Resource Allocation in NOMA-V2X Networks With Multi-Agent Parameterized Action Space Reinforcement Learning

The integration of non-orthogonal multiple access (NOMA) with vehicle-to-everything (V2X) communications has emerged as a crucial development in Intelligent Transportation Systems (ITS), driven by advances in vehicular networking technology. However, the differing quality of service (QoS) demands for vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) links pose challenges for effective spectrum management. This paper proposes a multi-agent reinforcement learning (MARL)-based resource allocation strategy aimed at maximizing the overall system throughput while ensuring the reliability of V2V links in NOMA-enabled V2X networks. We present a NOMA-based V2X transmission model and frame the joint resource optimization (JRO) problem, focusing on optimizing both V2I and V2V links to enhance system performance. Additionally, the JRO problem is decomposed into two sub-problems: independent optimization for V2I links and optimization for V2V links. For V2I links, we adopt the grouping approach and the convex optimization method in the NOMA scenario to solve the problem in order to reduce the system complexity. For V2V links, we design a residual multi-agent parameterized action space deep deterministic policy gradient (res-MAPDDPG) method, which allows each agent to select both discrete and continuous actions and achieve the global optimal allocation. Simulation results show that our approach offers significant advantages in terms of system capacity, outage probability and spectral efficiency over conventional NOMA and OMA schemes. This study not only improves the efficiency of V2X communication, but also provides a new technical approach for resource management in future intelligent in-vehicle networks.