A Deep Learning – Based Hybrid Pre-coder and Combiner for MIMO Communication

             Millimetre wave (mm Wave) communication systems promise high data rates for future wireless networks but are challenged by severe signal degradation due to high mobility and narrow beam requirements. Traditional beamforming techniques often suffer from significant training overhead, especially in highly dynamic environments. This paper proposes a novel deep learning-based coordinated beamforming framework to overcome these limitations. By leveraging deep neural networks (DNNs) trained on Omni-received uplink pilot signals, the method enables base stations (BSs) to directly predict the optimal beamforming vectors without requiring explicit channel estimation. The proposed approach not only reduces training overhead but also effectively captures complex beamforming patterns in multi-BS scenarios. Simulation results demonstrate that the deep learning model closely approximates the performance of traditional beam selection methods, achieving near-optimal spectral efficiency with significantly reduced latency. This framework offers a scalable and practical solution for enabling fast and efficient beamforming in highly-mobile mm Wave networks.

Keywords: Deep Learning, Hybrid Precoding, Millimetre Wave (mm Wave), Massive MIMO, Auto-Precoder, Channel Estimation, Beamforming, Compressive Sensing, Neural Networks, Low-Overhead Training.