Empirical Mode Decomposition aided by adaptive low pass filtering

Empirical Mode Decomposition (EMD) is an adaptive signal analysis technique which derives its basis functions from the signal itself. EMD is realized through successive iterations of a sifting process requiring local mean computation. For that purpose, local minima and maxima of the signal are assumed to constitute proper local time scales. EMD lacks accuracy, however, experiencing the so-called mode mixing phenomenon in the presence of noise which creates artificial extrema. In this paper, we propose adaptively filtering the signal in Discrete Cosine Transform domain before each local mean computation step to prevent mode mixing. Denoising filter thresholds are optimized for a product form criterion which is a function of the preserved energy and the eliminated number of extrema of the signal after filtering. Results obtained from synthetic signals reveal the potential of the proposed technique.