0
$\begingroup$

I am performing an Ensemble Empirical Mode Decomposition (EEMD). I wrote a MATLAB routine to perform the EMD. I noticed that when adding white Gaussian noise to the input signal, sometimes I obtain a different number of modes, even if the noise variance is not so high. Clearly, if this happens, I cannot sum compute consistent ensemble-averages for each mode.

Do you think this depends more on the nature of the input signal or on my EMD algorithm/routine?

$\endgroup$
  • $\begingroup$ "...sometimes I obtain a different number of modes...", than when? What is the comparison against? $\endgroup$ – A_A Jul 2 at 10:44
  • $\begingroup$ I mean, if I add two different White Gaussian noise realizations, I may obtain a different number of IMFs when I perform the EMD. For example I can obtain 11 IMFs or 12 IMFs for two added noise realizations. Maybe I didn't explain myself clearly. I hope it's clear now. $\endgroup$ – EmThorns Jul 2 at 12:14
  • $\begingroup$ Thank you, this is what I understood too but wanted to confirm. Can I please ask what sort of "problem" you are dealing with? What sort of signals are you working with? $\endgroup$ – A_A Jul 2 at 12:29
  • $\begingroup$ It is a transient signal, so basically it starts with constant value, then there is a linear increase to another value and then the value remains constant until the end of the signal. In addition I have a superimposed sinusoidal noise plus random white Gaussian noise. $\endgroup$ – EmThorns Jul 2 at 12:34
  • $\begingroup$ Alright, I meant what sort of domain (e.g. Audio, seismology, biomedical, etc). Also, which stopping criteria are you using? (Matlab already has hht (?)) $\endgroup$ – A_A Jul 2 at 12:47
0
$\begingroup$

It depends on both on the signal content and on the EMD implementation. EMD can be sensitive to time shifts, especially with strong transients, to noise power. Robust or constrained EMD on single signal or multivariate signals have been developed.

Your question relates to signal morphology, processing purpose, and quality metrics. I doubt they can be answered globally.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Hi Laurent, thanks for the answer. Indeed, I have chosen to use EEMD because it is more robust than the simple EMD. While computing the EEMD, for the moment being I decided to select only the EMDs that return a number of modes that occurs more frequently (say 10) and to discard the other ones. (for ex. those with 11 modes). I see there are robust algorithms, but I think it is difficult to find one that is suitable to a specific case. Is there any robust algorithm that you would suggest, especially for transient signals? $\endgroup$ – EmThorns Jul 6 at 10:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.