I have some signals with different lengths (from 1 to 129 ) and I need to compare them and find their similarities and differences. Actually, I need to extract features for the signals regardless the origin of them. (I do not know the origin, I just have the samples). What methods can be applied to analyze signals with different lengths and find their differences?
http://en.wikipedia.org/wiki/Sequence_alignment
You can get some ideas here. This is a standard problem in genetics. Maybe you want to contact someone in the field of bioinformatics.
If I were you I would try to use a mathematical program such as matlab or scilab (scilab is a free software). You can put there the samples of your signals (in vectors) and some additional commands to extract features - assuming that you also have time vector (without it you won't be able to assess the frequency). Those programs are also a good choice if you are looking for a correlation between signals.
I hope this is what you were looking for.
I think you problem is about correlation. There are many formulas to deal with the relationship between two signal, for example one of them is canonical correlation analysis (CCA).
Thank you all for the answers.
Marcin Walczak, i am already using matlab.
I had signals with the same lengths and i extracted lots of features for them (e.g. auto-correlation, cross correlation, average of wavelet coeffs, spectrogram, ... ) , but now i have signals with different lengths and i want to know which features are suitable for them and what are the methods to determine differences between them. There is some clue. Processing methods of speech or text or seismic signals are suitable since these signals are from different lengths too. But what are these methods? What are the common methods for analyzing speech or text signals for instance?
First of all since your signals are from different origin, doing a normalization (dividing by the square root of the signal energy) would be important. One, very recent method, that I have been using is to generate the signals time-frequency distribution, so that you have a 2D representation of the signal in time and frequency domains, simultaneously. Afterwards, the image processing feature extraction techniques tools could extract meaningful features. Though you might need to use some feature selection techniques to narrow down the feature set to small more informative set. You can find some of Dr. Behnaz Ghorani works on this matter.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Process