Angle of Arrival Estimation in Indoor Environment Using Machine Learning

Many localization techniques have been developed over the past decades. Angle of Arrival (AoA) is one of the most common techniques due to its high accuracy. In this paper, an AoA estimation framework for a multipath radio environment is proposed. A Convolutional Neural Network (CNN), which is a part of Deep Learning (DL), is employed to learn the mapping between the eigenvectors of the spatial covariance matrix of received array signals and angles of arrival. The CNN architecture is discussed with a detailed description of the hyper-parameters. The results present the AoA estimation with varied Signal-to-Noise Ratio (SNR), number of snapshots and path separation angle. Simulation results show that the proposed approach outperforms the traditional MUltiple SIgnal Classification (MUSIC) algorithm with less execution time especially in demanding scenarios of low SNR and limited snapshots. The proposed approach provides an improvement of at least 73% compared with MUSIC at a very low SNR.