Machine Learning Inspired Precoding for Multi-user mmWave 3D MIMO Systems

Abstract

In downlink transmission scenarios, power allocation at the transmitter and beam shaping design are critical. Considering the issue of precoding matrix selection in a multi-user mmWave 3D MIMO system, the traditional two-step zero forcing (ZF-ZF) algorithm and 3D DFT codebook algorithm is too complex to compute, low efficiency and high latency. To address these issues, this paper presents a fast beam shaping design method based on machine learning. By using the channel matrix set obtained by quantifying elevation angle and azimuth angle of the antenna in multi-user mmWave 3D MIMO system and DFT codebook as the training data to train machine learning model. In this way, the model can be used online after offline training, which saves the consumption of terminal's resources. The experimental results show that this method can approximate the performance of traditional precoding algorithm, while the computational complexity and time delay are greatly reduced.