Research on Next-Generation Wi-Fi Spatial Reuse Power Control Based on Federated Reinforcement Learning

With the proliferation of densely deployed access points (APs), competition among multiple APs has led to increased occurrences of co-channel interference and access conflicts between APs. The IEEE 802.11 working group has proposed multi-AP coordination schemes in the next-generation Wi-Fi standard to enhance Wi-Fi performance in multi-AP scenarios. We primarily investigate the coordinated spatial reuse (CSR) mechanism for multiple APs. In the CSR group, a set of APs utilize the same channel for data transmission within a single transmission opportunity (TXOP), during which power control for individual devices is necessary to mitigate interference between APs. We propose a Federated Learning Double Deep Q-Network (FL-DDQN) algorithm for power allocation among multiple users in the CSR environment. The simulation results indicate that the proposed algorithm can improve throughput performance by approximately 11.06% to 100.51% and reduce the latency by approximately 61.34% to 86.71 % compared to several baseline schemes.