A Moving Source Angle-of-Arrival Estimation in Real-Time Using Machine Learning
Jin Feng Lin, Charles Montes, Nathaniel D. Bastian, Ruolin Zhou · 2025
This paper presents a hybrid CNN-BiLSTM model for accurate Angle of Arrival (AoA) estimation in dynamic signal environments. The proposed model achieves an overall root mean square error (RMSE) of 4.77 degrees for a single moving source, demonstrating superior robustness in low signal-to-noise ratio (SNR) conditions (-20 dB), where the Multiple Signal Classification (MUSIC) algorithm’s performance significantly deteriorates. Additioanly, over-the-air (OTA) test validate the real-world applicability of our approach, where our model achieves an RMSE of 5.10 degrees compared to MUSIC’s 34.99 degrees. The results highlight the advantages of machine learning over traditional subspace-based methods for AoA estimation in challenging environments.