Empirical Mode Decomposition from a Filter Bank Viewpoint
After recalling the Empirical Mode Decomposition (EMD) principle, we will provide this entirely data-driven signal decomposition introduced by Huang et al. (1998) a filter bank interpretation from two complementary perspectives. First, a stochastic approach operating in the frequency domain evidences the spontaneous emergence of an equivalent dyadic filter bank structure when applying EMD to the versatile class of fractional Gaussian noise processes. Second, a similar structure is observed when operating in the time domain on a deterministic pulse. A detailed statistical analysis of the reported behaviour is carried out on the basis of extensive numerical simulations, allowing for a number of applications. New EMD-based approaches are outlined in different directions, namely estimating scaling exponents in the case of self-similar processes, performing a fully data-driven spectrum analysis and denoising-detrending signal + noise mixtures.
