Data-Driven Tuning for Weighted Least Squares of BLE-AoA-based Indoor Localization

This paper proposes a novel positioning algorithm based on angle of arrival (AoA) in Bluetooth low energy (BLE), utilizing machine learning (ML) to achieve high-precision indoor positioning. Although various positioning methods use BLE, such as radio fingerprinting and proximity methods, the AoA method is known for its superior positioning accuracy and has received significant attention. However, similar to other methods, it is susceptible to the degradation of the positioning accuracy caused by the reflection, obstruction, and diffraction of radio waves by walls and furniture. To address this issue, we exploit ML techniques in the weighted least squares (WLS) method in BLE-AoA systems. Using ML models trained on AoA datasets with supervised learning, the weights in the WLS method can be adjusted in a data-driven manner, thereby enhancing positioning accuracy. As the ML models, this paper utilizes regression model (RM), random forest (RF), and multi-layer perceptron (MLP) and validates their effectiveness through computer simulations using publicly available datasets.