Title:Cubic NK-SVD: An Algorithm for Designing Parametric Dictionary in Frequency Estimation

Abstract:We propose a novel parametric dictionary learning algorithm for line spectral estimation, applicable in both single measurement vector (SMV) and multiple measurement vectors (MMV) scenarios. This algorithm, termed cubic Newtonized K-SVD (NK-SVD), extends the traditional K-SVD method by incorporating cubic regularization into Newton refinements. The proposed Gauss-Seidel scheme not only enhances the accuracy of frequency estimation over the continuum but also achieves better convergence by incorporating higher-order derivative information. We rigorously establish the convergence of the proposed algorithm within the Block Coordinate Descent (BCD) framework. To the best of our knowledge, this is the first convergence analysis of BCD with a higher-order regularization scheme. Furthermore, this convergence framework can be applied to designing general alternating minimization algorithms with higher-order regularization. Extensive simulations demonstrate that cubic NK-SVD outperforms existing state-of-the-art methods in both SMV and MMV settings.