Deep Learning Based Hybrid Precoding Technique for Millimeter-Wave Massive MIMO Systems

Communications over millimeter-wave (mm-Wave) frequencies are considered as a new revolution of wireless communications, specifically with the official launching of 5G. Typically, mm-Wave with massive multiple-input multiple-output (MIMO) can be implemented by using the hybrid beamforming transceivers that consists of massive number of analog phase shifters and smaller number of RF chains. The power consumption and cost are reduced when the hybrid beamforming architecture is implemented by combining the digital and analog beamforming. The main motivation for this paper is to introduce a deep learning-based hybrid beamforming design to join optimization of the precoder and combiner in massive MIMO mm-Wave communication systems. Specifically, the joint optimization of the precoder and combiner is carried out by means of two convolutional neural networks (CNN) and through going into two stages of operation, namely training and prediction stages. The MATLAB simulation results show that the deep learning-based hybrid beamforming approach for the mm-Wave massive MIMO outperforms the legacy optimization-based hybrid beamforming approaches in terms of spectrum efficiency.