Deep Reinforcement Learning Based Wireless Resource Allocation for V2X Communications

The shortage and low utilization of air-interface spectrum resources have always been the bottleneck of the development of vehicle-to-everything (V2X) communications. In this paper, we investigate the issues of resource blocks (RBs) sharing and vehicle transmission power allocation under orthogonal frequency division multiplexing (OFDM) technology to improve the utilization of spectrum resources. In order to meet the high data rate of vehicle-to-infrastructure (V2I) links and the high reliability of vehicle-to-vehicle (V2V) links, the optimization problem is defined as maximizing a weighted utility function that can jointly represent the different requirements of two types of V2X links. Considering the high mobility of vehicle environments, we propose an online distributed multiagent reinforcement learning (MARL) method to solve the above non-convex optimization problem and design the three elements of reinforcement learning. Simulation results demonstrate that this method can effectively improve V2X networks performance under the dynamic channel environment.