Low Time-Complexity Learning-Aided Codebook-Based Beam Forming Selection in MmWave System

Abstract:
In millimeter-wave (mmWave) cellular networks, a large antenna array equipped at a transmitter or a receiver can form a narrow beam with high beamforming gain to compensate for the severe path loss at the mmWave frequencies. For the codebook-based analog beamforming, due to the narrower beamwidth, a larger codebook is needed to cover the same spatial area to guarantee that no deep sink of beamforming gain occurs in any direction. For the selection of the best entry in a codebook, hierarchical search (HS) has been widely adopted as it can considerably reduce the time complexity compared to exhaustive search (ES). However, owing to the hierarchical structure, a gap to the optimal performance obtained from exhaustive search exists in the noisy environment. In this paper, we proposed a learningaided search (LAS) beam selection scheme without the tree-searching structure that is intrinsic in the HS. With the aid of the convolution neural network (CNN) as well as a traditional detection module, our proposed scheme can achieve higher average spectral efficiency with low time complexity according to our simulation results.