Intelligent Resource Allocation Based on Reinforcement Learning for NOMA Vehicular Cooperative Communication Networks

Abstract

There exits difficulty in environment sensing for autonomous vehicles in Internet of vehicles (IoV) networks. The autonomous vehicle can obtain a full sense of the environment from connected vehicles by vehicle-to-vehicle (V2V) communications and make better decisions for driving safety. However, limited spectrum resources cannot adapt to the increasing number of communication links and the traditional frequency multiplexing will lead to serious co-frequency interference and fail to connect more vehicles. Therefore, we develop a resource allocation scheme through jointly optimizing sub-band selection and power control based on V2X Network. Meanwhile, the non-orthogonal multiple access (NOMA) technology is adopted for multi-V2V communication that shares the spectrum allocated to vehicle-to-infrastructure (V2I) links. Considering time-varying channel conditions caused by vehicle movement and power control over a continuous range in a vehicular environment, we investigate a deep deterministic policy gradient algorithm to maximize the sum transmission rate of the V2I links and V2V links while ensuring the quality of service of V2V links. Simulation results demonstrate the proposed solution can effectively optimize the total throughput.