Radio Resource Management for C-V2X Using Graph Matching and Actor–Critic Learning

Abstract

We propose a hybrid centralized-distributed radio resource management (RRM) scheme for cellular vehicle-to-everything (C-V2X), which is to mitigate the interference caused by radio resource sharing between vehicle-to-infrastructure (V2I) links and vehicle-to-vehicle (V2V) links. Specifically, it contains two decoupled aspects: centralized channel allocation and distributed power control. The former is conducted by graph matching to decide spectrum sharing strategy; the latter is performed via actor-critic learning, a deep deterministic policy gradient (DDPG)-based algorithm, to increase the total system capacity. Finally, the proposed scheme is numerically evaluated and outperforms other deep Q-network (DQN)-based schemes in terms of the system capacity.