Deep Reinforcement Learning Based Resource Allocation for 5G V2V Groupcast Communications

Cellular vehicle-to-everything (C-V2X) proposed by 3GPP aims to support vehicle safety information, real-time vehicle updates, and coordinated driving capabilities. One of the safety information transmission methods of C-V2X is to transmit messages via broadcasting or multicasting from a Road-Side Unit (RSU) or a base station (BS). However, the messages may not be received by some of the target vehicles due to the interference caused by other vehicle communications or the modulation and coding scheme (MCS) configuration for broadcast. In recent years more and more research shows that V2V communication technology can improve performance, therefore we propose using V2V groupcast in platoons to compensate for the failed transmission from the RSU to the vehicles to improve the quality of service (QoS). But effective resource allocation among the platoons has been a challenge due to the potential interference between the vehicle message transmissions caused by the platoons selecting the same uplink transmission resources. In this work, we propose a Deep Reinforcement Learning (DRL) scheme to properly allocate V2V communication resources and conFigure the MCS to improve message transmission reliability and maximize system utility. Specifically, each platoon leader acts as a DRL agent, which makes its V2V communication policy independently based on a centralized trained DRL model. Our simulation results verified that the probability of receiving messages for platoon members and the system utility are significantly increased by applying the proposed DRL model.