FusionNet: Enhanced Beam Prediction for mmWave Communications Using Sub-6 GHz Channel and a Few Pilots

In order to reduce the downlink training overhead of mmWave communications, we propose a novel beamforming strategy using uplink sub-6GHz channel and pilots that are sent from few active antennas. Specifically, design dual-input neural network architecture, called FusionNet, merge The proposed fusion model could intelligently adjust attention paid (by network) for by an mechanism. output FusionNet represents probability each beam being optimal one. We also antenna selection can choose better antennas send pilots, in which gradient vector is approximated vector. Simulation results demonstrate superior performance compared existing one purely relies on information or shallow directly adds uniform pilots.