AoA estimation based on Wi-Fi CSI and deep learning

It’s well known that the MUSIC algorithm can achieve high-resolution angle of arrival (AoA) estimation by utilizing the phase information contained in Wi-Fi channel state information (CSI). However, in real-world indoor environments, directly applying the MUSIC algorithm often leads to significant errors due to several challenges. These include the limited number of antenna arrays in commercial Wi-Fi devices, non-line-of-sight (NLOS) conditions caused by environments, and complex multipath propagation. To address these challenges, we propose a novel approach that combines the MUSIC algorithm with deep learning, incorporating a Bi-directional Long Short-Term Memory (BiLSTM) layer and an attention mechanism to capture temporal dependencies in dynamic environments. Our experimental results demonstrate that this hybrid approach outperforms traditional methods. On a publicly available dataset collected by access points (APs) with three receiving antennas, our method achieves an average AoA mean absolute error (MAE) ranging from 5.5 to 8.1 degrees across different scenarios.