Simulation setup of PCN application on the V2V domain
The simulation of vehicular communications is a valuable tool to assess the performance, requirements, and feasibility of the VANET applications. Since the implementation of many cooperative systems require high costs, a complex infrastructure and a high level of coordination and logistics, it is very common that these projects are first evaluated using a simulation approach. Recent literature of VANETs refers to hybrid simulation as the most accurate technique to analyze Inter-Vehicle Communications (IVC) [18]; for this sake, coupling a network and microscopic traffic simulator provides a representative model of the mobility and communication patterns in urban scenarios [19].
We have performed a cross-layer simulation, which includes the physical; media access control (MAC), and the routing and application layers to evaluate the performance of our design. Considering this approach, we want to validate three concepts: First, that DRG is an accurate protocol for safety applications with low latency demand, and that it is a scalable routing mechanism for other applications with similar dissemination requirements. Second, that DRG is effective even for larger urban areas. And third, what is the impact of the PCN application on road traffic, once an accident has been detected and the neighbor drivers are successfully notified. An accurate integration of the application functionality and the effectiveness of the routing protocol leads to the success of the application, hence, the applications can positively influence the traffic flow and the drivers' behavior and their interactions.
Simulation scenario
The simulation is built upon Veins [18], an open source vehicular network simulation framework. Veins makes use of the interface called Traffic Control Interface (TraCI) of traffic simulation suit SUMO Simulation of Urban MObility [20] and OMNeT++ [21]. Our city-scenarios and geo-data come from the OpenStreetMap project, an open data repository for geographical information. The microscopic model employed is the car-following model proposed by Krauss [22], this model allows the simulation of platooning of vehicles and reproduces the drivers' behavior. This model provides the driver’s perfection parameter σ to adjust the accuracy of the drivers' behavior; in our scenario σ was set to 0.5 to add dawdle to the drivers' responsiveness; furthermore, the accidents have been manually set up.
We have deployed vehicles of two types: cars and trucks. Table 2 depicts the road traffic parameters and describes characteristics of the car and truck traffic flows. For the lower layers of the wireless communication model we have employed the 802.11p MiXiM package, which implements the Dedicated Short Range Communications (DSRC) standard. We have considered all vehicles equipped with 802.11p DSRC radio on board and we assumed that all vehicles participate in the ad-hoc network. The Car-to-X (C2X) communication parameters are summarized in Table 3.
Table 2 Mobility Model Parameters
Table 3 Simulation Framework Parameters
Two metrics have been selected in order to assess the performance of the geocast routing mechanism: End-To-End Delay and Packet Reception Rate (PRR).
-
End-to-End Delay: it is the time elapsed since a packet is sent by the application layer at the source node until the recipient node’s application layer receives the packet [23]. The latency is a common metric used to show the effect of larger areas to cover and also the impact of nodes on the performance of the protocol.
-
Packet Reception Rate (PRR): it is defined as the percentage of nodes that successfully receive a packet from the tagged node given that all receivers are within transmission range of the sender at the moment the packet is sent out [24]. It is defined by Khan et al. [25] as a percentage given by:
$$ {\displaystyle {PRR}_{(ZOR)}}=\frac{{\displaystyle \sum NodesReceivingDataPackets}}{{\displaystyle \sum NodesWithinTheZOR}} $$
(1)
Simulation results
As mentioned before, the critical latency for safety applications is up to 100 ms. Figure 3 shows the latency obtained for diverse coverage areas that vary from 500 m to 4000 m; the impact on the end-to-end delay can be observed, as the zone of relevance is larger. Larger distance to the intended receivers require more hops, therefore it takes longer to propagate the messages over the network. It is important to mention that even though the latency increments, the nodes up to 4000 m are receiving the geocast packets within the expected time for the PCN application or other safety applications. This reflects the efficiency of the protocol for safety applications. We observe that for 1500 m – the zone of relevance for most safety applications – the efficiency is an 85 % lower (14.85 ms) than the maximum delay accepted at the ZOR.
The effect of the traffic density is depicted in Fig. 4. It can be observed that the end-to-end delay gradually increases as the number of nodes increments, which is an expected behavior. When the traffic is denser, DRG meets the expected delay requirement - less than 100 ms - in the zone of relevance for PCN application (1500 m). Moreover, the behavior of the end-to-end delay is similar to the latency variation as a function of the zone of relevance shown in Fig. 3, where large distances do not significantly affect the end-to-end delay; these characteristics show DRG to be a stable protocol, because it maintains a high performance for an incremental variation of the nodes density and the size of the zone of relevance.
The PRR has been compared versus the single hop broadcast scheme, where the maximum range for simple broadcast is up to 300 m. It can be observed in Fig. 5 that, when the zone of relevance is set as 500 m, the PRR achieved 100 %, and for 1500 m, the PRR decreases slightly to 98.8 %, showing a suitable performance for collision warning applications like our proposed PCN application. For larger distances, between 1500 m to 3500 m, the PRR gradually decreases as expected. Further, when the ZOR is set to 4000 m, the PRR is still near to 80 %, which is considerable for traffic applications demanding a bigger area of interest to cover. We have also tested the impact of the density k (veh/km) on the PRR for a zone of relevance set to 1500 m. The traffic density is a determinant macroscopic variable, which gives an idea of the degree of congestion of a road segment when it is analyzed in combination of the flow volume and average speed of the traffic stream. In this case, Fig. 6 shows that the increment of the number of nodes does not affect the PRR, which maintains a constant value of 98.8 %.
The performance of PCN application has been evaluated through the PRR also as a function of velocity, in order to validate the effectiveness of the application for different vehicular velocity. Fig. 7 shows the impact of the velocity of the vehicles in the zone of relevance. For urban areas where the typical speed is 60 Km/h, DRG protocol shows high rate of the recipient nodes up to 80 km/h. It can be observed that vehicles do not affect the PRR until 80 km/h; which means that drivers on urban roads and highways effectively received PCN notification messages.
We have assessed the velocity and acceleration of the vehicles nearby the accident’s location to show how the speed varies after the reception of the WaveAppMsg packets of the PCN application. Therefore, we have measured the variation of velocity as a function of the simulation time. Fig. 8 shows five nodes in the period of time starting in 390 s. At 400 s, the first node on the road (Node 0) abruptly stops, which is detected by the application layer as an accident, and it triggers the DRG forwarding mechanism. Nodes 1 to 4 receive the PcnAccidentMsg packets and decrease their speed until they completely stop. After 30 s, the accident is solved and Node 0 sends PcnAccidentResolvedMsg packets at the 430 s, then Nodes 1 to 4 accelerate and traffic returned to normal. Such a behavior demonstrates the effectiveness of the PCN design, regarding the harmonization of the velocity and acceleration of the vehicles within the zone of interest. The speed of the stream is gradually reduced, in contrast to the first vehicle, the acceleration of the vehicles in front the incident also decreases smoothly. It can be assumed that after the drivers received the notification messages (few seconds after the incident detection) the break system was activated; Fig. 8 demonstrates the aforementioned behavior.