If you break a 4 second signal into 0.5 second segments, add them together and then perform an FFT on the resultant 0.5 second data.

Will the FFT / frequency spectrum be the simlar to an FFT on the original 4 second data?

This would mean that you can store long amounts of data in smaller memory if all you cared about was retaining the frequency components... I'm not sure if I am missing something though?

  • $\begingroup$ No, as Fat32 explained, but what you can do is sum the magnitudes to get a non-coherent gain for the FFT spectrum that you would get in a 0.5 sec data segment. This is similar to getting a finer “Video Bandwidth” on a spectrum analyzer- it does not reduce the noise density as a longer sequence would but it does smooth (average) the noise floor. $\endgroup$ – Dan Boschen Sep 30 '18 at 14:52
  • $\begingroup$ @DanBoschen Hi Dan! Assuming that OP is indeed into spectral analysis of periodic signals (other then signal compression using some esoteric FFT tetchnique which I initially thought of), then I should delete my answer... And then, I consider that OP is asking a puzzle rather then a question and that is quite annoying isn't it :-))) $\endgroup$ – Fat32 Sep 30 '18 at 18:47
  • $\begingroup$ Nothing wrong with your answer or the question. It is very clear that you answered the question so I wouldn't delete it. I was just offering further insight reading between the lines and helping to further the understanding. All good! Never annoying. $\endgroup$ – Dan Boschen Sep 30 '18 at 18:50
  • 1
    $\begingroup$ @DanBoschen ok then, lets wait to see if OP clarifies what he actually meant to ask: 1-) Signal compression (seems weird to me now) or 2-) Spectral analysis of periodic signals (a much reasonable application)... $\endgroup$ – Fat32 Sep 30 '18 at 18:54

That's not true (unless the signal is periodic, and those pieces are the base periods of it).

If you divide a signal $x_L[n]$ of a length $L$ into smaller pieces $x_i[n]$ of length $N$ each,

$$ x_L[n] = x_1[n] + x_2[n-N] + x_3[n-2N] + x_4[n-3N]$$

and sum them into a shorter signal $x_N[n]$ of length N

$$ x_N[n] = x_1[n] + x_2[n] + x_3[n] + x_4[n]$$

and take $N$-point DFT/FFT of their resultant $x_N[n]$, you won't be able to reconstruct the original long singnal $x_L[n]$ back.

Indeed it's not even about FFT. You would not be able to retrieve the components $x_i[n]$ from their sum $x_N[n]$. So since there is no way to reconstruct back $x_L[n]$ in this manner.


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.