SAMPLING EFFECTS ON THE EMPIRICAL MODE DECOMPOSITION

Abstract

Standard exposition of Empirical Mode Decomposition (EMD) is usually done within a continuous-time setting whereas, in practice, the effective implementation always operates in discrete-time. The purpose of this contribution is to summarize a number of results aimed at quantifying the influence of sampling on EMD. The idealized case of a sampled pure tone is first considered in detail and a theoretical model is proposed for upper bounding the approximation error due to finite sampling rates. A more general approach is then discussed, based on the analysis of the nonlinear operator that underlies the EMD (one step) sifting process. New explicit, yet looser, bounds are obtained this way, whose parameters can be estimated directly from the analyzed signal. Theoretical predictions are compared to simulation results in a number of well-controlled numerical experiments.