Help!!! Feature Extraction or SIgnal processing for spectrum waveform!!!

Hey, I am doing a project, Appliance signature identification by using RF method!

Every electrical appliances emits it's own and unique signal when it consumed energy. I would like to capture and record the signature of electrical appliances This project is more towards signal processing where the recorded signature will be processed and analysed to be used with classifiers to classify the source of the signal which is hope in turn will identify the appliance being used.

 Three different model of blenders was tested and three different type signals was obtained on spectrum analyzer.

Blender A http://img828.imageshack.us/img828/7899/scren011.gif

Blender B http://imageshack.us/photo/my-images/803/scren007.gif/

Blender B http://img651.imageshack.us/img651/6499/scren008.gif

By using this signal, I have no idea of h ow to analyze it as unique feature of blender. How to elimated the Noise floors from the signal by using matlab?

Peak detection algorithms?

Analyze it by using fourier tranform? or any else?

Any expect here can gv me some idea? thanks!!

[Image Analyst]

What form does the signature take? A 1D signal? An image? A bar code? A feature vector? Something else? Also, check out this tip: http://www.mathworks.com/matlabcentral/answers/6200-tutorial-how-to-ask-a-question-on-answers-and-get-a-fast-answer

[Chan]

RF waveforms from the spectrum analyzer!

[Image Analyst]

I see that now that you've added the photos. I don't have any code for you though.

Answer 1

[Greg Heath]

help patternnet
doc patternet

Hope this helps.

Greg

[Chan]

do you any code of Fast Fourier Transformation (FFT)for RF waveform?

[Image Analyst]

Why would you want the FFT when your signal is already in the frequency domain? Do you want to go back to the time domain for some reason?

[Chan]

what is the next step to get the feature extraction from the signal?

[Image Analyst]

Defining exactly what features you want. For example total power in frequencies less than (or greater than) xxx hertz, overall shape of the spectrum (as compared to other applicances), etc. Use whatever it is that uniquely identifies an appliance. Do the spectra look different to you? If so why? There must be something that is different or else you wouldn't notice any difference. You need to look at how different an appliance is compared to itself at different times so you can know how different it needs to be from other appliances. If it looks wildly different from itself from one time to the next, then how can you compare it to other appliances? You need to find something that is stable from one measurement time to the next, yet uniquely identifies that appliance. I don't know what that is - that's why I asked you what looks different.