Other ways to improve fft amplitude

Are there any other ways to improve the amplitude accuracy of fft, other than filtering the signal and using a flat top window, which I've already tried? Thank you in advance.

Answer 1

[Star Strider]

Another option may be to zero-pad the fft using:

Fs = signal_sampling_frequency

L = signal_length;

NFFT = 2^nextpow2(L)

FTsignal = fft(signal.*window, NFFT)/sum(window)

Fv = Fs*(0:(NFFT/2))/NFFT; % Frequency Vector

or something similar.

Increase ‘NFFT’ by powers-of-2 (for the most efficient results) to increase the frequency resolution of the result. This may reduce spectral leakage, since that may also improve the amplitude (magnitude) estimation. Experiment with this to see if it works in your situation.

[Claudio Gerosa]

Hi, thanks for your answer. I tried that but it didn't really change much. Would changing my sampling frequency or the number of samples do anything to the spectrum?

[Star Strider]

My pleasure!

I doubt that changing the sampling frequency (for example using using the resample function) would change anything, since that would simply interpolate the existing data to a new sampling frequency. (This also creates data where none previously existed, so I generally use resample to change a signal with non-constant sampling to a signal with constant sampling at approximately the original samplilng frequency.)

If you have more samples in your signal that you originally used, then definitely use the entire signal vector with fft.

Increasing the frequency resolution is generally the preferred method, since it does not change the original data. I also experiment with different windows (I usually end up using the hann window), since this also may improve the result by correcting the finite nature of the discrete Fourier transform to something approximating an infinite Fourier transform. With sampled data, there are limits to the ability to analyse any signal.

[Claudio Gerosa]

Thank you very much.

[Star Strider]

My pleasure!

If my Answer helped you solve your problem, please Accept it!

.