Machine Learning for Massive MIMO

Abstract

This research has proposed a new system model intended to improve the throughput and performance of massive MIMO networks while simultaneously reducing the complexity of the detection process. This research has contributed towards the detection system at the receiver in addition to an adaptive modulation system at the transmitter. Within the detection system, the partial learning model is applied and consists of two core parts - the first, usage of a non-linear detector which utilizes the neural network technique, and the second, proposal of the application of a linear detector of low complexity. The goal of this model is to not only further add improvements towards the functional performance of the detector, but to also maintain an even lower level of complexity in comparison to other commonly used systems. As a result of this research, the proposed model is successfully simulated where machine learning techniques can be attributed towards improved system performance in comparison to conventional systems. Improvements in system throughput are observed, in addition to positive net changes regarding SNR, when utilizing the proposed system in massive MIMO.