A Reduced complexity hybrid precoding architecture and user grouping algorithms for downlink wideband massive MIMO channels

Kilcioğlu, Emre
In this thesis, an efficient hybrid precoding architecture is proposed for single-carrier (SC) downlink wideband spatially correlated massive multiple-input multiple-output (MIMO) channels. The design of two-stage beamformers is realized by using a virtual sectorization via second-order channel statistics based user grouping. The novel feature of the proposed architecture is that the effect of both inter-group-interference (due to non-orthogonality of virtual angular sectors) and the inter-symbol-interference (due to SC wideband transmission) are taken into account. While designing the analog beamformer, we examine the dimension reduction problem and proper subspace (beamspace) construction (by exploiting the joint angle-delay sparsity map and power profile of the multi-user channel) based on which a highly efficient spatio-temporal digital precoding is proposed. Before the beamforming stage, the user terminals are distributed into multiple groups by using their second order statistics via a user grouping algorithm named as merge and split based algorithm with K-means initialization and our own performance metric. This algorithm finds an optimal user grouping distribution to maximize the performance of the hybrid precoding structure. If the proposed joint angle-delay generalized eigen beamformer (JAD-GEB) type analog beamformer, reduced complexity spatio-temporal channel matched filter-spatially regularized zero forcing (CMF-SZF) type digital precoder and merge and split based user grouping algorithm with K-means initialization and our own performance metric are utilized together for the hybrid structure, the performance results reach very close to those of the fully digital CMF type precoding by using less number of RF chains.