Deep Learning-Based Hybrid Analog-Digital Signal Processing in mmWave Massive-MIMO Systems

Hybrid analog-digital signal processing (HSP) is an enabling technology to harvest the potential of millimeter-wave (mmWave) massive-MIMO communications. In this paper, we present a general deep learning (DL) framework for efficient design and implementation HSP-based systems. Exploiting fact that any complex matrix can be written as scaled sum two matrices with unit-modulus entries, novel analog neural network (ADNN) structure first developed which implemented common radio frequency (RF) components. This then embedded into extended hybrid (HDNN) architecture facilitates mmWave systems while improving their performance. particular, proposed HDNN enables transceivers approximate desired transmitter receiver mapping arbitrary precision. To demonstrate capabilities DL framework, new HDNN-based beamformer achieve same performance fully-digital beamforming, reduced number RF chains. Finally, simulation results are presented confirming advantages over existing beamforming schemes.