Direction of Arrival Estimation Based on Shifted Coprime Arrays and Deep Learning

To address the issue that existing direction of arrival (DOA) estimation methods have limited resolution capability in multi-target scenarios, this paper proposes an end-to-end DOA estimation network framework based on a multilayer perceptron (MLP). The framework first designs a subarray cyclic permutation array (SCPA) coprime array structure to expand virtual array elements and improve array degrees of freedom. Then, it uses physical array elements to receive snapshot information, calculates the covariance matrix, and further projects it onto the difference co-array to generate a virtual co-array spatial pseudo-spectrum estimate (PSE) containing target orientation discrimination information as the network input. Through the deep learning network, the pseudo-spectrum features are extracted and discriminated, thereby achieving high-resolution DOA estimation and spatial spectrum reconstruction of multiple targets. Experimental results show that the proposed method exhibits superior performance under different SNR and target density conditions, maintaining clear spectral peaks and accurate peak positions in various signal scenarios, thus realizing stable DOA estimation.