Covariance Matrix Reconstruction Based on Regularized Denoising Autoencoders for Direction-of-Arrival Estimation

In this paper, we present a new approach to improve direction-of-arrival estimation for uniform linear arrays (ULA) using deep learning. We propose to denoise the array covariance matrix estimated from the sample matrix using a fully connected autoencoder. This technique enhances target detection in low Signal-to-Noise Ratio (SNR) conditions by applying a MUSIC estimator to the reconstructed covariance matrix. The proposed method outperforms classical estimators and has benefits compared to recent proposed deep learning-based techniques. The performance of the architecture are assessed with the MUSIC pseudospectrum, the Signal Eigenvalues-to-Noise Eigenvalues Ratio (SENER) and the Maximum-to-Minimum Eigenvalue (MME).