Deep Learning Based Low-Resolution Hybrid Precoding Design for mmWave MISO Systems

Abstract

Millimeter wave (mmWave) communication has been considered as a key enabling technology for next generation cellular systems. In order to combat severe path-loss of mmWave channel, hardware-efficient hybrid beamforming schemes are widely considered in mmWave massive multi-input multi-output (MIMO) systems. However, the high computational complexity for obtaining accurate channel estimation and designing hybrid beamformer is a major challenge, which hurdles the deployment of mmWave massive MIMO systems. In this paper, we consider a multi-user mmWave communication system with low-resolution phase shifters (PSs) and aim to design a novel deep learning based low-complexity and robust hybrid precoding design algorithm. A convolutional neural network (CNN) is utilized to predict/design analog precoders with roughly estimated channels using a small amount of pilot signals. Analog precoder can be efficiently predicted from the imperfect channel convolutional neural network (IC-CNN). Once the analog precoder is obtained, the digital precoder is designed by minimum mean square error method. Simulation results demonstrate that the proposed algorithm is efficient and can achieve satisfactory performance with a significant reduction in computing time.