Title: Two-stage Beamformer Design via Deterministic Equivalents

Abstract: Complexity reduction of optimal linear receiver is considered in a scenario where both the number of single-antenna user equipments (UEs) $K$ and base station (BS) antennas $N$ are large. Two-stage beamforming (TSB) greatly alleviates the high implementation complexity of large scale multiantenna receiver by concatenating a statistical outer beamformer (OBF) with an instantaneous inner beamformer (IBF) design. Using asymptotic large system analysis, we propose a novel TSB method that adjusts the dimensions of user-specific OBF matrices based on the projection of the optimal minimum mean square error (MMSE) vectors into the beam domain. The beam domain is first divided into $S$ narrow sectors such that each sector contains $D$ DFT beams. Then, so-called deterministic equivalents are computed for the amplitude-projection of the optimal MMSE vectors into each sector in asymptotic regime where $N$ , $K$ and $D$ grow large with a non-trivial ratio $N/K = C$ and $N/D = S$. Given the approximations for the sector-specific values, the structure and dimension of each UE specific OBF vector are optimized based on the statistical channel properties and the amount of overlap among users in angular domain. The numerical analysis shows that the attained SINR values closely follow the optimal MMSE receiver while the computational burden is greatly reduced.