Improved coverage measurements through machine learning algorithms in a situational aware channel condition for indoor distributed massive MIMO mm-wave system

Abstract

In massive MIMO (Multiple Input Multiple Output) mm (millimetre) wave system, the channel conditions are measured and analysed for a better placement of reflectors or antennas. In order to increase the coverage area and to reduce interference among users factors such as pathloss and power delay profile are extracted from the Channel Impulse Response (CIR) i.e. from the received signal with respect to transmitter and receiver channel propagation conditions. In a distributed indoor massive MIMO mm-wave system, pathloss and power delay profile are evaluated for Line of Sight (LoS) and Non-Line of Sight (NLoS) environments at frequencies such as 28 and 39 GHz. Based on these factors, a dataset is constructed for 28 GHz. Algorithms such as Support Vector Machine, KNN and Fine Tree are considered. These algorithms are trained with a set of datasets and are tested for performance metrics such as Mean Absolute Error, Correlation Coefficient, Root Mean Squared Error, Relative Absolute Error, Root Relative Squared Error are evaluated. Simulation results show that an accuracy of 94% and 95% using support vector machine, 93.8% and 94.5% accuracy using KNN and an accuracy of 93.2% and 93.8% using fine tree algorithm for pathloss and power delay profile respectively.