<p>Direction of arrival (DOA) estimation plays a crucial role in various applications such as wireless communication, radar systems, and microphone arrays, enabling the localization and tracking of signal sources in space. To address the issue of performance degradation of traditional DOA estimation algorithms in impulsive noise environments, a novel deep learning-based DOA estimation algorithm is proposed. For the purpose of suppressing impulsive noise, this algorithm makes use of a bidirectional long short-term memory (biLSTM) network. By taking the covariance matrix of the array output signals as input, it generates a noise-free Toeplitz covariance matrix. Afterwards, the subspace technology is employed to carry out the DOA estimation. The proposed algorithm effectively addresses the challenge of impulsive noise interference and improves the accuracy of DOA estimation. Experimental results show that the proposed algorithm outperforms existing algorithms in terms of DOA estimation accuracy and robustness in the presence of impulsive noise.</p>