Multi-Agent Reinforcement Learning Approach for Interference Optimization in Wi-Fi 8

Interference management is becoming ever more complicated in multi-access networks such as Wi-Fi. Here, ultradense access point (AP) deployments add to the problem. With the introduction of beamforming, this can be in part reduced, however, only in combination with optimized nulling the operation of overlapping basic service set (OBSS) can be improved. In this paper, we present a machine learning (ML)-based solution for interference optimization for Wi-Fi 8. In particular, we present a multi-agent reinforcement learning (MARL) framework that uses a distributed approach to coordinate beamforming and nulling at nearby APs. We follow a two-step procedure: first, APs and nodes are clustered in order to reduce overhead due to exchange of channel state information; and secondly, the MARL framework configures beams for the next period. Simulation results confirm the advantages of our MARL approach.