Knowledge Distillation-Based MAPPO Approach of Wireless Power and Spectrum Resource Joint Allocation for 6G Networks

With the widespread deployment of 5G networks, the dense layout of base stations has led to significant interference issues. Consequently, optimizing spectrum and power planning has become crucial for enhancing system channel capacity. This paper proposes a multi-agent proximal policy optimization method based on knowledge distillation, utilizing a globally shared reward function to promote cooperation among agents. The method employs a centralized training and distributed execution framework, allowing each agent to make effective decisions based solely on local state information, thereby achieving efficient spectrum selection and power allocation. Additionally, knowledge distillation techniques are introduced to simplify the model structure, optimizing memory usage and inference speed. Simulation results demonstrate that the proposed KD-MAPPO method exhibits significant advantages in performance and storage efficiency.