10.2 Related Works on Broadcasts in VANETs
Earlier works on broadcast protocols in VANETs mostly assumed the ideal propagation model, i.e., whether a packet reception is successful merely depends on a fixed transmission range. Examples include the UMB (Korkmaz and Ekici 2004) and SB (Fasolo et al. 2006). They designate the furthest node that receives a packet as the relay node to maximize the one-hop progress (Korkmaz and Ekici 2004) or minimize the one-hop delay (Fasolo et al. 2006).
However, as shown by empirical studies, the wireless channel in VANETs is far from perfect (Taliwal et al. 2004). Instead, channel fading is the primary challenge and has a major impact on broadcast reception rates (Torrent-Moreno et al. 2004, 2006). The suggested realistic propagation model on the highway is the Nakagami model (Torrent-Moreno et al. 28–25 Sept., 2005), where the packet reception probability (PRP) of single broadcast decreases with the distance, which aggravates the “broadcast storm” problem. The probability-based methods (Ni et al. 1999; Wisitpongphan et al. 2007b) simply let each node rebroadcast a packet with some probability. However, the probabilistic forwarding decision still results in redundant rebroadcasts and does not solve the broadcast storm problem.
10.2.1 Opportunistic Forwarding in VANETs
As we have seen in the previous chapters, opportunistic forwarding (Biswas and Morris 2005; Zorzi and Rao 2003) is a promising way to deal with lossy links in multihop wireless networks. It exploits the “spacial diversity” enabled by the broadcast nature of the wireless medium, so that the probability of at least one node forwarding a transmitted packet is greatly increased. In this way, each transmission is useful with high probability, and the forwarding delay and the number of transmissions can be reduced.
The concept of opportunistic forwarding has recently been applied to event-driven emergency message broadcast in VANETs. Existing works include contention-based dissemination (CBD) (T-Moreno 2007), emergency message dissemination for vehicular environments (EMDV) (Torrent-Moreno et al. 2009), contention-based forwarding (Torrent-Moreno et al. 2006), OB-VAN (Blaszczyszyn et al. 2008) and location-based flooding (Oh et al. 2006). In these works, in order to maximize the “hop progress” of the message in its each rebroadcast, the common idea is to employ a contention process where the furthest node in the message dissemination direction is opportunistically selected. This is often achieved by letting each node that receives the same rebroadcast packet set up a contention timer, in which the backoff time is inversely proportional to the distance of that node to the sender. As a result, the nodes located nearer to the sender backoff longer times and quit contention whenever they hear rebroadcasts (or ACK signals) from a node that has larger hop progress.
However, under the presence of lossy wireless links, the previous schemes could incur unnecessary duplicate rebroadcasts (especially under congested channel) (Torrent-Moreno et al. 2009), which takes up the previous bandwidth of the VANET and may adversely affect the performance of other messaging services such as periodical safety beacon messages. This is because the rebroadcast of a vehicle located furthest in the WM dissemination direction may not be heard by other contending nodes due to channel fading. Although these duplicate rebroadcasts can enhance the reliability of WM reception, currently the “duplicate level” is often controlled by adjusting the threshold count of received duplicate messages for each contending node to decide whether to quit the contention. However, it is very difficult to select the appropriate parameters so that a desired WM reception reliability level is achieved, without causing more redundant rebroadcasts.
In addition, the contention processes could result in undesirably large end-to-end WM propagation delay. For nodes to decide whether to participate in contention, there is always some “contention range” chosen by each forwarder upon its rebroadcast. If this range is too large, even the furthest node would wait for a long time before rebroadcast since the contention count-down time tends to be prolonged. Towards solving this problem, in EMDV (Torrent-Moreno et al. 2009) Torrent-Moreno et. al. proposed to first designate the furthest node (who would rebroadcast without delay) within a “forwarding range” that is shorter than the communication range, and if that node does not receive the rebroadcast then other nodes continue participating in normal contention processes like that in CBD (T-Moreno 2007). However, the issue of how to select an optimal forwarding range has not been addressed.
In summary, the inefficiency of the existing schemes results from two aspects. 1. Nodes make forwarding decisions based on heuristic guesses of whether neighbors have received the same packet or not. The reliability requirement (PRR) of the WM propagation is not considered, and no optimizations have been made so far. 2. In the coordination mechanisms, rebroadcast messages are employed as “implicit acknowledgements”, which always subject to channel fading and collisions. Thus it is hard to suppress unnecessary redundant rebroadcasts effectively. In this chapter, we solve these two problems accordingly, by 1. explicitly considering the reliability (PRR) as one of the relay node selection criteria, in that we minimize the number of rebroadcasts to satisfy a given PRR requirement; 2. designing a more reliable and efficient broadcast coordination mechanism, where a short, explicit “broadcast acknowledgement” (BACK) is send out (at the base rate) before each WM's rebroadcast, which has larger communication range than the normal WMs. The BACK not only effectively suppresses redundant rebroadcasts but also clears the channel for the rebroadcast. With BACK, the contention (relay selection) processes can be optimized for higher reliability and lower end-to-end delay.
10.2.2 The Reliability Issue in VANET Broadcast
ElBatt et al. 2006 studied one-hop periodic broadcast in cooperative collision warning applications. They characterized the tradeoffs between the packet inter-reception latency, application broadcast rate and transmission range. For multihop WM broadcasts, Resta et al. (2007) analyzed theoretically the tradeoff between vehicles' probability to receive a WM within time t and the link-level reliability. But their channel model is oversimplified, and no distributed protocol has been proposed. On the other hand, in this chapter we cast insight on the application-level tradeoff between packet reception performances and the overall transmission overhead, under a realistic channel model.
10.2.3 Broadcast in Partitioned VANETs
VANETs turn out to be partitioned (or disconnected) frequently, especially under sparse vehicular traffic, which falls into the delay-tolerant network (DTN) paradigm. Ilias and Cecilia (2007) proposed an opportunistic event dissemination protocol that employs cache and periodic replay mechanisms to keep a message alive in an area. In Abuelela et al. (2008), the authors proposed a routing protocol that uses local routing in connected clusters and store-carry-forward at cluster boundaries in order to reduce latency and overheads. In this chapter, we also extended OppCast into a DTN-compatible broadcast scheme, where the protocol adaptively switches between normal dissemination and store-carry-and-forward modes based on vehicles' local traffic densities. Recently, Tonguz et al. (2010) proposed DV-CAST, which extended the work in Wisitpongphan et al. (2007b) to handle network disconnections. The rebroadcast decisions in DV-CAST are also made based on local topology information; however, there are no reliability (or broadcast success rate) guarantees in DV-CAST. Unlike all the previous works, in our extension we focus on finding the optimal switching traffic density beyond which unnecessary redundant transmissions will be incurred, and below which desired PRR requirements cannot be fulfilled.