Angle-of-Arrival Estimation Using an Adaptive Machine Learning Framework

Angle-of-arrival (AoA) estimation is of great interest, particularly for using radio to localize a device; good estimates of angles result in good estimates of location. In this letter, we propose a signal processing and machine learning combined tool for the AoA estimation. In particular, we utilize regression models trained using the snapshot data collected using multiple antennas for estimating the angle of arrival. Based on a set of simulation and real measurements under the Bluetooth 5 low-energy system in an indoor environment, the proposed method is able to provide a considerable and consistent improvement without significant additional computational effort. We show that the proposed approach for AoA estimation provides an improvement of at least 20% compared with the baseline approach of traditional Multiple Signal Classification algorithm. We evaluate the performance of the proposed methods and show a consistent improvement using a range of channel parameters, including elevation angles, SNRs, and channel configurations.