Hybrid Deep Learning for Angle of Arrival Estimation: A Case Study on the H-WiLD Dataset

Achieving accurate indoor localization is challenging due to multipath interference, especially under dynamic conditions. Traditional methods, such as 2D-FFT and Matrix Pencil (MP), often struggle in non-line-of-sight scenarios. To overcome this limitation, we propose a hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model integrating MP estimates, Channel State Information (CSI) data, and environmental metadata. Experiments conducted on the H-WiLD dataset demonstrate a 30 % RMSE improvement, highlighting our method's robustness for practical Angle of Arrival (AoA) estimation.