I have a noisy signal with frequency between 1/3 and 1/5 of the sampling window.
DFT has a very low resolution for low frequency components, what is the best way to find the frequence?
3
-
$\begingroup$ Do you mean you only get 1/3 to 1/5 of a period in the sampling window, or that you only get 3 to 5 periods in the sampling window? $\endgroup$– Peter K. ♦Commented Apr 27, 2013 at 1:30
-
1$\begingroup$ I get 3-5 periods in the sampling window. I would like to find the frequency with a better precision then the FFT resolution. $\endgroup$– AndreACommented Apr 27, 2013 at 14:07
-
$\begingroup$ @user175348: Have you tried interpolating the FFT output? ccrma.stanford.edu/~jos/sasp/… gist.github.com/endolith/255291#file_parabolic.md $\endgroup$– endolithCommented May 1, 2013 at 18:58
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
2
Check out this article by Eric Jacobsen and Peter Kootsookos: http://www.ingelec.uns.edu.ar/pds2803/Materiales/Articulos/AnalisisFrecuencial/04205098.pdf
I had some success using this method for doing "sub-bin" frequency estimation when only one sinusoid is present.
-
$\begingroup$ +1 for Eric's work! Check out his frequency estimation page here, too: ericjacobsen.org/fe.htm $\endgroup$– Peter K. ♦Commented May 4, 2013 at 13:47
-
$\begingroup$ Updated paper link: academia.edu/download/70249550/04205098.pdf $\endgroup$– MBazCommented Sep 20, 2022 at 2:03
$\begingroup$
$\endgroup$
Have you tried just zero-padding?
Suppose I have:
SINE = sin(2*%pi*0.001923492369*[0:1023] + rand(1)2%pi);
And I zero-pad it:
SINE_ZERO_PADDED = [SINE zeros(1,10240)];
Then I can take the FFT of both:
where the blue plot is the absolute value of the FFT of the original, and the red (dashed) plot is the absolute value of the FFT of the zero-padded signal.