Multi-Agent Deep Reinforcement Learning for Access Point Activation Strategy in Cell-Free Massive MIMO Networks

Cell-free massive MIMO network is a promising solution to support 5G communication. However, as an ultra-dense network (UDN), it suffers from two major issues: power consumption and scalability, which hinders it to be widely used in practice. The increasing and time-varying user demand, the dynamic propagation environment, and the huge amount of access points (APs) make it a challenging task to address these issues. For this purpose, we propose a distributed solution to solve the access point activation (APA) problem in cell-free massive MIMO networks to reduce power consumption considering dynamic user demand. We leverage the user-centric characteristic and design a multi-agent deep reinforcement learning (MADRL) algorithm by which each AP independently decides whether it needs to be switched on or off. The simulation results show that the MADRL outperforms the centralized and the random strategies. This research demonstrates the ability of distributed learning in solving the APA problem in cell-free massive MIMO networks.