Dynamic Allocation of C-V2X Communication Resources Based on Graph Attention Network and Deep Reinforcement Learning

Vehicle-to-vehicle (V2V) and vehicle-to-network (V2N) are two communication types in intelligent transport systems (ITSs) that can share spectrum through in-band overlay. V2V primarily ensures traffic safety, while V2N focuses on entertainment and information sharing. Ensuring reliable V2V communication and high-rate V2N in resource-constrained and dynamically changing traffic environments poses a significant challenge for resource allocation. To address this, we propose a novel Graph Attention Network (GAT)-Advantage Actor-Critic (GAT-A2C) reinforcement learning (RL) framework in the paper. In the framework, we build a graph based on V2V links and their potential interference relationships, in which the V2V link between vehicles is regarded as the node, and connect nodes with interference relationships to form edges. GAT is employed to model significant interference relationships among neighboring vehicles while accounting for real-time mobility and channel conditions. The feature output by GAT and the feature of link itself are combined to form the environment state, which is then fed into the agent to optimize the allocation of resource blocks and transmission power for V2V and V2N. Simulation results demonstrate that the method significantly improves V2N rates and V2V communication success ratios across varying vehicle densities. The proposed solution exhibits high scalability, making it suitable for future large-scale intelligent vehicular networks in dynamic traffic scenarios.