Multi-Domain Extended ELM-DOA Subarray Localization Method Based on Distributed Coherent Aperture Radar

In distributed coherent aperture radar (DCAR), the fundamental problem in ensuring the full coherence of the radar system is how to realize the precise positioning of the subarray. The subarray positioning scheme is designed based on the multi-domain extended extreme learning machine (ELM)-direction of arrival (DOA) estimation for the mobile DCAR system. The multi-domain partition scheme improves the ELM model, making the algorithm effective and low-complex in the changing electromagnetic environment. Firstly, the array covariance matrix is constrained by L2 regularization to decrease the influence of different feature measurement scales on the model. An airspace extension ELM (AEELM) method, which reduces the original matrix by the spatial partition layer and the sequential process, is proposed. This method diminishes the size of the model's solving matrix and enhances the DOA estimation accuracy while increasing or refining the samples. Secondly, an extended ELM method based on partition projection (PPEELM) is proposed to unravel the problems of high dimensions, small samples, and complex electromagnetic environments in radar array features. The radar covariance matrix is analyzed by partitioned multi-space projection according to the changes in the scene's electromagnetic environment. The operation can ameliorate the model's ability to solve nonlinear problems in complex environments. Therefore, the fully coherent system's position estimation accuracy and signal-to-noise ratio (SNR) gain are ameliorated. Numerical simulations show that AEELM can effectively augment the estimation accuracy and curtail the maximum matrix dimension of a single training. The PPEELM method can ensure the system coherence performance is more excellent than 95% when the subarray signal's SNR is greater than 0 dB in small scenarios.