Steps after ICA on unknown signals

Can anyone please suggest any algorithm to extract the signals after performing ICA on them. I have just mixture of signals coming from respective sites in certain order but after applying ICA the order gets mixed and I find it difficult to classify that which signal is for which site. So, can anyone suggest any algorithm that can do it automatically for me and tells me which signal is for which site?

Answer 1

[Star Strider]

ICA is not going to tell you that, because it does not know. It simply separates the signals. You must then so whatever sort of classification that is appropriate to characterise the signals ICA recovers.

[Aslam Khan]

Thanks for the response. After considering the factor you mentioned about ICA, I am asking for any algorithm that will help me to do so. Can you please suggest me any classification algorithm so that I can match the signals with before and after ICA and conclude which signals are for which one. Currently I am trying to use xcorr funtion to find the similarity between them and trying to judge through it, but I dont want to rely on it further.

Thanks again.Suggestions will be highly appreciated!

[Star Strider]

As always, my pleasure!

If you know something about the original sources of the signals, you can likely use those characteristics to classify them. Any classifier should work, although the classify linear discriminant may be easiest. The important thing is that you must know something about the characteristics of the signals from the source of each signal for this to work. If you do not, then the best you can do is to simply separate them using ICA. You can likely identify them only if each signal has something specific about it to identify it.

[Star Strider]

Aslam Khan’s Answer moved here:

I don't know much about the original signal, it's coming out from the bunch of nano electrodes(currently five). After getting it I am separating them using electrode numbers as a tag for an individual column of the data matrix. I am able to plot them individually before and after ICA but I am not sure which signal belongs to which one by visualizing them after performing ICA. If I am able to get your email I can sned you the plots that I have.Thank you again!

[Star Strider]

I do not understand. If the signals are from different electrodes, and you are recording them in different columns of your matrix, that would seem to identify them and their sources.

[Aslam Khan]

Yes, so we are using the electrodes implanted in five different locations which are hardly mm apart from each other. I am applying ica on them to minimize the effects of the other signals. So for each electrode the signals are recorded in each columns of 5 by 200k matrix(approx). The signals are very low and highly sensitive. Hope I made the point to get some idea about it.

[Star Strider]

The best classifier in that instance may be the -nearest neighbor classifier, fitcknn or the related knnsearch function. (Choose the most appropriate one for your data.) You can use the data from the electrode whose identity you know to find the closest match to waveforms in the ICA results.

[Aslam Khan]

Thanks a lot,I will try it !

Thank you for your time and suggestions, appreciated! I always need mentors like you for my research to be moving.

If we can connect. My email: [e-mail].

[Star Strider]

As always, my pleasure!