Deep Reinforcement Learning based Congestion Control for V2X Communication

In release 14 (Rel-14) Long Term Evolution (LTE), the 3rd generation partnership project (3GPP) standard has introduced Cellular Vehicle to Everything (C-V2X) communication to pave the way for future intelligent transport systems (ITS). C-V2X communication envisions supporting a diverse range of use cases with varying quality of service (QoS) requirements. For example, cooperative collision avoidance requires stringent reliability, while infotainment use cases require a high data throughput. C-V2X communication remains susceptible to performance degradation due to network congestion. This paper presents a centralized congestion control scheme for C-V2X communication based on the Deep Reinforcement Learning (DRL) framework. A performance evaluation of the algorithm is conducted based on system-level simulation based on TAPASCologne scenario in the Simulation of Urban Mobility (SUMO) platform. The results show the effectiveness of a DRL-based approach to achieve the packet reception ratio (PRR) as per the packet's associated QoS while maintaining the average measured Channel Busy Ratio (CBR) below 0.65.