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

Area of Science:

• Wireless Communications
• Artificial Intelligence
• Intelligent Transport Systems

Background:

• Vehicle-to-vehicle (V2V) and vehicle-to-network (V2N) communications are crucial for intelligent transport systems (ITSs).
• Sharing spectrum resources for V2V (safety) and V2N (infotainment) presents significant resource allocation challenges in dynamic traffic environments.
• Existing methods struggle to balance reliable V2V transmission with high-rate V2N services.

Purpose of the Study:

• To propose a novel reinforcement learning (RL) framework for joint resource allocation in V2V and V2N communications.
• To address the challenges of resource-constrained and dynamic vehicular network environments.
• To optimize resource blocks and transmission power for improved communication performance.

Main Methods:

• Developed a Graph Attention Network (GAT)-Advantage Actor-Critic (GAT-A2C) reinforcement learning framework.
• Constructed a graph representing V2V links and interference relationships, with V2V links as nodes and interference as edges.
• Utilized GAT to capture interference patterns and combined them with link characteristics for the RL environment state.
• Employed the RL agent to jointly optimize resource blocks allocation and transmission power for V2V and V2N.

Main Results:

• The GAT-A2C framework significantly improved V2N data rates.
• The proposed method substantially increased V2V communication success ratios across various vehicle densities.
• Demonstrated substantial improvements in both V2N rates and V2V communication success ratios.

Conclusions:

• The GAT-A2C approach offers a promising solution for resource allocation in intelligent vehicular networks.
• The framework exhibits strong scalability for future large-scale, dynamic traffic scenarios.
• Effective joint optimization of V2V and V2N resources is achievable with advanced RL techniques.