Deep Networks for Direction of Arrival Estimation With Sparse Prior in Low SNR
Yanhua Qin
Abstract
This work introduces direction of arrival (DOA) estimation considering the sparsity prior in the low signal to noise ratio (SNR) using deep learning (DL). The sparsity of the representation can be prompted to realize super-resolution DOA estimation of the spatial spectrum. The undersampled noisy linear measurements of the spatial spectrum, which is formulated by the columns of the array covariance matrix, can be given as input to the neural network for DOA estimation.The convolutional layers enable feature extraction and then the nonlinear relationship between the spatial spectrum and the corresponding DOA can be learned by training the neural network. After offline training phase, the proposed method performs well in response to the input signals for any amount of snapshots in low SNR. The results of experiments have demonstrated the superior DOA performance of the proposed method compared to state-of-the-art methods including traditional superresolution DOA estimation methods and existing DL-based DOA estimation methods.