How to interpret FFT output, Spectrogram and calculate Spectral centroid

Hi how do i interpret the FFT output, spectrogram (the figures below) and how do I calculate the spectral centroid ? The way i calculate spectral centroid is using summation(amplitude * frequency) / summation(frequency) Thanks in advance

Answer 1

[Star Strider]

Plotting your fft as a loglog plot would help you interpret it. You have a large d-c (constant) offset to your signal, and unless you eliminate that by subtracting the mean of your signal before you take the fft, only a loglog plot will let you see the details.

You can calculate the Spectral centroid relatively easily now that you have correctly calculated and plotted your fft.

The spectrogram is difficult to interpret. The only reasonable interpretation is that your signal has completely different frequency content at different times. There does not appear to be any pattern in it with respect to time.

[crixus]

When you say log log plot, i assume the y axis has to be of log scale as well. Can I know what do you mean by D-C offset ?

I followed the tutorial at http://jwilson.coe.uga.edu/emt668/EMT668.Folders.F97/Patterson/EMT%20669/centroid%20of%20quad/Centuse.html to calculate the spectrum centroid which I assume to be the same as the one at wikipedia ??

(Sorry for so many questions, i have bought the book that you recommended but have yet to arrive so my understand at signal processing is still very poor)

[Star Strider]

My use of ‘d-c offset’ is an electrical engineering term. It refers to the direct-current component of an otherwise oscillating signal. In practice, it is simply the mean of any signal.

You assume correctly. The definition of ‘spectral centroid’ you’re using is the same as the Wikipedia definition.

No worries with the questions! I’ve taken several signal processing courses over the years (not covering all the same information, since some involved wavelets, neural networks, hardware implementations of filters, and other techniques), and I know from my experience it takes time to learn them. If it was possible to learn signal processing overnight, there would not be year-long courses in it! Take your time, and understand it.