Density Based Clustering for Downlink User Grouping in FDD Massive MIMO

Abstract:
Recently, the two-stage Joint Spatial Division and Multiplexing (JSDM) precoding scheme was introduced as a valuable solution for implementing, with a reduced complexity, Massive MIMO transmission techniques in emerging 5th Generation of mobile networks. Among its main features, JSDM requires to partition mobile users into groups, based on the channel covariance similarity. To this end, the Density-Based Clustering of Applications With Noise (DBSCAN) clustering algorithm has already proved to be reasonably suitable. Unfortunately, DBSCAN falls short in challenging situations, like a large homogeneous crowd of users. Here, in fact, all the users are put in the same group despite experiencing different channel covariances, which degrades the behavior of JSDM. Based on these premises, the contribution presented herein proposes a modified version of DBSCAN, which enforces a maximum size on detected clusters by limiting the difference in channel covariances between any pair of users in the same group. In line with the requirements of JSDM, the modified DBSCAN breaks down excessively large clusters into smaller ones, thus extending its applicability to a much wider range of scenarios (from small dispersed user groups to large compact ones). Computer simulations confirm the behavior predicted by the theoretical formulation: the proposed approach reaches a better user and system spectral efficiency, while showing low sensitivity against parameter tuning.