Abstract:
Vehicular networks are characterized by high mobility, and frequent topology changes occur in them. Due to this, the links among the vehicles are volatile and can exist for a short amount of time compared to other networks. Thus, in order to develop efficient communication among vehicles without packet losses, knowledge of the lifetime of links is very important. Existing works in this domain have considered only sensor-based readings to predict the link lifetimes. However, we foresee that predicting link lifetimes only using sensor readings can bring less correct estimations since they do not have any information on how the vehicle is likely to behave in the future time step. In order to cater to this problem, we propose to obtain driving action outputs: steering angle and throttle with brake values to optimize sensor readings such that the relative motion information is more futuristic. Specifically, we compute jerk using changes in throttle and compute average values for acceleration and velocity by integrating them and combining them with the sensor readings. Next, when the steering angle changes compared to the previous timestep, we compute new components of the motion components considering the steering angle change. We proposed to use non-linear optimization to model the link lifetime estimation task, considering relative motion between vehicles, incorporating jerk, and adjusting with driving outputs. However, due to the high computational complexity of that approach, we also propose a deep neural network-based suboptimal approach in order to reduce the computational complexity. The system is simulated using CARLA for autonomous driving using a pre-trained model and NS3 for vehicular communication. The results show that the proposed models’ link lifetime predictions are much closer to real link lifetimes (mean absolute error < 150 ms) compared to existing approaches; thus, the proposed technique can be utilized to improve vehicular communication.
