Communication resource allocation method in vehicular networks based on federated multi-agent deep reinforcement...

Area of Science:

• Vehicular Networking
• Communication Systems
• Machine Learning

Background:

• Traditional resource allocation in vehicular networks suffers from low spectral efficiency due to lack of global optimization and slow response to dynamic environments.
• Sharing spectrum resources between Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) links presents significant challenges for efficient resource management.

Purpose of the Study:

• To propose a novel resource allocation method for vehicular networks using federated multi-agent deep reinforcement learning.
• To improve system spectral efficiency, V2V transmission success rate, and V2I link capacity in dynamic vehicular communication scenarios.

Main Methods:

• Fusing Asynchronous Federated Learning (AFL) with Multi-Agent Deep Deterministic Policy Gradient (MADDPG) for synergistic resource allocation.
• Vehicles act as agents optimizing spectrum access, power control, and bandwidth allocation based on local channel states.
• An asynchronous federation architecture enables independent model parameter uploads, dynamic weight adjustment based on channel quality, and global model optimization.

Main Results:

• Achieved an average improvement of 19.1% in system spectral efficiency compared to existing algorithms.
• Increased the average transmission success rate of V2V links by 9.3%.
• Enhanced the average total capacity of V2I links by 16.1%.

Conclusions:

• The proposed federated multi-agent deep reinforcement learning method effectively optimizes resource allocation in vehicular networks.
• The approach significantly improves key performance metrics, demonstrating its superiority over traditional and other advanced algorithms.