Deep Spatio-temporal Beam Training for mmWave Communications with Human Self-blockage

Human self-blockage can severely attenuate the mmWave signal and degrade the throughput, even in the absence of environmental blockages. Compared with environmental blockages, the human self-blockage is highly related to the direction of human movements, which has strong spatio-temporal correlations, and can be used to reduce beam training overheads meanwhile improve the throughput. In particular, we propose a convolutional long-short term memory (ConvLSTM) based deep spatio-temporal beam training algorithm, which can accurately infer the optimal beam by probing only a small portion of beams. Simulation results demonstrate that the proposed algorithm can provide a higher average throughput than the state of the arts.