Lightweight Beamforming Through Random Embedding in Massive MIMO Systems

For massive MIMO transceivers with subspace decomposition, we propose a novel and practical solution leveraging randomized numerical linear algebra. The loss in covariance matrix estimation and subspace decomposition can be minimized while reducing complexity and memory requirements, by adjusting the random embedding levels based on multivariate statistical analysis. An interference-resilient null beamforming structure with random embedding is proposed for both DMRSbased beamforming with/without equalizers (i.e., DMRS-BFEQ and DMRS-BF-NEQ) supported in uplink performance improvement (ULPI). Complexity in the proposed interference null space projection (INSP) based beamformer can be significantly reduced compared to full dimensional MMSE-IRC by removing received signal whitening and exploiting an effective rank of the interference subspace for INSP. It is shown that the proposed INSP based beamformer provides noticeable performance gain against the conventional low complexity receivers, and equivalent or even enhanced performance than full dimensional MMSE-IRC with practical sample size in both interference-free and highly interference-limited environments through link level simulation.