0
$\begingroup$

I have developed my own window function for pre-processing signals prior to a DFT (FFT). How do I measure the performance of the window so that I can specify how good (or bad) it is? Maybe other people would be interested in the window but only if I can quantify its performance I think.

If I can give an example... When I use a Blackman Harris window on a pure dc synthetic signal using a 48,000 sample DFT (with 1Hz resolution), the Blackman Harris window measures around 4dB below for the 1Hz bin. A Hanning window output is around 6dB down. My window gives a 1Hz bin value of around -170dB.

$\endgroup$
6
  • $\begingroup$ A window is measured by the coverage of all frequencies. I am not sure it is too complicated to design a filter for a very specific signal with high attenuation. 170 db means about 57 bits quantization noise. Where you see it used and for what purpose? $\endgroup$
    – Moti
    Commented Sep 11, 2020 at 4:49
  • $\begingroup$ The sum of all other frequencies is around -210dB/Hz relative to the dc bin. I wanted a window where I could still discern small frequency components closely adjacent to large components yet still measure both with high accuracy. At the moment seem to be getting accuracy better than 0.1 dB but I am sure this could be improved. $\endgroup$
    – Richard
    Commented Sep 11, 2020 at 5:29
  • $\begingroup$ It seems to be an attenuator where the frequency not of interest are 40 db below. $\endgroup$
    – Moti
    Commented Sep 11, 2020 at 5:32
  • $\begingroup$ The window function is traditionally used before an FFT - if you need, you can look up why and how they are used. My window function does not attenuate the signal or the other bins. While it does leak signal from the bin where the signal exists to others, that leakage is better than -170dB. This is superior to any other window I have come across - and like 160dB+ better than most Moti. $\endgroup$
    – Richard
    Commented Sep 11, 2020 at 7:19
  • $\begingroup$ Everything you need is in this great paper by fred harris: web.mit.edu/xiphmont/Public/windows.pdf $\endgroup$ Commented Sep 11, 2020 at 15:19

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.