EFFICIENT TWO-STAGE BEAM TRAINING AND CHANNEL ESTIMATION FOR RIS-AIDED MMWAVE SYSTEMS VIA FAST ALTERNATING LEAST SQUARES

Abstract

This paper proposes a two-stage beam training and a channel estimation based on fast alternating least squares (FALS) for reconfigurable intelligent surface (RIS)-aided millimeter-wave systems. To reduce the beam training overhead, only selected columns and rows of the channel matrix are observed by two-stage beam training. This beam training produces a partly observed channel matrix with low coherence, which enables the low rank matrix completion technique to recover unobserved entries. Unobserved entries are recovered by FALS, which alternatingly updates the left and the right singular vectors that comprise the channel. Simulation results and analysis show that the proposed algorithm is computationally efficient and has superior accuracy to existing algorithms.