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

Zhijuan Li, Guohong Li, Zhuofei Wu, Wei Zhang, Alessandro BazziSensors 2025 · DOI

TL;DR

This paper introduces a novel framework, GAT-A2C, that uses Graph Attention Networks (GAT) and Deep Reinforcement Learning (DRL) to dynamically manage communication resources in connected vehicles (C-V2X). By intelligently learning interference patterns and adapting to changing traffic conditions, GAT-A2C significantly improves both safety-critical V2V communication reliability and infotainment-focused V2N data rates, especially in dense traffic.

What Problem Does This Paper Address?

Intelligent Transport Systems (ITS) rely heavily on Vehicle-to-Everything (V2X) communication, including Vehicle-to-Vehicle (V2V) for safety and Vehicle-to-Network (V2N) for services like traffic updates. However, efficiently allocating limited radio resources (like spectrum and transmission power) to balance the ultra-low latency and high reliability demands of V2V with the high throughput needs of V2N in dynamic, dense traffic environments is a major challenge. Traditional methods struggle with the complexity of rapidly changing network conditions and the interference between numerous vehicles. Existing reinforcement learning approaches often fail to capture the intricate spatial relationships and interference patterns among vehicles, leading to suboptimal resource allocation.

Methods

The core innovation is the GAT-A2C framework, which combines Graph Attention Networks (GAT) with the Advantage Actor-Critic (A2C) deep reinforcement learning algorithm.

  • Graph Construction: The vehicular network is modeled as a graph where each V2V communication link is a node. Edges represent potential interference between these links. This graph is dynamically updated based on real-time vehicle positions and channel conditions to accurately reflect the current interference landscape.
  • Graph Attention Network (GAT): The GAT module processes this graph to learn "node embeddings." It uses an attention mechanism to weigh the importance of neighboring links' features, effectively capturing spatial relationships and interference patterns. This provides the RL agent with a richer, context-aware understanding of the environment.
  • Advantage Actor-Critic (A2C): This DRL algorithm is used for decision-making. The "Actor" network learns a policy to select resource blocks and transmission power levels, while the "Critic" network evaluates the quality of these actions. The GAT's output embeddings are integrated into the state representation for the A2C agent, enabling more informed decisions.
  • Dynamic Resource Allocation: The system jointly optimizes resource block allocation and transmission power for both V2V and V2N communications to maximize V2N rates and V2V reliability while minimizing interference.

Datasets

Not discussed in the paper. The experiments were conducted using simulations.

Key Insights and Ideas

  • Graph Structure is Key: The paper strongly emphasizes that treating vehicles/links as isolated entities is insufficient. Modeling the network as a graph and using GAT to learn interference patterns is crucial for effective resource management in dense V2X environments.
  • Dynamic Graph Updates Matter: Static graph representations are inadequate for the fast-paced nature of vehicular networks. Dynamically updating the graph based on real-time conditions allows the system to adapt to changing interference.
  • GAT Enhances RL Performance: Integrating GAT embeddings into the RL agent's state significantly improves its ability to make optimal decisions, leading to better performance than a pure A2C approach.
  • Balancing V2V and V2N: The framework successfully navigates the trade-off between V2V reliability and V2N throughput, demonstrating improved performance in both metrics compared to baseline methods.

Key Results

  • Improved V2N Rate: The GAT-A2C strategy achieved significantly higher V2N rates, especially in high-density scenarios (e.g., over 20% higher than traditional DQN with 100 vehicles).
  • Robust V2V Reliability: V2V success ratios remained consistently high (above 95%) across various vehicle densities, demonstrating strong robustness even as interference increased.
  • Scalability: The approach showed good scalability, maintaining performance as the number of vehicles increased from 20 to 100.
  • Power Control Strategy: The GAT-A2C strategy consistently favored higher transmission power levels (75-82%), which enhances data delivery range and robustness, further supported by the GAT's interference awareness.
  • Ablation Studies: Experiments showed that using eight attention heads and two attention layers in the GAT provided an optimal balance for performance.

Limitations and Future Work

  • Computational Complexity: The paper acknowledges that the GAT-A2C model's complexity scales with the cube of the number of vehicles ($n^3$) and attention heads ($H$). This makes it most suitable for scenarios with a relatively small number of vehicles, and real-time computation might be challenging for very large networks.
  • Pretrain-and-Deploy: A potential solution to the complexity issue is a pretrain-and-deploy strategy, where the model is trained on smaller scenarios and then applied to denser ones, potentially with a slight accuracy trade-off.
  • Full Vehicle Penetration Assumption: The current study assumes all vehicles are equipped with communication devices. Future work aims to extend the framework to scenarios with mixed connectivity (some vehicles connected, some not) to better reflect real-world traffic.
  • Integration with MARL: The authors plan to integrate the GAT-A2C framework with advanced multi-agent reinforcement learning techniques (like mean-field MARL or cascaded MADDPG variants) to further enhance its capabilities.

Summary updated

Abstract

Vehicle-to-vehicle (V2V) and vehicle-to-network (V2N) communications are two key components of intelligent transport systems (ITSs) that can share spectrum resources through in-band overlay. V2V communication primarily supports traffic safety, whereas V2N primarily focuses on infotainment and information exchange. Achieving reliable V2V transmission alongside high-rate V2N services in resource-constrained, dynamically changing traffic environments poses a significant challenge for resource allocation. To address this, we propose a novel reinforcement learning (RL) framework, termed Graph Attention Network (GAT)-Advantage Actor-Critic (GAT-A2C). In this framework, we construct a graph based on V2V links and their potential interference relationships. Each V2V link is represented as a node, and edges connect nodes that may interfere. The GAT captures key interference patterns among neighboring vehicles while accounting for real-time mobility and channel variations. The features generated by the GAT, combined with individual link characteristics, form the environment state, which is then processed by the RL agent to jointly optimize the resource blocks allocation and the transmission power for both V2V and V2N communications. Simulation results demonstrate that the proposed method substantially improves V2N rates and V2V communication success ratios under various vehicle densities. Furthermore, the approach exhibits strong scalability, making it a promising solution for future large-scale intelligent vehicular networks operating in dynamic traffic scenarios.
BibTeX
@inproceedings{li2025dynamic,
  author = {Zhijuan Li and Guohong Li and Zhuofei Wu and Wei Zhang and Alessandro Bazzi},
  title = {{Dynamic Allocation of C-V2X Communication Resources Based on Graph Attention Network and Deep Reinforcement Learning}},
  booktitle = {Sensors},
  year = {2025},
  doi = {10.3390/s25165209},
}

Loading related content...

Figure