Deep Reinforcement Learning-Based Access Point Selection for Cell-Free Massive MIMO with Graph Convolutional Networks

In this paper, we investigate the association problem between access points (APs) and users in cell-free massive multiple-input multiple-output (MIMO) systems. To address the limitations of received signal reference power (RSRP) based user-centric approach, we develop a deep reinforcement learning (DRL) approach with graph convolutional networks (GCN) to extract pertinent features encompassing connection status and topological information. The relative positioning of nodes is encoded within an adjacency matrix, while the connection states are captured within a feature matrix. Our empirical results validate the efficacy of the proposed approach, which consistently outperforms both the baseline method without GCN and the conventional user-centric approach.