Deep Reinforcement Learning for Dynamic Access Point Activation in Cell-Free MIMO Networks

The cell-free network architecture suppresses intercell interference by eliminating cell boundaries through the joint operation of distributed access points (APs). While a dense deployment of these APs may lead to performance gains, the corresponding increase in energy consumption poses environmental and economic issues. One way to improve energy efficiency is to turn off underutilized APs. In this work, we present a deep reinforcement learning-based framework that derives the set of active APs given the spatial user information. The flexible design of the reward function for AP selection allows easy adjustment of performance targets in terms of quality of service and power consumption, as well as studying their trade-off. We also demonstrate how the proposed framework intelligently selects and activates only a subset of APs that contributes significantly to user performance, thereby enabling the cell-free network to provide good service while achieving power savings.