Adaptive Intervehicle Communication Control for Cooperative Safety Systems

8/11/2019 Adaptive Intervehicle Communication Control for Cooperative Safety Systems

http://slidepdf.com/reader/full/adaptive-intervehicle-communication-control-for-cooperative-safety-systems 1/8IEEE Network • January/February 20106 0890-8044/10/$25.00 © 2010 IEEE

ntelligent transportation systems (ITSs), empowered by wireless communication, are expected to significantlyimprove the safety and efficiency of our transportationnetwork. The idea is to interconnect all the physical com-

ponents of a transportation system in cyber-space. Vehiclesequipped with IEEE 802.11p transceivers [1], dubbed dedicat-ed short-range communication (DSRC) or wireless access for vehicular environments (WAVE), can exchange informationeither vehicle-to-vehicle (V2V) or vehicle-to-infrastructure(V2I). While V2I communication requires huge infrastructureinvestment, V2V communication is a more viable solution forthe near future. Such V2V communication is the basis for thesetup of vehicular ad hoc networks (VANETs). VANETs areexpected to host new safety, navigation, and automation appli-cations in vehicular environments.

The DSRC technology, based on IEEE 802.11p, is the cen-terpiece of VANETs. Formally, IEEE 802.11p defines high-speed short-range wireless communication among vehiclesand surface transportation infrastructure. Similar to IEEE802.11a, 802.11p radio is based on matured orthogonal fre-quency-division multiplexing (OFDM) technology. The medi-um access control (MAC) layer functionality is slightlymodified to include provision for rapid communication of DSRC devices with no need for authentication or authoriza-tion processes as in the original 802.11 standard. In additionto the 802.11p standard, IEEE 1609 [2] defines higher-layerfunctionalities such as networking and multichannel opera-tions for VANETs. The protocol stack depicted in Fig. 1describes how different protocols are involved in enabling

vehicular communication. As part of these standards, a newtype of message, WAVE short message (WSM), is defined in1609.3, which supports frame-by-frame power and modula-tion assignment, and thus provides new possibilities for cross-

layer optimization. To ensure that DSRC applications areinteroperable, other standards may be employed in this archi-tecture (e.g., SAE J2735,1 which describes DSRC messagecontent). The standardization effort is equally supported byindustry and government entities. For example, the U.S.Department of Transportation supports such activitiesthrough the IntelliDrive initiative.2

DSRC and Cooperative Active Safety Currently, prototype DSRC-based applications are undertest by different car manufacturers and researchinstitutions.3 While these efforts show that DSRC-basedtechnology works in principle, deployment and performanceissues are still under investigation, especially for the highmarket penetration case. For example, the issues facingcooperative safety applications are still active research sub- jects. Unlike research on mobile ad hoc networks(MANETs), which focuses more on maximizing throughputor maintaining network connectivity, the distinct mission of VANETs is to enhance roadway safety through V2V com-munication [3, 4]. To name a few, active safety applications

I

I

Ching-Ling Huang, Yaser P. Fallah, and Raja Sengupta,

University of California at Berkeley Hariharan Krishnan, General Motors R&D Center

AbstractVehicular ad hoc networks play a critical role in enabling important active safetyapplications such as cooperative collision warning. These active safety applicationsrely on continuous broadcast of self-information by all vehicles, which allows eachvehicle to track all its neighboring cars in real time. The most pressing challenge insuch safety-driven communication is to maintain acceptable tracking accuracywhile avoiding congestion in the shared channel. In this article we propose a trans-mission control protocol that adapts communication rate and power based on thedynamics of a vehicular network and safety-driven tracking process. The proposed

solution uses a closed-loop control concept and accounts for wireless channel unre-liability. Simulation results confirm that if packet generation rate and associatedtransmission power for safety messages are adjusted in an on-demand and adap-tive fashion, robust tracking is possible under various traffic conditions.

Adaptive Intervehicle CommunicationControl for Cooperative Safety Systems

Research supported in part by General Motors R&D Center through con-

tract #TCS 70709 to U. C. Berkeley. The views expressed here are those of

the authors and not of the research sponsors.

1 SAE J2735 is a message set dictionary (i.e., common language) for

DSRC applications to understand each other.

2 http://www.intellidriveusa.org/

3 For example, see California PATH's demonstration at ITS World

Congress in New York at http://www.path.berkeley.edu/


http://slidepdf.com/reader/full/adaptive-intervehicle-communication-control-for-cooperative-safety-systems 2/8IEEE Network • January/February 2010 7

include cooperative collision warning (CCW), electronic

emergency brake light (EEBL), and slow/stopped vehiclealert (SVA). In general, these active safety applicationsrequire that a subject vehicle have a good estimate of theposition and state of the cars in its proximity. The availabili-ty of such proximity awareness allows various safety applica-tions to provide advisory services to the driver (via audio or vi sual human-machine interfaces ) or perform emergencyreactions to avoid hazardous situations.

For this safety/tracking purpose, each vehicle is assumed tobe equipped with a DSRC radio, a Global Positioning System(GPS) receiver, and onboard sensors. As shown in Fig. 2, each vehicle is designed to continuously report its own statu (e.g.,position, speed, and direction) by broadcasting safety mes-sages in WSM format. At the same time, each vehicle alsotracks movements of neighboring vehicles based on informa-

tion received from them over the shared channel. The ques-tion of how often and how far these messages should bebroadcast is one that drives this research.

In this article we propose a rate-power control algorithmthat describes how state information of a vehicle should bebroadcast to its neighbors. Specifically, for different trafficscenarios, we compare the tracking accuracy of the proposedalgorithm with that of the de facto solution (beaconing at aconstant interval, e.g., 100 ms and 500 ms) suggested by theVehicle Safety Communications Consortium (VSCC) [4]. Sim-ulation results indicate that our proposed algorithmis robust and achieves better tracking accuracy thanthe beaconing method in almost all traffic scenarios.In the context of safety-driven communicationdesign for VANETs, this work provides a new per-spective, different from the existing methods thattry to maximize throughput, achieve fairness, orreduce message rate; the objective of our design isto reduce tracking error and improve the robustnessand performance of cooperative vehicle safety sys-tems under different traffic conditions.

Overview of Existing ApproachesFor information propagation in wireless ad hoc networks, the work in [5] gives a fundamental boundon the product of rate and distance (due to theshared nature of the considered wireless medium).The implication of such a capacity notion is that

when the network becomes denser, one needs to

either throttle the data rate or reduce transmissionpower so that the limited channel resource can beproperly shared by all nodes. This bound appliesto VANETs too and most existing designs forbroadcasting safety messages focus on either rateor power adaptation so that the network can oper-ate properly when traffic density (number of vehi-cles) changes.

For safety applications in VANETs, [6] propos-es to fairly allocate transmission power across allcars in a max-min fashion, which helps reduce bea-con load at every point of a formulated 1D highway and thus reserves bandwidth for emergencymessages with higher priorities. This methodassumes a predefined maximum load as the target.By utilizing the water-filling concept, [6] proves acentralized algorithm that achieves this fairnessgoal and gives a decentralized algorithm thatapproximates the operations of a centralized one.

Since many applications require the same dataelements from other vehicles, the message dis-

patcher in [7] is proposed to reduce the required data rate by

removing duplicate elements. The message dispatcher at thesender side will group data elements from the applicationlayer (i.e., the source) and decide how frequently each dataelement should be broadcast. At the receiver side, the mes-sage dispatcher pushes received data elements to designatedapplications (i.e., the sink). This design is similar to compres-sion ideas in source coding.

Since our primary focus is on cooperative safety, ourapproach is to correlate communication behaviors with track-ing error magnitude. In [8] error threshold crossing is used asthe transmission trigger; broadcast of the state informationhappens only when a threshold is violated. In [9] we propose arate control design that has advantages over others in ashared channel:• It increases the communication rate when a vehicle suspects

the estimation error of neighboring vehicles toward itself hasincreased.

• It throttles rate during channel congestion to avoid furtherefficiency degradation and message loss.These two principles and their combined effect govern

transmission behaviors on each vehicle and provideimproved robustness to the tracking process. Our proposedrate-power control in this article follows this line of designrationale, and adds another adaptation dimension: thetransmission range.

Figure 1. Protocol stack for WAVE, mainly composed of IEEE 802.11p [1] and IEEE 1609 family standard [2]: 1609.1 (resource manager), 1609.2(security), 1609.3 (networking services), and 1609.4 (multichannel opera-tions). Note that 1609.2 works jointly with 1609.3 and thus is not shown inthe above architecture.

WSMP:

UDP:

TCP:

IPv6:

LLC:MAC:

MLME:

PHY:PLME:

WME:

WAVE ShortMessage ProtocolUser DatagramProtocolTransmissionControl ProtocolInternet ProtocolVersion 6Logical lnk controlMedium accesscontrolMAC layer

management entityPhysical layerPHY layermanagement entityWAVE managemententity

Applications (1609.1 and others)

Multichannel operations(1609.4)

MAC (802.11p)

PHY (802.11p)PLME

(802.11)

MLME

(1609.4,802.11)

LLC

UDP or TCP

IPv6

WSMP(1609.3)WME (1609.3)

Figure 2. To enable active safety applications, each vehicle broadcasts its own

state information to neighboring vehicles via the DSRC channel.

Broadcast of safety messagesvia shared DSRC channel

S


http://slidepdf.com/reader/full/adaptive-intervehicle-communication-control-for-cooperative-safety-systems 3/8IEEE Network • January/February 20108

VANET Tracking Problem and Proposed Design

Channel access in DSRC, especially for safety applications, isperformed in a random manner, following carrier sense multi-ple access with collision avoidance (CSMA/CA) rules [1] withno centralized coordination. Therefore, the vehicle trackingproblem is essentially a remote estimation problem over arandom access channel. In this section we first state theVANET tracking problem at an abstract level. As shown inFig. 3, each vehicle is designed to contain a communicationcontrol logic, a bank of estimators to track other vehicles, anda plant (producing vehicle state information). The estimatedstates of neighboring vehicles will be fed to active safety appli-cations, which in turn will provide warnings to the driver ortake emergency control of the vehicle in case of imminentdanger. The proposed rate and power control scheme isimplemented in the communication control logic. The internalarchitecture of the communication control logic is shown inthe upper right corner of Fig. 3. Note that the produced safetymessages are in WSM format, which allows specifying per-message power level for 802.11p radio. For the rest of thisarticle, we describe our proposed algorithm for decentralizedself-information dissemination based on the design in Fig. 3.

Tracking over a Multi-Access Network To better understand the problem at hand, assume that anumber of vehicles ( n ≥ 2) share the same wireless channel,and each vehicle contains a transmission control logic, a plantgenerating state information x j(t), and a bank of model-basedestimators for tracking neighboring vehicles. On each vehicle

there are a balanced control input u j(t) to model the driver’s

accelerating/decelerating behavior and a noise input ε j(t) tomodel the mechanical disturbance within a car. The transmis-sion control logic decides whether to broadcast self-informa-

tion vector x j(t) (i.e., the state) at each moment. Each subject vehicle will try to estimate the states of other cars based onreceived information from the shared channel.

The estimated state (e.g., position) of sender j at receiver i, x~ij(t), is the expectation conditioned on all the previousreceived information over the lossy channel. Packet loss in thechannel could be due to collision, or channel error and fading.For presentation convenience, let’s assume the dynamics of a vehicle can be sampled at discrete time step and approximat-ed as a linear-time-invariant (LTI) system of appropriatedimensions as follows:

x j(t) = a j × x j(t – 1) + b j × u j(t – 1) + e j(t – 1), (1)

where a j and b j represent the mechanical characteristics andphysical laws that govern vehicle j, respectively. The model-based estimator at receiver i switches between two modes. Attime t – 1, if state information of vehicle j is received, thereceiver uses this information to reset the estimate: x~ij(t) = a j

× x j(t – 1). Otherwise, if no information regarding vehicle j isreceived, receiver i uses known model Eq. 1 to carry on andestimate the new state: x~ij(t) = a j × x~ij(t – 1).

In the tracking problem described above, the arrival of anew measurement from sender j to receiver i resets the track-ing error similar to a renewal process. Between measurementarrivals, the real-time tracking error could grow due to noiseand disturbance. Note that the above calculation of x~ij(t) is aminimum mean squared error (MMSE) optimal estimate if we assume u j(t) and ε j(t) have zero mean. The tracking error

eij(t), the ith vehicle’s estimation error toward vehicle j, is

Figure 3. Functional blocks of the in-vehicle unit, including the DSRC communication control module.

Internal block of DSRC comm. control

Plant measurements

Suspectedtracking error

PHY/MAC info

Modelestimator

Safety messages

Channel access Channel access

DSRC channel shared by all vehicles

The kth vehicle

CCW

Emergencycontrol

EEBL

Active safety applications

SVA . . .

Estimates of neighboring vehicles

Received info from DSRC channel

Safety messages and PHY/MAC info

DSRC communication control

IEEE 802.11p

The (k

+ 1)th vehicle

Rate/poweradaptation algorithm

Vehicle dynamics(plant)

Sensor measurements Rate/power adaptation for safety messages

Warnings of advisory services to the driver

Vehicle neighborhood mapping

A bank of estimators for tracking vehicles

The (k

+ 2)th vehicle. . .



defined as eij(t) = x j(t) – x~ij(t). Due to the hidden node prob-lem or channel fading, this tracking error might vary for dif-ferent vehicles.

Given the multi-access nature of the DSRC channel, thepacket loss ratio and therefore the real-time tracking errordepend on the number of vehicles that share the channel.This dependence means that at high transmission rates andranges, the random access channel will become congested and

consecutive collisions will make the channel useless. On theother hand, higher transmission rate and range, if successfullyachieved, will directly result in better tracking performanceand lower error. Therefore, the question is how to controleach vehicle’s transmission behavior to avoid channel collisionfailure while maximizing the use of the wireless medium fortracking purposes. Moreover, the maximum achievable valuesof rate and range before excessive collision happens are dif-ferent for different vehicular traffic densities. This fact moti- vates the need for an adaptive transmission control protocolor algorithm, which is described in the next subsection.

Proposed Adaptive Transmission Control In this subsection we present a joint rate-power control algo-

rithm for the VANET tracking problem. The algorithm com-prises two parts:• Rate control, which decides how frequently a subject vehicle

should broadcast its own state information• Power control, which decides how far the state information

should be broadcast, thus determining the power level forthe 802.11p radio

The functional components of the proposed a lgorithm areshown in Fig. 3. The following describes these methods for-mally and in more detail.

Transmission Rate Control — At each time step t ∈ N , the jth vehicle’s communication logic calculates transmission proba-bility p j(t) based on suspected tracking error e~ j(t) on neighbor-ing vehicles toward its own position in a Euclidean sense (i.e.,

the usual distance definition for a Cartesian coordinate sys-tem). If suspected error e~ j(t) is smaller than eth (the errorthreshold), there is no transmission at all from vehicle j (i.e., p j(t) = 0). This threshold design leaves the channel to be usedby other vehicles when the suspected error of vehicle j is with-in the tolerable range. Otherwise, if the suspected error islarger than this threshold, the probability of transmission inthat time step t is computed as follows:

p j(t) = 1 – exp(–α × | e~ j(t) – eth|2), (2)

where α > 0 is the sensitivity to the suspected tracking error.The rationale behind this formula is that a subject vehicle jincreases transmission probability p j when it suspects trackingerror on other vehicles (toward its own state) has exceededthe threshold. The more this error violates the threshold, thehigher the transmission probability becomes.

The on-demand nature of Eq. 2 responds to the fact that ahigher transmission rate, and thus probability, is required fora subject vehicle that has more unexpected movements (as esti-mated by other vehicles). In information theoretical terms, a vehicle with higher entropy (a measure of surprise) needs ahigher communication rate to describe its stochastic behavior.On the contrary, when the suspected error is small or underthe threshold, a vehicle tends to stay quiet, and this allows thechannel to be used by those vehicles having larger suspectederror. This design also implicitly acts as transmission errorcorrection as explained below.

Since there is no explicit acknowledgment (ACK) for

broadcast in DSRC, after each transmission, we use the pack-

et erasure rate (PER) to stochastically decide the suspectederror e~ j(t) in Eq. 2: e~ j(t+) = (1 – ζ) × e~ j(t), where ζ is aBernoulli trial (similar to coin tossing) with success probability1 – PER to address potential channel loss. If successful, thesuspected error is reset; otherwise, suspected error accumu-lates from e~ j(t) based on the known state model (e.g., Eq. 1).Note that this suspected error is not the actual estimationerror; instead, it is only a measure used by a subject vehicle to

adjust its own communication rate (the higher this probability,the higher the communication rate of the vehicle). Based on p j(t) in Eq. 2, a subject vehicle stochastically generates a pack-et (i.e., a safety message containing its own state information)and places it in the queue of the 802.11 MAC layer.

Finally, the PER used by the above Bernoulli trials is esti-mated on the fly by checking the inconsistency in sequencenumbers of recently received packets (by a subject vehicle)from all corresponding senders (with at least two messagesreceived) within a 1 s history log. That is, a subject vehicleuses the number of lost packets divided by the number of total packets initiated from a certain sender to infer recentchannel loss rate; PER is this measure averaged over all itssenders within a geographical area. Assuming network sym-

metry, PER tells a subject vehicle the likelihood of the loss of its previous transmission at the receivers. Note that this PERis in fact an average and rough measure of the real errorprobability of individual links between the subject vehicle andother vehicles within the spatial neighborhood. The exact value of link error probabili ty is not practical ly measurableunder the assumptions of the proposed active safety systemsin Fig. 3 since all safety messages are in the broadcast mode.

Transmission Power Control — The transmission range L j(t) isadjusted based on the observed channel status. Here, insteadof using explicit feedback, usually done through piggybackinformation exchange as in [6], we propose a simple design toscale transmission range linearly with average channel occu-pancy. This method is scalable and easy to implement. The

average channel occupancy, U j(t), is a real number between 0and 1. U j(t) is computed for each vehicle j by observing theclear channel assessment (CCA) reports that are availablefrom the PHY to the MAC layer of 802.11. This occupancymeasure is the time average of recently sampled CCA withina 1 s time window. Since there is no explicit ACK for broad-cast in DSRC, we use this side information, channel occupan-cy, to infer channel status and adjust targeted communicationrange.

In the proposed algorithm, if channel occupancy is higherthan U max , minimum transmission range Lmin is used; if chan-nel occupancy is lower than U min, maximum transmissionrange Lmax is used; otherwise, the range is selected as follows:

(3)

In Eq. 3 Lmin and Lmax are communication range lower/upperbounds, which are usually determined from safety specifica-tion for vehicular environments. Besides, U min and U max rep-resent the desired linear operating range of channel occupancyand are selected based on experimental or analytical data.Since the granularity of power levels for DSRC is 0.5 dBm [1],to convert L j(t) to power, we apply a power-range mapping(with step size 0.5 dBm) based on the empirical channelmodel reported in [3]. The basic assumption of proposedpower control is that vehicles in proximity would perceiveroughly the same CCA and the same amount of average chan-

nel occupancy. Therefore, in our design vehicles in proximity

L t L

U U t

U U L L j j

( )

( )

( ).minmax

max minmax min= +

−

−× −



coordinate their transmission power by using the same sideinformation (i.e., the observed channel occupancy).

The idea behind our rate-power control algorithm is thatnearby vehicles are assumed to be physically more related to asubject vehicle. In fact, it is potentially more dangerous if asubject vehicle does not know the state (e.g., position andspeed) of its immediate neighbors. In proposed power control, when the channel is congested, a vehicle tries to make surethat its nearby cars can still hear its state information byreducing the transmission power, which results in temporarilygiving up faraway cars. According to [5], adapting power prop-erly can help reduce interference and increase successfulreception probability of nearby nodes. In short, the proposedalgorithm does not compromise information rate intensity

while facing channel congestion; instead, it reduces power tomaintain decent tracking accuracy of its immediate neighbor-ing vehicles to ensure safety.

Performance EvaluationTo evaluate the performance of the proposed design, we haveconducted several simulation experiments with realistic set-tings using a network simulator and a microscopic traffic sim-ulator. The main stage for simulations is a bidirectional 2 kmhighway, with each direction containing four lanes of identicaltraffic. We only collect statistics from vehicles within a 0.5 kmto 1.5 km segment to avoid boundary effects in simulations.Our simulations consider the traffic scenarios listed in Table1: cases H1–H4 represent scenarios with homogeneous trafficflows in both directions, while cases M1–M6 contain mixedtraffic conditions. Mean flow speed, in the unit of miles perhour, is also listed for reference.

Simulation SettingsFor network simulations in OPNET,4 we us e a mo di fi ed802.11a PHY module working at 5.9 GHz with 10 MHz bandwidth. We fol low the DSRC channel model reported in [3]and simplify the far distances as Rayleigh fading instead of pre-Rayleigh. The reason for this simplification is that we areconsidering a straight highway scenario, while [3] considersurban scenarios with intersections and corners, which lead to

pre-Rayleigh fading observations. In addition, we assume thatthe path loss exponent is always 2.31. The transceiver operates with –87 dBm receiver sensitivity, 3 Mb/s raw data rate, and afixed 28 dBm transmission power (which roughly covers theradius of 250 m) for 100 ms and 500 ms beaconing, which isthe standard solution suggested by VSCC [4]. The payloadsize of each safety message is 300 bytes. Throughout ourOPNET simulations, the CSMA/CA of the 802.11 MAC layeris kept intact.

For vehicle dynamics, we use trajectory files of [position,speed, heading] sampled at 20 Hz, produced by a microscopictraffic simulator, SHIFT.5 Total simulation time is 30 s foreach run. During the simulation, at 50 ms time steps, each vehicle gets a measurement of its own status from onboard

sensors and decides whether or not to generate a packet,according to the specified communication policy. The onboardmeasurement noise is modeled exactly the same way as in [8], wh ic h wa s in tu rn ba se d on ex pe ri me nt da ta [1 0] . Up onreceiving information from the shared channel, each vehicleupdates its estimation of cars in proximity. We choose a sim-ple first-order kinematic model (i.e., a constant speed predic-tor) for tracking neighboring vehicles. By using this predictor,a vehicle is assumed to run at the same speed and in the samedirection after its last successful information broadcast isreceived by this predictor.

Tracking Accuracy Measure After each simulation run, statistics are collected from neigh-bors within the indicated distance interval in Fig. 5, and wecalculate Euclidean tracking error over all vehicles to explorethe law of large numbers. See Fig. 4a for an example of typi-cal tracking error distribution. Specifically, 95 percent cutoff error (indicated by the arrow in Fig. 4a) means that 95 per-cent of Euclidean errors fall below this number (i.e., only 5percent population is on the right side of the arrow), whichrepresents the tracking accuracy of the used communicationpolicy. We use this measure, 95 percent cutoff error, as themain performance metric for comparisons in Fig. 5. The mainadvantage of this performance measure over others (e.g.,mean or standard deviation of error) is that it gives a statisti-cal sense similar to that of a confidence interval. Note that the

Table 1. Simulated bidirectional highway traffic scenarios.

Case ID Direction #1 status Direction #1 speed Direction #2 status Direction #2 speed

H1 Congested 14 mph Congested 14 mph

H2 Low speed 30 mph Low speed 30 mph

H3 Medium speed 53 mph Medium speed 53 mph

H4 Free flow 74 mph Free flow 74 mph

M1 Congested 14 mph Low speed 30 mph

M2 Congested 14 mph Medium speed 53 mph

M3 Congested 14 mph Free flow 74 mph

M4 Low speed 30 mph Medium speed 53 mph

M5 Low speed 30 mph Free flow 74 mph

M6 Medium speed 53 mph Free flow 74 mph

4

OPNET modeler and its wireless module, http://www.opnet.com/ 5

California PATH SHIFT simulator, http://path.berkeley.edu/SHIFT/



presented tracking error in Figs. 4a and 5 is the ground-truthtracking error (in meters) during simulations instead of thesuspected error e~ j(t) perceived by vehicle j in Eq. 2.

Robustness of the Proposed DesignIn this subsection we present simulation results and comparethe performance of the 100 ms and 500 ms beaconing of safe-ty messages with that of the proposed communication algo-rithm for different traffic conditions listed in Table 1. Thefollowing parameters are used for the proposed rate-powercontrol: α = 2 m–2, eth = 0.2 m, Lmin = 50 m, Lmax = 250 m,U min = 0.4, and U max = 0.8 in Eqs. 2 and 3. In this subsection we focus on the robustness of the proposed design: with a fixed set of parameters, our design can work well in differenttraffic scenarios compared to non-controlled beaconing.

Figure 4b shows a typical distribution of consecutive lossesin one particular simulation run for each communicationdesign. Note that the shown percentage is calculated based on all communication losses instead of all communication instants.The reason for analyzing this data is to understand how effi-ciently the channel is shared by all vehicles and how effectivethe broadcasting behaviors of vehicles are. It is seen that the

100 ms beaconing method causes significantly more consecu-

tive losses because each vehicle keeps transmitting the sameamount of messages without considering whether the channelcould support such a high volume of data or not. Using ourproposed design, there are fewer consecutive losses in thechannel. Figures 4c and 4d show an example of the adaptivebehavior of transmission rate and power for three vehicles inproximity for 10 simulation seconds. For simplicity of presen-tation, only values on integer seconds are plotted in Figs. 4cand 4d; during simulation, the transmission power and ratefor each vehicle is adapted every 50 ms based on our algo-rithm.

The tracking accuracy (95 percent cutoff Euclidean error;Fig. 4a) for both 100 ms and 500 ms beaconing, and our pro-posed solution is plotted in Fig. 5 with respect to various traf-fic conditions in Table 1. Each subplot shows the trackingaccuracy of three communication policies at different dis-tances, presented in 30 m bins between 0–240 m. The reasonfor such presentation is to illustrate how tracking accuracydegrades as the distance from a subject vehicle increases. Dueto space limitations, only results from cases H1, H2, M3, M5,and M6 are presented in Fig. 5. For cases H1 and H2, onlytracking accuracy in direction #1 is presented since traffic in

both directions is homogeneous; for cases M3, M5, and M6,

Figure 4. a) Histogram of tracking error (in meters) collected within 150 m radius of each subject vehicle while using 100 ms beaconing (in H4 case); b) percentage of consecutive losses collected within 30 m radius of each subject vehicle during simulation (in M3 case); c) samples of three vehicles’ adaptive transmission probability for 10 s (in M3 case); d) samples of three vehicles’ adaptive transmission power for 10 s (in M3 case).

(a) (b)

95% cutoff error

1.80 2

1

0

P o p u l a t i o n

o f t r a c k i n g e r r o r

2

3

4

5

6

7x105

1.61.41.210.80.60.40.2

100 ms beaconing500 ms beaconingProposed design

(c)

11

0.1

0.05

A d a p t i v e t r a n s m i s s i o n p r o b a

b i l i t y

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

12 13 14 15 16 17 18 19 20

(d)

11

0.09

0.08

A d a p t i v e t r a n s m i s s i o n p o w e

r ( W )

0.085

0.095

0.1

0.105

0.11

0.115

0.12

0.125

0.13

12 13 14 15 16 17 18 19 20

Vehicle #1Vehicle #2Vehicle #3

1

10%

0%

P e r c e n t a g e a m o n g a l l p a c k e t l o s s e s

2 3 4 5 6 7 8 9 10 11 12 13 14 15

20%

30%

40%

50%

60%

Vehicle #1Vehicle #2Vehicle #3



tracking accuracy in both direction #1 (left) and direction #2(right) is presented.

The message rate for 100 ms beaconing is fixed at 10 pack-ets/s, while it is 2 packets/s for 500 ms beaconing. The mes-sage rate of our proposed design is also shown in Fig. 5 foreach case, ranging from 2 to 3 packets/s. On average, our pro-posed design achieves better (or equal) tracking accuracy thanbeaconing. As observed from Fig. 5, the 100 ms beaconingmethod does not scale well when facing different traffic condi-tions, especially when there is a high density of cars. Thisuncontrolled method suffers from more consecutive losses (asindicated by Fig. 4b) and thus higher tracking error.

For the 500 ms beaconing, it does not suffer channel con-gestion in most cases (see one example in Fig. 4b) because itstransmission behavior is not as frequent as that of 100 ms bea-coning. Therefore, in some cases 500 ms beaconing achievesbetter tracking performance than 100 ms beaconing due tofewer collisions in the channel. However, its sending interval(one message every 0.5 s) is too long; thus, not enough infor-mation intensity is available for neighboring vehicles to tracka subject vehicle in real time. Therefore, the tracking error of 500 ms beaconing is still larger than our proposed algorithmin most scenarios. Although the channel congestion level of our proposed algorithm is similar to that of 500 ms beaconing(e.g., Fig. 4b), our proposed algorithm is more efficient anduses timely transmission of state information to eliminatelarge tracking error on neighboring vehicles. A higher trans-

mission probability is utilized when a subject vehicle suspects

other vehicles have higher error toward itself; otherwise, asubject vehicle leaves the shared channel to other vehiclesthat have larger suspected tracking error to broadcast safetymessages.

Furthermore, our proposed design adapts its power accord-ing to channel occupancy; thus, its tracking error stays flat fora certain range or distance (this range depends on vehiculartraffic condition and network congestion). Take the H1 case(Fig. 5), for example; the 95 percent Euclidean tracking errorstays roughly the same within a 90 m radius; beyond that, thetracking error goes up quickly as distance increases. In other words, the immediate neighbors have better tracking accuracythan others. This result is simply because the proposed designtemporarily gives up faraway cars during channel congestionand tries to maintain the same amount of state informationavailable to its closer neighbors. Our design pays the price of higher tracking error on remote vehicles to gain better track-ing accuracy on nearby vehicles to ensure safety.

Concluding RemarksCooperative active safety applications are among the mostimportant services provided by ITSs. Such cooperative safetysystems require that each vehicle track its neighboring vehi-cles in real time and detect hazardous situations. The uncon-trolled transmission of state information by each vehicle isshown to produce excessive data traffic that could choke the

vehicular wireless network and fail all applications. In this

Figure 5. Tracking accuracy (95 percent Euclidean error, in meters) in eight distance ranges for the 100 ms beaconing method (blue dashed curve), the 500 ms beaconing method (pink dashed curve), and the proposed rate-power control (green solid curve).

Case H1 (14 mph)

30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m

Case H2 (30 mph)

30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m

Case M3, Dir#1 (14 mph)

30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m


30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m


30 m0

4

0 9

5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m


30 m0

4

0 9

5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m


30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m


Rate of proposed design: 2.95 packets/s Rate of proposed design: 3.02 packets/s




30 m0

4

0 9 5 % e r r o r ( m )

3

2

5

1

60 m 90 m 120 m 150 m 180 m 210 m 240 m

100ms beaconing 500ms beaconing Proposed design



article we propose a joint rate-power control algorithm forbroadcast of self-information that enables neighbor trackingin VANETs. We evaluated the presented solution throughrealistic network and microscopic traffic simulations. We alsodefined a statistical performance measure for tracking accura-cy. To verify the robustness of the algorithm, we observed thetracking performance of the proposed algorithm that adaptstransmission power and rate in different traffic scenarios. Sim-

ulation results confirm that the proposed design is robust andcan considerably reduce the tracking error compared to thatof the de-facto solution (beaconing at 100 ms or 500 ms inter- vals). Our future work includes further performance analysis,implementation of the proposed design on a DSRC radiotestbed, and conducting hardware-in-the-loop evaluation.

References[1] Wireless Access in Vehicular Environments (WAVE) in Standard 802.11,

Specific Requirements: IEEE 802.11p/D2.01, Mar. 2007.[2] IEEE 1609.3 and 1609.4 trial standards for Wireless Access in Vehicular

Environments: Networking Services and multi-channel operations, Apr.2007.

[3] L. Cheng et al., “Mobile Vehicle-to-Vehicle Narrow-Band Channel Measure-ment and Characterization of the 5.9 GHz Dedicated Short Range Commu-nication Frequency Band,” IEEE JSAC , vol. 25, no. 8, Oct. 2007, pp.1501–16.

[4] Vehicle Safety Communications Consortium (VSCC), “Vehicle Safety Commu-nications Project, Task 3 Final Report: Identify Intelligent Vehicle Safety

Applications Enabled by DSRC,” 2005.[5] P. Gupta and P. R. Kumar, “The Capacity of Wireless Networks,” IEEE Trans.

Info. Theory , vol. 46, no. 2, Mar. 2000, pp. 388–404.[6] M. Torrent-Moreno et al., “Vehicle-to-Vehicle Communication: Fair Transmit

Power Control for Safety-Critical Information,” IEEE Trans. Vehic. Tech., vol.58, no. 7, Sept. 2009, pp. 3684–703.

[7] C. Robinson et al., “Efficient Coordination and Transmission of Data for Vehicular Safety Applications,” Proc. 3rd ACM VANET, Sept. 2006.

[8] S. Rezaei et al., “Tracking the Position of Neighboring Vehicles Using Wire-less Communications,” Elsevier Journal Transportation Research Part C:Emerging Technologies, special issue on Vehicular Communication Networks,

June 2009.[9] C. L. Huang et al., “Information Dissemination Control for Cooperative Active

Safety Applications in Vehicular Ad-Hoc Networks,” Proc. IEEE Globecom2009 , Nov. 2009.

[10] R. Sengupta et al., “Cooperative Collision Warning Systems: Concept Defi-nition and Experimental Implementation,” J. In te ll igen t T ransportat ionSystems, vol. 11, no. 3, July 2007, pp. 143–55.

BiographiesCHING-L ING HUANG ([email protected]) received B.S. and M.S. degreesfrom National Taiwan University, both in electrical engineering. His Master’sstudy in the Computer Science group of NTUEE focused on networking andmobile communication. He is currently pursuing his Ph.D. degree in the Systems

Engineering group, Civil and Environmental Engineering Department, University of California at Berkeley (UC Berkeley). His current research interests includecontrol and estimation over an unreliable channel, vehicular networks, and coop-erative active safety designs for intelligent transportation systems.

Y ASER POURMOHAMMADI F ALLAH is a postdoctoral research fellow at UC Berkeley,Institute of Transportation Studies (EECS and CEE Departments). His current research activities are in the areas of networked cyber physical systems and

wireless networking for intelligent t ransporta tion systems. He obtained his Ph.D.in electrical and computer engineering from the University of British Columbia in2007. His research has been supported by numerous awards including anNSERC postdoctoral fellowship, NSERC postgraduate scholarship, and BellCanada graduate scholarship.

R AJA SENGUPTA received his Ph.D. degree from the Electrical Engineering andComputer Science Department, University of Michigan, Ann Arbor, in 1995. Heis an associate professor in the Systems Engineering Program within Civil andEnvironmental Engineering, UC Berkeley, where he is also director of the PATH

Wireless Laboratory and deputy di rector of the Center for Col laborative Controlof Unmanned Vehicles. He is an Associate Editor of IEEE Control Systems andthe Journal o f In telligent Transportation Systems. He was Program Chair of theIEEE Conference on Autonomous Intelligent Networked Systems 2003 and Co-General Chair of the 1st ACM MOBICOM Workshop on Vehicular Ad-hoc Net-

works (2004).

H ARIHARA N K RISHNAN received his Ph.D. from the University of Michigan, M.S.from the University of Waterloo, and B.E. from Anna University, India. Currently,he is the thrust area lead on the GM research program on vehicle-to-vehicle(V2V) and vehicle-to-infrastructure (V2I) communications. He works on variousresearch projects in the area of V2X communication, including the Vehicle Safety Communications — Applications project conducted by GM, Ford, Mercedes,Toyota, and Honda under a cooperative agreement with the U.S. Department of Transportation. Prior to this, he was an assistant professor at the National Uni-

versity of Singapore from 1993 to 2000. He has over 60 publications in interna-tionally refereed journals and conference proceedings and over 300 citations.He serves as an Associate Editor for the IEEE Control System Society and Trans-portation Research — Part C.

Adaptive Intervehicle Communication Control for Cooperative Safety Systems

Documents

Transcript of Adaptive Intervehicle Communication Control for Cooperative Safety Systems