Dynamically sampled multivariate empirical mode decomposition

Abstract

A method for accurate multivariate local mean estimation in the multivariate empirical mode decomposition algorithm by using a statistical data-driven approach based on the Menger curvature measure and normal-to-anything variate-generation method is proposed. This is achieved by aligning the projection vectors in the direction of the maximum 'activity' of the input signal by considering the local curvature of the signal in multidimensional spaces, resulting in accurate mean estimation even for a very small number of projection vectors.