Evaluation of Machine Learning Algorithms on Power Control of Massive MIMO Systems

Abstract

Power control (PC) plays a crucial role in massive multiple-input-multiple-output (m-MIMO) networks. Several heuristic algorithms, like the weighted mean square error (WMMSE) algorithm are used to optimise the PC. In order these algorithms to perform the power control they require high computational power. In this paper, we address the problem through the application of machine learning (ML)-based algorithms as they can produce close to optimal solutions with lower computational complexity. We evaluate use of several different machine learning (ML) methods such as deep neural networks (DNN), deep Q-learning (DQL), support vector machines (SVM) with radial basis function (RBF), K-nearest neighbours (KNN), linear regression (LR), and decision trees (DT) to maximise the sum spectral efficiency (SE). The evaluation results demonstrate that the ML based approaches can approximate near to the WMMSE based method.