congestion and awareness control in cooperative vehicular...

1

Congestion and Awareness Control in CooperativeVehicular Systems

Miguel Sepulcre, Jens Mittag, Paolo Santi,Member, IEEE, Hannes Hartenstein,Member, IEEE, Javier Gozalvez

Abstract—Cooperative vehicular systems have been identifiedas a promising solution to overcome the current and futureneeds for increasing traffic safety and efficiency, while providinginfotainment and added-value services on the move. To achievetheir objectives, cooperative vehicular systems will be basedon wireless communications between vehicles and with otherinfrastructure nodes, and will have to deal with highly dynamicnodes, challenging propagation conditions, and stringent appli-cation requirements. By looking at cooperative applications andtheir data traffic, as well as the current and foreseen spectrumallocations for cooperative vehicular systems, there is a risk thatthe corresponding radio channels could easily be saturatedifno control algorithms are used. The saturation of the radiochannels would result in unstable vehicular communications, andthus in an inefficient operation of cooperative systems. As aprime example of upcoming ubiquitous networks contributingto the vision of “a thousand radios per person”, cooperativevehicular systems need to be designed to scale to high densitiesof radios without centralized coordination, while at the same timeguaranteeing the requirements of the implemented applicationsand services, for example the stringent needs of active trafficsafety applications. In this paper, we survey and classify variousdecentralized methods to control the load on the radio channelsand to ensure each vehicle’s capacity to detect and communicatewith the relevant neighbouring vehicles, with a particular focuson approaches based on transmit power and rate control. Finally,we discuss the open research challenges that are imposed bydifferent application requirements and potential existing contra-dictions.

Index Terms—Cooperative Vehicular Systems, CongestionControl, Awareness Control, Power Control.

I. I NTRODUCTION

Foreseen cooperative systems for Intelligent TransportationSystems (ITS) address the current and future needs of in-creasing traffic safety, efficiency and comfort. Despite thepredicted growth ratesin the number of motorized vehiclesand the volume of transported goods, transportation shouldbecome safer, cleaner, more efficient and more comfortable.Tohelp to reach these goals, cooperative vehicular systems willenable the direct exchange of information between vehicles,and between vehicles and Road Side Units (RSUs), usingthe IEEE 802.11p [1] technology on the 5.9GHz band. This

M. Sepulcre is with the Uwicore Lab., University Miguel Hernandez ofElche, Spain (email: [email protected])

J. Mittag is with the Institute of Telematics, Karlsruhe Institute of Tech-nology, Germany (email: [email protected])

P. Santi is with the Istituto di Informatica e Telematica del CNR, Italy(email: [email protected])

H. Hartenstein is with the Steinbuch Centre of Computing, KarlsruheInstitute of Technology, Karlsruhe, Germany (e-mail: [email protected])

J. Gozalvez is with the Uwicore Lab., University Miguel Hernandez ofElche, Spain (email: [email protected])

technology is based on the CSMA/CA (Carrier Sense MultipleAccess with Collision Avoidance) access protocol, and isbeing adapted to the European context in the ETSI ITS-G5standard [2]. The operation of cooperative vehicular systemsis currently based on the exchange oftwo primary typesofmessages. Ontheone hand, Cooperative Awareness Messages(CAMs), also known as beacons, are broadcasted by allnodes on theso-calledcontrol channel, to provide and receivestatus information aboutthe presence, geographical positionand movementof neighoring nodes, and service announce-ments to/fromthosenodes. On the other hand, event-drivenemergency messages are transmitted when an abnormal ordangerous situation is detected, in order to inform surroundingnodes about it.

As the technology becomes more widely adopted, andcooperative applications and services are deployed, the sharedradio channels could easily be saturated. It is well knownfrom wireless local area networks based on CSMA/CA, thatcommunication performance might not degrade gracefully ifthe network is saturated, butwill in fact drop significantlyonce the maximum capacity is exceeded. Since the exchangeof periodic CAMs alone could already saturate the channel,methods are required to control and limit the load on theradio channel. Congestion control has beenstudied in depthin variousareas of computer networking. The term congestioncontrol typically goes together with the Transport ControlProtocol (TCP) of the Internet protocol suite. Here, cooperativecontrol is used to ensure that each TCP connection gets afair share of the available network resources. The mechanismsfor congestion control in vehicular networks typically showanalogous concepts, like decentralized control and fairness, butdiffer significantly due to the specific constraints of wirelesscommunications in highly mobile and potentially harsh radioconditions.

In addition to guaranteeing a channel load level that ensuresstable system operation, cooperative vehicular systems willbe required to ensure connectivity amongthe vehicular nodesimposed by the implemented applications. To ensure network-wide connectivity through the dynamic adaptation of eachnode’s transmission parameters, topology control protocolshave been proposed for wireless ad-hoc and sensor net-works [3]. However, thepresence of highlydynamic vehicularmobility, along with the harsh radio propagation conditionsstrongly challenge the establishment of stable vehicular con-nections, andthusreduce the feasibility of traditional topologycontrol protocols. In addition, cooperative vehicular systemsdo not require network-wide connectivity and the establish-ment of links in the traditional sense, but rather accurate

2

��

��

Fig. 1. Highway scenario with a traffic jam in one direction ofdriving andfree flow conditions in the other direction. This example represents a typicaltraffic situation in which congestion control and awarenesscontrol protocolsmight be needed.

and updateddata oneach vehicle’s local environment (e.g.position, speed and direction of movement of neighboringvehicles) to support upper-layer protocols and cooperativeapplications. In this context, this paper defines awarenesscontrol protocols as those techniques aimed at ensuring eachvehicle’s capacity to detect, and possibly communicate withthe relevant vehicles and infrastructure nodes present in theirlocal neighborhood, through the dynamic adaptation of theirtransmission parameters. Awareness control protocols can, forexample, adapt each vehicle’s transmission power to success-fully transmit a message at a given distance, or dynamicallymodify each vehicle’s packet generation rate to increase theprobability of receiving at least one packet at a certain dis-tance during a given time window. Given their similarities,awareness controlcan be seenas a geo-localized adaptationof topology control.

Based on the previous definitions and their fundamentallydifferent objectives, congestion and awareness control can beeasily differentiated. For instance, congestion control aims tolimit the observed load on the wireless channel for all nodesin order to provide fair and harmonized access to the wirelessmedium. As such, congestion control algorithms reduce thetransmission power or rate of all nodes in order to avoidscenarios in which neighboring nodes,which are part of thesame traffic situation, use (on average) significantly differentpower levels or beaconing rates. Considering the exampleillustrated in Figure 1, the high density of vehicles in thetraffic-congestedarea would require the use of congestioncontrol protocols to control and limit the channel load. Unlikecongestion control protocols, awareness control algorithmsadjust the power or rate of only a selectedsubsetof nodes,with the objectiveof fulfilling the requirements of a partic-ular application. In the scenario illustrated in Figure 1, therequirements of the applications run by the vehicles in thetraffic jam are notably differentfrom those of the applicationsrun by the vehicles underfree-flow conditions moving in theopposite direction, with different speeds and distances betweenneighbouring vehicles. For example, while vehicles underfree-flow conditions would require their communication settings toallow for a safelane-change maneuver, such a maneuver couldbe completely unexpected, or be less dangerous, for vehicles inthe traffic jam. Awareness control protocols would be thereforerequired to dynamically adapt each vehicle’s communicationsparameters to efficiently satisfy their individual requirements.

In this context, this paper focuses on congestion andawareness-control techniques with a special emphasis ontransmit power control and application-driven design policies.

Recently, various researchers contributed approaches andper-formance evaluations that address the issues of controlling theload on the radio channel, and guaranteeing each vehicle’scapacity to communicate with its local neighborhood. In thispaper, we survey various key approaches and key findingsin a coherent manner to present and categorize the pool ofideas for upcoming standardization and deployment activities.We will discuss in some detail three approaches based ontransmit power control, one targeting the issue of congestioncontrol, and two addressing an efficient awareness control.The paper is structured accordingly. We first introduce somebackground information in Section II, followed by a discussionof congestion and awareness control through the perspectiveof control theory in Section III. In Section IV, we surveycongestion control approaches that have been proposed forvehicular communications, and discuss in detail a proposalbased on transmit power control, in addition to evaluating itsperformance vs. effectiveness trade-offs. Section V surveyskey contributionsto awareness control, before discussing geo-opportunistic and traffic contextual approaches designed toensure the strict requirements imposed by cooperative appli-cations, in particular traffic safety applications. Section VIdiscusses open research challenges deriving from the jointstudy of congestion and awareness control protocols, as well asmulti-application scenarios. Finally, Section VII summarizesthe main contributions made by this paper.

II. BACKGROUND

A first generation of future cooperative vehicular systemswill be based on the IEEE 802.11p standard, according tocurrent standardization activities. As such, previous studiesdealing with congestion and awareness control have beenperformed on top of IEEE 802.11p. In order to support theexplanations and descriptions given in the following sectionsand to clarify the system setup and the assumptions made inthis paper, we will briefly elaborate on the relevant aspectsofthe communication system for cooperative vehicular systems.

IEEE 802.11p [1] is specified to operate in the 5.9 GHzfrequency band. At the MAC (Medium Access Control) layer,it employs the CSMA/CA mechanism to coordinate mediumaccess by multiple stations. In CSMA/CA, each station hasto listen to the channel and check whether it is free beforebeing allowed to transmit.This operation is calledcarriersensing, and it is performed by comparing the detected energyon the channel with a pre-defined threshold, called thecarriersensingthreshold. The region of space where a certain ongoingtransmission can be detected by a device tentatively accessingthe channel is called thecarrier sensing region. Note that, ingeneral, this region has irregular shape due to non-isotropicradio signal propagation. However, it is a common practicein the wireless networking literature to consider the carriersensing region as nearly circular, hence the notion ofcarriersensing rangeused to refer to the distance up to which anongoing transmission can be sensed by a device attemptingto access the channel.If the channel is busy, the station hasto defer, wait until the channel is free again and choose arandom backoff timer that determines the additional waiting

3

time that has to elapse after the channel is sensed idle. Despitethe backoff mechanism, two or more stations can transmitsimultaneously, therefore producing a packet collision anda possible data loss due to interferences. In particular, twostations can transmit simultaneously mainly due to the wellknown hidden-terminal problem. The hidden-terminal problemoccurs when two (or more) stations cannot detect each other’stransmissions, but their transmission ranges are notdisjoint. Ithasbeen widely studied in the literature.

Due to its robustness against fast fading channels, theIEEE 802.11p amendment adapts the OFDM (Orthogonal Fre-quency Division Multiplexing) transmission technology usedin IEEE 802.11a and g, with the exception that 10 Mhz insteadof 20 MHz channels are used by default. The reduction from20 Mhz to 10 Mhz was necessary to account for the increasedDoppler and RMS delay spreads (as reported by [4] or[5]) which would otherwiselead to inter-symbol interference(ISI) and inter-carrier interference (ICI), and thus significantlychallenge the successful reception of packets. Apart from theproblem of ISI and ICI, a receiver is also challenged by thefast fading channel conditions that are observable due to thehigh relative mobility of vehicles. For instance, the coherencetime of the channel, i.e. the time during which the channelimpulse response is essentially invariant, can be smaller thanthe duration of a single packet transmission [6], which couldresult in an increased probability of bit and packet errors.Thisis an issue, since the IEEE 802.11p frame format providesonly a preamble to fully estimate the channel and only fourpilot sub-carriers to partially track the state of the channel [7].Hence, the initial estimate can become invalid at the end of thereception leading to an increased probability of bit and packeterrors [6]. As a consequence, the channel impulse response oftwo consecutive packet transmissions will most likely not becorrelated, and the channel can only be considered symmetricfor an instant of time, but not for slightly different timestamps.By symmetric, we mean that the propagation characteristicsof the radio channel are approximately the same in bothdirections of the wireless communication link.

As defined by theFederal Communications Commission ofthe USA (FCC)[8], a spectrum of 75 MHz has been allocatedat 5.9 GHz. Similarly, a spectrum of 50 MHz has been allo-cated at the same frequency band in Europe. In both cases, theentire spectrum is divided into several 10 MHz channels, outof which one channel, commonly calledthe Control Channel(CCH), is used as a reference for the exchange ofsafety-relatedinformation. The remaining channels are known as servicechannels and are used for safety and non-safety applications.The data rates provided by IEEE 802.11p [1] with such 10 Mhzchannels range from 3 to 27 Mbps. While the lower datarates are the most robust ones and require lower signal-to-interference and noise ratio to correctly decode a packet,higher data rates come with the benefit of reduced transmissiontimes, and thus with the possible gain of a reduced packetcollision probability in situations of higher channel loads.Obviously, there is trade-off between increased robustness andreduced congestion. To investigate this trade-off, a simulationstudy was performed in [9] in order to determine the mostrobust data rate for broadcast communication.The study’s

Level Max. density Avg. Speed Vehicles withinof Service (vehicles/km/lane) (km/h) 1000 m comm. range

A 7 100.0 84B 11 100.0 132C 16 98.4 192D 22 91.5 264E 25 88.0 300

TABLE ITHE CAPACITY OF MULTILANE HIGHWAYS AND THE CORRESPONDING

AVERAGE SPEEDS ACCORDING TO[10]. IN ADDITION , THE NUMBER OF

VEHICLES WITHIN THE COMMUNICATION RANGE ARE LISTED FOR A3LANE PER DIRECTION HIGHWAY AND A 1000M COMMUNICATION RANGE.

results revealthat a 6 Mbps data rate turns out to be thebest selection for safety related communication. Since recentstandardization activities [2] have taken up these findings, wewill assume a fixed data rate of 6 Mbps in the rest of thispaper.

Apart from the key technology characteristics, it is also im-portant to understand howit is envisioned thatthe cooperativevehicular systemwill operate and what its dimensions willbe. According to the common agreement among researchersand industry, vehicles will periodically broadcast cooperativeawareness messages (CAM) in order to establish a mutualawareness. CAM messagesprovide information onposition-ing, speed, and heading, among other fields. The establishmentof a mutual awareness can be considered the fundamentalsafety service in cooperative vehicular systems, on top ofwhich advanced safety applications, e.g. cooperative forwardcollision warning or intersection collision warning, willbedeployed. However, in order to fulfill the requirements ofsuch advanced safety applications, the transmitted informationmight needto be updated several times per second, possiblyrequiring a periodic CAM rate of up to ten messages persecond [11]–[13]. Hence, solely the establishment of a mutualawareness could saturate and congest the wireless channel,especially according to the following considerations. First,each CAM message could have asize of between 250 and800 Bytes, due to digital signatures and certificates that secureand authenticate the information contained in those messages.Second, the communication system is expected to cover dis-tances of up to 1000 m. And third, vehicledensities ofupto 25 vehicles/km/lane are not an exception (cf. the reportedcapacity of multilane highways as listed in Table I). As aresult, the total amount of traffic generated per second formutual awareness could easily exceed the available data rate of6 Mbps. As a consequence, the performance of the communi-cation system will degrade significantly if nocountermeasuresare taken.

III. A C ONTROL THEORY APPROACH

The process of restricting the load on the wireless channel,and thereby the congestion in the wireless network, andthe process of adapting the communications parameters toguarantee a certain awareness level are very closely relatedto traditional control theory. In particular, due to the sharedwireless communication channel and the lack of a centralized

4

TrafficSituation Controller

inputparameter

Objective

targetdescription

Resultaction

optional feedback

1

2 3

4

Fig. 2. Analysis of congestion and awareness control algorithms from acontrol-theory-basedperspective. In general acontroller adapts the trans-mission parameters, based on itsobjectiveand the detectedtraffic situation,in order to achieve a particularresult. Optionally, the controller makes useof some sort offeedbackwith regard to the observed result to optimize itsperformance. Depending on the objective (1), e.g. a network-wide limitationof the channel load or anawareness-rangerequirement, the scope of the open-or closed-loop controller (2) is either global or local. Consequently, optionalfeedback (4) about the resultcancome from the network or from individualnodes only (3).

coordination entity in vehicular communications, both pro-cesses are representatives of the distributed control discipline.For instance, congestion in the network can not be avoided orreduced if only one single node is decreasing its transmissionpower and/or rate, and more importantly, the result of theselected action can not be observed by the node itself, but onlyby its neighbors. That implies that all nodes should — at least,for an optimal and reliable control — act cooperatively andprovide feedback about the result of their actions to each other.Similarly, the success of an increased transmission power withregard to a desired awareness range can only be determinedby the receiving node, and not by the transmitter itself.

Due to the relationshipof congestion and awareness controlto traditional control theory, this paper discusses existing pro-posals for congestion and awareness control with respect totheconcepts and notions typically used in control theory.For thispurpose, both methods are analyzed and compared accordingto the general framework sketched in Figure 2: an algorithmmight use some sort of detection to classify the traffic situationor scenario which the node is currently in, and which might beusedproactivelyby the controller as feedforward input. Thecontroller itself decides how the transmissionwill be adjusted,of course depending on the situation and the correspondingtarget description, i.e. the current objective. The selected actionthen leads to an observable result, which can be fed back tothe controller in order to improve its accuracy.

Based on Figure 2, particular control algorithms are alsoclassified intoopen-and closed-loop controllers. The formerdo not make use of feedback to correct and optimize thedecisions made in the past, and typically incorporate a systemmodel to derive the actions to be taken. The advantage of sucha control loop is thenonexistingoverhead, but, obviously, theperformance and robustness of such a controller depends onthe accuracy of thesystem model used. On the contrary, closedloop controllers employ feedback to determine howwellthe objective has been achieved. An often used closed loopcontroller is the generic Proportional-Integral-Derivative (PID)controller, which uses the present error (P), the accumulationof past errors (I), a prediction of future errors (D), or onlyasubset of those measures to control the system. Comparedwith

open loop controllers, closed-loop controllers can improve thecontrol due to the use of feedback data, at the cost of commu-nications overhead. In addition, and particularly if a genericPID controller is used, they do not incorporate any systemmodel, have no direct knowledge of the underlying processand perform poorly in non-linear systems. With respect tofeedback, it is also necessary to distinguish between explicitfeedback, i.e.,first-orderfeedback with regard to the desiredresult, and implicit feedback, i.e.,second-orderfeedback thatis obtained by using different observations that are to someextent correlated to the actual observation. A possible implicitfeedback could be, for instance, the locally observed numberof neighbors or MAC layer reception statistics.

Another important aspect is how the design objective canbe quantified and how the achievement of the objective ismeasured.When awareness control protocols are used, theobjective could be defined as the reliability with which acertain vehicle’s awareness range is guaranteed. However,forcongestion control, the question of how to describe the objec-tive is more difficult to answer, since more than one metricfor channel congestion exists, and all have their advantagesand disadvantages. For instance, theuseof a metric such asthe channel busy time ratio, i.e., the fraction of time duringwhich the channel is considered busy by the access layer, orthe channel load, i.e., the fraction of time during which thesensed energy exceeds a specific threshold,cannot accountfor overlapping transmissions, but has theadvantageof beingeasily implementable by the communication hardware. On theother hand, a metric such as the beaconing load [14],thoughit does quantify the amount of overlapping packets, is notdirectly measurable by the hardware.

IV. CONGESTIONCONTROL

Congestion control techniques for vehicular communica-tions can be classified according to several criteria. The majorclassificationcriterion considers the information basefromwhich congestion control mechanisms derive their decisionto adjust the transmission parameters. The first class, whichin the literature is sometimes also referred toas reactivecongestion control, usesfirst-orderinformation about the chan-nel congestion status to decide whether and how an actionshould be undertaken. Due to their nature, actions to lessenchannel load are undertaken onlyafter a congested situationhas been detected. Using control theory terminology, reactivecongestion control approaches can be classified as an instanceof feedback controlmechanisms. The second class, sometimesalso referred to asproactive congestion control, uses modelsthat, based on information such as number of nodes in thevicinity and data generation patterns, try toestimatetrans-mission parameters which will not lead to congested channelconditions, while at the same time providing the desiredapplication-level performance. In particular, such mechanismstypically employ a system model to estimate the channel loadunder a given set of transmission parameters, and make use ofoptimization algorithms to determine the maximum transmitpower and/or rate setting that will adhere to a maximumcongestion limit. Using control theory terminology, proactive

5

congestion control approaches can be classified as an instanceof feedforward controlmechanisms.

Another criterion used to classify congestion control tech-niques is what type of information is used to feed the controlsystem, whichis typically only locally availableinformation,or alsoinformation provided by neighboring vehicles– dubbeddistributed informationin the following.

Finally, existing solutions can further be classifiedwithreference tothe means through which congestion is controlled,which is typically achieved by adjusting thetransmissionpower, thepacket generation rate, thecarrier sense thresholdor a combinationof a subset of the transmission parameters.

Let us start briefly discussing the relative advantages anddisadvantages of proactive vs. reactive approaches. Giventheirability to preventcongestion, proactive approaches are veryappealing for vehicular environments, where radio commu-nications are primarily used for safety applications, whoseperformance would be seriously threatened by congested chan-nel conditions. However, proactive approaches come with twomajor drawbacks. First, in order to estimate the expected loadgenerated by neighboring vehicles, such approaches requirea communication model that maps individual transmissionpower levels to deterministic carrier sense ranges. However,this mapping is reasonable only as long as it reflects theaverage propagation conditions of the wireless channel. Thus,propagation conditions should be either dynamically estimatedas the vehicle moves, which is very difficult to do in apractical scenario, or they should be statistically estimatedto build specific profiles for different environments, e.g.,urban and highway. A second major drawback of proactiveapproaches is the needto carefully estimatethe amount ofgenerated application-layer traffic in a certain period of time.Although in some cases this is indeed possible (e.g, in thecase of applications built on top of periodic beacon exchange),accurate application-layer traffic estimation is a challengingtask in general.

Reactive approaches, which do not suffer of the drawbacksthat accompanyproactive mechanisms,nonethelesshave thenotable disadvantage of undertaking control actions onlyaftera congested channel condition has been detected. Consideringthat some time is needed to recover from a congested channelsituation, this means thatreactive approaches expose safety-related applications to the risk of not being able to fulfilltheir design goal, due to the poor (temporary) performanceof the underlying radio channel. Another disadvantage ofreactive approaches is that important design goals such asfairnessandpacket prioritizationare more difficult to achievethan in a proactive approach. We remark that fairness isimportant in vehicular networks to ensure that all vehiclesin the network have similar opportunitiesto communicatingwith nearby nodes. In fact, if congestion controlwere tobeobtained by sacrificing, say, a specific node in the network isforced to set its transmission power to a very low value, thisnode would not have a chance to communicate with nodesin its surrounding, impairing application-level performance.Most importantly, in safety-related applications, every vehiclein the network should be able toreceive fresh informationabout the status of the other vehicles in the surrounding, as

well as to communicateits own status to the surroundingvehicles. Hence, fairness becomes a major design goal insafety-related applications. As for prioritization, providing astrict prioritization of different classes of packets is anim-portant requirement for vehicular networking, which is partlyaddressed in the drafted IEEE 802.11p standard by adoptingthe EDCA (Enhanced Distributed Channel Access) mechanismdefined within IEEE 802.11e.

A. Related Work

Before describinga relevant congestion control approach,we briefly survey the most representative studies aimed at op-timizing the packet generation rate and transmission powerofbeaconing applications. While this body of work is not directlyconcerned with controlling congestion on the wireless channel,it has the merit of giving very useful insightsinto the effectsof varying rate and transmission power on beaconing perfor-mance. These insights can be considered as knowledge baseupon which state-of-the-art congestion control approaches arebuilt. Further, with respect to the literal interpretationofcongestion control in this paper, those approachesmight ratherbe termedcongestion reductiontechniques, which, on theirown, are of course able to reduce thenumberof messagestransmitted to the channel, but which are not actually able toeffectively avoid congested and overloaded channel conditions.

In [15], the authors present a performance evaluation studyof cooperative collision warning applications based on periodicbeaconing. The major contributionmade by this study isthe introduction of a novel parameter to measurethe perfor-mance of cooperative collision warning applications, namelythe packet inter-reception time. This metric, defined as thetime elapsed between two successful reception events at avehicle referring to beacons sent by another, specific vehicle,is motivated by the observation that what is relevant for activesafety applications is the freshness of the status informationgathered from surrounding vehicles. Thus, a few consecutivefailed receptions are much more harmful to active safetyapplications thanareseveral scattered failed receptions. Afterintroducing the novel metric, the authors go onto performanextensive simulation-based performance study on the effectsof using different beacon generation rates and transmissionpower values on the packet inter-reception time.

In [16], the authors investigate the effect of different bea-coning strategies on active safety application performance.More specifically, the authors considertracking accuracyasthe performance metric, which is defined as the error (asperceived by the active safety application) in tracking thepositions of neighboring vehicles. After having defined theperformance metric, the authors present different beaconingstrategies aimed at minimizing tracking error, and identify anadaptive beaconing policy with repetitions as the best perform-ing one. According to this policy, a beacon is sent only if thepredicted tracking error of the own position at surroundingvehicles exceeds a threshold.If the threshold is exceeded, thebeacon is sent a few times (the number of repetitions is atunable parameter) to increase the probability that the beaconis correctly received by neighboring vehicles, thus improving

6

tracking accuracy. As mentioned in the beginning of thissection, such a mechanism will help to reduce the congestion,but it does not control the congestion of the channel in thefirst place.

Another example of congestion reduction can be found in[17]. The authors focus on emergency warning messages, thatare sent whenever a vehicle shows an abnormal behavior (e.g.,it broke down and is blocking the road/lane, or it lost controland it is changing lanes unexpectedly). The authors aim ofoptimizing the transmission of warning messages is based onthe observation that messages should be repeatedly sent outuntil the “abnormal” behavior stops and the vehicle returnsto“normal” behavior. The authors further state that if several ab-normal vehicles are sending out emergency warning messagesat a constant rate, the average delivery delay will increaserapidly due to channel congestion. Consequently, the numberof simultaneous emergency warning transmissions should becarefully controlled. To achieve this goal, the authors proposea “multiplicative rate decreasing algorithm”, which decreasesthe retransmission rate of an emergency warning messageovertime. As a result, several emergency warning messages can beserved and delivered by the system with limited delay. Theabove transmission strategy is further optimized by definingstrategies to freezethe generation of emergency messageswhen certain conditions are met (e.g., redundant transmissionsfrom following vehicles).

According to the terminology defined in Section III, theapproach of [17] belongs to the class of proactive approaches,and acts on packet generation rate to prevent congestion. Yet,the approach of [17] is mostly an open-loop controller, sincethe multiplicative rate decreasing algorithm that is used totune the packet generation rate is based only on predictedperformance based on suitable models of the communicationchannel. On the other hand, a form of primary feedback (e.g.,reception of redundant transmissions from following vehicles)is used in the decision rules to freeze emergency messagetransmission.

Apart from the cited studies above, other congestion reduc-tion solutions that adapt the transmission power and generationrate based on the current velocity exist as well (e.g. [18],[19]). Since the paper focuses on actual congestion controltechniques, wewill skip their detailed presentation here andinstead surveya collection of representative congestion controlapproaches for cooperative vehicular systems. One of theseapproaches,called Distributed Fair Power Adjustment forVehicular Environments or D-FPAV[20], will be describedin detail in the next section.

On the reactive side of congestion control, Khorakhun et. aldeveloped an algorithm that adjusts either the transmissionpower or the packet generation ratewith relation tothe locallymeasured channel busy time ratio [21]. The channel busy timeis the fraction of time during which the channel was sensedbusy. Depending on whether the local measurement is belowor above a pre-defined threshold, the transmission power orgeneration rate is either increased or decreased by one step.In order to achieve a higher level of fairness, the authors statedthat it is necessary to exchange the local measurements amongneighboring vehicles, and allow an increase of the transmission

power/rate only if the currently used value is below the averagepower/rate configuration used by thevehicle’s neighbors.Comparedwith proactive approaches, this reactive approachis not able to avoid congestion on the wireless channel, andsupports no prioritization of different classes of messages. Inaddition, a simple analysis shows that the proposed algorithmis not able to prevent oscillations in the adjustment process.The issue is systematic and fundamental: since not all vehiclesperform the transmit power adjustment at the same point intime, it can easily happen that the transmit power reductionata few nodes leads to a reduced channel busy time observationfrom the perspective of neighboring nodes that have yet notreduced their transmit power. As a result, those nodes willpossibly increase their transmit power (instead of decreasingit as well), and amplify the transmit power reduction of nodesthat have already decreased their transmit power. It is obviousthat some sort of additional feedback is needed to indicate thereason why the measured channel busy time has decreased orto determine who should reduce first.

A hybrid approach that attemptsto combinethe advantagesof both proactive and reactive approacheswas proposed byBaldessari et. al in [22]. Their solution consists of an improvedrate control, an improved power control and a combinedpower and rate control algorithm,all of which use channelbusy time observations to derive the number of neighbors inthe surroundingarea(optionally, also through an additionalexchange of local vehicle density estimations). Based on thenumber of neighbors and a pre-defined channel busy timethreshold, the authors theneither derivea packet generationrate directly, or start with a fixed packet generation rate andderive the maximum transmission power which will not violatethe threshold. In the latter case, the authors assume that thevehicles in the surroundingareaare distributed uniformly and,typical for a proactive approach, make use of a communicationmodel that maps carrier sense ranges to individual transmis-sion power levels.

Another hybrid congestion control approachwas recentlyproposed in [23], where the authors adaptively change bothbeacon generation rate (in a proactive way) and transmissionpower (in a reactive way) with the goal of reducing channelcongestion, and consequently improving a vehicle’s ability toaccurately track the position of surrounding vehicles. Twoslightly different control approaches are appliedto the tuningof beacon generation rate and transmission power. Beacongeneration rate is tuned based on a predicted tracking er-ror of own position. The prediction accounts for channelunreliability, i.e. packet losses, by including the observedfraction of successfully received beacons sent by surroundingvehicles. Thus, a closed-loop feedforward controller based onsecondary feedback is used for setting the beacon generationrate.Additionally, transmission power control is applied basedon the observed channel status (more specifically, based on thechannel busy time). This part of the algorithm isthusa closed-loop feedback controller based on secondary feedback. Notethat both beacon generation rate and transmission power useinformation locally available at the vehicles (i.e., direct obser-vations) to control transmission parameters. As a consequence,this mechanism bears the same fundamental issue observed for

7

[21]: without knowing the channel congestion status of thesurrounding nodes, the transmission power adaptation mech-anism cannot know why the channel isno longercongestedand which vehicleshould reduce or increase its power valuefirst.

B. Contribution

In this section, we present the Distributed-Fair Power Ad-justment for Vehicular Environments (D-FPAV) approach toproactive, distributed congestion control in vehicular envi-ronments. D-FPAV achieves congestion control by varyingthe node transmission power, where a node’s transmit powersetting depends on predictions of application-layer traffic andthe observed number of vehicles in the surrounding.

D-FPAV is designed to pursue all the optimization goalsdescribed in the previous sub-section:

1. congestion control: limit the load on the wireless mediumin order to prevent congestion generated by application-layer traffic. The metric used to assesstheeffectiveness ofcongestion control is channel access time. Furthermore,the authors showthe benefits of congestion control onthe performance of multi-hop emergency message prop-agation.

2. fairness: maximize the minimum transmit power valueover all transmission power levels assigned to the nodesforming the vehicular network, subject to Goal 1.

3. prioritization: improve the basic IEEE 802.11 EDCAmechanism to provide a better prioritization of higherpriority over lower priority messages.

We observe that the D-FPAV approach is aimed at limitingthe amount of traffic that is generated by vehicles, with thegoal of keeping this load under a specified congestion limit(the MAL value defined in the following). By tuning thecongestion limit, the optimal point of the “interference level”vs. “reception rate” tradeoff can be found. The optimal tuningof this tradeoff is however out of D-FPAV scope.

The D-FPAV protocol is periodically executed on the nodesforming the vehicular network, in order to adjust node trans-mission power in response to changes in the network topologyor application-layer traffic patterns. Before presenting D-FPAV,we introduce some notation and basic definitions. We denoteby N = {u1, . . . , un} the set of nodes in the vehicularnetwork. Each of these nodes can set its transmission powerin the interval [Pm, PM ], where Pm is the minimum andPM is the maximum possible transmission power. Given aset N of nodes as above, apower assignmentfunction PAfor N is a function that assigns to every nodeui ∈ N avalue PA(i) ∈ [0, 1]. The power used by nodeui to sendapplication-layer messages ispi = Pm +PA(i) · (PM −Pm).

For any nodeui in the network, we useCS(PA, i) to denotethe carrier sensing range of nodeui at transmission powerpi = Pm+PA(i) ·PM , andCS(MAX, i) to denote the samerange at maximum transmission powerPM .

A fundamental notion in D-FPAV is that of Application-layer Load (AL) generated by a node, and of Channel Load(CL) experienced by a node under a certain power assignment

Algorithm D-FPAV: (algorithm for nodeui)INPUT: AL of all the nodes inCS(MAX, i)OUTPUT: a power settingPA(i) for nodeui, such that

the resulting power assignment is an optimalsolution to CCF

I(MAX, i) denotes the set of nodesuj such thatui ∈ CS(MAX, j)

1. Based on the AL of the nodes inCS(MAX, i),compute the maximum common tx power levelPi s.t. the MAL threshold is not violated at anynode inCS(MAX, i)

2a.DisseminatePi to all nodes inCS(MAX, i)2b. Collect the power level values computed by

nodesuj ∈ I(MAX, i); store the received values inPj

3. Compute the final power level:PA(i) = min

{

Pi,minuj∈I(MAX,i){Pj}}

Fig. 3. The D-FPAV algorithm.

PA. Formally,AL(i) denotes the application-layer load (ex-pressed inbytes/sec) that nodeui is expected to generatein the nextperiod, where the period represents the intervalof time before the next D-FPAV execution. Based on thesedefinitions, the CL experienced by a nodeui can then becomputed based on the AL generated by the nodes in thesurrounding as follows:

CL(PA, i) =∑

uj∈I(PA,i)

AL(j) ,

whereI(PA, i) = {uj ∈ N : ui ∈ CS(PA, j)}.The intuition behind our definition of channel load is that

the load observed atui can be estimated as the sum of theapplication-layer load generated by nodes in the setI(PA, i),i.e., those nodes havingui within carrier sensing range at thecurrent transmit power levels. This is because a transmissionfrom a node in setI(PA, i) preventsui from accessing thechannel.

The Congestion Control under Fairness constraints (CCF)problem we are attempting to solve is defined as follows:

Definition 1 (CCF problem):Given a set N ={u1, . . . , un} of nodes, and a valueMAL of the MaximumApplication-layer Load admitted on the wireless channel(expressed inbytes/sec), solve the following optimizationproblem

maxPA∈PA (minui∈N PA(i))subject to

CL(PA, i) ≤ MAL ∀i ∈ {1, . . . , n},

wherePA is the set of all possible power assignments.Solving CCF addresses design goals1 and2 above, where

MAL (a choice of the network designer) is used to control thecongestion generated by application-layer load. As we showin the following, goal3. can be achieved by transmitting low-priority messages using the transmit power computed solvingCCF, and by transmitting high-priority messages at maximumpower.

The D-FPAV algorithm is reported in Figure 3, and iscomposed of the following steps:i) gather information aboutAL for nodes within (maximum) carrier sense range;ii)

8

based oni), locally execute the FPAV algorithm from [14]to compute the optimal CCF solution for the nodes within(maximum) carrier sense range;iii) exchange the locally com-puted transmission power values with surrounding vehicles;iv) select the minimum transmission power value among theone locally computed and those computed by surroundingvehicles in order to build the network-wide optimal solutionto CCF.

The FPAV algorithm of [14] is a centralized algorithm basedon the well-known “max-min” principle: the transmissionpower of all vehicles in a surroundingarea is ‘virtually’increased step-by-step (starting at lowest possible powerlevel),while estimating the resulting application load at each vehicleafter each step. As long as the MAL threshold is not violatedat any vehicle, and the maximum allowed transmission powerhas not been reached, power levels are further increased. Upontermination, FPAV has thus computed the highest commontransmission power level which did not violate the MALparameter in the whole network.

In [20], it is formally proven that D-FPAV computes anoptimal solution to CCF under the following assumptions:a)carrier sense ranges of nodes are symmetric;b) each node isable to accurately estimate the AL for the next period; andc) each node is able to gather AL information fromall nodeswithin maximum carrier sense range. In practical scenarios,these assumptions are unlikely to hold, due tothe complexityof the propagation environment (a), and difficulties in accu-rately predicting AL and gathering AL information (b) and(c). Yet, in [20] it is shown through extensive simulation thatD-FPAV successfully solves the CCF problem at least whenassumptionc) is released, i.e., when nodes have only partialknowledge of the AL generated by nodes within maximumCS range.

In [20], D-FPAV is evaluated in a scenario1 in whichapplication-layer load at each node is generated by a beaconingapplication, which periodically generates packets to reportvehicle status to nodes in the surroundingarea. Beacon mes-sages are considered low-priority messages in this scenario,and transmitted using D-FPAV computed transmission powerwith lowest EDCA priority class. Besides beaconing messages,event-driven emergency messages are randomly generatedwithin the network. These are high-priority messages that areseldom generated, and, given their safety-critical nature, arenot subject to congestion control. Emergency messages aresent at maximum transmission powerPM using the highestpriority EDCA traffic class.

An important issue to understand in the D-FPAV approachis the tradeoff between accuracy of channel load estimationona vehicle, and additional overhead which is put on the channel.In fact, as the carrier sense rangeis typically larger than thetransmission range, the onlyway to acquire knowledge aboutpresence of vehicles located outside the transmission range isby making use of a multi-hop strategy, i.e., having vehicles re-transmit the position of their neighbors. Clearly, propagatingthis information in a multi-hop manner puts an additional load

1For details on the simulation scenario, including features of the radioenvironment, please see [20].

0

0.2

0.4

0.6

0.8

1

100 200 300 400 500 600 700 800 900 1000

Pro

babi

lity

Of M

essa

ge R

ecep

tion

Distance to Sender [m]

Event-driven D-FPAV-OnEvent-driven D-FPAV-Off

Beacon D-FPAV-OnBeacon D-FPAV-Off

Fig. 4. Probability of successful reception of periodic beacon and emergencymessages at varying distances. MAL is set to2.5Mb/s.

on the channel, which can be considered as control overhead.In order to optimally tune the above described tradeoff, the

following design decisions have to be made: how often thestatus of neighboring vehiclesshould be forwarded, what rangeof neighbors must be included, and which transmission powermust be used to transmit this information.

The following strategies have been considered in [20]:piggyback the aggregated status information (position of sur-rounding vehicles) to 1) each beacon, 2) every 5th beacon or3) every 10th beacon, and transmit it with powerPA(i) (thetransmit power value as computed by D-FPAV). The authorsfound that piggybacking aggregate status information in 1 outof 10 beacon messages results in the best compromise betweencontrol overhead and effectiveness of congestion control.

The probability of correctly receiving a beacon or emer-gency message as a function of distance with and withoutD-FPAV is reported in Figure 4. As seen from the figure,EDCA alone is not sufficient to clearly prioritize emergencyover beacon messages. On the other hand, D-FPAV achievesa clear prioritization of emergency over beacon messages,i.e. emergency messageshave a consistently higher receptionprobability in the whole range of transmission distances. Itis also interesting to observe that D-FPAV congestion controlmechanism has beneficial effects not only on high prioritytraffic, but also on low-priority traffic (beacons): in fact,theirreception probability at close distances from the transmitter(within 150m) is considerably higher thanwhenno congestioncontrol exists.

The effectiveness of the D-FPAV approach in achievingfair channel access opportunities is shown in Figure 5, whichreports the channel access time of vehicles as a function oftheir position on the road: without D-FPAV, channel accesstime dependshighly on the density of vehicles in the surround-ings, and it is thus unfairly distributed. On theother hand,when D-FPAV is active, the load generated by the beaconingapplication is kept under control, and channel access time isnearly constant throughout the network.

In a follow-up study [24], we showed that in order toeffectively guarantee a strict enforcement of an upper channelload limit and to provide fairness with respect to channelaccess opportunities, it isnecessaryto propagatethe position

9

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

1000 2000 3000 4000 5000 6000 7000

Cha

nnel

Acc

ess

Tim

e [s

]

X-Position on Highway

D-FPAV OffD-FPAV On

Fig. 5. Average channel access time experienced by periodic beaconmessages as a function of vehicle position on the highway. MALis set to2.5Mb/s.

of neighbor vehicles for at least two hops. As described above,the D-FPAV protocol provides this information by piggyback-ing this information only in 1 out of 10 beacon messages inorder to reduce the overhead, yet, the overhead can still growto 40 % – compared to the actual AL data. In [24], we thereforedeveloped a distributed algorithm that adjusts the transmissionpower based on averaged values for the neighbor informationinstead of using detailed neighbor information about eachsingle node. By using only this averaged information, we wereable to reduce the overhead down to less than 1 %, at the costof only slighty exceeding the pre-defined MAL limit.

C. Standardization

In Europe, congestion control in vehicular communicationsis considered to be a building block and seen as mandatoryin order to guarantee a reliable communication performancefor safety-related applications. Thiswasacknowledged by theCar-2-Car Communication Consortium (C2C-CC)as early asin 2009 through the establishment of a task force on transmitpower control. The results from this task force have helpedto persuadethe European Commission to establish a Spe-cialist Task Force (STF) on the configuration and validationof decentralued congestion control methods for IntelligentTransportation Systems (ITS) and techniques to enable the co-existence of cooperative ITS and Dedicated Short Range Com-munication (DSRC) within the European TelecommunicationsStandard Institute (ETSI). Since March 2010, the STF withits six experts from the industry and research communityhavebeenworking on a technical specification for a standardizedcongestion control algorithm to be used by ITS.

The specification defines a mandatory basic congestioncontrol algorithm based on a controller that uses no feedbackand only information that is locally available. In addition, thespecification describes an enhanced control algorithm thatusesfeedback from neighboring nodes, e.g. their observed channelload and their currently used transmit power. Both approacheswill use the channel load metric to define the congestion limit,since it can be implemented by the hardware.

V. AWARENESS CONTROL

Following the previous discussions, awareness control tech-niques are aimed at ensuring each vehicle’s capacity to detectas well as to communicatewith the relevant vehicles in theirlocal neighbourhood. Awareness control protocols are neededto reliably and efficiently support higher-layer protocolsandapplications, for example, ensuring that traffic safety appli-cations obtain, at a minimum, the level of awareness that isrequired to detect dangerous traffic situations in advance andact accordingly.

Cooperative vehicular systems impose very stringent ap-plication requirements, with most of the applications beingsupported by the periodic exchange of beacons. It is generallyassumed in the literature that the cooperative applicationrequirements can be defined in terms of dissemination area (orrange), latency (or delay) and reliability [25]. This extends thepreliminary requirements defined by VSC [26] and ETSI [27],which consider onlysome of these requirements. While thedissemination area can be defined as the geographic areawhere a given message should be received, the latency isthe maximum allowed delay to deliver such message, andthe reliability is the minimum probability of receiving such amessage (usually estimated during a certain time window). Therequirements imposed by cooperative applications representthe basis of awareness control protocols.

A. Related Work

Initial studies have been conducted to evaluate the com-munications and applications performance in VANETs, andserve as fundamental studies for the design of awarenesscontrol protocols. For example, the work in [28] presents aperformance and sensitivity analysis of different MAC layerprotocols, based on the idea of repetitive transmissions overCSMA. In this work, the authors propose that a packet beretransmitted multiple times during its lifetime, and studydifferent repetition mechanisms by means of simulation anda detailed mathematical anlalysis. The conducted study iden-tifies the operating and communications conditions (numberof interferers and packet generation rates) under which theapplication requirements are satisfied and the channel loadismaintained undera certain limit. In [15], the authors conducta performance evaluation study of cooperative collision warn-ing applications under different traffic densities, and exploredifferent packet generation rates and transmission ranges.The conducted study shows the importance of consideringappropriate metrics to evaluate the performance of cooperativeapplications, such as latency orpacket inter-reception time.This observation results from the importance of the freshnessof the information received from surrounding vehicles. Therelevance of adequately evaluating the performance of coop-erative vehicular systems has also been emphasized in otherstudies. For example, the work in [29] highlights the needtodifferentiatecommunication and application performance orreliability. In particular, in [29] the authors demonstrate thesuitability of cooperative vehicular systems to improve trafficsafety based on real-world experimental dataon highways.

10

Following the control-theory perspective discussed in Sec-tion III, awareness control protocols can be classified asopen-loop or closed-loop approaches, and can use implicitor explicit feedback. Different existing open-loop awarenesscontrol protocols use power-range maps to dynamically adapteach vehicle’s transmission power as a function of its trans-mission range requirements. For example, the work in [30]proposes an OPportunistic-driven adaptive RAdio resourceManagement (OPRAM) mechanism, that adapts each vehicle’stransmission parameters to reliably and efficiently exchange amessage before reaching a critical safety area, for exampleanintersection. The OPRAM mechanism is an application-drivenawareness control protocol that is based on radio propagationestimates to dynamically calculate the required transmissionpower levels as a function of the distance to the critical safetyarea. Despitehaving beendesigned as a power-range mapbased technique, OPRAM could be evolved to dynamicallyadapt the transmission power and packet generation rate tothe experienced channel load, in order to compensateforthe negative effects of packet collisions on the application’sreliability. OPRAM will be described in detail in the nextsection.

Multi-hop beaconing protocols represent an alternativeopen-loop solution for awareness control. With these proto-cols, broadcast messages transmitted by a vehicle are relayedby neighboring vehicles to achieve the target probability ofreception at high distances within the required delay. In [31],the Multi-Hop Vehicular Broadcast (MHVB) protocol isin-tended to efficiently relay broadcast packets over multiplehops, and satisfy the target dissemination area within theallowable latency. With MHVB, only the vehicle that correctlyreceives a given broadcast message and is located at thehighest distance from the transmitter will relay such message.A similar approach was proposed in [32], where a vehicle canrelay multiple beacons duringits lifetime in a single packetas long as it hasreceivedeach one of these beacons lessthan a established maximum number of times. With multi-hop beaconing protocols, the beacon’s transmission powerwas able tobe reduced compared to single hop beaconingprotocols. In this context, the work in [33] compares single-hop and multi-hop beaconing protocols. This study showsthat under simplified propagation and multi-hop operatingconditions, the channel load in multi-hop beaconing protocolscan be reduced using packet multiplexing techniques. Withthese techniques, when a vehicle has to relay a broadcastpacket it will attach its own broadcast packet to the relayedmessage. However, in realistic environments where packetscan be lost due to radio channel errors and packet collisions,the reduction of the channel load obtained with multi-hopbeaconing cannot be achieved. It is worth mentioning that,in scenarioswhere obstaclesblock the radio signal, such asbuildings or trucks, multi-hop beaconing protocols could berequired to successfully satisfy the application requirements.

Closed-loop awareness control protocols make use of ex-changed broadcast messages to dynamically adapt to the vary-ing propagation and channel load conditions. In some existingsolutions, the feedback isimplicitly obtained from receivedmessages without the needto transmitextra information. An

example of closed-loop solution with implicit feedback is thework in [34], in which the authors propose the use of thereceived messages to estimate in real-time thepath lossoraverage signal attenuation, by subtracting the power estimatedat the receiver from the transmitted power (included by defaultin the header of all beacons in cooperative vehicular systems).The proposed algorithm dynamically selects the power anddata raterequired to successfully transmit a packet to a givenvehicle, while minimising the interference generated to othervehicles. In the joint rate-power control algorithm proposedin [23], two different control approaches are applied to adaptthe packet generation rate and the transmission power. In orderto detect abnormal drivingmaneuvers in advance, the packetgeneration rate is dynamically adapted to bound the longitudi-nal and lateral position tracking errors of surrounding vehicles.A new packet is transmitted when the estimated trackingerror of surrounding vehicles exceedsa certain threshold. Toestimate the tracking error, the algorithm takes into accountthe channel reliability by dynamically estimating the packeterror probability from the packets received from surroundingvehicles. The transmission power is adapted based on theobserved channel status. ConsideringLmin and Lmax asthe lower/upper transmission range bounds dictated by thesafety applications, the transmission range is linearly adaptedbetweenLmin and Lmax as a function of the experiencedchannel load. The transmission power is then calculated con-sidering power-range maps based on empirical measurements.

Other closed-loop awareness control solutions are based onexplicit feedback. In this case, vehicles inform each otherabout the correct reception of messages at a certain distance.Using this explicit information, vehicles can decide whetherthey should increase or decrease their transmission powerand/or packet generation rate. For example, the work in [35]includes the target range of the packet, and the IDs of thenodes from which a message was successfully receivedwhentheir separation distance was larger than the target range,asextra information in the beacon’s header. When a vehiclereceives more thanN beacons containing its ID, it decreasesits transmission power, since at leastN vehicles beyond thetarget range are receiving its broadcast messages. This type ofclosed-loop technique depends heavilyon the correct receptionof a beaconto adequately adapt its operational parameters.As a result, if some of these messages are not correctlyreceived, for example due to packet collisions under hightraffic densities, the vehicles can incorrectly increase theirtransmission power and augment packet collisions.

The accuracy or precision of the awareness informationreceived from surrounding nodes is also analysed in differentprotocols proposed in the literature. A representative exampleis reported in [16], where the authors propose an open-loop approachthrough which the packet generation rate isdynamically adapted to bound the tracking errors of surround-ing vehicles. To thisend, a new packet is transmitted by avehicle only when its movement changes (speed, heading,etc.). In this case, the authors propose that each of thesepacketsbe retransmittedseveral times to ensure its receptionby surrounding vehicles, which results in a decrease of thetracking error in realistic propagation environments, at the

11

expense of increasing the channel load. In fact, as shownin [36], the increase of the packet generation rate can augmentthe probability of successfully receiving a packet at the targetdistance within the required time window, as long as thechannel load is maintainedat reasonable levels. In particular,the work reported in [36] proposes the rapidrebroadcastingof each packet during its lifetime to increase its probabilityof reception, extending the work presented in [28]. To thisaim, variousrebroadcastingschemes are proposed based ondifferent open-loop strategies: synchronous and asynchronousdesigns, repetition with and without carrier sensing, fixednumber, andp-persistent repetition. A different perspective isreported in [37]with regardto the accuracy of the awarenessinformation required by each vehicle. The work in [37] con-siders that the packet generation rate should depend on eachvehicle’s mobility characteristics, and those ofthe vehiclessurrounding it, as well as the traffic context/situation. Inparticular, the authors propose the use of situation-adaptivebeaconing to achieve adequate levels of accuracy or updatedawareness information received from neighboring vehicles.

B. Contribution

The consideration of application requirements in the de-sign of awareness control protocols for cooperative vehicularsystems is particularly important due to the critical nature ofsafety applications. Abasicapplication-driven awareness con-trol approach is summarized in Figure 6. Its operation is basedon each vehicle’s application requirements. As an example,inthe case of an intersection collision warning application,therequirements would correspond to the minimum distance to theintersection at which two potentially colliding vehicles needto exchange a message to alert the driver with sufficient timeto avoid the accident. Since such requirementsdepend heavilyon the vehicles’ position, speed and acceleration, each vehiclewould need to continuously adapt its application requirementsbased on its positioning and movement information. Oncethe application requirements have been updated, each vehiclewill accordingly modify its communications parameters (e.g.transmission power and packet generation rate) to satisfy themwith certain reliability imposed by the application (papp).The adaptation of the communications parameters could bebased on some of the protocols and algorithms described inthe previous section. While thisbasic approach could satisfythe required vehicle’s awareness level, the following sectionsdescribe two application-driven awareness control approachesaimed at further improving the communications efficiencythrough the use of geographic and traffic context information.

1) Geo-Opportunistic approach:OPRAM [30] is an ex-ample of an application-driven awareness control techniqueaimed at efficiently adapting each vehicle’s communicationsparameters (transmission power and packet generation rate) toguarantee the transmission range and reliability requirementsimposed by traffic applications. To further improve the effi-ciency of thebasicawareness control approach previously dis-cussed, OPRAM proposes a geo-opportunistic approach thatmakes use of the geographic positioning and the knowledgeof potentially critical safety areas. To illustrate its operation

Basicapplication-driven awareness control approachINPUT: Established application reliabilitypappOUTPUT: A power and rate setting for vehiclei, such that

the resulting assignment satisfies the applicationrequirements

1. Update positioning and movement information.2. Adapt application requirements.3. Obtain required power and rate based onpapp

and application requirements.

Fig. 6. Basicapplication-driven awareness control approach.

and benefits, an intersection collision warning application isconsidered here. However, OPRAM’s operation could be eas-illy adapted to other applications such as cooperative mergingassistance and left turn assistance applications.

Intersection collision warning applications in urban inter-sections represent one of the most challenging scenarios forawareness control protocols due to the strict traffic safetyapplication requirements and the challengingnon-line-of-sightpropagation conditions. In a typical intersection scenario, twoapproaching vehicles A and B might collide at the intersectiondue to the driver’s lack of attention, wrong/hidden trafficsignals, or any other reason that could provoke the accidentdespite the driver’s ability and perception capabilities (seeFigure 7). To detect each other’s presence, the two vehiclesperiodically broadcast beacon messages. The intersectioncol-lision warning application requires that both vehicles exchangeat least one packet before the critical distance (CD). TheCD distance is the minimum distance to the intersection atwhich a vehicle needs to receive a broadcast message froma the potentially colliding vehicle to alert the driver of apotential road danger with sufficient time to react, and stopbefore reaching the intersection. TheCD distance typicallydepends on the vehicle’s speed, the driver’s reaction time andthe vehicle’s emergency deceleration. The presence of build-ings may require the use of high transmission power levelsand/or packet generation rates to guarantee the communicationbetween the two vehicles before the target distanceCD, andhence avoid the potential accident. However, the constant useof high transmission powers and rates by multiple vehiclescould create channel congestion, and increase the system’sinstability. To reduce the channel load while satisfying theapplication requirements, OPRAM is designed to dynamicallyincrease the transmission power and packet rate of each vehicleonly in a small region beforeCD, calledtheAlgorithm Region(AR), as illustrated in Figure 7. With this geo-opportunisticincrease, OPRAM aims to guarantee with high probability thecorrect reception of at least one packet from a potentiallycolliding vehicle before reachingCD, while minimizing theoverall channel load. OutsideAR, OPRAM operates witha low transmission power, sufficient to communicate withthe vehicles moving along the same street in line-of-sightpropagation conditions. The definition of theAR region allowsOPRAM to adapt its communication parametersonly whenapproaching a critical safety area, such as an intersection, lanemerging zones, entrance ramps,blind curves, etc.

12

��

��

��

��

��

��

CDPosition of B

Position

of A

Algorithm region (AR)

Transmission power

Probability of reception p e

NT messages

A

B

Fig. 7. OPRAM operation in intersection scenarios

To define the operation of the OPRAM mechanism, weconsider that each vehicle transmitsNT broadcast packets inAR. Avoiding the intersection collision requires that at leastone of these packets is correctly exchanged with high probabil-ity beforeCD. Considering the challenging and probabilisticradio propagation conditions, OPRAM has been configuredto successfully receive at least one of these messages from apotentially colliding vehicle before reachingCD in 99% ofthe cases (pappA = pappB = 0.99). An intersection collisioncould be avoided if at least one of the two vehicles receives abroadcast message with sufficient time to react. If we assumethat the success of transmission from vehicle A to vehicle Bis independent of the successof transmissionfrom B to A,the overall application’s reliability could be thenpapp = 1 −(1− pappA)(1− pappB) = 0.9999. However, it is important tonote that such full independence is difficult to achieve despitethe potentially different interference conditions experiencedby each vehicle, and the different obstacles present in theirrespective local environments. To reach the target reliability,OPRAM initially considers that the probability that a singlepacket will be successfully received by the potentially col-liding vehiclepe is constant and independent inAR. HavingdefinedNT and each vehicle’s application reliability, suchpeprobability can be calculated through a Binomial distributionconstructed byNT Bernoully experiments [30].

To dynamically calculate the required transmission powerlevel for each of theNT packets transmitted inAR, the initialOPRAM implementation considers an open-loop approach,based on propagation models obtained through empiricalradio channel measurements conducted under the EuropeanWINNER project [38]. In particular, OPRAM computes thetransmission power for each of theNT packets based onthe current distance to the intersection, and thepath loss,shadowing and multipath fading propagation effects2. Thetransmission power is then selected so that each of theNT

packets transmitted withinAR is received with a probabilitype. This will ensure that at least one of theNT transmittedpackets will be correctly received with probabilitypappA (or

2For details on the calculation methodology, plaese see [30].

pappB). It is interesting to note that the increase ofNT reducesthe requiredpe and the consequent required transmissionpower levels for a given application reliability.

Under realistic operating conditions, the probability ofpacket reception from a potentially colliding vehicle woulddepend not onlyon the radio channel propagation effects, butalso on the channel load, and consequent packet collisions.Packet collisions reduce the probability of packet receptionpe, and hence decrease the application’s reliability. The workin [39] proposed two different packet collisions compensationtechniques. These techniques are based on the evaluation ofthe experienced channel load, and the consequent adaptationof each vehicle’s transmission power or packet generation ratein AR to combat the negative effect of packet collisions onthe OPRAM performance. With these compensaiton policies,OPRAM couldtherefore beextended to follow a closed-loopapproach based on the feedback received from neighbouringvehicles.

The OPRAM technique was initially designed consideringthat theNT transmitted messages are received independently.However, such independence cannot be guaranteed undercorrelated radio channel conditions. Although such correlationeffects can be simplified for certain system level investigations,their impact on the instantaneous performance of cooperativevehicular systems cannot be neglected, in particular for criticaltraffic safety applications. In this context, the work in [40]proposes and evaluates various compensation policies thatcan efficiently overcome the negative communication effectscaused by the radio channel correlation.

For the scenario illustrated in Figure 7, Figure 8 depictsthe cumulative distribution function (CDF) of the distancetothe intersection at which the first message from a potentiallycolliding vehicle is received, considering the use of OPRAMand the constant transmission power level (Pt=2W) neededto ensure the same application’s reliability under the sameoperating conditions. As it can be observed, using OPRAM3 ora 2W fixed transmission power can guarantee that at least onemessage is exchanged beforeCD with the target reliability.However, as demonstrated in [41], the application of OPRAMat a system scale results in a more efficient use of the radiochannel, since it is able to considerably reduce the channelload (by nearly70%), while guaranteeing the awareness levelneeded to ensure the same application reliabilityas withfixedtransmission power policies.

2) Traffic contextual approach:The previous subsectionillustrated the benefits of enhancingbasic application-drivenawareness control approaches through the use of geographicalinformation. This subsection is aimed at demonstrating thatthe operation ofbasic awareness control techniques can alsobe improved through the use of traffic context information.To this end, a lane change assistance application in highwayscenarioswasconsidered. This application informs the driverabout whether a potential lane changemaneuvercan beperformed in a safe way or not based on the proximity ofother vehicles. Such proximity can be detected through thereception of broadcast messages transmitted by neighboring

3The same OPRAM performance is achieved for differentNT values.

13

0 25 50 75 1000

0.2

0.4

0.6

0.8

1

Distance to the intersection [m]

CD

F

OPRAM, N

T=10

Fixed Pt=2W

CD

AR

Probabilityof notreceiving apacketbefore CD

Fig. 8. CDF of the distance to the intersection at which the first messagefrom a potentially colliding vehicle is received

vehicles. Following the illustration in Figure 9a, vehicleBwould consider its lane change unsafe if another vehicle Awasapproaching on the left lane and theywerecloser than a certaindistanceDw (Warning Distance).Dw represents the minimumseparation distance between the two vehicles allowing vehicleB to change lane without making vehicle A reduce its speed,and can be computed as:

Dw = −1

2

(vB − vA)2

aA − aB+ L+Ds (1)

wherevA andvB represent the vehicles speed inm/s, aAand aB their acceleration inm/s2, L is the vehicle lengthin m, andDs is the safety distance. It is important to notethat neither of the two vehicles knows the speed of the othervehicle before receiving its first message. Consequently, theyneed to assume the worst case scenario in terms of speed tocalculate their respectiveDw. This corresponds to vehicle AcalculatingDw considering that vehicle B is moving at theminimum speed allowed on the road, and it has the lowestpossible acceleration in the overtakingmaneuver(vB = vmin,aB = amin = 1m/s2). Vehicle B will consider that vehicleA is moving at the maximum constant speed allowed on theroad (vA = vmax, aA = 0m/s2). This results in differentDw

distances for vehicles with different driving context situations.It is interesting to note that vehicles with a speed outside the(vmax, vmin) limits could be configured to transmit with ahigher transmission power to warn surrounding vehicles withenough time for the driver to avoid dangerous situations. Sinceonly a few vehicles would be driving with a speed outside thelimits, this results in a more efficient use of the radio channelthan considering all vehicles calculatingDw based on speedshigher thanvmax or lower thanvmin.

Based on the proposed application and previous definitions,Dw is the minimum distance at which vehicles A and B wouldneed to communicate to avoid a dangerous situation. As aresult,Dw represents the application requirement according onthe basic application-driven awareness control approach pre-viously described. Following thisbasicapproach, each vehicleautonomously adaptsDw based on its own vehicular speed.

��

CB

��

CB(a)

(b)

DwA DWindowA

A

Df

A D

DwBDWindowB

Fig. 9. Lane change assistance application. (a) Scenario. (b) Traffic contextualinformation.

The transmission power is accordingly adapted to satisfy thetarget reliability following the OPRAM transmission powerestimation methodology. In this case, the application reliabilityhas been defined as the probability of receiving at least onebroadcast message beforeDw and during a given time windowTWindow (see Figure 9a, where theTWindow is mappedto theDWindow distance following the vehicle’s speed). Tocombat the negative effects of packet collisions and radio chan-nel correlation, the compensation policies proposed in [39]and [40] could be considered. However, reduced correlationlevels have been observed in highway scenarios [42], and thechannel correlation compensation techniques have not beenrequired in this case. Following thisbasic approach, eachvehicle is able to autonomously and dynamically configureits transmission power to the minimum value that satisfies theapplication reliability.

Considering a highway scenario withsix lanes, the com-bination of transmission power and packet generation ratethat allows meeting the application requirements with thetarget reliability is illustrated in Figure 10. In particular, thetransmission power levels shown in this figure correspond tothe vehicles experiencing the highestDw, i.e. vehicles movingat vmax = 120km/h and vmin = 60km/h. In this case,the application’s reliability for each vehicle has been settopappA = pappB = 0.99. In this scenario, the multipath fadingeffect has been modeled by a Nakagami model, followingthe observations for highway scenarios in [42]. As shown inthis figure, when increasing the packet generation rate, thetransmission power can be decreased to maintain the sameapplication reliability.

To reduce the channel load and unnecessary interference, thedescribedbasic awareness control approach can be improvedthrough the use of traffic context information following aclosed-loop approach. To this end, each vehicle could utilizethe specific positions of neighbouring vehicles to reconfigureits application requirements and the resulting transmissionparameters. An example of the use of traffic context infor-mation for the lane change assistance application is illustratedin Figure 9b. Considering the previously explained approach,vehicle A would broadcast its beacon message at itsDwA

14

1 2 3 4 5 6 7 8 9 1010

11

12

13

14

15

16

17

18

Packet generation rate [Hz]

Tra

nsm

issi

on p

ower

[dB

m]

TWindow = 2sTWindow = 4sTWindow = 6s

Fig. 10. Communications configurations that satisfy the application re-quirements with the target reliability (Dw and pappA = pappB = 0.99),considering a traffic density of 15 veh/km/lane.

distance. However, if vehicle A is aware of the presence ofvehicle D through the reception of one of its beacon messages,it can assume that vehicle B would have been informed byvehicle D that it cannot conduct alane-change maneuver. Asa result,if vehicle D is located at a distance lower thanDwA

from vehicle A, thenvehicle A does not need to transmitwith the power level required to guaranteepapp at DwA.In this context, vehicle A can reduce its transmission powerto that needed to communicate with vehicle D, located at adistanceDf from vehicle A (see Figure 9b). Consequently,each vehicle can configure its transmission parameters basedon the minimum of theDw and Df distances. This resultsin vehicle A configuringits transmission power to directlycommunicate with vehicle B only when there is no vehicleD in the same lane ahead located atDf < Dw. Therefore,the use of traffic context information obtained through theperiodic exchange of broadcast messages allowsthe reductionof unnecessary interference and limits the channel load withrespect to abasicapproach.

Figure 11 shows the average channel busy time for thebasicand traffic contextualapproaches previously described. Theresults shown in the figure have been obtained for a highwayscenario with 6 lanes, different traffic densities (D1 = 7.2,D2 = 9.6, and D3 = 14.4 veh/km/lane) and packetgeneration rates of 2Hz and 10Hz. It is important to notethat all the configurations reported in this figure wereable tosatisfy the target application’s reliability. As it can be observed,the traffic contextualextension of awareness control policiescan significantly reduce the channel load (more than 50% insome cases) while guaranteeing the application requirements.The obtained results also demonstrate that the reduction ofthe packet generation rate can also decrease the channel loadgenerated, despitethe fact that it would require a highertransmission power to meet the application requirements (seeFigure 10).

VI. OUTLOOK

The design of future cooperative vehicular systems couldcertainly require the joint consideration of congestion and

D1 D2 D30

5

10

15

20

25

30

35

Traffic density

CB

T [%

]

D1 D2 D30

2

4

6

8

10

12

Traffic density

CB

T [%

]

Basic Traffic contextual

(b)(a)

Fig. 11. Channel busy time for the communications configurationthat satisfythe application requirements with the target reliability (Dw and pappA =pappB = 0.99), considering a payload of 500 Bytes. (a) 2Hz (b) 10Hz

awareness control protocols. Under certain conditions, the con-gestion control limitationscould prevent the proper functionof multiple applications running on neighboring vehicles. Thiscouldarisein scenarios such as the one illustrated in Figure 1.In this scenario, the requirements of the lane change assistanceapplication run by the vehicles under free flow conditionswould be notably differentfrom the requirements of the appli-cations run by the vehicles in the traffic jam. While awarenesscontrol protocols would adapt each vehicle’s communicationsparameters to efficiently satisfy their individual requirements,congestion control protocols would limit the channel loadgenerated, given the high density of vehicles in the scenario.As a result, the requirements of all the different applicationsmight not be simultaneously satisfied. This example indicatesthe obvious challenge of how to integrate both control aspectsinto one system, in particular if the selected actions andadjustments are contradictory. The fundamentally differentobjectives lead to the issue that a joint realization might bedifficult to realize or even mutually exclusive. One potentialapproach to address this problemis based on the use ofadditional policies to prioritize among different applicationsand control the amount of information sent to the wirelesschannel. An example of this type of policy was proposedin [43], based on application-specific utility functions and aprioritization andreschedulingtechnique.

In addition to the joint consideration of congestion andawareness control, future cooperative vehicles might needtorun different applications simultaneously. As a result, theyshould be able to simultaneously support potentially different(and maybe contrary) communication and application require-ments. How to efficiently satisfy these requirements whileefficiently using the communications channel is a challeng-ing aspect that would need to be carefully investigated inthe coming years. Considering only safety applications forillustration purposes, these applications might need to detector monitor neighboring vehicles in the various safety areasshown in Figure 12. For the example shown in this figure,vehicle A might need to simultaneously run a cooperativeforward collision warning (CFCW) application with vehicle

15

��

AB

C

AheadBehind

Ahead leftBehind left

Fig. 12. Multi-application scenarios

B, and an overtaking vehicle warning (OVW) application withvehicle C [27]. In this context, vehicle A would experience avery different relative speed with vehicles B and C due totheir opposite directions. This would result in very differentwarning distances and communications settings for the twosimultaneous applications run by vehicle A.

In this context, it is also worth highlighting the necessityof jointly taking into account the requirements of safetyand non-safety applications. For example, considering onlythe requirements of traffic safety applications in the designof awareness control protocols could compromise the con-nectivity requirements of non-safety applications employingmulti-hop transmissions. This example can be illustrated withthe highway scenario considered in subsection V-B2. In thisscenario, the vehicles adapt their communication parametersto support the lane change assistance application. As demon-strated in in subsection V-B2, the use of a traffic contextualapproach satisfying the safety requirements can reduce theriskof channel congestion. However, the decrease of the trans-mission power without considering the requirements of otherapplications could compromise, for example, the connectivityrequirements. This effect can be observed in Figure 13. Thisfigure shows the minimum number of neighboring vehiclesthat each vehicle has in its neighbor list 99% of the time. Theresults shown in this figure correspond to thebasicapplication-driven awareness control technique previously explained andits traffic contextualadaptation. Following [44], a vehicle isremoved from another vehicle’s neighbor list after 5s withoutreceiving any 1-hop broadcast packet from it. As it can beobserved, the use of a contextual approach to reduce the riskof channel congestion reduces the number of neighboring ve-hicles, which could compromise the connectivity requirementsof the different vehicles.

To efficiently support various simultaneous applications,each vehicle should dynamically define the minimum com-munications parameters (e.g. transmission power and packetgeneration rate) that are able to satisfy the requirements ofeach application, following the awareness control proposalsdiscussed in this paper. Then, the communication requirementsshould be efficiently and safely combined to minimize thechannel load generated and satisfy the requirements of all thedifferent applications. To thisend, the definition of a Com-munications Adaptation Layer (CAL) would be needed, andits operation could be as follows. Assume that all applicationsrequire the same information to be sent/received in the periodicbroadcast packets transmitted in the communications channel.Let us further assume that a given vehicle is runningNapplications, each of them with communication requirementsPti (transmission power) andRi (packet generation rate),with 1 ≤ i ≤ N . To satisfy theRi requirements of the

D1 D2 D30

5

10

15

20

25

30

35

40

45

50

Traffic density

Num

ber

of n

eigh

bors

D1 D2 D30

5

10

15

20

25

30

35

40

45

50

Traffic density

Num

ber

of n

eigh

bors

Basic Traffic contextual

(a) (b)

Fig. 13. Minimum number of neighboring vehicles detected during 99% ofthe time for the communications configuration thatsatisfiesthe establishedapplication reliabilitypappA = pappB = 0.99, considering a payload of500 Bytes. (a) 2Hz (b) 10Hz

Pti, transmission ni, number of packets transmittedpower [dBm] per second usingPti

Pt1 n1 = R1

Pt2 n2 = max(0, R2 − n1)Pt3 n3 = max(0, R3 − n2 − n1)... ...

PtN nN = max(0, RN − nN−1 − nN−2 − ...− n1)

TABLE IICOMMUNICATIONS ADAPTATION LAYER CONFIGURATION FOR

MULTI -APPLICATION SCENARIOS

different applications, the total number of packets transmittedper second by this vehicle would be:

R = max(R1, R2, ..., RN )

To satisfy thePti requirements of the different applications,these applications should first be ordered as a function oftheir transmission power requirements so thatPt1 ≥ Pt2 ≥... ≥ PtN . Then, the transmission power of theR packetstransmitted per second could be distributed as indicated inTable II, so that at leastRi packets per second are transmittedwith a transmission power equal or higher thanPti. Anexample of the operation of the CAL is provided in Figure 14,where three applications are being run by a vehicle, and eachof them has different transmission power and packet gener-ation rate requirements. Following the proposed adaptation,R = 5 packets would need to be transmitted per second tosatisfy the requirements of the three applications in termsofpacket generation rate. Two of these packets would need to betransmitted with at least 20dBm to satisfy application 1, threepackets with at least 10dBm to satisfy application 2, and fivepackets with at least 6dBm to satisfy application 3.

Although this example clearly shows that the considerationof a CAL is certainly more efficient than treating each ap-plication separately, it also highlights the needfor further re-search investigatingoptimization approaches to address multi-application scenarios. For example, considering the exampleshown in Figure 14, the communications settings could be

16

20dBm

20dBm

10dBm

10dBm

10dBm

6dBm

Application 1(Pt1=20dBm,

R1=2Hz)


R2=3Hz)


R3=5Hz)

6dBm

6dBm

6dBm

6dBm

Communications Adaptation Layer

20dBm

20dBm

10dBm

6dBm

6dBm

Transmission power and packet generation rate requirements

Packets transmitted per second

Fig. 14. Example of the communications adaptation layer to combine the dif-ferent communication and applications requirements of various simultaneousapplications

redefined if transmitting five packets with just 6dBm coulddirectly satisfy the requirements of applications 1 and 2.This would reduce the channel load and increase the systemcapacity with respect to the solution discussed in the figure.In addition, the proposed CAL should be designed consid-ering its interaction with the contribution from [45], whichproposes to combine the information to be transmitted bydifferent applications to reduce the channel load generatedby each vehicle. This contribution focused on the payload ofdifferent application messages, but not on the configurationof the communication resources. The efficient combinationof the required transmission parameters and the informationto be transmitted by the different applications could then bepart of the optimization of the proposed adaptation layer.The integration of all these multi-application considerationsconstitutes an interesting and open research field that shouldbe addressed by the cooperative vehicular systems researchcommunity.

VII. C ONCLUSIONS

Congestion and awareness control techniques represent rele-vant building blocks in cooperative vehicular communications,since they are essential mechanisms to ensurethe stable andreliable operation ofcommunications system, while efficientlyusing the limited channel bandwidth. This paper has presenteda unified viewof the underlying control issues, and has alsoclarified the different perspectives that congestion control andawareness control proposals have taken in the past. Differentapproaches such as D-FPAV and OPRAM served as specificexamples for congestion control and awareness control, re-spectively.

When drawing conclusions from the survey of options wehave presented, one should differentiate between a first-stagedeployment and full and wide deployment. To get a ‘first gen-eration’ of cooperative vehicular communications deployed, itmight be helpful to assume a small set of different applicationclasses. In this case, congestion control is well understood,

and relatively simple methods like the reactive approachesbased on distributed information about the currently usedtransmit parameters and channel congestion, provide a goodperformance versus complexity trade-off somewhat similarto overprovisioning strategies in other networks. The designof awareness control policies could also be simplified andan individual application approach could be feasible. A full-scale deployment in large and multi-application scenariosmight require the use of advanced congestion and awarenesscontrol policies. However, tackling joint congestion controland awareness control can be highly complex, in considerationof thedifferent and possibly contradicting requirementsof thevehicles involved. In this context, advances in informationtheory for local broadcast networks as well as applicationof operations research techniques, for example, game theory,might help as foundations for a theory of optimal communica-tions setting in cooperative vehicular systems: an informationtheory for local broadcast networks could define what exactlyis possible in these types of networks from a local broadcastcapacity point of view, while the operations research relatedtreatment could help in dealing with contradicting system andapplication requirements.

Finally, the issue of congestion control and awarenesscontrol depends heavily on allocated frequency bands, mediumaccess control techniques, and available technology, and willdefinitely have to be revised with regulatory and technicaladvances.

ACKNOWLEDGMENT

Miguel Sepulcre and Javier Gozalvez acknowledge thesupport of the FP7 ICT Project iTETRIS (No. FP7 224644),the SpanishMinisterio de Fomento(T39/2006) and the Gen-eralitat Valenciana (BFPI06/126). Jens Mittag acknowledgesthe support of the Ministry of Science, Research and the Artsof Baden-Wurttemberg (Az: Zu 33-827.377/19,20), the KlausTschira Stiftung, the INIT GmbH and the PTV AG for theresearch group on Traffic Telematics for the research groupon Traffic Telematics. He also acknowledges the support ofthe European Telecommunications Standard Institute (ETSI)for the Specialist Task Force (STF) 395 on Configurationand validation of channel congestion control methods of ITSas well as all members of this STF for many interestingdiscussions.

REFERENCES

[1] “IEEE P802.11p/D11.0, Draft Amendment for Wireless Access in Ve-hicular Environments (WAVE),” April 2010.

[2] ETSI TC ITS, Intelligent Transport Systems (ITS); European ProfileStandard on the Physical and Medium Access Layer of 5GHz ITSs.Draft ETSI ES 202 663 V0.0.6, Oct. 2009.

[3] P. Santi, “Topology Control in Wireless Ad Hoc and SensorNetworks,”ACM Computing Surveys, vol. 37, no. 2, pp. 164–194, 2005.

[4] L. Cheng, B. Henty, R. Cooper, D. Stancil, and F. Bai, “Multi-pathpropagation measurements for vehicular networks at 5.9 GHz,”inWireless Communications and Networking Conference, April 2008, pp.1239–1244.

[5] I. N. L. Tan, W. Tang, K. Laberteaux, and A. Bahai, “Measurementand analysis of wireless channel impairments in DSRC vehicularcommunications,” EECS Department, University of California,Berkeley,Tech. Rep. UCB/EECS-2008-33, Apr 2008. [Online]. Available:http://www.eecs.berkeley.edu/Pubs/TechRpts/2008/EECS-2008-33.html

17

[6] Y. Zhang, I. L. Tan, C. Chun, K. Laberteaux, and A. Bahai, “Adifferential ofdm approach to coherence time mitigation in dsrc,” inVANET ’08: Proceedings of the fifth ACM international workshop onVehiculAr Inter-NETworking. New York, NY, USA: ACM, 2008, pp.1–6.

[7] P. Alexander, D. Haley, and A. Grant, “Outdoor mobile broadband accesswith 802.11,” Communications Magazine, IEEE, vol. 45, no. 11, pp.108–114, November 2007.

[8] “Federal Communications Commision. FCC 03-324. FCC Report andOrder,” February 2004.

[9] D. Jiang, Q. Chen, and L. Delgrossi, “Optimal Data Rate Selection forVehicle Safety Communications,” inVANET ’08: Proceedings of the fifthACM international workshop on VehiculAr Inter-NETworking. NewYork, NY, USA: ACM, 2008, pp. 30–38.

[10] T. R. Board,Highway Capacity Manual. National Academy of Science,2000.

[11] Q. Xu, R. Sengupta, and D. Jiang, “Design and Analysis ofHighwaySafety Communication Protocol in 5.9 GHz Dedicated Short RangeCommunication Spectrum,” inProceedings of the IEEE 57th VehicularTechnology Conference (VTC-Spring), Jeju, Korea, October 2004, pp.2451–2455.

[12] H.-J. Reumerman, M. Roggero, and M. Ruffini, “The Application-BasedClustering Concept and Requirements for Intervehicle Networks,” IEEECommunications Magazine, vol. 42, no. 4, pp. 108–113, April 2005.

[13] VSC, “Task 3 Final Report - Identify Intelligent Vehicle Safety Appli-cations Enabled by DSRC,” Public Document, Vehicle Safety Commu-nications Project, Crash Avoidance Metrics Partnership, 2004.

[14] M. Torrent-Moreno, P. Santi, and H. Hartenstein, “FairSharing ofBandwidth in VANETs,” in Proceedings of the ACM InternationalWorkshop on Vehicular Inter-Networking (VANET), Cologne, Germany,Sept. 2005, pp. 49–58.

[15] T. ElBatt, S. K. Goel, G. Holland, H. Krishnan, and J. Parikh, “Co-operative Collision Warning Using Dedicated Short Range WirelessCommunications,” inProceedings of the ACM International Workshopon Vehicular Inter-Networking (VANET), Los Angeles, CA, USA, Sept.2006, pp. 1–9.

[16] S. Rezaei, R. Sengupta, H. Krishnan, X. Guan, and R. Bhatia, “Trackingthe Position of Neighboring Vehicles Using Wireless Communications,”Transportation Research Part C: Emerging Technologies, vol. 18, no. 3,pp. 335–350, June 2010.

[17] X. Yang, J. Liu, F. Zhao, and N. Vaidya, “A Vehicle-to-Vehicle Com-munication Protocol for Cooperative Collision Warning,” inProceedingsof the First Annual International Conference on Mobile and UbiquitousSystems, Networking and Services (MobiQuitous), Bosto, USA, Aug.2004, pp. 1–10.

[18] H. Xu and M. Barth, “A Transmission-Interval and Power-Level Mod-ulation Methodology for Optimizing Inter-Vehicle Communications,” inProceedings of the ACM International Workshop on VehicularInter-Networking (VANET), Philadelphia, PA, USA, Oct. 2004, pp. 97–98.

[19] M. Artimy, W. Robertson, and W. Phillips, “Assignment of DynamicTransmission Range Based on Estimation of Vehicle Density,” inProceedings of the ACM International Workshop on VehicularInter-Networking (VANET), Cologne, Germany, Sep. 2005, pp. 40–48.

[20] M. Torrent-Moreno, J. Mittag, P. Santi, and H. Hartenstein, “Vehicle-to-Vehicle Communication: Fair Transmit Power Control for Safety-CriticalInformation,” IEEE Transactions on Vehicular Technology, vol. 58, pp.3684–3703, Sept. 2009.

[21] C. Khorakhun, H. Busche, and H. Rohling, “Congestion Control forVANETs based on Power or Rate Adaptation,” inProceedings of theInternational Workshop on Intelligent Transportation (WIT), Hamburg,Germany, March 2008.

[22] R. Baldessaro, D. Scanferla, L. Le, W. Zhang, and A. Festag, “JoiningForces for VANETs: A Combined Transmit Power and Rate ControlAlgorithm,” in Proceedings of the International Workshop on IntelligentTransportation (WIT), Hamburg, Germany, March 2010.

[23] C.-L. Huang, Y. Fallah, R. Sengupta, and H. Krishnan, “AdaptiveIntervehicle Communication Control for Cooperative Safety Systems,”IEEE Network, vol. 24, no. 1, pp. 6–13, Jan. 2010.

[24] J. Mittag, F. Schmidt-Eisenlohr, M. Killat, J. Harri, and H. Hartenstein,“Analysis and Design of Effective and Low-Overhead TransmissionPower Control for VANETs,” inProceedings of the ACM InternationalWorkshop on Vehicular Inter-Networking (VANET), San Francisco, CA,USA, Sept. 2008, pp. 39–48.

[25] W. Zhang, A. Festag, R. Baldessari, and L. Le, “Congestion Control forSafety Messages in VANETs: Concepts and Framework,” inProceedingsof the International Conference on ITS Telecommunications(ITST),Pucket, Thailand, Oct. 2008, pp. 199–203.

[26] VSC Consortium,Vehicle Safety Communications Project Task 3 -Final Report: Identify Intelligent Vehicle Safety Applications enabledby DSRC. US DOT HS 809 859, March 2005.

[27] ETSI TC ITS, Intelligent Transport Systems (ITS); Vehicular Commu-nications; Basic Set of Applications; Definitions. ETSI TR 102 638V1.1.1, June 2009.

[28] Q. Xu, T. Mak, J. Ko, and R. Sengupta, “Vehicle-to-Vehicle Safety Mes-saging in DSRC,” inProceedings of the ACM International Workshopon Vehicular Inter-Networking (VANET), Philadelphia, PA, USA, Oct.2004, pp. 19–28.

[29] F. Bai and H. Krishnan, “Reliability Analysis of DSRC WirelessCommunication for Vehicle Safety Applications,” inProceedings ofthe IEEE Intelligent Transportation System Conference (ITSC), Toronto,Canada, Sept. 2006, pp. 355–362.

[30] J. Gozalvez and M. Sepulcre, “Opportunistic Techniquefor EfficientWireless Vehicular Communications,”IEEE Vehicular Technology Mag-azine, vol. 2, no. 4, pp. 33–39, Dec. 2007.

[31] T. Osafune, L. Lin, and M. Lenardi, “Multi-Hop Vehicular Broadcast(MHVB),” in Proceedings of the International Conference on ITSTelecommunications (ITST), Chengdu, China, June 2006, pp. 757–760.

[32] L. Yang, J. Guo, and Y. Wu, “Piggyback Cooperative Repetition forReliable Broadcasting of Safety Messages in VANETs,” inProceedingsof the IEEE Consumer Communications and Networking Conference(CCNC), Las Vegas, NV, USA, Jan. 2009, pp. 1165–1169.

[33] J. Mittag, F. Thomas, J. Harri, and H. Hartenstein, “A Comparison ofSingle- and Multi-hop Beaconing in VANETs,” inProceedings of theACM International Workshop on Vehicular Inter-Networking(VANET),Beijing, China, Sept. 2009, pp. 69–78.

[34] M. Ruffini and H.-J. Reumerman, “Power-Rate Adaptation inHigh-Mobility Distributed Ad-Hoc Wireless Networks,” inProceedings ofthe IEEE Vehicular Technology Conference (VTC-Spring), vol. 4, Stock-holm, Sweden, May 2005, pp. 2299–2303.

[35] X. Guan, R. Sengupta, H. Krishnan, and F. Bai, “A Feedback-BasedPower Control Algorithm Design for VANET,” inProceedings of theMobile Networking for Vehicular Environments Workshop (MOVE),Anchorage, AK, USA, May 2007, pp. 67–72.

[36] Q. Xu, T. Mak, J. Ko, and R. Sengupta, “Medium Access Control Pro-tocol Design for Vehicle-Vehicle Safety Messages,”IEEE Transactionson Vehicular Technology, vol. 56, no. 2, pp. 499–518, March 2007.

[37] R. Schmidt, T. Leinmuller, E. Schoch, F. Kargl, and G. Schafer,“Exploration of Adaptive Beaconing for Efficient Intervehicle SafetyCommunication,”IEEE Network, vol. 24, no. 1, pp. 14–19, Jan. 2010.

[38] WINNER consortium,WINNER II channel models. WINNER EuropeanResearch project Public Deliverable D1.1.2., Sept. 2007.

[39] M. Sepulcre and J. Gozalvez, “Wireless Vehicular Adaptive Radio Re-source Management Policies in Congested Channels,” inProceedings ofthe IEEE International Symposium on Wireless Communication Systems(ISWCS), Trondheim, Norway, Oct. 2007, pp. 380–384.

[40] M. Sepulcre and J. Gozalvez, “Adaptive Wireless Vehicular Commu-nication Techniques under Correlated Radio Channels,” inProceedingsof the IEEE Vehicular Technology Conference (VTC-Spring), Barcelona,Spain, April 2009, pp. 1–5.

[41] J. Gozalvez and M. Sepulcre, “Channel Efficiency of Adaptive Trans-mission Techniques for Wireless Vehicular Communications,” in Pro-ceedings of the International World Congress on ITS, New York, NY,USA, Nov. 2008.

[42] V. Taliwal, D. Jiang, H. Mangold, C. Chen, and R. Sengupta, “EmpiricalDetermination of Channel Characteristics for DSRC Vehicle-to-VehicleCommunication,” inProceedings of the ACM International Workshopon Vehicular Inter-Networking (VANET), Philadelphia, PA, USA, Oct.2004, pp. 88–88.

[43] T. Kosch, C. Adler, S. Eichler, C. Schroth, and M. Strassberger, “TheScalability Problem of Vehicular Ad Hoc Networks and how to Solveit,” IEEE Wireless Communications, vol. 13, no. 5, pp. 22–28, Oct. 2006.

[44] Geonet Consortium,Final GeoNet Specification. GeoNet EuropeanResearch project Public Deliverable D2.2, Jan. 2010.

[45] C. Robinson, D. Caveney, L. Caminiti, G. Baliga, K. Laberteaux, andP. Kumar, “Efficient Message Composition and Coding for CooperativeVehicular Safety Applications,”IEEE Transactions on Vehicular Tech-nology, vol. 56, no. 6, pp. 3244–3255, Nov. 2007.

18

Miguel Sepulcre received a TelecommunicationsEngineering degree in 2004 and a Ph.D. in Commu-nications Technologies in 2010, both from the Uni-versity Miguel Hernandez of Elche (UMH), Spain.In 2004, he spent six months at the European SpaceAgency (ESA) in Noordwijk (The Netherlands)working on the communications physical layer ofearth exploration satellites. He then joined in 2005the University Miguel Hernandez of Elche as anetworks manager and teaching assistant. In March2006, he obtained a Ph.D. fellowship from the

Valencian regional government and joined the Uwicore research laboratory. In2009, he spent three months at the Karlsruhe Institute of Technology (KIT) inKarlsruhe (Germany) working on cooperative vehicular communications. Heis currently a research fellow at the Uwicore Research Laboratory of UMH,working on cooperative vehicular communications, radio resource manage-ment, and modeling and simulation of wireless communications systems.

Jens Mittag received the Diploma in computersience from the University of Karlsruhe, Germany, in2008. He is currently working towards the Ph.D. de-gree within the Decentralized Systems and NetworkServices Research Group at the Karlsruhe Instituteof Technology (KIT), Germany.

His research interests include vehicular mobilenetworks, accurate simulation environments and themodeling of the wireless lower physical laayers. Heparticipated in the NOW: Network on Wheels project(2004-2008) and contributed to the standardization

of intelligent transportation systems (ITS) in Europe as a member of theEuropean Telecommunications Standards Institute (ETSI) specialist task forceon Configuration and Validation of Channel Congestion ControlMethods ofITS.

Paolo Santi received the Laura Degree summa cumlaude and Ph.D. Degree in Computer Science fromthe University of Pisa in 1994 and 2000, respectively.He has been researcher at the Istituto di Informaticae Telematica - CNR in Pisa, Italy, since 2001, andSenior Researcher since 2009.

During his career, he visited Georgia Institute ofTechnology in 2001, and Carnegie Mellon Univer-sity in 2003. His research interests include fault-tolerant computing in multiprocessor systems (dur-ing Ph.D. studies), and, more recently, the inves-

tigation of fundamental properties of wireless multihop networks such asconnectivity, topology control, lifetime, capacity, mobility modeling, andcooperation issues. He has contributed more than 50 papers and a book inthe field of wireless ad hoc and sensor networking, he has beenGeneralCo-Chair of ACM VANET 2007/2008, Technical Program Co-Chair of IEEEWiMesh 2009, and he is involved in the organizational and technical programcommittee of several conferences in the field. Dr. Santi is Associate Editor forIEEE Transactions on Mobile Computing and IEEE Transactionson Paralleland Distributed Systems. He is a member of IEEE Computer Society and aSenior member of ACM and SIGMOBILE.

Hannes Hartenstein received the Diploma inmathematics and the Ph.D. degree in computerscience from Albert-Ludwigs-Universitat, Freiburg,Germany, in 1995 and 1998, respectively.

He is a full professor for decentralized systemsand network services at the Karlsruhe Institute ofTechnology (KIT), Germany, and executive directorof the KIT Steinbuch Centre for Computing. Hisresearch interests include mobile networks, virtualnetworks and IT management. Prior to joining theUniversity of Karlsruhe, he was a senior research

staff member with NEC Europe. He was involved in the FleetNet - Internet onthe Road (2000-2003) and NOW: Network on Wheels (2004-2008) projects,partly funded by the German Ministry of Education and Research (BMBF), heactively participated in the EU FP7 project PRE-DRIVE-C2X (2008-2010),and is now contributing to the follow-on project called DRIVE-C2X. He hasbeen TPC co-chair and general chair of various highly selective ACM andIEEE international workshops and symposia on vehicular communications.

Javier Gozalvez received an electronics engineer-ing degree from the French Engineering SchoolENSEIRB and a Ph.D. in mobile communicationsfrom the University of Strathclyde, Glasgow, U.K.Since October 2002, he has been with the UniversityMiguel Hernandez of Elche, Spain, where he iscurrently an Associate Professor and Director of theUwicore Research Laboratory. At Uwicore, he isleading research activities in the areas of wirelessvehicular communications, radio resource manage-ment, heterogeneous wireless systems, and wireless

system design and optimization. He currently serves as Mobile Radio SeniorEditor of IEEE Vehicular Technology Magazine, and previously served as AEof IEEE Communication Letters. He was TPC Co-Chair of the 2009 IEEEVTC Spring, and General Co-Chair of the 3rd Int. Symposium on WirelessCommunications Systems (ISWCS’2006). He is also the founder andGeneralCo-Chair of the IEEE Int. Symposium on Wireless Vehicular Communications(WiVeC) in its 2007, 2008 and 2010 editions. He has recently been electedto the Board of Governors of the IEEE Vehicular Technology Society (2011-2014).

congestion and awareness control in cooperative vehicular...

Documents