a context-based vehicular communication protocol - IDA.LiU.se

Accepted for publication at The 15th IEEE International Symposium on Personal, Indoor and Mobile Radio Communication (PIMRC2004), to be held in Barcelona, Spain, 5-8 September 2004

A CONTEXT-BASED VEHICULAR COMMUNICATION PROTOCOL Ioan Chisalita and Nahid Shahmehri Department of Computer and Information Science, Linköping University, S-581 83, Linköping, Sweden Phone: +46-13-28-2452, Fax: +46-13-281422, E-mail: {ioach, nahsh}@ida.liu.se Abstract – Research in traffic safety has indicated that active safety systems provide a better service to drivers when they use data about nearby vehicles. For supplying such information inter-vehicle communication is employed. In this paper we propose a distributed communication protocol that allows the vehicles to organize the communication network in a decentralized manner. For information dissemination we use an anonymous contextbased broadcast protocol. The receivers determine whether they are the intended destination of sent messages based on knowledge about their local environment. Simulation results indicate that the proposed protocol performs well in terms of communication performance and filtering of the received information. Keywords: inter-vehicle communication, context-based protocol, anonymous broadcast. I. INTRODUCTION Accident statistics reveal that every year half a million people loose their lives due to traffic accidents [1]. Extensive costs are also related with car crashes [1][2]. Therefore, traffic accidents became a major issue of concern for the society and their reduction is considered very important by the automotive industry and transport safety administrations [3]. Consequently, efforts were directed towards studying and designing technologies that make it possible to develop systems that can help in avoiding accidents [4][5]. Within these developments, the intervehicle communication, used for providing the exchange of data between vehicles, is considered a key technology [2][6]. The use of onboard safety systems that analyze not only data describing the immediate surrounding of a vehicle, but also data provided by other vehicles for accurately detecting hazards in traffic, is predicted to lead to a major reduction of the amount and severity of crashes [6][7]. The design of an efficient vehicular communication system poses a series of technical challenges. The communicating hosts are moving at high speed and the communication links need to be established frequently. This requires a highly dynamic management of the links. Strict requirements on the latency of the data dissemination need to be fulfilled for traffic safety applications. The communication may also imply exchange of data between hosts whose identities are not known by default. Other important aspects that need to be considered when designing a vehicular communication system are the extent of the service area, the required bandwidth, the forwarding of information and the adaptability to the environment [8].

Several approaches using dedicated infrastructures, cellular networks or direct communication have been considered for developing vehicular communication systems. In [8] we have provided an investigation of these solutions, and selected the direct communication between vehicles as the most appropriate alternative for supporting safety applications. Within this solution a vehicle equipped with a communication device (e.g. a transceiver) is referred to as a host. These hosts exchange safety-related data (e.g. vehicle location, road status) that is used by in-vehicle active safety systems (e.g. collision warning and avoidance systems) for determining risks associated with road traffic. II. RELATED WORK The vehicular network we address in this work is an ad-hoc network. For this type of network, there exist proactive and reactive routing protocols intended to provide efficient data dissemination. We introduce these solutions and discuss impediments that can hinder their use in vehicular networks. Examples of proactive protocols are Wireless Routing Protocol (WRP) and Destination-Sequenced DistanceVector routing protocol (DSDV) [9]. These protocols require a host to maintain consistent routing information that describes how a packet can be transmitted throughout the network from one host to another. Each host maintains a number of routing tables reflecting the host’s view of the network. For consistency reasons, each host needs to announce to the other hosts the modifications of its view. Examples of reactive protocols are Ad-hoc On-demand Distance Vector routing protocol (AODV) and Temporally Ordered Routing Algorithm (TORA) [9]. These protocols create routes only when a sender host needs to transmit data to another host. The establishment of a route is performed using route discovery mechanisms that are specific to each protocol. A route is maintained until the destination is no longer reachable or until the route is not needed anymore. An important feature of some routing protocols for adhoc networks is the use of geographic information (i.e. host position) when forwarding data [10]. An example is the Location-Aided Routing (LAR) where the destination of a packet is indicated by a combination of an identifier of the destination and an estimation of its position. One of the issues that hinder the applicability of the above mentioned protocols to vehicular communication is the anonymity of the hosts. In traffic safety applications a vehicle needs to receive information from a number of vehicles in its proximity rather than receiving data from a specific vehicle. It is not feasible for a vehicle to maintain


the identities of any other vehicle that may receive data sent by it. Consequently, it is an advantage in not requiring a sending vehicle to know the addresses of nearby vehicles. It is still possible to provide a vehicle with the identities of neighboring vehicles using dedicated packets. However, this may lead to communication overload due to the transmission and forwarding of a large number of packets. It also introduces latency in providing the needed safetyrelated data to vehicles. Therefore, routing protocols that require the sender to know the identity of the receiver do not work well for safety-oriented vehicular communication. The vehicular environment is very dynamic as vehicles frequently change their driving orientation and randomly exit and join the roads. Consequently, the vehicular network can often be fragmented. Therefore, the maintenance of a consistent view of the network as required by proactive protocols would be extremely difficult. It would also lead to communication overload and to high delays in data provision since a large number of route updates need to be performed. The dynamic of traffic also hinders the use of protocols that require position estimation when forwarding data since these can be inaccurate. The issues mentioned above indicate the need for developing a new vehicular communication protocol that does not require knowledge of the network topology (e.g. neighboring hosts addresses) or updates of routes. III. PROPOSED PROTOCOL The vehicular communication protocol needs to provide the exchange of safety-related data between vehicles. Since this information is used for determining the risks in traffic, it needs to be disseminated in a timely manner. The protocol also needs to provide reliable communication between a possibly large number of hosts. The random establishment of links and frequent link failures need to be considered. Forwarding of data may be necessary if line-of-sight communication is not possible. Further, the vehicles should be able to select the important data from the received information. Also, the size of the exchanged data should be reasonably low for not overloading the communication. Furthermore, the vehicles need to be able to send both regular data describing their momentary view of the traffic and diverse notifications about events in traffic. Considering these requirements, we propose an anonymous context-based broadcast protocol. Our protocol can be seen as a reactive protocol where the communication is performed using anonymous broadcast for disseminating short messages between hosts. Each host broadcasts messages that can be received and accepted by other hosts within its transmission range. We did not consider the addresses of the communicating hosts to be a priori known and our protocol requires the receivers to analyze the content of the received messages for determining if the embedded data is of importance. When the information in a received message is important for traffic safety, the receiver

accepts the message and uses the data contained by it. Otherwise, the message is dropped. Based on the content of the message, the receiver can decide to forward the message. We refer to the process of analyzing, considering, dropping or forwarding a message by a host as information filtering and forwarding. This process is implemented by algorithms that take into consideration elements related with the momentary traffic situation of the receiving host (e.g. host position, relative distance to sender). Therefore, the communication protocol is called context-based. The proposed protocol makes it possible to achieve a small latency in data provision and to accommodate random communication between hosts. The use of short messages facilitates the efficient utilization of the communication channel. A reliable dissemination of safety data can also be achieved when messages are sent frequently enough for the hosts to receive them at short intervals. By defining different types of messages the protocol provides both regular data exchange and event-based communication. The information filtering assures that the receiving hosts are provided with data of importance for their momentary traffic situation. Based on the received data, the hosts aggregate in local networks. In this work we have considered a host-oriented approach for organizing these networks [11]. In this approach, each host defines, constructs and maintains its own local network. Thus, each host takes the role of an organizer host by analyzing the information received from other hosts and deciding which of them should belong to its own local network. This group of hosts is continually updated. With regard to diverse traffic situations it is common that vehicles in the close vicinity have data of interest for each other. Therefore, we assumed a limited space and composition for a local network. The number of hosts that can coexist at any moment in a local network was limited to a Maximum Number of Hosts (MNH, e.g. 15). The geographical extent of a local network was limited to a Service Area Threshold (SAT, e.g. 300 meters). In our communication protocol we defined different types of messages for data dissemination. Since the constant update of data describing the traffic is needed for each host, we defined Basic Safety Messages (BSM) that are sent over regular time intervals. These messages contain information that describes the sending host’s momentary view of the traffic situation. Further, there are traffic events that can trigger the sending of notifications. Examples are the detection of an accident or when the driver requires some specific data. For a host to be able to send notifications about such events we defined dedicated messages (e.g. warning, infotainement, and service request messages). In our proposal we assumed that each host is equipped with accurate positioning devices such as Global Positioning System (GPS) receivers. This assumption is feasible since the market penetration of positioning devices is rapidly growing [2], and the accuracy of these devices can be less than one meter [3][6].


Security and privacy are issues of concern in vehicular networks and we have identified different security requirements for our proposed protocol. However, in this paper we did not address security aspects and rather focus on the functionality of the communication protocol. IV. THE PROTOCOL FUNCTIONALITY The functionality of the protocol relies on the following rules for information management: • Acceptance of a message by a host: a received basic safety message is accepted if the information filtering process indicates that the message contains data of interest. Messages other than basic safety messages and warning messages are not accepted if the receiving host does not maintain data about the sender at the receiving moment. • Local maintenance of data describing senders: a host H maintains data about a sending host S as long as within a time interval TR another basic safety message from S is accepted by H. If this event does not happen, H removes all information about S. Each time a new basic message is received and accepted by a host H, the time interval TR related with this record is reinitialized. • Transmission of basic safety messages: a host regularly transmits basic safety messages at short intervals. A timer TB is associated with the transmission of these messages and initialized each time a message is transmitted. When this timer triggers, a new BSM is created and transmitted. • Transmission of messages other than basic safety messages: messages other than BSM are transmitted as the result of events that appear in traffic. The conceptual functionality of the communication protocol for reception of messages is illustrated in figure 1. Received Message

Information Filtering

Message Not Accepted

Accept Message

Drop message

Communication System Database

Write data in database Communication System Timer

Delete outdated records

Figure 1. Communication protocol – conceptual functionality

In this paper we focus on the exchange of Basic Safety Messages (BSM), which structure is presented in figure 2. TID is a type identifier of the host, SID is the sender host identity and MSqn is the message sequence number. The type identifier can be used for characterizing hosts other than vehicles (e.g. roadside servers [11]). However, in this paper we consider all hosts to be vehicles. PS1 and PS2 are two consecutive most recent positioning data of the host and VSp is its speed and VHd its heading. VSt is the status of the host - we currently consider two generic values, good and poor. HIDs are the identities of the other hosts from the same local network as the sender, and NH is their number. RT is the road type (i.e. divided or undivided), RID is the road identity and RSp is the speed limit on the road. RSl is

the road slipperiness indication - we currently consider two values, slippery and dry. The fields marked with “*” are used if the data can be provided by some onboard system. TID

SID

MSqn PS1

PS2

VSp

VSt

VHd*

HIDs

RID*

RSl*

NH

RSp*

RT*

Figure 2. Basic safety message structure

Another important type of messages are the Warning Messages (WAM), which provide indications about events that appear in traffic. These messages include the position of such events and a description of them. The proposed protocol provides a flexible mechanism for the host to accept warnings based on their importance [8]. The warning messages are subject to retransmission based on counters. Due to the anonymous broadcast and the dynamic selfmanagement of the network, the need for forwarding messages (i.e. routing) is reduced to those cases when the hosts can not directly exchange data due to transmission problems (e.g. interference, shadowing). Still, routing is necessary and we propose the use of an automatic mediation mechanism based on the characteristics of hosts that are able to directly communicate. As shown in the BSM structure, each host indicates the other hosts that are part of its local network. When a host accepts a message, it analyzes its local network composition to determine if there are hosts that may be interested in this message. These are hosts that also indicate that they did not receive (yet) the message. As analysis method we used a basic approach that compares the distance between the current sender and the other hosts with a threshold value that was set to SAT. Thus, if this distance is lower than SAT the accepted message is retransmitted. The information filtering is performed using decision mechanisms implemented as IF-THEN rules based on analyses of traffic situations and guidelines provided for collision avoidance systems (e.g. as in [3][4][5]). These rules take into consideration a number of parameters. Primary parameters are provided by the receiver’s internal sensors and by data contained within accepted BSMs. They are the identity, velocity, heading, position records and status of the sender and the receiver, the properties (i.e. type, slipperiness and speed limit) of the roads they travel on, and the number of hosts in the receiver’s local network. Derived parameters, obtained by analyzing primary parameters are: • Relative distance between sender and receiver. • Similar heading: we estimated if two vehicles travel in the same direction by evaluating the difference between their headings and compare it with a threshold value. If the difference is smaller than this value, the vehicles are considered to travel in the same direction. We have used 90o as the threshold value (the reference was North). • Hosts on the same road: we estimated if two hosts are situated on the same road. The employed algorithm analyses the movement vectors of the vehicles between their two successive positioning records. The angle between these vectors is evaluated and, if close to 0o or 180o, the vehicles are considered to be on the same road.


• Relative positioning of two hosts: we evaluated if a vehicle is behind or ahead of another vehicle. We used an algorithm that analyses the angle between the vector given by the two consecutive positions of a vehicle and the vector between the latest positions of the considered vehicles. This angle is compared with 90o and indicates if a vehicle is ahead of or behind the other vehicle. • Possibility of route contention: we evaluated if two vehicles can meet at an intersection. Using the two consecutive positions of the vehicles we first determined if an intersection point exists and calculated its coordinates. We then evaluated using the latest values of the vehicles’ velocities when they can arrive at the intersection point and compare these values. We then relaxed the requirement on strict equality of these time moments, and introduced a contention interval for modeling the dimension of vehicles and small variation of their speed. We note that some of the derived parameters can also be determined using positioning data and digital maps if these are available on vehicles [8]. A set of information filtering rules is presented in figure 3. The filtering process is intended to prevent a vehicle from receiving useless or misleading data. We note that the rules presented here are a “proof of concept”, being an example that we used in our work. The main aspects considered for defining these rules were: a limited number of vehicles in close proximity usually have important data, vehicles in front or behind on the same lane have data of interest, vehicles coming from an opposite direction may constitute a danger on undivided roads, and vehicles may collide if they arrive at an intersection at the same time (e.g. [3][4][12]). Exclusion rule: The Euclidean distance between the sender and the receiver is less than SAT. Regular rules: • The sender and the receiver travel on the same road and have similar heading and [the receiver maintains data about less that MNH hosts or the sender is closer to the receiver than the farthest host previously considered]. • The sender and receiver travel on the same undivided road and have different headings and the sender is ahead of the receiver and [the road is slippery or the sender status is poor or the sender speed is excessive] and [the receiver maintains data about less than MNH hosts or the sender is closer to the receiver than the farthest host previously considered]. • The sender and the receiver travel on different roads and a route contention is detected and [the receiver maintains data about less than MNH hosts or the sender is closer to the receiver than the farthest host previously considered].

Figure 3. Information filtering rules

When both the exclusion rule and one of the regular rules hold at the same time, the sender is considered of interest and the receiver accepts the message and uses the data.

which were inputs for GloMoSim. We have generated these traces by constructing a traffic simulator that implemented the car-following model proposed in [13]. We defined four metrics for evaluating the performances of our protocol: message delay, packet collision, send errors and information filtering rate. These metrics were evaluated for exchange of basic messages and are described below. The message delay represents the time between the sending of a message and the moment when it is accepted by a host. The send errors represents the number of messages that could not be sent by a host due to transmission problems. The packet collision represents the sum of all collisions observed by all nodes during a simulation run. We note that the same collision can be counted several times. The information filtering rate is the ratio between the number of messages accepted and the number of messages received by a host. It is a measure of how much useful information can be “extracted” from the received data. For each simulation run, the results were averaged over the number of hosts. As free parameters we used the transmission interval for basic safety messages (i.e. TB), the network load (i.e. density of hosts) and the service communication area (i.e. SAT). We discuss in the following the results obtained by simulating for 120 seconds the communication between vehicles traveling on a 5-km bi-directional road and having the maximum values for speed, acceleration and deceleration as 36 m/s, 1 m/s2 and -4,5 m/s2 respectively. MNH was 15 and SAT was 300 m. When not modified, the vehicle density was 6 vehicles/km/lane. All vehicles were considered equipped with transceivers modeling signal-tonoise bounded radios at 2.4 GHz with a 2 Mbits bandwidth, 9dBm as transmission power and non-persistent CSMA as the medium access control (MAC) scheme. The propagation model was free space. The initial time point for sending basic safety messages was randomized between 0 and 0.1 s. Figure 4 shows the message delay for modification of the communication service area and the transmission interval. We usually obtained small values for delay, which indicates that timely provision of safety-related data is possible. Higher delay values were obtained for low values of the transmission interval (i.e. less than 0.1 s). This indicates that the hosts were not able to efficiently use the medium and the sent messages collided frequently. For small transmission intervals, an extensive use of routing was noticed and led to an increase of traffic and larger delays. The message delays increased when the service area was larger than 300 meters and when the load density was high. This shows that the protocol works better for small (e.g. 300m) local networks. 20

20 15 10 5 0

15

V. SIMULATION ENVIRONMENT AND RESULTS

10

To investigate the performances of our protocol, we implemented it within the application layer of the GloMoSim mobile network simulator. Within our simulations we used realistic movement traces of vehicles,

0

5

0 0

200

400

C o m m unic a tio n A re a [m ]

600

1

T ra ns m is s io n In te rva l [s e c ]

Figure 4. Delay for accepted basic safety messages

2


The collisions and send errors for modification of the transmission interval are represented in figure 5. The number of collisions and send errors followed the same pattern. Since the vehicles tried to send their data very often when using small values of the transmission interval (i.e. less than 0.05 sec), a large number of collisions and send errors occurred. However, when the transmission interval was increased (i.e. over 0.1 s), the number of collisions and send errors decreased significantly. The number of collisions for modification of the load density and the communication service area are presented in figure 6. The number of collisions was low for communication areas less than 300 meters. An extensive increase of the number of collisions was obtained when increasing the load density. This was expected since more hosts competed for the transmission medium. Still, the large number of collisions obtained for high load densities indicates that more advanced MAC schemes and techniques (e.g. jittering) for avoiding synchronization of hosts need to be investigated. 2 50 0 0 20000 150 0 0 10 0 0 0 50 0 0 0

10 0 0 0 0 80000 60000 40000 20000 0 0

1

2

0

1

2

Figure 5. Send errors and Collisions – Transmission Interval 12 0 0 0 10 0 0 0 8000 6000 4000 2000 0 0

10

Lo a d De ns ity [ve h/km /la ne ]

20

10 0 0 800 600 400 200 0 0

200

400

600

C o m m unic a tio n Are a [m ]

Figure 6. Collisions – Communication Area and Load Density 0 ,4

0 ,6

0 ,3 0 ,2

0 ,4 0 ,2

0 ,1 0

0

0

200

400

600

C o m m unic a tio n Are a [m ]

Simulations were conducted for variations of parameters of the communication protocol, and variations of the road traffic dynamics. The obtained results indicate that the communication between vehicles can be efficiently performed under certain settings of the proposed protocol. Future work will focus on extensions of the communication protocol. Techniques that allow an improvement of the communication performances and the dissemination of warning messages need to be further investigated. Extensions and a formal definition of the information filtering rules are also of interest. REFERENCES

Tra ns m is s io n Inte rva l [s e c ]


VI. CONCLUSION AND FUTURE WORK We have presented an approach to disseminating safetyrelated information in vehicular networks. We proposed a protocol that requires the receivers to analyze the content of exchanged messages for deciding if they are the intended destination of a message. This filtering of received data is performed considering the momentary traffic situation of the receiving vehicle.

0

1

2


Figure 7. Information filtering rate

Figure 7 presents the information filtering rate when modifying the communication area and the transmission interval. The metric had lower values for small transmission intervals and stabilized around 0.25 for intervals larger than 0.1 s. For small communication areas, the metric exhibited small values, and then stabilized around 0.30 for larger areas (i.e. more than 300 meters). Further, the increase of the load density (not presented in figures) determined a decrease of the information filtering rate since more messages of no interest arrived at a host. Generally, in all simulations a major filtering of messages (i.e. 25%-35%) was noticed. This indicates the advantage of using the filtering technique when receiving data about vehicles in traffic. We also noticed an improvement of the bandwidth utilization when comparing our proposed protocol with a classic floodingbased broadcast protocol under the same conditions.

[1] International Road Traffic and Accident Database, “Road Traffic Data” and “Accident Traffic Data”, Feb 2004. [2] Jones W. D., “Keeping cars from crashing”, IEEE Spectrum, vol. 38, pp. 40-45, Sep. 2001. [3] Miller R., Huang Q.” An adaptive peer-to-peer collision warning system”, Vehicular Technology Conference, Birmingham, USA, pp. 317-321, May 2002. [4] R. Kiefer, D. LeBlanc; M. Palmer; J. Salinger; R. Deering; "Development and Validation of Functional Definitions and Evaluation Procedures for Collision Warning/Avoidance System", National Highway Traffic Safety Administration, USA, Aug 1999. [5] Talmadge S., Chu R., Eberhard C., Jordan K., Moffa P., "Performance Specifications for Collisions Avoidance Systems for Lane Change Crashes", TRW Space and Defense, USA,Aug. 2000. [6] Kato S., Tsugawa S., Tokuda K., Matsui T., Fujii H., “Vehicle control algorithms for cooperative driving with automated vehicles and intervehicle communications”, IEEE Transactions on Intelligent Transportation Systems, no. 3, pp. 155-161, Sept. 2002. [7] Andrisano O., Verdone R., Nakagawa M., “Intelligent transportation systems: the role of third generation mobile radio networks”, IEEE Communication Magazine, pp. 144-151, 09/2000. [8] Chisalita I, “Safety-oriented communication in mobile networks for vehicles” (tentative title), Licentiate Thesis, Linköping University, Sweden (forthcoming 2004). [9] Royer E.M., Toh C.K., “A review of current routing protocols for ad hoc mobile wireless networks”, IEEE Personal Communications, Vol. 6, Issue 2, pp. 46-55, Apr 1999. [10] Mauve M., Widmer A., Hartenstein H., “A survey on positionbased routing in mobile ad hoc networks”, IEEE Network, Vol. 15, Issue 6, pp. 30-39, Nov.-Dec. 2001. [11] Chisalita I., Shahmehri N. “A peer-to-peer approach to vehicular communication for the support of traffic safety applications”, ITSC 2002, Singapore, pp. 336-341, Sept. 2002. [12] Najm W., Sen B., Smith J., Campbell B., “Analysis of Light Vehicle Crashes and Pre-Crash Scenarios Based on the 2000 General Estimates System”, National Highway Traffic Safety Administration, USA, Feb 2003. [13] Krauss S., “Microscopic Modeling of Traffic Flow”, PhD thesis, University of Cologne, Apr. 1998.