Deep learning-based beamforming design for MIMO systems with hardware impairments

Abstract

We address the beamforming design problem for multi-input multi-output (MIMO) communication systems with hardware impairments. Most beamforming technologies for wireless communication systems are designed assuming that transceiver hardware is ideal. However, a practical hardware behaves differently from ideal mathematical modeling, and these differences cause degradation of communication performance. In particular, the nonlinearity of the power amplifier is a major impairment which causes interference between data streams and distorts the beam pattern, deteriorating the performance of beamforming technique. In this paper, we aim to design beamforming considering nonlinearity in order to improve communication performance which is affected by nonlinearity of power amplifiers. Specifically, 1) a beamforming technique for nonlinear MISO systems, 2) a beamforming technique for nonlinear multi-user MIMO systems, 3) a cooperative beamforming technique for nonlinear distributed networks, 4) a joint channel estimation, feedback and beamforming technique were developed. A mathematical analysis of the system is provided to determine the characteristics of nonlinear MISO and MIMO systems, and a deep learning-based methodology is applied to design a complicated distributed network and closed-loop FDD system. It is confirmed that the performance of a wireless system with nonlinear power amplifiers can be dramatically improved through the proposed techniques.