Dolph-Chebyshev 150 window noise floor

[Braca]

When testing a Topping D10b DAC at different output voltages in loopback with a Cosmos E1DA ADC I noticed that the spectral noise floor of the Dolph-Chebyshev 150 window varies with the signal amplitude.
Both the noise level and its variance, i.e the line thickness, are affected.
No other in-built spectral window displays such behaviour.
The spectra in the attachment were obtained at the respective levels of -1.26, -7.28, and -60dBFS (top to bottom), FFT 512k, overlap 87.5%, 64 averages.

 

[John Mulcahy]

You can see that effect with several window types when the stimulus is not periodic within the window, dramatically with rectangular of course, but also with Blackman-Harris 4 and Flat-top, for example. Choose a window to suit your measurement noise floor, such as Blackman-Harris 7, Dolph-Chebyshev 200 or Cosine Sum 9-235.

 

[Braca]

I'm not using D-C 150 window in my work (I'm using Blackman-Harris 7 most of the time), but I ran across this peculiarity of the D-C150 and thought it worth mentioning.

In addition to the signal amplitude, the Dolph-Chebyshev 150 window appears to be also dependent on the overlap fraction.
The first attachment shows a series of ten successive measurements at the same signal level where the noise floor randomly varies between -151 and -162dBFS. While the THD is correct and consistent in all 10 runs, THD+N varies between -99.4 and -107.2dBFS.
Using the overlap of 87.5% the results remain consistent (att. 2, violet tracks, 10 runs), but the noise track is wrong as regards the value, and in comparison with the result obtained with the Blackman-Harris 7 (the orange track) doesn't look right.

This behaviour could be quite a burden for an inexperienced user.

BTW, REW has been my standard tool for a few years, and as an ex-developer of large simulation software systems I appreciate very much John's efforts to make it better with every new release.

 

[John Mulcahy]

This behaviour could be quite a burden for an inexperienced user.

Perhaps, but that's how windows behave. D-C 150 is optimised for all sidelobe amplitudes to be 150 dB down. If the measurement noise floor is lower than that it isn't a good choice.

 

[Braca]

Thank you for the quick response.

It might be called nit-picking, and I promise not to push this matter any more, but all other windows in REW do not have any problems with my low-noise floor.
Among them are the humble Hann window with its first sidelobe at -31dB and Blackman-Harris 4 at -92dB.
The attachment shows a succession of measurements at -1.3dBFS with all REW windows, with the exception of the rectangular and Dolph-Chebyshev 150.
In terms of V/rtHz they all converge to the same noise floor value, as expected.