Fast Multibeam Training for RIS-Assisted Millimeter Wave Massive MIMO

For reconfigurable intelligent surface (RIS)-assisted mmWave massive MIMO, we propose a fast multibeam training (FMT) scheme with two stages. In the first stage, we find a multibeam together with the RIS reflection coefficients best matching the LoS links or the cascaded LoS links, where all the RISs use the same reflection coefficients at the same time to reduce the overhead of beam training. In the second stage, we find a DFT codeword best matching the LoS links or the cascaded LoS links for the BS instead of the multibeam. Different from the FMT scheme that uses a single-layer DFT codebook, in the proposed fast hierarchical multibeam training (FHMT) scheme we use a multi-layer hierarchical codebook, where in each layer only two wide beams are used to cover the angle space of interest, and the wide beams are narrowed down from top layer to bottom layer. Simulation results show that significantly smaller training overhead can be achieved by the FMT and FHMT schemes than the baseline scheme with slightly worse performance.