I have to find FFT of difference between two voltages in my circuit. ltspice just give me FFT of one node voltage, though I transport my data to matlab and then write a code to re-sample data and then find FFT of this re-sampled data. I check it for a spice AM source and also a nonuniform sampled signal that I've been created in matlab, but it doesn't work good enough (Of course I considered Nyquist and sampling frequency).
Could anyone help me out in this?
Thanks in advance for your consideration.
Dear Paniz,
Resampling is okay. But you have to constrain your sampling such that it is not less than the highest sampling rate of your signal. The other condition is that the original samples must be included in their original positions. In this way you will get the expected FFT results.
The other important thing is that there is a basic difference between the Fourier transform and the DFT which is that the DFT is the sampling of the Fourier transform in the frequency domain.
Best wishes
nonuniform data you mean your X is data have different value between neighbor digits ?
make it uniform
% x,y, - your nonuniform data
Yuni=interp1(x,y,Xuni); % Xuni - uniform data points that you previously describe
Thank you Arkady.
Yes, X(here the time) is non uniform. it means that time intervals are not equal. The "re-sample" command worked better than interp1 in my case. I think re-sampling part is okay but I have no idea why the FFT is not the one that I've expected.
You need to understand for yourself what is happening as a result of this uneven spacing of signal samples :
-to make it tangible, let me transpose your problem in the audio domain.
You have a music pop band with double bass (very low frequency mainly), and a very high pitched flute (and maybe guitar).
If you have low-pass filtered your signal, you can sample it with fairly long intervals between two samples (sampling at 2f if low pass below f: see Nyquist rules to avoid spectral aliasing ).
If you high pass your signal, you need higher sampling rate hence shorter interval between samples.
Low-pass: you get the double-bass, you miss the flute
Few samples, highly spaced...
High-pass: you get the flute but the double-bass is not quite there.
Many samples, short time interval between samples.
With the FFT you get everything in the frequency domain.
You mix areas with
-numerous samples, and short intervals between them, giving possibly the higher frequency representation ok
-few samples, highly spaced.
If you had been able to low-pass filter your signal before producing these samples, you might have a somewhat Ok representation of the lower frequencies part, and not the higher one.
Without the low-pass filtering, you get aliasing, that is artificial frequency components: with signals at f1 and f2, you get , (f1+f2)/2 and (f2-f1)/2. Very horrible noise to your ears if it's audio.
That probably explains your surprising FFT.
Try to isolate the signal segments [t1, T2] with high sampling rate (short dt between samples)
There your windowed FFT will cover from low to high frequency properly.
For intervals [t'1,t'2] where the sampling rate is low and the spacing dt is high, you should do your interval restricted FFT with no accuracy on high frequencies components...
Dear Paniz,
Resampling is okay. But you have to constrain your sampling such that it is not less than the highest sampling rate of your signal. The other condition is that the original samples must be included in their original positions. In this way you will get the expected FFT results.
The other important thing is that there is a basic difference between the Fourier transform and the DFT which is that the DFT is the sampling of the Fourier transform in the frequency domain.
Best wishes
- Badges
- Science topic