Spectrum analyzer: windowing

snipped-for-privacy@gmx.de a écrit :

Windowing is done in time domain, by multiplying the time series with a weighting function which spans all the acquired time length. Now, the acquired time length is related to the intended frequency span. The less span, the more frequency resolution, the longer the time series. As the weighting function weights the time series in the same manner in terms of samples, whatever its time lenght, its effects in the frequency domain are the same WRT to the total freq span. IOW the shape of the windowing effects in frequency domain are constant in terms of frequency bins, whatever the frequency span.

You can easily verify this by injecting a single frequency into your analyser, say 640Hz, starting at 6.4kHz span. Being a single ray and neglecting frequency leakage, it should be within only one bin (you'll have to precisely adjust frequency so as to visualy reduce frequency leakage as much as possible). Now switch the window in, and observe how it modifies (widening) the visualized spectrum. Then switch the frequency span to 12.8kHz or 25.6kHz and notice that the spectrum lines shape and (percentage span) width has not changed.

I remember having seen a very good appnote on the subject, but I can't find it ATM.