Fast Super-Resolution DOA Estimation for Unknown Number of Radiation Sources Based on Complex Convolutional Neural Networks

To address the high computational complexity of traditional Direction of Arrival (DOA) estimation techniques and the limitations of previous neural networks in handling fixed signal sources, a new signal processing model was developed. This model employs sparse representation methods and deep learning through complex convolutional neural networks(CCNN). Focused research on the array’s performance under unknown radiation sources using the complex convolutional neural network algorithm for super-resolution DOA estimation was conducted in a PyTorch environment. The results indicate that the complex convolutional model has a lower DOA computation complexity and higher precision in handling unknown radiation sources. After incorporating sparse representation with sampled data through the CCNN model for DOA estimation, an improvement in estimation accuracy and a reduction in computational complexity were observed compared to methods relying solely on sparse representation. Additionally, this model can effectively detect incoming waves from unknown radiation sources, unlike previous neural networks.