Deep Reinforcement Learning-Based Enhancement of SATMAC for Reliable Channel Access in VANETs

Vehicular ad-hoc network (VANET) is a core technology of the intelligent transportation system, enabling new applications such as road safety. The SATMAC is a recently proposed distributed TDMA-based MAC protocol for VANET scenarios. It provides a self-adaptive and reliable channel access by time slot occupancy adjustment and adaptive frame length. However, the actions of adjustment only relies on the limited and current network state to make decisions, which exists the problem of unstable decision-makings and leads to more packet collisions. In this paper, we utilize the deep reinforcement learning to revise the decision-making processes of SATMAC. Simulation result shows that the enhanced protocol has an average of 8% fewer packet collisions than the SATMAC protocol.