development and testing of a 3g/lte adaptive data collection … · drira et al.: developmentand...

240 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 1, JANUARY 2016

Development and Testing of a 3G/LTE AdaptiveData Collection System in Vehicular NetworksWassim Drira, Kyoungho Ahn, Hesham Rakha, Member, IEEE, and Fethi Filali, Senior Member, IEEE

Abstract—Vehicular ad hoc networks (VANETs) are a spe-cial case of mobile ad hoc networks (MANETs). The distinctivecharacteristics of VANETs include high-speed vehicular nodesand significant variability in node density. Collecting data fromVANETs is important in monitoring, controlling, and managingroad traffic. However, efficient collection of the needed data is chal-lenging because vehicles are continuously moving and generatinga significant amount of events and data. The focus of this paper ison vehicle data collection using 3G/LTE. Initially, a comparison ofproactive and reactive data collection schemes is conducted usingsimulation. The results show that proactive schemes produce thelowest delay and bandwidth usage but the highest loss ratio. Toefficiently use the available bandwidth, an adaptive data collectionscheme is developed and described. This adaptive data collectionscheme is based on a proactive scheme using variable pollingperiods depending on the vehicle positions in the network andtravel time to provide accurate traffic and travel time informationto the Traffic Management Center (TMC). Simulation results,using taxi traces in Qatar, show that the proposed algorithm con-sumes an acceptable amount of megabytes (≈31 MB) per monthwhen the basic polling interval is set to 10 s. Furthermore, trafficsimulation results demonstrate that the proposed scheme has aminimum impact on delay and travel time estimates (relative errorless than 2.5%), but can produce significant degradations in fuelconsumption and emission estimates if computations are made atthe TMC (relative error greater than 28% and 65%, respectively).These errors can be eliminated if computations are done on thevehicle.

Index Terms—Mobile communication, data collection, auto-motive applications, intelligent transportation systems, wirelessnetworks, traffic information, data aggregation, travel time esti-mation, vehicle emission estimates.

I. INTRODUCTION

V EHICULAR ad hoc networks (VANETs) share several sim-ilarities with Mobile ad hoc networks (MANETs). Both

types of networks are often characterized by node movement,infrastructure-less (ad hoc) operation and self-organization.However the distinctive characteristics that distinguishVANETs from MANETs include highly dynamic networktopology that changes rapidly and frequently and higher node

Manuscript received December 17, 2014; revised April 6, 2015 and June 22,2015; accepted July 29, 2015. Date of publication August 27, 2015; date ofcurrent version December 24, 2015. This work was supported by NPRP Grant5-1272-1-214 from the Qatar National Research Fund (a member of the QatarFoundation). The Associate Editor for this paper was Q. Kong.

W. Drira and F. Filali are with Qatar Mobility Innovations Center, Doha,Qatar (e-mail: [email protected]; [email protected]).

K. Ahn and H. Rakha are with Virginia Polytechnic Institute and StateUniversity Transportation Institute, Blacksburg, VA 24061 USA (e-mail:[email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TITS.2015.2464792

density. This is mainly due to the higher speeds of vehicularnodes and variability in node density. A significant numberof efforts are published on data collection techniques sincethey make inter-vehicle communications more efficient andreliable while minimizing bandwidth utilization. In fact, mostVANET applications are based on the dissemination process[1]–[5] on which information must be propagated over a ratherlong distance so that drivers can be alerted in advance. Sinceeach vehicle in a vehicular environment can detect a hazardoussituation or a congestion zone, the number of messagespumped through the network might increase dramatically.Hence, data collection, as a technique aiming at gatheringpackets from moving vehicles and aggregating the vehiclebeacons, is regarded as an important approach to circumventthese problems.

The emergence of V2X (Vehicle-to-Vehicle (V2V) andVehicle-to-Infrastructure (V2I)) networks lends significant sup-port for Intelligent Transportation Systems (ITS) to improvemany applications for different purposes such as safety, trafficefficiency and added-value services. Typically, such an environ-ment is distinguished by its mobility and topology dynamicsover space and time. To enable these applications, vehiclesneed to broadcast periodic beacons every 100 ms containingvehicle positional and kinematic data [6], [7]. To allow trafficdata analysis, the obvious solution is to collect these beaconsfrom the vehicular network through 3G/LTE or Road Side Units(RSUs) using 802.11p at a centralized Traffic ManagementCenter (TMC). Subsequently, the latter can perform traffic dataanalysis to characterize traffic patterns and detect incidents inorder to take some actions such as controlling traffic lights, alertpolice or provide traffic information to drivers using Radio DataSystem-Traffic Message Channel (RDS-TMC) or TransportProtocol Experts Group (TPEG) standards [8].

Qatar Mobility Innovations Center (QMIC) has a mobileapplication, named iTraffic [9], that provides current trafficstatus and routing information to drivers in Qatar. Currently,the system is using probe data collected from more than200 devices deployed along the Qatar road network. Conse-quently, the system requires the efficient collection of data fromthe application users to cover zones without devices. At thesame time the application should consume a minimum amountof data from the user data plan.

In this paper, the focus is on vehicular data collection through3G/LTE technology using either a smartphone application oran embedded component in the car that can send its data tothe TMC. The contributions from this work can be summa-rized as follows. First, we compare the proactive and reactivedata collection approaches using LTE technology. Then, we

1524-9050 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

DRIRA et al.: DEVELOPMENT AND TESTING OF A DATA COLLECTION SYSTEM IN VEHICULAR NETWORKS 241

propose an adaptive data collection system (ADCS) [10] tocollect data while efficiently using the bandwidth. Next, theprotocol performance in terms of bandwidth usage and savingwas analyzed using real taxi traces. Finally, the algorithm wastested in a simulation environment to quantify the impact of thedata aggregation on the estimates of various measures of effec-tiveness (MOEs). Our proposed idea deals with non-sensitivedelay and saves bandwidth consumption for emergency trafficcontrol applications (hazardous situation, congestion zone, etc).The basic idea of the ADCS scheme is to provide a locationaware adaptive data collection protocol. Our results show thatthe proposed scheme reduces the amount of data transmittedin the network while not adversely affecting the performanceof travel time prediction applications, but potentially impactingenvironmental applications.

The remainder of this paper is organized as follows.Section II presents some related work, followed by a compar-ison between proactive and reactive data collection schemes.Subsequently, the proposed adaptive data collection algorithmis described in Section IV. The impact of the data aggregationscheme on the estimation of measures of effectiveness (MOEs)is then quantified. Finally, the paper conclusions are presentedin Section VI.

II. RELATED WORK

Data collection in vehicular networks can be either central-ized or distributed. Several publications have investigated thedata collection concept by adopting different approaches asdiscussed here.

A. Soua et al. in [11] introduced the ADOPEL protocol,dedicated data collection technique for vehicular networks.ADOPEL is based on a distributed learning approach wherevehicles adaptively choose the forwarding relays to maximizethe ratio of collected data without losing the way to the des-tination node. They introduce a data collector node, whichcollects data from a random number of neighbors and aggre-gates it while simultaneously each vehicle uses reinforcementlearning as a learner. Their simulation results corroborate theeffectiveness of the proposed scheme compared to non-learningversions.

GeoVanet [12] is a routing protocol for query processing inVehicular networks. It ensures that the sender of a query canobtain a consistent answer within a bounded time. In order toresolve the issue of high mobility and network dynamics, whena driver submits a query, a mailbox is assigned to the query.Then, query results are routed to this mailbox instead of thesender. Consequently, the sender can retrieve the results fromthe mailbox. The mailbox can refer to an entity of the fixedinfrastructure or to an area.

Periodic Floating Car Data (FCD) collection using 3G/LTEnetworks was tackled in some research efforts, for example[13], [14]. To enable such an integrated architecture, vehicles areclustered according to different metrics. Then, some vehiclesact as gateways to report the data of the cluster members.LTE4V2X [13] proposes a centralized clustering algorithmwhile Rémy et al. [14] provides a decentralized clusteringprotocol.

Terroso-Saenz [15] put forward an event-driven architectureas a novel mechanism to derive insight into beacon messages todetect different levels of traffic jams. The system can be eitherin the TMC or distributed in vehicles.

Liu et al. in [16] proposed Compressive Sensing Based DataCollection (CS-DC) as a scalable and reliable data collectionscheme for VANETs. In CS-DC, the distance and mobilitybased clustering protocol is proposed to achieve stable clusters,which collects local data reliably and achieves data spatialrelevance. Meanwhile, the CS theory is applied to effectivelyencode/decode the in-network data in data collection, ensuringefficient communication and accurate data recovery. More-over, performance evaluations show that CS-DC can guaranteedesirable performance with higher efficiency, scalability andreliability.

Another effort carried out by Dieudonne et al. in [17] focuseson a distributed collection information for VANETs. It collectsdata produced by vehicles using inter-vehicle communicationsonly. It is based on the operator ant allowing to construct a localview of the network and therefore to collect data in spite ofthe network topology changes. A theoretical proof of correct-ness and experiments confirm the efficiency of the proposedtechnique.

Placzek [1] introduced a novel method of selective datacollection for traffic control applications, which provides asignificant reduction in the amount of data transmitted throughVehicular sensor network (VSN) and observed data are trans-mitted from vehicles to the control node only at selected timemoments. The offered approach is based on an original conceptof uncertainty dependent data collection. According to theconcept the necessity of data transfers can be detected by theevaluation of uncertainty of traffic control decisions. The basicformulas for the uncertainty calculations were derived using aformal description of generalized traffic control procedure andthis is the means of fuzzy arithmetic.

Cherif et al. [18] presented a demonstration of a platformfor data Collection and Aggregation for vehicular NETworks(CANET) developed by France Telecom R&D. The aim of thisplatform is to collect a set of information issued from differentdata sources (sensors, GPS, etc.) and aggregate them in areliable manner. The vehicular platform holds: (i) an on-boardcomputer connected to the Internet via an UMTS PCMCIAcard (ii) a Crossbow sensor which sends temperature, humidityand acceleration data using ZigBee protocol, (iii) a HoluxGPS receiver which sends GPS data using Bluetooth protocol.Collected data is then stored in MySQL database and presentedthough user-friendly webs interface.

Cherif et al. [3] proposed the Road Oriented Dissemination(ROD) protocol that consists of two modules: (i) an OptimizedDistance Defer Transfer module, and (ii) Store and Forwardmodule. The proposed protocol increases the delivery ratio andoptimizes the bandwidth use by limiting the number of vehicleshaving to relay each packet. ROD also implements a Store andForward module allowing the storage of packets when no relayis found.

Salhi et al. [4] proposed a novel data gathering and dis-semination system Clustered Gathering Protocol (CGP) basedon hierarchical and geographical dissemination mechanisms on


TABLE ICOMPARISON BETWEEN DIFFERENT DATA COLLECTION PROCESS

vehicular sensor networks. Designed for a hybrid VANETarchitecture, it allows telecommunication/service providers(WiMax access point, 2.5/3G base station) to get valuable in-formation about the road environment in a specific geographicalarea. In operated VSNs, providers tend to reduce the traffic loadon their network, using an unlicensed spectrum communicationmedium i.e. IEEE 802.11p.

A comparison between the used techniques, in terms of net-work organization and data aggregation, in these data collectionprotocols is given in Table I.

Data collection using V2V communication [11], [12] canbe limited due to slow response times and disjoint networks.While data collection using 3G/LTE [13], [14] can be costlyfor drivers. In this paper, the unique communication technologyavailable for vehicles to send their data will be 3G/LTE and wewill provide an adaptive algorithm to collect the data and savethe used bandwidth.

The literature provides many works that propose adaptivedata collection techniques to collect data efficiently from dis-tributed networks. The push/pull envelope optimization pro-posed in the paper [19] adjusts dynamically the base push/pullrate and data sampling for each sensor according to the relativecharacteristics of sensor requests and sensor data change. Theadaptive data collection approach, presented in [20], gives anefficient adaptive sampling approach based on the dependenceof conditional variance on measurements varies over time. Theproposed FCD data collection in [21], using mobiles, reducesthe negative impact on the utilization of the LTE air interface aswell as on the energy consumption of the mobile devices. Themain concern of most of these protocols is to reduce the energyconsumption. Conscious that energy is not the main concernin V2X, and that data collection optimization is a big concernof telecommunication operators [22] and the role accordedto application in improving the bandwidth usage [23], weproposed a location-aware protocol to adapt data collection ofvehicles and tried to test its effects on data analytic algorithmsthat compute delay and travel time and fuel consumption andemission estimates.

III. REACTIVE VS. PROACTIVE DATA COLLECTION

Proactive data collection is considered when vehicles areconfigured to report their parameters (speed, heading, position,fuel consumption, etc.) every period p to the TMC. In cur-rent V2X standards, a vehicle broadcasts a beacon messageevery 100 ms. The reactive data collection is considered whenTMC requests, either periodically or when needed, all or some

Fig. 1. Messages in proactive data collection approach.

selected vehicles to provide their instantaneous parameters or ahistorical report.

In our previous works, a Pub/Sub framework [24] was pro-posed to allow vehicles to publish their periodic data and aquery language [25] was used to provide a Request/Responseframework to collect data from vehicles. Such frameworks canbe easily adapted to 3G/LTE networks. In this work, we focuson simulating and evaluating proactive and reactive data col-lection approaches to collect data from vehicles using 3G/LTEnetworks.

In order to evaluate both approaches the SUMO traffic sim-ulation software was initially used to generate a random gridnetwork of 2 Km × 2 Km and a random traffic flow of vehiclesrunning at a variable speed between 0 and 60 Km/h. NS-3was then used to evaluate the communication performance.Two eNodeB(s) were placed in the network at locations (1000,500) and (1000,1500), respectively. The TMC communicated tovehicles through the Internet. The penetration rate was varied tomodel different traffic demand levels.

In the proactive simulation, vehicles are configured to sendtheir data every 100 ms (p = 100 ms) to the TMC in a messagehaving a 512 bytes payload size. In the reactive data collection,the TMC requests data from vehicles every 100 ms where therequest and response payloads are of 10 and 512 bytes, respec-tively. Both simulations were executed for a simulation timeof 60 s.

The evaluation results are shown in Figs. 1–4. Figs. 1 and 2presents the number of sent, received and lost messages in bothapproaches. As expected, the total number of messages sentin the reactive data collection scheme, which is the sum ofsent requests and sent responses (i.e. the sum of received andlost responses), is double that sent in the proactive data collec-tion mode.

Fig. 3 describes the ratio of lost messages in both datacollection approaches. It shows that the Proactive approachsuffers from a greater loss ratio, which may be due to messagecollisions when vehicles send data at the same instant in time.In the reactive data collection scheme, however, requests are


Fig. 2. Messages in reactive data collection approach.

sent sequentially to vehicles which reduces the probabilityof collisions. Note that it is possible to use E-MBMS, theEvolved Multicast Broadcast Multimedia Services, to broadcastor multicast requests to vehicles in an attempt to reduce thenumber of messages in the network, but losses will be greater asmany vehicles attempt to respond at the same time. In addition,the used bytes in driver data will remain same. The E-MBMSis not implemented in the used LTE module of NS-3.

In contrast, Fig. 4 shows that the reactive data collectionsuffers from a greater delay which is totally reasonable astwo messages are required to collect the data from vehicles.As minimizing the used bandwidth is a main concern in ourrequirements, the proactive approach is selected. In the nextsection, an adaptive data collection algorithm will be provided.It adapts the period between messages intelligently or after aTMC request to handle the message loss and performance issuewhere the penetration rate is low and more data is required toestimate and/or predict the traffic status.

IV. PROPOSED DATA COLLECTION MODEL

A. Network Model

The communication network is mainly composed of vehiclesand a TMC. Vehicles are equipped with a system that can readvehicle parameters and communicate with the TMC using only3G/LTE communication technologies. In other words, they cancommunicate with TMC using the Internet. We assume thatthe road network is divided into macro-segments, which aresegments between roadway junctions. This road map is storedin vehicles along with a free-flow speed estimate associatedwith every macro-segment to allow the computation of a baseuncongested travel time when needed. Moreover, vehicles canrun the function FindCurrentMS() to find their currentmacro-segment using their GPS coordinates.

B. Proposed Method

The main idea is to have two dynamic timers and a conditionto select the sending frequency of data towards the TMC.

Fig. 3. Loss ratio in proactive and reactive data collection approaches.

Fig. 4. Delay in proactive and reactive data collection approaches.

The first timer Imax defines the maximum time between twosuccessive messages. It is set by the TMC and can be updatedas needed. The defined value can differ between vehicles indifferent zones. Thus it can be reduced when and where moredata are required and increased when and where there is asufficient penetration rate or many sources are in a specific area.The second timer is fired when the travel time of a vehicle alonga specific roadway segment exceeds the maximum estimatedtravel time (emax). This will help the TMC to update theestimated travel time based on the position of vehicles in thesegment knowing the traveled and the remaining distances. Thecondition checks if the current macro-segment is different fromthe last seen one.

C. Proposed Data Collection Model

We consider a network of vehicles that periodically sampleslocal measurements (e.g., position, speed, etc.) at an adaptiverate that is dynamically updated by the TMC. The adaptive


Fig. 5. Adaptive data collection approach.

data collection algorithm can be described as two distributedprocesses running in the TMC and the vehicles.

1) Process at TMC: The TMC collects data from all theconnected vehicles in order to compute the estimated traveltime using real-time and historical data. The algorithm used tocompute the estimated travel time is beyond the scope of thispaper but has been developed in numerous research papers [26].In addition, the TMC can detect congestion, road derivation,and new segments.

In order to develop navigation instructions, vehicles canrequest the routing information from the TMC along with themean and maximum estimated travel time, e and emax, respec-tively. Otherwise, if the vehicle does not use the TMC routingservice, every time it moves to a new segment, it requests therelevant information from nearby segments. This information isrequired to run the adaptive data collection algorithm. Note thatthe TMC should be able to update the vehicular informationwhenever it is needed.

2) Process in Vehicle: The vehicle attempts to update trafficand road status information by collaborating and sending therequired data to the TMC. Thus, sending a beacon every 100 msis not the best solution as it consumes the user’s data plan.Instead, each vehicle uses an adaptive data collection algorithm,shown in Fig. 5, to aggregate and reduce the amount of data sentto the TMC.

The algorithm starts when the vehicle is ON and the GPSsignal is available. Subsequently, it initializes some required

variables: once to False and Sc to the current macro-segment.Next, it evaluates the expression:

(ct − L > Imax ∨ S �= Sc ∨ (Tt > emax ∧ once = False))(1)

which is used to check if the last message to the TMC wassent before Imax, or the current segment is different from thelast saved one, or the travel time has exceeded the maximumestimated travel time. If all the conditions are not satisfied, thevehicle waits for Ip, the interval period between two successivechecks which may be set to 100 ms by default. Subsequently,the vehicle obtains its current parameters (e.g., position, speed,heading, etc.) using the function GetCurrentParams().Next, it attempts to find the current macro-segment (MS) usingthe function FindCurrentMS(). This function may returna NULL value if it was not able to associate a MS to thecurrent position, which may indicate that the GPS signal is lostor there is a new segment. Thus, coordinates are saved andthe algorithm continues to run until finding a MS. Saved newsegment coordinates or a diversion are sent to the TMC to allowit to update its GIS database, if approved. Subsequently, if thefunction FindCurrentMS() associates a MS with the vehicleGPS coordinates, the algorithm returns to check Eq. (1).

If the new observed segment Sc is different from the previousone S, the new segment is saved, once is set to False and thecurrent time ct is saved in L, the Last sent message time. Subse-quently, a message is sent towards the TMC. The message maycontain among other information: the previous and current MS,the travel time along the segment, ct, the entry and exit positionin the previous segment, etc. Otherwise, if the maximum periodImax between two successive messages is exceeded, a messageis sent to the TMC. Otherwise, if the current travel time exceedsemax and once is False indicating that no messages have beensent to inform the TMC about this timeout.

V. ADCS EVALUATION

A. Experimental Evaluation

In order to evaluate the proposed algorithm, traces of ninetaxis running in Qatar were collected over a 58 hour period(≈2.5 days). Each GPS datum is associated with a timestamp,a macro-segment and the distance of the vehicle from the seg-ment start and end points. The estimated travel time is generatedby the QMIC backbone every 60 s using data collected from thedeployed sensors in the Qatar road network. Subsequently, theproposed algorithm runs using these traces for different Imax

values: 60 s, 10 s and 1 s. The evaluation goal is to estimatethe number of messages sent by vehicles and their consumedbandwidth in the selected period. Note that in this evaluation,we ignore messages coming from the TMC that contain emax

values because, while this is a requirement of the routingservice, it is optional because a free-flow travel time can beestimated using the segment length and the free-flow speedalong the segment. The TMC commands to update Imax willbe very few and thus can be neglected without impacting theresults.


TABLE IIEVALUATION RESULTS

TABLE IIIESTIMATED USED BYTES

Table II provides an overview of the obtained results based onthe taxi traces and QMIC travel time data. The results show thatincreasing Imax reduces the number of messages sent by thevehicles. Depending on the used Imax, the proposed algorithmcan reduce significantly the number of generated messages byeach vehicle as demonstrated in Table II. Using Imax = 10 sproduces efficient results and consumes an acceptable amountof mega bytes, as shown in Table III. For example a taxi mayconsume at most 31 MB per month to report its parametersto the TMC when Imax = 10 s. Whereas, Imax = 10 s can notbe increased to an infinite value in order to allow the TMCto mitigate packet losses. In the case of the typical proactivealgorithm, in which a taxi sends a message every 100 ms, thetaxi having the ID 290 would send a total of 513,910 messages.Thus, consuming more than 3 GB per month to transmit beaconmessages to the TMC which currently costs an average valueof 22$.

B. Experimental Validation

The study evaluates the impacts of the various aggregationlevels on the accuracy of the various measures of effectiveness(MOEs) estimates. The study uses the INTEGRATION micro-scopic traffic simulation software to quantify the system-wideimpacts of various field testing. The INTEGRATION softwarewas used because the simulation model has not only beenvalidated against standard traffic flow theory but has also beenutilized for the evaluation of many real-life applications. Themodel is designed to trace individual vehicle movements froma vehicle’s origin to its destination at a level of resolution of onestatus update every 1/10th of a second. The MOEs consideredin the validation include vehicle speed, delay, fuel consumptionlevels, and emissions levels.

The INTEGRATION software is a microscopic traffic as-signment and simulation software that was developed over thepast three decades [27]–[29]. It was conceived as an integratedsimulation and traffic assignment model and performs trafficsimulations by tracking the movement of individual vehicles

every 1/10th of a second. This allows detailed analysis oflane-changing movements and shock wave propagation. It alsopermits considerable flexibility in representing spatial and tem-poral variations in traffic conditions. In addition to estimatingstops and delays, the model can also estimate the fuel con-sumed by individual vehicles, as well as the emissions [30],[31]. Finally, the model also estimates the expected number ofvehicle crashes using a time-based crash prediction model. Themodel has not only been validated against standard traffic flowtheory, but also has been utilized for the evaluation of real-lifeapplications [32].

The INTEGRATION model updates vehicle speeds everydeci-second using the Rakha-Pasumarthy-Adjerid (RPA) car-following model [33], [34]. The RPA model, which integratesa user-specified steady-state speed-spacing relationship (VanAerde model), collision avoidance constraints, and vehicleacceleration constraints, was validated against empirical data[35]. In order to ensure realistic vehicle accelerations, the modeluses the driver throttle input together with a vehicle dynamicsmodel that estimates the maximum vehicle acceleration level.The model computes the vehicle speed as the minimum of threespeeds, namely: the maximum vehicle speed based on vehicledynamics considering the driver throttle level input, the desiredspeed based on the Van Aerde steady-state car-following modelformulation, and the maximum vehicle speed that the vehiclecan travel at ensuring that it does not collide with the vehicleahead of it. The Van Aerde steady-state model was comparedto other steady-state models and has been demonstrated tobe superior to other functional forms in replicating empiricalobservations [36]–[40].

The computation of vehicle accelerations within the simula-tion environment is significant because it ensures that the fuelconsumption and emission level estimates are accurate.

1) Network Description: Fig. 6 illustrates the test networkthat was utilized in the study. The total length of the corridor is1 km composed of 8 links, each 0.25 km in length. The corridorhas three signalized intersections with a fixed-time traffic signalplan. The free-flow speed was set to 60 km/h and the saturationflow rate was 1800 veh/h/lane for the entire corridor. The totalcycle time was 60 second and the green/Cycle ratio was set to0.5. Demand from the start node to the end node was 900 veh/hand was loaded for 30 minutes. Thus 450 vehicles were injectedinto the simulation.

The simulation ran in application level and assumed thatthere is no communication delay, interference, and lost package.


Fig. 6. Test network.

TABLE IVTOTAL DATA COLLECTED

TABLE VRESULTS OF DATA AGGREGATION USING ADAPTIVE DATA COLLECTION

The simulation study only focused to identify the traffic relateddata quality impacts on the intervals of data collection. Theintroduction of communication errors including different com-munication delays, interferences, lost packages may introduceadditional variability on traffic data quality in this study.

2) Simulation Results: Table IV demonstrates the systemneeds to send a total of 43 317 messages at 1 second com-munication interval. However, if the data are collected every10 second interval, the system sends only 4552 messages. Thus,the adaptive data collection can reduce the communication loadand data storage requirements considerably.

Table V explains the traffic operational impacts of adaptivedata collection. The results clearly demonstrate that there isno significant impact on delay estimation with various datacollection intervals but in case of the speed, as the data col-lection interval increases the vehicle speed increases by up to3.59 percent.

TABLE VIFUEL CONSUMPTION AND EMISSION ESTIMATES

The total delay is computed as:

β∫α

[1 − min(uf , ui)

uf

]Δt

where dk is the delay incurred at intersection k, Δt is theduration of the time interval, α is the time when a test vehicleentered the corridor, β is the time when the vehicle exits thecorridor, uf is the free-flow speed, and ui is the vehicle speedat instant i.

While the adaptive data collection can effectively collectvehicle speed and delay data, the accuracy of fuel consumptionand emissions is significantly affected by an adaptive data col-lection method as demonstrated in Tables VI. Fuel consumptionand emissions were estimated using the VT-Micro model [30].The table explains the aggregated data reduces the accuracyof the fuel consumption and emission estimates, particularlywhen the data are aggregated at 10 s intervals. The 10 s ag-gregated data produced much smoothed vehicle speed profilesand reduced sharp acceleration and deceleration operations.Highest fuel consumption and emissions are typically producedunder high acceleration operations. Thus, the fuel consumptionand emission estimates from 10 s aggregated data produce lessaccurate energy and emission results than 1 s aggregated modelestimates. The accuracy could be enhanced if the computationsare done on the vehicle and then the estimates are sent at eachpolling interval. This would not reduce the accuracy of theestimates.

The MOEs are significantly affected by the level of conges-tion at the three intersections (250 m, 500 m, and 750 m) asdemonstrated in Table VII. Interestingly Roadway Section 8 isthe most congested section based on the delay and vehicle speedestimates but Roadway Section 3 is the least environmentallyfriendly section.

Additionally, Table VIII shows that there is no significantdifference for the delay estimation considering 1 s, 5 s, and 10 sdata aggregation intervals. However, in case of CO emissions,1 s aggregation results are significantly different than the resultsfor 10 s aggregation. Again, as was mentioned earlier, this couldbe resolved by doing the computation and aggregation in thevehicle and then just transmitting the data each polling interval.Such an approach would not result in any degradation of thesystem outputs.


TABLE VIITHE IMPACTS OF ROADWAY SECTIONS USING ENTIRE DATA (1 S INTERVAL)

TABLE VIIICOMPARISON OF ROADWAY SECTION RESULTS

VI. CONCLUSION

The paper presented and compared proactive and reactivedata collection mechanisms. Subsequently, an adaptive datacollection algorithm to collect vehicle data efficiently usingonly a 3G/LTE communication network was developed. Sub-sequently, the system was validated using in-field taxi traces toevaluate the system performance in terms of bandwidth savings.Finally, the algorithm was tested in a simulation environmentto quantify the impact of the data aggregation on the estimatesof various MOEs. The results demonstrate that computationsdone using 10 s data has minimal impact on delay estimates(error less than 2.5 percent) and travel time estimates (errorless than 18 percent), however, it has a significant impact onvehicle fuel consumption and emission estimates (over 38 and65 percent error, respectively). A potential solution to thisproblem would be for an on-board system to aggregate the dataand then send the estimates to the TMC at the polling intervals.This would ensure that the MOE estimates are not degradedfor different levels of resolution. The results are encouragingwarranting further testing of the algorithm in real world devicesand comparing the travel time estimates at the TMC based to theground truth.

ACKNOWLEDGMENT

The statements made herein are solely the responsibility ofthe authors.

REFERENCES

[1] B. Placzek, “Selective data collection in vehicular networks for trafficcontrol applications,” CoRR, vol. abs/1112.4620, 2011.

[2] X. Tang and J. Xu, “Adaptive data collection strategies for lifetime-constrained wireless sensor networks,” IEEE Trans. Parallel Distrib. Syst.,vol. 19, no. 6, pp. 721–734, Jun. 2008.

[3] M. Cherif, S.-M. Secouci, and B. Ducourthial, “How to disseminatevehicular data efficiently in both highway and urban environments?” inProc. IEEE 6th Int. Conf. WiMob Comput., Netw. Commun., Oct. 2010,pp. 165–171.

[4] I. Salhi, S. Senouci, and M. Cherif, “A new framework for data collectionin vehicular networks,” in Proc. IEEE ICC, 2009, pp. 1–6.

[5] R. Singh and A. Gupta, “Information dissemination in vanets using zonebased forwarding,” in Proc. IFIP WD, Oct. 2011, pp. 1–3.

[6] Basic Set of Applications; Part 2: Specification of Cooperative AwarenessBasic Service, ETSI TS 102 637-2 V1.2.1, 2011.

[7] J. Kenney, “Dedicated Short-Range Communications (DSRC) standardsin the United States,” Proc. IEEE, vol. 99, no. 7, pp. 1162–1182,Jul. 2011.

[8] TISA—Traveller Information Services Association, Accessed: 2014-02-12. [Online]. Available: http://www.tisa.org

[9] iTraffic—Mobile Application, Accessed: 2014-02-12. [Online].Available: http://www.tisa.org

[10] W. Drira, D. Puthal, and F. Filali, “ADCS: An adaptive data collectionscheme in vehicular networks using 3G/LTE,” in Proc. IEEE ICCVE,Vienna, Austria, 2014, pp. 753–758.

[11] A. Soua and H. Afifi, “Adaptive data collection protocol using rein-forcement learning for VANETs,” in Proc. 9th IWCMC, Jul. 2013,pp. 1040–1045.

[12] T. Delot, N. Mitton, S. Ilarri, and T. Hien, “GeoVaNET: A routing protocolfor query processing in vehicular networks,” Mobile Inf. Syst., vol. 7,no. 4, pp. 329–359, Dec. 2011.

[13] G. Remy, S. M. Senouci, F. Jan, and Y. Gourhant, “LTE4V2X: LTE fora centralized VANET organization,” in Proc. IEEE GLOBECOM, 2011,pp. 1–6.

[14] T. Taleb and A. Benslimane, “Design guidelines for a network architec-ture integrating VANET with 3G amp; beyond networks,” in Proc. IEEEGLOBECOM, 2010, pp. 1–5.


[15] F. Terroso-Saenz, M. Valdes-Vela, C. Sotomayor-Martinez,R. Toledo-Moreo, and A. Gomez-Skarmeta, “A cooperative approachto traffic congestion detection with complex event processing andVANET,” IEEE Trans. Intell. Transp. Syst., vol. 13, no. 2, pp. 914–929,Jun. 2012.

[16] C. Liu, C. Chigan, and C. Gao, “Compressive sensing based data collec-tion in VANETs,” in Proc. IEEE WCNC, Apr. 2013, pp. 1756–1761.

[17] Y. Dieudonne, B. Ducourthial, and S. M. Senouci, “COL: A datacollection protocol for VANET,” in Proc. IEEE IV Symp., Jun. 2012,pp. 711–716.

[18] M. Cherif, S. M. Senouci, and G. Remy, “A data collection and aggrega-tion platform for vehicular networks,” in Proc. 35th Int. Conf. LCN, 2010,pp. 172–175.

[19] Y. Xu, S. Helal, M. Thai, and M. Scmalz, “Optimizing push/pull envelopesfor energy-efficient cloud-sensor systems,” in Proc. 14th ACM Int. Conf.MSWiM, 2011, pp. 17–26.

[20] D. Laiymani and A. Makhoul, “Adaptive data collection approach for peri-odic sensor networks,” in Proc. 9th IWCMC, Jul. 2013, pp. 1448–1453.

[21] C. Ide, B. Dusza, and C. Wietfeld, “Energy efficient LTE-based floatingcar data collection for dynamic traffic forecasts,” in Proc. IEEE ICC,Jun. 2012, pp. 6303–6307.

[22] ABIresearch—The Importance of Mobile Video & Data Optimization, Ac-cessed: 2014-04-01. [Online]. Available: https://www.abiresearch.com/blogs/impportance-mobile-video-data-optimization/

[23] telecoms.com—Why Optimisation Matters, Accessed: 2014-04-01. [On-line]. Available: http://telecoms.com/20220/why-optimisation-matters/

[24] W. Drira and F. Filali, “A Pub/Sub extension to NDN for efficient datacollection and dissemination in V2X networks,” in Proc. 15st Int. Symp.Worl Wireless, Mobile Multimedia Netw., Sydney, N.S.W., Australia,Jun. 2014, pp. 1–7.

[25] W. Drira and F. Filali, “NDN-Q: An NDN query mechanism for efficientV2X data collection in smart cities,” in Proc. 11th Annu. IEEE Int. Conf.SECON Workshops, Singapore, Jun. 2014, pp. 13–18.

[26] Fundamentals of Transportation-Traffic Flow, Accessed: 2014-05-12.[Online]. Available: http://en.wikibooks.org/wiki/Fundamentals_of_Transportation/Traffic_Flow

[27] H. Rakha and M. V. Aerde, Integration Release 2.40 for Windows:User’s Guide—Volume I: Fundamental Model Features. Blacksburg,VA, USA: M. Van Aerde & Assoc., Ltd., 2007.

[28] H. Rakha and M. V. Aerde, Integration Release 2.40 for Windows: User’sGuide—Volume II: Advanced Model Features. Blacksburg, VA, USA:M. Van Aerde & Assoc., Ltd., 2007.

[29] M. Van Aerde and S. Yagar, “Dynamic integrated freeway/traffic signalnetworks: Problems and proposed solutions,” Transp. Res. A, Gener.,vol. 22, no. 6, pp. 435–443, Nov. 1988.

[30] K. Ahn, H. Rakha, A. Trani, and M. V. Aerde, “Estimating vehi-cle fuel consumption and emissions based on instantaneous speed andacceleration levels,” J. Transp. Eng., vol. 128, no. 2, pp. 182–190,Mar. 2001.

[31] K. Ahn, H. Rakha, and A. Trani, “Microframework for modeling of high-emitting vehicles,” Transp. Res. Rec.—J. Transp. Res. Board, no. 1880,pp. 39–49, 2004.

[32] H. Rakha, A. M. Flintsch, K. Ahn, I. El-Shawarby, and M. Arafeh, “Evalu-ating alternative truck management strategies along interstate 81,” Transp.Res. Rec.—J. Transp. Res. Board, no. 1925, pp. 76–86, 2005.

[33] H. Rakha, H. Pasumarthy, and S. Adjerid, “A simplified behavioral vehiclelongitudinal motion model,” Transp. Lett.—Int. J. Transp. Res., vol. 1,no. 2, pp. 95–110, Apr. 2009.

[34] J. Sangster and H. Rakha, “Enhancing and calibrating the Rakha-Pasumarthy-Adjerid car-following model using naturalistic drivingdata,” Int. J. Transp. Sci. Technol., vol. 3, no. 3, pp. 229–248,Sep. 2014.

[35] J. Sangster, H. Rakha, and J. Du, “Application of naturalistic drivingdata to modeling of driver car-following behavior,” Transp. Res. Rec.—J.Transp. Res. Board, no. 2390, pp. 20–33, 2013.

[36] H. Rakha and B. Crowther, “Comparison of greenshields, pipes, and VanAerde car-following and traffic stream models,” Transp. Res. Rec.—J.Transp. Res. Board, no. 1802, pp. 248–262, 2002.

[37] H. Rakha and B. Crowther, “Comparison and calibration of Fresim and in-tegration steady-state car-following behavior,” Transp. Res. A, Pol. Pract.,vol. 37, no. 1, pp. 1–27, Jan. 2003.

[38] H. Rakha and M. Arafeh, “Calibrating steady-state traffic stream and car-following models using loop detector data,” Transp. Sci., vol. 44, no. 2,pp. 151–168, May 2010.

[39] H. Rakha, “Validation of Van Aerde’s simplified steadystate car-followingand traffic stream model,” Transp. Lett., vol. 1, no. 3, pp. 227–244,2009.

[40] N. Wu and H. Rakha, “Derivation of van aerde traffic stream modelfrom tandem-queuing theory,” Transp. Res. Rec.—J. Transp. Res. Board,no. 2124, pp. 18–27, 2009.

Wassim Drira received the M.Sc. degree in com-puter engineering from the National School ofEngineers, Sfax, Tunisia, in 2009 and the Ph.D.degree in computer science jointly from the Uni-versity of Pierre and Marie Curie (Paris 6), Paris,France, and the Institut Mines-Télécom/TélécomSudParis, Evry, France, in 2012. He is currently aResearch Scientist with Qatar Mobility InnovationsCenter (QMIC), Doha, Qatar, where he has been aResearch Scientist on CoSMob, a Qatar National Re-search Fund funded research project, since February

2013. In addition, he was working on different Masarak-related tasks to enhanceor extend some QMIC-related products. He has participated in several Frenchand European projects during his studies and Ph.D. thesis. These projectswere in various fields such as the cold chain supervision using wireless sensornetworks and the data collection and management of data from a wireless sensornetwork for water quality supervision and cloud networking. His researchinterests are in connected vehicles, wireless sensor networks, cloud computing,and data analytic.

Kyoungho Ahn received the Ph.D. degree in civiland environmental engineering from Virginia Poly-technic Institute and State University (VirginiaTech), Blacksburg, VA, USA, in 2002. He is cur-rently a Research Scientist with the Center for Sus-tainable Mobility, Virginia Polytechnic Institute andState University Transportation Institute. He havedeveloped and conducted extensive research studiesas a Principal Investigator and a Faculty Investiga-tor in vehicle energy and environmental modeling,connected vehicle research, traffic simulation, traffic

flow theory and modeling, Intelligent Transportation Systems, advanced trafficsignal control systems, and traffic impact studies. He has authored/coauthoredover 60 refereed publications, of which 25 are fully refereed journal publica-tions.

Hesham Rakha (M’04) received the B.Sc.(Hons.)degree in civil engineering from Cairo University,Cairo, Egypt, in 1987 and the M.Sc. and Ph.D.degrees in civil and environmental engineering fromQueen’s University, Kingston, ON, Canada, in 1990and 1993, respectively. He is currently the SamuelReynolds Pritchard Professor of Engineering withthe Charles E. Via, Jr. Department of Civil and En-vironmental Engineering and a Courtesy Professorwith the Bradley Department of Electrical and Com-puter Engineering at Virginia Polytechnic Institute

and State University (Virginia Tech), Blacksburg, VA, USA, and the Directorof the Center for Sustainable Mobility at Virginia Tech Transportation Institute.He has authored/coauthored over 302 refereed publications in the areas oftraveler and driver behavior modeling, traffic flow theory, large-scale trafficsimulation and modeling, dynamic traffic assignment, traffic control, energyand environmental modeling, and safety modeling. Dr. Rakha is a memberof the Institute of Transportation Engineers, the American Society of CivilEngineers, and the Transportation Research Board. He is also a ProfessionalEngineer in Ontario, Canada.


Fethi Filali (SM’14) received the Computer Sci-ence Engineering and Master’s degrees from the Na-tional College of Computer Science (ENSI), Tunis,Tunisia, in 1998 and 1999, respectively. In 1999, hejoined the INRIA research institute, Sophia Antipolis,France, to prepare a Ph.D. in computer science,which he has defended in November 2002. During2003, he was an Assistant Professor at UniversitéNice Sophia Antipolis. From September 2003 to De-cember 2009, he was an Assistant/Associate Profes-sor with the Mobile Communications Department,

EURECOM, Sophia Antipolis. He is currently the Head of Technology Devel-opment and Applied Research at Qatar Mobility Innovations Center (QMIC),Doha, Qatar. He is leading the technology development for QMIC’s Masarak,Labeeb IoT/M2M, WaveTraf, and Connected Cars platforms and services.He is/was involved in several French-funded (Dipcast, Constellation, Rhodos,Cosinus, Airnet, WiNEM) and IST FP6/7 (Widens, Newcom, Daidalos, E2R,Multinet, Unite, Chorist, iTetris, Newcom++) projects. In the context ofsome of these projects, he designed and developed an open, flexible, andefficient architecture for the support of heterogeneous radio technologies. Thisarchitecture was integrated in EURECOM’s wireless software-radio platform.He is/was the Ph.D. Supervisor of eight Ph.D. students in the area of computernetworking, wireless sensor and mesh networks, vehicular communications,and mobility management. His current research interests include the design,development, and performance evaluation of communication protocols andsystems for ubiquitous networking, communications for intelligent transporta-tion systems (particularly vehicle-to-vehicle and vehicle-to-infrastructure com-munications), WIMAX wireless networks, sensor and actuator networks, andwireless mesh networks. Dr. Filali is a Member of IEEE CommunicationSociety and a Member of IEEE Intelligent Transportation System Society. Hehas served as a technical reviewer of several international conferences andjournals. He was a recipient of the “Habilitation à Diriger des Recherches”(HDR) from Université Nice Sophia Antipolis in April 2008 for his research onwireless networking.

development and testing of a 3g/lte adaptive data collection … · drira et al.: developmentand...

Documents