0
$\begingroup$

enter image description here

My FFT spectrum analyzer works fine in general except when the frequency is an exact multiple of the frequency bin. In the pic at the left the frequency is 5999Hz and at the right 6000Hz. The lines that appear when the frequency is an exact multiple, appear randomly (i.e. when I change the frequency to 5999Hz and back to 6000Hz the lines may be fewer or in some cases non-existent at all ).

How should I interpret the FFT results when the frequency is an exact multiple of the frequency bin?

$\endgroup$
1
$\begingroup$

Short term FFTs apply a windowing function at moving intervals, do an FFT, and average. In this case, the interval with which the windowing function is applied is phase-locked to the signal, so what tends to be a process averaging out over the integration time no longer does. You get the FFT of this intermediate product without averaging and it depends on just where you locked phase what comes out.

However, you did not state just what kind of signal you are using. For a sine wave, I still would expect the result to be pretty close to a single peak. It may be that this peak is absent due to fixed point overflow since the analyzer might not be prepared for the full concentration of frequency. If that is the case, just reducing the amplitude might change the outcome to something more expected.

$\endgroup$
  • $\begingroup$ +1 Thanks. I' don't understand the first part of your answer. I do a simple FFT on a real time signal. The waveform is a sine, and yes the amplitude seems to affect the appearance of these artifacts but affects these in the random way I described in the question. Yes it probably is a float overflow problem. I guess I'll have to print the FFT results and check the values of the real and the imaginary parts at these frequencies I get the artifacts so to try to understand what causes them and if I get overflow exceptions in the computation. $\endgroup$ – John Am Jun 15 at 9:46
  • $\begingroup$ I'm not sure what are the advantages or when to select STFT over FFT for my applications (real time audio analysis). Also does FFT with a buffer resembles STFT? $\endgroup$ – John Am Jun 15 at 10:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.