Antenna selection in massive MIMO using machine learning

Abstract

In massive MIMO (Multiple Input Multiple Output) systems the overall performance (bit/s/Hz/cell) is significantly improved by equipping the base stations with arrays of a hundred antennas; which becomes one of its most significant challenges; economically and technically due to the high power consumption. To solve this, Antenna selection (AS) is increasingly gaining more interest, as it strategically reduces the hardware complexity while maximizing efficiency and throughput by selecting a specific subset of antennas to activate in each transmission slot. In this report, we examine the application of multi-label learning (MLL) based algorithms in AS, such as problem transformation methods, including first order binary relevance; and high order chain classification. Additionally, we investigate the Deep neural networks (DNN) based algorithms, namely Multi-Label Convolutional Neural Networks (MLCNN) and Multi-Layer Perceptron (MLP) classifier, and multi-View based algorithm. These proposed methods are rigorously evaluated based on their maximum capacity, performance and the computation time across various scenarios. Our work concludes that, in comparison with the convex relaxation based method, the Multi-view MLL achieves comparable results.