Deep Learning based Optimal Hybrid Precoder for Millimetre Wave MIMO Communication

Abstract

Hybrid beamforming is a vital technique for making large-scale MIMO systems a reality in millimeter-wave and Terahertz communication systems. Common methods like Orthogonal Matching Pursuit (OMP) and Simultaneous OMP (SOMP) are frequently used to create the beamforming weights, but they require extensive computation due to repeated matrix calculations. To address this problem, this paper introduces a deep learning method for designing ideal hybrid beamformers. We use a CNN to understand the intricate relationship between the communication channel information and the best possible hybrid beamforming configurations. The effectiveness of our method is assessed by measuring data throughput and accuracy across various signal strength levels. Simulations show that the deep learning approach performs very well, achieving about four times the data throughput at a signal-to-noise ratio of 20 dB and a much higher degree of accuracy compared to standard OMP and SOMP methods. This research highlights the promise of using data-driven techniques to optimize future wireless systems.