Research on Multiple Cross-Correlation DOA Estimation Methods Under Hybrid Analog-Digital Architectures on Deep Learning

Direction of Arrival (DOA) estimation of spatial signals is crucial in array signal processing. To address the defects of uniform linear array far-field narrowband incoherent multi-target estimation algorithms, such as poor adaptability to low signal-to-noise ratios, few predefined angles, high computational complexity, and low accuracy, and to overcome the difficulty of existing deep learning methods in effectively extracting complex-valued features of data, this paper proposes a multiple cross-correlation DOA estimation method based on deep learning with a hybrid modulus structure. This method transforms DOA estimation problem into an inverse mapping problem from the signal average power to the target arrival angle, extracts the modulus of the signal average power to construct the input data of the network, and constructs a convolutional neural network combined with an attention mechanism to extract data features. The labels of the network correspond to the arrival angles of the target, thereby achieving the DOA estimation of multiple signal sources. Simulation results indicate that this method can fully extract spatial features, improve the accuracy of DOA estimation, and reduce algorithm complexity. Meanwhile, it significantly outperforms traditional algorithms in estimation accuracy under the conditions of low signal-to-noise ratios and few predefined angles.