Machine Learning Based Beam Selection With Low Complexity Hybrid Beamforming Design for 5G Massive MIMO Systems

In this paper, we present an energy-efficient joint machine learning based beam-user selection and low complexity hybrid beamforming for the multiuser massive multiple-input multiple-output (MIMO) downlink system. Massive MIMO system is a key enabler of
$5^{th}$
generation (5G) technology which is being used in vehicle-to-everything (V2X) communications and other Internet-of-Things (IoT). The large number of antennas at the base-station side causes high power consumption in the radio frequency (RF) chain. Hybrid beamforming techniques are used to reduce the number of RF chains with minimal performance loss. This paper proposes a low complexity orthogonal hybrid beamforming design with the aid of a machine learning based beam-user selection scheme. The
Householder
(HH) reflectors are used to generate the orthogonal analog beamforming (ABF) matrix. We use a feedforward neural network (FFNN) beam-user selection scheme. The proposed HH ABF + FFNN scheme provides better energy-efficiency (EE) performance in the ill-conditioned massive MIMO channel. It has been shown that the green SE-EE point (41.35
b/s/Hz
, 0.8
b/Hz/J
) can be obtained by the proposed HH ABF + FFNN scheme as compared to the (33.77
b/s/Hz
, 0.63
b/Hz/J
) from the state-of-the-art Discrete Fourier transform based techniques.