User Clustering and Hybrid Precoding Design for mm-Wave Multi-User Massive MIMO Systems

Interference cancellation in hybrid precoded downlink massive multi-user multi-input multi-output (MU-MIMO) systems, e.g., block diagonalization (BD), has a complexity that increases rapidly with the number of users. Spectral clustering into user groups based on mutual interference is proposed to reduce the complexity of servicing a large group of users in the same channel. By assigning users with the lowest mutual interference into separate groups, only intra-group interference cancellation needs to be considered. A relaxed suboptimal user grouping solution is proposed by characterizing MU-MIMO downlink interference using the pairwise angle between channel subspaces as graph weights for spectral clustering. Group assignment is achieved by discretization via constrained spectral clustering. The proposed approach, applied to mmWave multi-user massive MIMO systems, offers new system performance versus complexity trade-offs compared to conventional methods, as demonstrated by numerical results.