Machine Learning Aided Demodulator in MISO Beamforming Systems

Abstract

In this study, a machine learning (ML) aided demodulator was designed for equalization and demodulation of the signal at the receiver in a multiple-input single-output (MISO) beamforming system. To observe the accuracy of MLaided demodulator in the presence of Rayleigh channel; XGBoost, LightGBM, and Linear Support Vector Machine (SVM) algorithms are used. Several scenarios were discussed where the number of antennas, signal-to-noise ratio (SNR), channel perfection, and channel information availability changed. As a result of 72 different measurements, it was observed that Linear SVM operates at slightly lower bit error rate (BER) levels among 3 algorithms. However, the SVM has higher time complexity and could not provide a significant advantage over other algorithms. Therefore, the use of the LightGBM algorithm, with a very low amount of BER sacrifice, greatly reduces time complexity.