Reinforcement Learning-based Interference Coordination for Distributed MU-MIMO

Abstract

In our previous studies, we proposed a graph coloring algorithm (GCA) based on heuristics to solve the interference coordination problem for distributed multi-user multi-input multi-output (MU-MIMO). In this paper, along with the recent advances of machine learning, we propose a reinforcement learning (RL) based GCA for cluster-wise distributed MU-MIMO. The computer simulation confirms that our newly proposed RL-GCA can significantly improve the downlink link capacity compared with other non-intelligent GCAs. Also, an interesting conclusion has been obtained in terms of chromatic number (required minimum number of colors). It is shown that the less chromatic number does not necessarily lead to a better interference coordination. Under the propagation environment assumed in this paper, the best chromatic number which maximizes the achievable link capacity is shown to be 4.