Deep Reinforcement Learning for Spatial User Density-based AP Clustering

Cell-free massive MIMO combines the benefits of massive MIMO and network densification to provide a uniformly good service throughout the coverage area. This is achieved by the joint transmission from multiple distributed access points (APs)/antennas, as well as by bringing them closer to the users. However, its canonical form where all APs are connected to only a single centralized processing unit (CPU) is not scalable and hard to realize in practice. Motivated by this, we propose a deep reinforcement learning-based approach for partitioning the APs in a multi-CPU cell-free MIMO network. We exploit the available spatial user density information when deciding which APs form the disjoint clusters that are associated to the CPUs. Our simulation results show that our framework dynamically allocates more APs (forms bigger AP clusters) in areas of larger user density, leading to a better performance when compared to small cells and predefined static AP groupings.