Millimeter Wave Beam Prediction Based on Compressed Sensing Features at Sub-6 GHz

Abstract

AbstractThe extremely large bandwidth and ultra-high speed of millimeter wave (mmWave) have attracted widespread attention in the research for its exploiting in next generation wireless networks. However, mmWave suffers from serious problems such as large path loss and high cost, which make it prohibitive in implement. To compensate for the severe path loss in the mmWave band, beamforming with large antenna array is usually used to provide high directivity. Nevertheless, while bringing high beamforming gain, beam search also increases the overhead in terms of time delay and power consumption. Aiming at improving the beam search efficiency, we propose a beam predicting scheme that combines deep neural network with compressed sensing (CS) features of sub-6 GHz channels. The distinctive features of the proposed scheme include: 1) the use of sub-6 GHz channels features estimated based on CS; 2) the realization of mapping from sub-6 GHz features to optimal mmWave beam by neural network. Owing to the efforts in harnessing the sparsity of the mmWave channel and the correlation with sub-6 GHz channels, the proposed scheme ensures that the beamforming gain can be improved with less overhead. The simulation results highlight the ability of the proposed scheme to predict the best mmWave beam. Moreover, compared with schemes which make direct prediction from sub-6 GHz channels, the obvious gain in broader range of SNR is indicated by the proposed scheme.KeywordsMillimeter waveCompressed sensingDeep learningBeam prediction