Convolutional neural network-based hybrid beamforming design based on energy efficiency for mmWave M-MIMO systems

Abstract

Millimeter-wave (mmWave) massive multiple-input multiple-output (M- MIMO) technology brings significant improvements in data transmission rates for communication systems. A key to the design of mmWave M-MIMO systems is beamforming techniques, which focus signals toward specific directions but rely on expensive, energy-intensive radio frequency (RF) chains. To address this issue, hybrid beamformers (HB) have been introduced as a partial solution, and deep learning (DL) has proven effective for HB design. However, previous works utilizing machine learning (ML) networks have primarily focused on the spectral efficiency (SE) metric for constructing HB. In this paper, we present a convolutional neural network (CNN) architecture whose loss function is defined to maximize energy efficiency (EE) directly. The network jointly learns analog and digital beamformers by evaluating EE (throughput per total power, including phase shifters, switches, digital-to-analog converters (DACs), and RF chains) and selecting the configuration that yields the highest EE. The CNN takes a channel matrix as input and outputs RF and baseband beamformer matrices. Simulation results validate the effectiveness of the proposed M-MIMO EE scheme, achieving significant EE improvements by optimizing hybrid precoding and reducing RF chain usage.