# Is there any problem with my FFT method?

I am trying to analysis a system by Fourier Transform. my system has 3 dominant modes. (1,0.05) (0.51,0.01),(0.46,0.01) the pairs are defined as (freq,damping_ratio)
I've got those result with eigenvalue analysis.the system contains 3 signal.so these frequency[1,0.51,0.46] must be find in these three signals too. so I've run FFT but I didn't see these frequency in the plot. what is the problem in your mind? why these frequencies do not exist in FFT plots?
Here you can get the signal this txt file contain 3 signal in it. I put last signal FFT(PSD) here. Tnx
as you see those frequencies don't exist in the 3rd-signal. its same for other 2 signals too. Also I will put my codes here.

t=linspace(0,30,3012);
fS=ceil(inv(t(2)-t(1)));
x=data(:,3);
Nfft = 10 * 2^nextpow2(length(t));
psd = 20.*log10(fftshift(abs(fft(x,Nfft))));
freqs=[0:Nfft - 1].*(fS/Nfft);
freqs(freqs >= fS/2) = freqs(freqs >= fS/2) - fS;
freqs=fftshift(freqs);
figure(1);
plot(freqs, psd);
xlim([-1.5  1.5]);
xlabel('Frequency / Hz');
title (sprintf('PSD'));

• Try removing the DC component, e.g. x = data(:,3) - mean(data(:,3)); and redo the FFT. – geometrikal Aug 21 '13 at 21:30
• The peak separation delta F is roughly between 2 and 3 times the sample rate divided by the length of data fed to a FFT. For finer resolution between closer frequencies, use more data. – hotpaw2 Aug 22 '13 at 2:35
• Try zero padding the signal if you can't get more data as @hotpaw2 suggests. See dsp.stackexchange.com/questions/741/… – geometrikal Aug 22 '13 at 4:04
• I don't think zero padding will help resolve two components in the same bin, it will only interpolate the existing values. A longer data series will help though. – Speedy Aug 22 '13 at 9:39
• Are you windowing before doing the FFT? Your start and end sample are certainly not the same, so they are creating a discontinuity, which produces wide skirts around the frequencies you're interested in. Zero padding does not give you any better resolution. The only way to improve resolution is to use more time domain samples. – endolith Aug 27 '13 at 20:27