases’ which are adjacent whilenon-overlapping. The first phrase is an emptyphrase O. A new phrase is then created as soonas a prefix of the unparsed part of the stringdiffers from all preceding phrases. Figure 3shows an instance of road map, according towhich a road sequence acdacbacdabdc isgenerated. We parse the sequence and getphases O, a, c, d, ac, b, acd, ab, dc.

Step 2 Learning phase. A multi-way parsing tree isconstructed to display the parsed phrases. Eachnode in the tree carries a counter that holdstatistics of r1

n and each internal node has exactly|R| children (|R| = 4 in the above example). Eachphrase can find a path in the tree starting fromthe root while ending at some internal node. Bygoing through the parsed sequence starting withO we add each phrase to the tree as follows. First

Figure 3 An instance of road map.

empty phrase O is added to the tree as root andthen its |R| children are added to it as leaf nodes.The counter of each leaf node in the tree is alwaysset to 1. The counter of internal node is updatedto ensure it is always equal to the sum of all itschildren’s counter. Then for each phrase, wetraverse the tree starting at the root. Once a leafnode is reached, it is transformed into an internalnode by adding |R| leaf children to it.

Step 3 To estimate the probability f (rik) = Pr (ri+k–1|ri

k–1).We traverse the parsing tree starting from the rootO according to the sequence ri

k–1. If we reach a leafnode before ending the sequence ri

k–1, we willjump to the root to continue the traverse until weuse up the sequence. We then go one stepfurther according to ri+k–1 and reach the finalnode d. Thus we can compute the estimationPr(ri+k–1|ri

k–1) = c(d)/c(Parent(d)), where c(d)denotes the counter of node d.

A pseudo code of our VMP generating algorithm is givenin Figure 4. Denote P as the final pattern set, Adj(ri) as theset of adjacent roads of road ri. We show a parsing treeaccording to the above sequence instance acdacbacdabdcin Figure 5. To estimate Pr(d|ac), we traverse the tree inthe following order: O→ a→ c→ d and get the resultPr(d|ac) = 4/7 = 0.57. For Pr(c|da), we traverse in theorder: O→ d→ a→O→ c and get Pr(d|ac) = 4/28 = 0.14.To well measure VMP, one aspect that needs to be

considered is how often VMP can be used. In otherwords, it is about the proportion of the number of roadshaving VMP to the total times of roads being visited(one road may be visited more than one times). Higherfrequency means VMP can be more effective for predictingthe future route.

Figure 4 Mobility pattern generating algorithm.


Definition 2Denote V = {V1, V2, . . . Vv} as the set of roads havingbeen visited, and S = {S1, S2, . . . Ss} as the set roads thatcan be applied VMP, S, V ∈ R. The pattern utilization(PU) is given by PU ¼ Psj¼1fSj=

Pvi¼1fV i , where fr is the

times of road r (∈ R) being visited by vehicles.

PSR designPSR assumes that vehicle nodes are equipped with802.11 devices, forming a VANET. These GPS-enablednodes are equipped with pre-loaded digital maps whichprovide road-level map and traffic statistics. They alsoknow their position and velocity information on digitalmap of the city. Vehicle nodes have limited transmissionbandwidth, but unlimited storage and power. PSRmainly includes three components: Location InformationPropagation, Data Packet Forwarding, and PredictionError Recovery.

Location information propagationBoth beacon messages and announcement messages arediffused proactively. Periodic single hop beacon messages(Table 1) with information including current position,

Figure 5 A parsing tree.

velocity, and current road ID are sent to inform the neigh-borhood the sender node’s location with high frequency(e.g., 1 per second).Each node maintains a local time stamp (TS) for bea-

con message, which is utilized to make freshness judg-ments of geographical information by other nodes.When neighbor nodes receive beacon messages, theycache them for two different durations that one forneighborhood management, and another for evolutionalstate management, serving as evolutional state in routingtable. In this design, we set the neighborhood durationas 3–5 T, where T is the beacon period. TE is used torepresent the valid time span for different beacon mes-sages. As shown in Figure 6, for the vehicle node whichis currently in position C, moving with velocity v in roadr, the information sent from C would go invalid after thenode leaves road r. Taking the communication range(CR) into account, TE =Δt1 + CR/v, where Δt1 = L1/v.To efficiently disseminate one node’s evolutional states

to other nodes over the network, we randomly selectfour orthogonal directions and pick four correspondingpoints outside the network along these directions. Weuse GPSR [7] to forward announcement to these points.In ORRP [8], the authors show that a pair of orthogonal

Table 1 Entries in beacon messages

Notation Description (beacon)

ID Sender’s ID

(X,Y) Current position of sender

v Velocity with direction of sender

Cr.ID Current road’s ID of sender

Table 2 Entries in announcement messages

Notation Description (announcement)

ID Sender’s ID

(X,Y) Current position of sender

v Velocity with direction of sender

Cr.ID Current road’s ID of sender

Nr.ID Next road’s ID of sender


lines would intersect at a point within the network areawith a high probability. Hence, all the receivers of theannouncement can get the sender’s location information.Unlike the state dissemination process in WSR [9], we

do not diffuse announcement for updating state infor-mation periodically with high frequency. Taking advan-tage of the longer TE of evolutional state under specificShanghai Urban Vehicular Network (SUVnet) topology,nodes update their state information only when their lastsent evolutional states go invalid. Consider the scenarioin Figure 6, where a vehicle node currently at position Cis moving from left to right on road r. Its TE of evolu-tional state is at least Δt1 + CR/v, the same value as inbeacon. A step forward is to leverage the possible routeinformation according to VMP. Each vehicle node locallyrecords its historical passing route and applies predic-tion scheme to generate its VMP. If road r has VMP,e.g., rc, we can further presume that the node willmove on to the end of road c with speed v and thusTE =Δt1 +Δt2 + CR/v, where Δt2 = L2/v. We include thenode’s position, moving velocity, road ID, and next road IDin announcement packets as shown in Table 2. If the nodedoes follow the pattern rc, we update the node’s evolu-tional state when it moves to road e. Otherwise, refer tosolutions in later prediction error recovery section.Once an intermediate node receives an announcement

packet, it records the carried source information includ-ing node ID, position, current road ID, next road ID in

Figure 6 A taxi moving on road r.

its routing table, and set the TE of the newly recordedentry to be 0. As time elapses, the invalid entry will beremoved from the routing table according to its TS.

Data packet forwardingAnnouncement dissemination will distribute the nodeevolutional states among the network. To transmit adata packet from source-to-destination, we use a similarforwarding strategy as WSR. If the source node has thelocation information of the destination node, data pack-ets will be sent along the direction where the destinationcurrently locates. Otherwise, the data packet will be sentin a random direction as in WSR, employing the rendez-vous concepts in earlier work like ORRP, then biased inintermediate nodes. After the data packet is received bythe destination node, no acknowledgement is sent back.Figure 7 presents the data packet forwarding algorithm

in detail. During the forwarding process, intermediatenodes bias the next forwarding direction if they havefresher state information about the destination. Theheader of a data packet includes items shown in Table 3.The header of a data packet includes TS, Destination’sID, position, pre-estimated position, road ID, and nextroad ID. When an intermediate node receives a packet,it checks whether to update the contents of routing tableentry to the data packet header. If its routing table con-tains an evolutional state about the destination and the

Algorithm 2 Data Packet Forwarding Algorithm

ForwardDataPacke t(p) 1: d Destination(p) 2: t TS(p) 3: (x, y) Pre-estimated(d) 4: r LookupRoutingTable(d) 5: If r is not null6: t1 TS(r) 7: If (t1

(a) Destination goes ahead

(b) Destination lags behind

(c) Wrong prediction of possible route

Figure 8 Scenarios of wrong prediction.


selected from the 1000 nodes as source–destinations.The velocities of vehicular nodes vary from 0 to 30 m/s.The average velocity of all nodes is 9.75 m/s.

Using the trace, we conduct simulations on the Net-work Simulator (NS2) [10] platform together with theIEEE 802.11 MAC layer extension [11]. MAC layer

Figure 9 SUVnet CR = 500 m.

0.4

0.45

0.5

ds


failure notification from the IEEE 802.11 MAC layer ex-tension is applied when data packets transmission fails.In this simulation, we do not consider the effect of theobstacles, such as buildings and trees on the both sides ofthe streets, on the communication. In each round, wesimulate PSR, WSR for 1000 s and DSR for 500 s as DSRsimulations require too much memory. Other parametersin the simulations are listed in Table 4. All nodes have500 m omni-directional CR. Each source node sends onedata packet per second for 100 s. The size of a data packetis 512 bytes. The beacon interval is set to 1 s to keep thestate information of neighborhood up-to-date. The TTLvalue of data packet is set to 128. The setting of transmis-sion range is related to the connectivity of SUVnet. Figure 9presents the profile of SUVnet. With the CR of 500 m, theSUVnet is almost full connected.

VMPs generationFigure 10 shows the PU of selected three taxies. Obvi-ously, as the amount of training data increase (displayedas the time period of data increases), more VMP can bediscovered and PU goes up. Higher PU means greaterproportion of roads are enable to apply VMP for predict-ing the next roads according to the past route. Asillustrated in Figure 10 that as the time period of train-ing data increases from 1 to 6 months (2006.10.01–2007.03.31), PU of taxi 1 grows up to 44.88% (with pat-tern generating threshold 0.6).Figure 11 plots the distribution of length of VMP,

which indicates that a majority of VMP is of length 2 to3. As the value of pattern length gets larger than 5, thenumber of increased patterns almost approaches to zero,which will be ignored in order to reduce the computingcomplexity. Hence, we set the value of K to be 5.The prediction accuracy of VMP would affect the final

performance of PSR. We use the traces of taxies during theperiod from 2006.10.01 to 2007.03.31 to generate patterns,and random select three taxies of another one month(2007.04.01–2007.04.30) to test the accuracy. Figure 12shows various prediction accuracies with different patterngeneration thresholds. Generally, as the pattern generating

Table 4 Simulation setup

Parameter Value

Simulation area 4000 × 6000 m2

Number of nodes 1000

Data packet size 512 bytes

Average velocity 9.75 m/s

CR 500 m

Beacon interval 1 second

Data packet sending rate 1 packet per second

TTL of data packet 128

threshold increases, the prediction accuracy increases ac-cordingly. However, for taxi 1 and 3, the prediction accur-acy with threshold 1.0 is lower than that with threshold0.9. The reason is that the cardinal number for generatingthe patterns with the threshold 1.0 is so small that evenone prediction error may decrease the prediction accuracya lot. The patterns generating threshold decides the num-ber of patterns generated. In Figure 13, the number of pat-terns declines rapidly as the pattern generating thresholdincreases. While larger threshold means higher predictionaccuracy, it also means fewer VMP can be applied forprediction.Clearly, there is a trade-off in the selection of thresh-

old. However, as the amount of training data increase,

1 2 3 4 5 60

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Patterns Training Time(months)

Pat

tern

s W

ith R

oads

/Tot

al R

oa

Taxi1Taxi2Taxi3

Figure 10 Pattern utilization.

2 3 4 5 6 76000

6200

6400

6600

6800

7000

7200

7400

7600

7800

8000

Pattern Length (


2. Routing overhead: The total number of control trafficpackets transmitted. Figure 15 shows the controloverhead incurred per second in terms of the totalnumber of control packet transmissions. As DSRuses broadcasting to require information of thedestination node and local routing maintain scheme,when the network area becomes larger and nodes getmore dynamic, the cost increases dramatically.

The routing control packets of PSR and WSR in-clude beacons and announcements. In WSR, since thebeacon and announcement messages are generatedperiodically and their TTL are fixed, the overhead ofWSR increases linearly with the number of nodes inthe network. Unlike WSR, our PSR is based on detect-ing the route maneuver of vehicles and does notpropagate announcement periodically. Without VMP,vehicular nodes only propagate announcement (alongfour directions) when move to a new road. With VMP,if the possible route is predicted correctly, nodes onlypropagate announcements (along four directions) everyother road. Even if the possible route is wrongly pre-dicted, we only need to send announcement (alongone direction) to the wrongly predicted route. As a re-sult, PSR significantly reduces the control overheadand achieve the lowest overhead among the threeapproaches as shown in Figure 15.By analyzing the statistics of the total roads visited

by all the 4000 taxies in a whole day (2006-12-27), wefigure out the average length of road to be around650 m. Given the average velocity of 9.75 m/s, the ve-hicular node’s average travel time per road is 67.2 s.The roads having patterns account for 40% and theprediction accuracy is up to 70% (based on 6-monthdata, with the threshold 0.6). Suppose the averagetravel time per road is 65 s and the roads havingpatterns account for 40% with prediction accuracy

600 700 800 900 10000

2000

4000

6000

8000

10000

12000

14000

16000

Number of Nodes

Tot

al O

verh

ead

per S

econ

d (N

umbe

r of P

acke

ts)

DSRWSRPSR

Figure 15 Control packet overhead.

70%, the frequency of propagating announcement iscomputed as

FPSR ¼ 1� 40%ð Þ � 4=65ð Þ þ 40%� 4=130ð Þ þ 1� 70%ð Þ � 1=130ð Þ½ �

¼ 0:049=s

For WSR, the propagation frequency is

FWSR ¼ 1=5 ¼ 0:2=s

The results show that the number of announcementpackets PSR sends is only about 1/4 of that in WSR. Theactual number of announcement of PSR would be evenfewer considering the case with velocity 0.

3. Average path length: Average number of transmissionhops for all successfully delivered data packets.

Figure 16 shows the transmission hops per successfullyreceived packets. DSR has less transmission hops thanWSR and PSR, since it finds a complete routing path be-fore sending a packet using flooding scheme whichincurs huge overhead. WSR and PSR need more trans-missions to successfully send data packets. The reason isthat source nodes would send data packets in randomdirections if they have no knowledge about the destin-ation nodes, which will prolong the distance betweenthe source nodes and destination nodes. Although thefrequency of announcement propagation in PSR is lowerthan WSR, the longer TE of predictive state informationin PSR offers more available state information amongthe network, and PSR sends announcements in four or-thogonal directions which enable the data packets findan intermediate forwarder having the destination node’supdated information faster.

600 650 700 750 800 850 900 950 10000

2

4

6

8

10

12

14

16

18

Number of Nodes

Num

ber o

f Tra

nsm

issi

ons

per R

ecei

ved

Pac

ket

DSRWSRPSR

Figure 16 Transmission hops.

600 650 700 750 800 850 900 950 10000

500

1000

1500

2000

2500

3000

Number of Nodes

Tot

al O

verh

ead

per S

econ

d (N

umbe

r of P

acke

ts)

Without PatternsUsing Patterns

Figure 18 Control packet overhead.


4. Average end-to-end delay: The average delay betweenpackets sending and receiving for all successfullydelivered application packets. Figure 17 shows theresults about end-to-end delay of the protocols. Thedelay values are only calculated for successfully receivedpackets. DSR has a much higher end-to-end delay thanthe other two methods due to its flooding-based routediscovery, so in this test we only compare PSR with theWSR approach. For WSR and PSR, they have no routediscovery phase before sending data packets, so thedelay is mainly due to propagation delay. Since the stateinformation in WSR decays over time quickly, a datapacket may travel a long path before it finds theavailable information of the destination. Instead, PSRemploys the predictive state information with muchlonger TE and provides more available stateinformation in the network. Also, the lower frequencyof control packets delivered in PSR reduces thepossibility of transmission collision. As a result, PSRoutperforms WSR significantly.

5. The impact of VMP: In this experiment, we analyzethe impact of VMP on routing performance ofPSR. The result in Figure 18 shows that therouting control information in PSR is much lessthan 20–30% compared to PSR without VMP.Figure 19 shows the results about end-to-end delayof PSR with and without VMP. PSR has a lowerend-to-end delay than PSR without VMP. Thelower frequency of control packets delivered in PSRdecreases the possibility of transmission collision.The result in Figure 10 shows that the PU of taxisin our experiments reaches up to 44.88% (based on6-month data, with the threshold 0.6). Theprediction accuracy is up to 70%. The experimentalresults show that VMP can be applied with high

600 650 700 750 800 850 900 950 10000

0.5

1

1.5

2

2.5

3

Number of Nodes

End

to E

nd D

elay

(s)

WSRPSR

Figure 17 Data packet delay.

accuracy, which offers significant enhancement toPSR in cutting control overhead.

6. State information: Localization accuracy relies heavilyon the amount of available information aboutlocation constraints. In this experiment, we comparethe total number of states stored in the network ofPSR and WSR. The result is illustrated in Figure 20.The announcement propagation frequency of PSRis much lower than that of WSR, so in the first500 s WSR disseminates more state informationthan PSR. As the weak state used in WSR becomesstale quickly, the amount of available stateinformation increases slowly or even decreases.Instead, PSR applies the predictive soft state whichhas much longer TE, so more state information ismaintained by PSR after a short period (500 s inthis experiment).

600 650 700 750 800 850 900 950 10000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Number of Nodes

End

to E

nd D

elay

(s)

Without PatternsUsing Patterns

Figure 19 Data packet delay.

Figure 20 Total number of valid routing states maintained pernode versus the number of nodes.


Related studyThe idea of utilizing historic records for prediction isnot a new concept. The authors of [12,13] designed analgorithm adaptively to predict the future positions ofmobile terminals in wireless networks using historicrecords. The authors of [14] addressed location and ac-tivity tracking problems across multiple inhabitants insmart homes through prediction of contexts. Vinod andLixin [15] proposed a prediction-based routing protocolto predict route lifetimes to preemptively create newroutes before existing ones fail in VANET. Nevertheless,these algorithms are specifically designed for concreteenvironments and thus can not play their roles in thisstudy.Mobility models are critical for evaluating applications

for vehicular networks, and proper employment of mo-bility model is often considered as a significant step forvehicular networks simulation. Synthesized mobilitytraces have widely been used to evaluate vehicular net-works performance. Nevertheless, many mathematicalmodels used for generating synthesized traces are farfrom practical as they ignore key characteristics in therealistic scenarios [16], such as traffic jams, traffic rules,and driver behavior.Recently, real-traffic trace and maps and even traffic

pattern have been introduced to assist routing in vehicu-lar networks. By using map data for real US cities,authors of [1] present STeet RAndom Waypoint mobil-ity model which constrains node movement to streetsand limits their mobility according to traffic conditions.Connectivity-aware routing (CAR) [2] uses realistic mo-bility traces and road maps of Switzerland to locate des-tination and path, and data packets are then forwardedalong the found path. CAR uses preferred group broad-casting in data dissemination mode and vehicle-assisted

data delivery (VADD) [3] utilizes snapshot of a realstreet map to derive the street layout. Based on existingcarry and forward mechanism, VADD makes use of pre-dictable vehicle mobility, and a vehicle finds the bestnext road to forward packets according to existing trafficpattern. To share content with fellow travelers, authorsof [17] propose a user-centric prediction scheme collect-ing historical collocation information to determine thebest content sources based on a real dataset consistingof traces of people moving in a large city’s mass transitsystem. Hongzi et al. [18] presented the systematic ap-proach to perceiving metropolitan traffic using a cost-effective system of taxi sensors.Large-scale dynamic wireless networks have intro-

duced new challenges to routing scheme, a well-studiedproblem. Traditional shortest path-based link state ordistance vector routing protocols are no longer effect-ively scale to large mobility networks due to costly con-trol overhead for maintaining the up-to-date stateinformation in networks. Recently, routing protocols inWSR [9] and EASE [19] aim to address the problems ofrouting in large-scale mobile ad-hoc networks. As one ofthe most efficient routing protocols, WSR is a position-aware protocol designed for large-scale dynamic net-works. It aggregates information about a set of remotelocations in a geographic region by mapping a set of IDto the region, indicating the state information. The rout-ing task is conducted using unstructured random direc-tional walks at first. Intermediate nodes then bias theforwarding direction according to their partial weak stateinformation about the destination node. However, thestate information diffusion in WSR inevitably usingflooding mechanism which incurs excessive control traf-fic overhead [19] exploits the node mobility by keepingin each node the encounter history with every othernode, used in finding routes. Although our algorithmshares some similarity with EASE that we also utilize thehistoric records for mining some useful information forroute discovery, we go further to predict the next routebased on each node’s individual records.

ConclusionsIn this article, we study the impact of regular moving beha-viors of vehicles on applications and protocol designs in anurban vehicular network environment. We find that insuch environments the mobility of vehicles are well pat-terned which can be used to predict the locations of thevehicles in a large degree of accuracy. We use the real tracecollected from more than 4,000 taxies at Shanghai over6 months to identify the VMP by VOM model scheme. Bytaking the advantage of VMP, we propose PSR. In PSR, thedisseminated state information carries vehicular node’scurrent state and the predictive states. As the locations ofvehicles can be accurately predicted, a large amount of


control overhead can be saved. The results show that PSRoutperforms existing state-of-art designs in terms of con-trol overhead while retaining high data packet deliveryratio and low delay, at the cost of a little longer averagepath length. Under certain scenarios, the control overheadcan be reduced by 90% compared with DSR, and 75% com-pared with WSR.The future work will be conducted along following

directions. We will first build a prototype system toevaluate the performance of PSR in real settings. Thisstudy is still trace-driven simulations and many practicalissues are not clear yet. We need field study to validatethe assumptions we made in this work. In the next, wewill then investigate how the predicted locations canbenefit other applications such as the in-network storageand management. We have found that the vehicle trafficpatterns will vary based on different day of the week orthe time of the day in our other work [20]. We plan toimprove PSR based on time-varying vehicle traffic pat-terns in future. At last, we will study the MAC-layer de-sign issues in a practical VANET environments.

Competing interestsThe authors declare that they have no competing interests.

AcknowledgementsThis study was supported in part by the National 973 project under GrantNo. 2012CB316106, the China NSFC Grants 60970106, 61170237, and theTechnology Major Project of China under Grant No. 2009ZX03006-004, theChina 863 program under Grant No. 2011AA010500.

Received: 15 February 2012 Accepted: 2 July 2012Published: 18 July 2012

References1. DR Choffnes, F.E. Bustamante, An Integrated Mobility and Traffic Model for

Vehicular Wireless Networks (Proceedings of the Second ACM InternationalWorkshop on Vehicular Ad Hoc Networks (VANET), Cologne, Germany,2005), pp. 69–78

2. V Naumov, TR Gross, Connectivity-aware routing (CAR) in vehicular ad hocnetworks (Proceedings of the 26th IEEE International Conference on ComputerCommunications (INFOCOM), Anchorage, AK, 2007), pp. 1919–1927

3. J Zhao, G Cao, VADD: vehicle-assisted data delivery in vehicular ad hocnetworks (Proceedings of the 25th IEEE International Conference onComputer Communications, Barcelona, Catalunya, 2006), pp. 1–12

4. S Li, YH Liu, X-Y Li, Capacity of large scale wireless networks under Gaussianchannel model (Proceedings of the 14th Annual International Conference onMobile Computing and Networking (MobiCom), San Francisco, California,2008), pp. 140–151

5. J Rissanen, A universal data compression system. IEEE Trans. Inf. Theory 29(5), 656–664 (1983)

6. MC Gonzalez, CA Hidalgo, A-L Barabasi, Understanding individual humanmobility patterns. Nat. J. 453(7196), 779–782 (2008)

7. DB Johnson, DA Maltz, Dynamic source routing in ad hoc wireless networks, inProceedings of the Second Annual International Conference on Mobile Computingand Networking (MobiCom), Rye. New York 353, 153–181 (1996)

8. B-N Cheng, M Yuksel, S Kalyanaraman, Orthogonal rendezvous routingprotocol for wireless mesh networks. IEEE/ACM Trans. Netw. 17(2),542–555 (2009)

9. UG Acer, S Kalyanaraman, AA Abouzeid, Weak state routing for large-scaledynamic networks. IEEE/ACM Trans. Netw. 18(5), 1450–1463 (2010)

10. Network simulator (2011). ns-2. http://www.isi.edu/nsnam/ns11. CMU Monarch Project (1999). ftp.monarch.cs.cmu.Edu/pub/monarch/

wireless-sim/ns-c mu.ps

12. IF Akyildiz, W Wang, The predictive user mobility profile framework for wirelessmultimedia networks. IEEE/ACM Trans. Netw. 12(6), 1021–1035 (2004)

13. T-E Lu, K-T Feng, Predictive mobility and location-aware routing protocol in mobilead hoc networks, vol. 2 (Proceedings of IEEE Global TelecommunicationsConference (GlobeCom), St. Louis, MO, 2005), pp. 899–903

14. R Nirmalya, R Abhishek, SK Das, Context-aware resource management in multi-inhabitant smart homes: a Nash h-learning based approach (Proceedings of theFourth Annual IEEE International Conference on Pervasive Computing andCommunications (Percom), Pisa, Italy, 2006), pp. 11–158

15. N Vinod, G Lixin, Prediction based routing protocol for vehicular ad hocnetworks. IEEE Trans. Veh. Technol. 56(4), 2332–2345 (2007)

16. D Son, A Helmy, B Krishnamachari, The effect of mobility induced locationerrors on geographic routing in ad hoc networks: analysis andimprovement using mobility prediction. IEEE Trans. Mobile Comput. 3(3),233–245 (2004)

17. L McNamara, C Mascolo, L Capra, Media sharing based on colocationprediction in urban transport (Proceedings of the 14th ACM internationalconference on Mobile computing and networking (MobiCom), SanFrancisco, California, 2008), pp. 58–69

18. Z Hongzi, Z Yuanchen, L Minglu, LM Ni, SEER: metropolitan-scale trafficperception based on lossy sensory data (Proceedings of the 28th Conferenceon Computer Communications (INFOCOM), Rio de Janeiro, Brazil, 2009), pp.217–225

19. M Grossglauser, M Vetterli, Locating mobile nodes with ease: learningefficient routes from encounter histories alone. IEEE/ACM Trans. Netw. 14(3),457–469 (2006)

20. G Xue, Z Li, H Zhu, Y Liu, Traffic-known urban vehicular route predictionbased on partial mobility patterns (Proceedings of 2009 InternationalConference on Parallel and Distributed Systems (ICPADS2009), ShenzhenChina, 2009), pp. 369–375

doi:10.1186/1687-1499-2012-222Cite this article as: Xue et al.: A novel vehicular location predictionbased on mobility patterns for routing in urban VANET. EURASIP Journalon Wireless Communications and Networking 2012 2012:222.

Submit your manuscript to a journal and benefi t from:

7 Convenient online submission7 Rigorous peer review7 Immediate publication on acceptance7 Open access: articles freely available online7 High visibility within the fi eld7 Retaining the copyright to your article

Submit your next manuscript at 7 springeropen.com

http://www.isi.edu/nsnam/nshttp://ftp.monarch.cs.cmu.Edu/pub/monarch/wireless-sim/ns-c mu.pshttp://ftp.monarch.cs.cmu.Edu/pub/monarch/wireless-sim/ns-c mu.ps

AbstractIntroductionVMPDefinition 1VMP in the real traceVMP generationDefinition 2

PSR designLocation information propagationData packet forwardingPrediction error recovery

Performance evaluationMethodologyVMPs generationPSR results

Related studyConclusionsCompeting interestsAcknowledgementsReferences

Documents

RESEARCH Open Access A novel vehicular location prediction ... · RESEARCH Open Access A novel vehicular location prediction based on mobility patterns for routing in urban VANET