Time domain Signal decomposition

[JanIR]

I am trying to Identify different noise components in a signal which is in the time domain. There are atleast two different processes which contribute to the noise in my target signal. My problem is to identify one from the other and the severity of each one. One of the types of noise is a "brown" or "pink" type noise which has a negative "linearish" behaviour in the frequency domain. The other form of noise has a "step and Discharge" type response in the time domain. The problem is that the latter type of noise is a totally random occurence. Also the relative magnitude of the first type of noise is also unknown. If anybody has any ideas on how to identify and seperate these two noise processes by looking at the combined output, it will be greatly appreciated.

 

[melone]

What is this for? It sounds interesting, but I don't see the final application. Perhaps if you gave a little more information on the final application, we can suggest some cool ways to solve the problem.

 

[nbucska]

Often using the cepstrum ( spectrum of spectrum ) may be
informative if there are multiple fundamentals with their harmonics.

<nbucska@pcperipherals.com>

 

[JanIR]

This is for a research project of mine, trying to identify, quantify and create a model for different kinds of noise seen in a 10-25 MHz signal. There are different processes at work creating noise in my system (thermal, External coupling, and most I don't know about).The noise created by different processes can be seen emperically but I have about a terabyte of data (millions of waveforms)which I need to analyse. So I need to create a computer program to look at each waveform and try and identify where different kinds of noise can be seen. When this is done the next step is to create a statistical model for each noise phenomenon.

 

[nbucska]

Well, it doesn't look to difficult. You can use FFT tho get the spectra and cepstra (FFT of FFT ). I am sure you can find FFT on the WEB. Is this SETI ? if not, they may have related experience.

<nbucska@pcperipherals.com>

 

[hacksaw]

FFT will make mince meat of your data.

The spectral energy density is useful where you are trying to quantify the processes involved and the sources are not correlated.