There are many algorithms for computing the EMD of a given signal, such as those developed by Flandrin and Huang. An important drawback of EMD is its computational load, which increments for longer signals. It is therefore highly desirable to calculate the EMD in short segments of the signal. I would like you to recommend me who is working in this kind of algoritms. Thank you!
Hi,
I am working on developing a toolbox for the Hilbert-Huang Transform. I've written some codes in MATLAB, will be happy to send them over, if you like. Currently I'm working on the same algorithm in Python.
My code bottlenecks at the cubic spline interpolation - that's the most complex mathematical piece in the whole algorithm.
Please consider going through my blog post (see attached link), if you're interested in more discussion.
Thanks
http://brocabrain.blogspot.com/2011/04/empirical-mode-decomposition-riding.html
Ok, thank you very much Jaidev. I am very interested in your matlab code for estimating the HHT. In this stage, what algorithm do you use for determining the instantaneous frequency? Maybe the "instfreq.m" function from INRIA's time-frequency toolbox?
Have you worked in online EMD? My interest is to analyze very long acoustic signals, where classic EMD is impractical.
I want to process any 1D signal with EMD. One option is to analyze the whole signal, another option is to analyze the input signal in short segments. I want to do this, but have not found any algorithm for this purpose. How do you analyze very long signals with EMD?
Which algorithm is best for estimating the instantaneous frequency? After applying EMD to the input signal, I estimate the Hilbert transform for each IMF and then compute the phase of the corresponding analytic signal. An obvious option for obtaining the frequency is to differentiate numerically the phase. Since this result is highly affected by noise, which other options do you recommend for calculating the instantaneous frequency?
Does Wigner distribution suffer from uncertainty principle? In order to avoid this, I estimate the instantaneous frequency by means of the Hilbert transform. But the result is too sensitive to local errors, i.e. with many 2*pi jumps. Any other suggestions?
Fernando, from the Wigner distribution you can obtain the full complex amplitude of a deterministic signal, i.e., its f(x)=|A(x)|exp(i·P(x)). The algorithm used is quite simple and direct, at least for 1D signals which is what I believe you want to do. They basically involve a FFT.
I mainly do this kind of things for characterization of 1D/2D stochastic/deterministic optical signals. I am interested in the phase of deterministic signals and the coherence properties of stochastic signals, but I believe it could be directly used to obtain the frequency over time in your case.
You could read some of the articles we have published about the topic, especially the one from 2009 and the Radon-Wigner display one from 2011. I do not have many articles published, so it wont be hard to find them.
I hope this helps :-)
Gracias Alejandro. ¿Me podrías enviar el artículo donde calculás la frec instantánea a [email protected] ?
Abrazo.
Fer
My undegrad thesis was about online (streaming-data) calculation of the Empirical Mode Decomposition. You can find it here, with MATLAB code: http://www.penwatch.net/cms/?p=24
Some implementations of more or less "production" grade are beginning to appear in the wild, too. See this for one example: http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5518428
- Badges
- Science topic
- Similar topics
- Thesis Research