Hanning vs Flat top window in Extract Single Tone

[viScience]

I was wondering if someone could explain why the Hanning window is used since the Flat Top window is reputed to be more accurate for FFT based amplitude measurement.

[Darin.K]

It does not help that they are so bloody similar, but in the pantheon of window functions there is the Hann window and the Hamming, but no Hanning (at least that I know of).  The statement about flat-top windows being the best really depends on the particular figure or figures of merit.  It does, for example have the best noise bandwidth, so when I look for signals buried in noise I usually go with that one.  Extracting a single tone is typically a different application, and you are trying to minimize bin leakage.  Here the flat-top is about as bad as it gets (unless you are really trying to be bad), and so most of the named window functions do better.  Arguing for Hann versus Hamming versus Bartlett versus whatever is trickier, the bottom line is they are all better than the Flat Top.

Other places to read:

http://forums.ni.com/t5/LabVIEW/Extract-Single-Tone-Information-from-Hann-Spectrum/td-p/139345

This particular VI has me a little scarred:

http://forums.ni.com/t5/LabVIEW/Scary-LabVIEW-images/m-p/1580676#M578210

[viScience]

Well I am certainly no expert, just been scouting around and found stuff like this...

From the Bruel & Kjaer Website:

Flat Top Window

The flat top window is a special time window with low ripple that is used in some FFT analyzers in addition to the more common Hanning Window and Rectangular Time Window. The flat top window does not allow as fine a frequency resolution as the Hanning window, but it will accurately measure the amplitude level of a signal at any frequency, even if the frequency is between the lines of the FFT analysis. The maximum picket-fence-error is 0.008 dB. It is used in transducer calibration systems to increase amplitude accuracy.

.

.

If you google 'Flat top window tranducer calibration' you will get lots of similar references.  Since I am using this in a tranducer calibration system for fundamental amplitude accuracy I thought I should investigate.

[Darin.K]

It sounds like in your case the most important figure of merit is amplitude, and so you will get slightly better results by using the lowest noise bandwidth window (flat top).  You lose a lot of frequency information unless your signal frequency is exactly commensurate with your sampling frequency (ie. falls into a single bin).  If you do not care then this will work.

Extract Single Tone is trying to make amplitude and frequency measurements so the Hann window is a very reasonable choice to give up a tiny amount of amplitude resolution to gain a lot of frequency resolution.

[viScience]

Since I have lots of CPU BW I think I can do both 🙂  I think a custom Extract Single Tone with the option of window type would be useful.

Thanks for the illuminating discussion.