Deep Reinforcement Learning Assisted Multi-Operator Spectrum Sharing in Cell-Free MIMO Networks

Abstract

This letter introduces a deep reinforcement learning (DRL)-assisted multi-operator spectrum sharing methodology in cell-free multi-input multi-output (CF-MIMO) networks. Existing schemes mainly focus on cognitive communications or rely on strong operators’ coordination. Our approach leverages DRL agents at the access points (APs) of each mobile network operator (MNO) for dynamic spectrum sharing across separate frequency bands. These DRL agents operate independently, selecting the spectrum resources from other MNOs’ bands with minimal inter-operator information exchange between the central processing units (CPUs). Performance evaluation under various scenarios demonstrates that DRL agents at APs can effectively learn optimal resource allocation, leading to improvements in delay, network throughput, and user-perceived throughput performance compared to the conventional non-spectrum sharing scheme.