Quasi-Newton algorithm for adaptive minor component extraction

Abstract

An adaptive quasi-Newton algorithm is first developed to extract a single minor component corresponding to the smallest eigenvalue of a stationary sample covariance matrix. A deflation technique instead of the commonly used inflation method is then applied to extract the higher-order minor components. The algorithm enjoys the advantage of having a simpler computational complexity and a highly modular and parallel structure for efficient implementation. Simulation results are given to demonstrate the effectiveness of the proposed algorithm for extracting multiple minor components adaptively.