THD calculation & THD vs frequency graph smoothing

[_redfox]

Hello!
I've recently downloaded a new REW beta (5.40 beta 98 to be specific) and now I'm puzzled. It seems that new beta not only uses different algorithm to plot THD vs freq. data (graph seems less smoothed), but the entire calculation process has changed!

These are 2 exact files of sweep recordings I imported. The THD value at higher frequencies is almost one magnitude higher on the older version and the new one doesn't have noticeable "steps" from reaching Nyquist.

Please enlighten me on what's happening because this change alone can render an entire community's database of measurements basically useless.
REW 5.40 beta 98 and 5.31.3 were used. I couldn't find any release notes regarding this or settings in REW itself...

 

[John Mulcahy]

V5.40 beta 45 27th July 2024
Changed: Use frequency-dependent windows to reduce the noise floor of sweep distortion measurements

The above change eliminated the rising noise floor with frequency that previously affected sweep measurements. It does not affect genuine distortion that exceeds the noise floor. The traces you attached are noisy because the contributions are mainly noise.

 

[_redfox]

V5.40 beta 45 27th July 2024
Changed: Use frequency-dependent windows to reduce the noise floor of sweep distortion measurements

The above change eliminated the rising noise floor with frequency that previously affected sweep measurements. It does not affect genuine distortion that exceeds the noise floor. The traces you attached are noisy because the contributions are mainly noise.

Oh, I've read that note, but was unsure, whether it gives that effect or not. Thanks!
Precision aside, how can one compare older and newer results? You were absolutely correct about those graphs: it was a compander simulation test with distortion buried under the noise floor...

 

[dcibel]

Precision aside, how can one compare older and newer results?

Best way is to make sure "capture noise floor" is checked for all measurements so you can compare it, and differentiate between noise and distortion signal. Without noise floor captured, you will have to make an assumption of whether the distortion level is a result of the measurement noise floor or not.

The update to REW's measurement capabilities that John highlighted above is a huge improvement, thank you again @John Mulcahy for this, you gave everyone a free upgrade in measurement capability!

 

[sm52]

Use frequency-dependent windows to reduce the noise floor of sweep distortion measurements

Does it work automatically with every measurement? Or should it be included somewhere?

 

[_redfox]

Best way is to make sure "capture noise floor" is checked for all measurements so you can compare it, and differentiate between noise and distortion signal. Without noise floor captured, you will have to make an assumption of whether the distortion level is a result of the measurement noise floor or not.

The update to REW's measurement capabilities that John highlighted above is a huge improvement, thank you again @John Mulcahy for this, you gave everyone a free upgrade in measurement capability!

The problem is: this is a file import, not a realtime measurement...

 

[_redfox]

Does it work automatically with every measurement? Or should it be included somewhere?

At this moment it uses new window system by default and I think you can't change it

 

[John Mulcahy]

The problem is: this is a file import, not a realtime measurement...

If there is sufficient data before the captured sweep starts (about 1.5 seconds or so) REW will use that to determine the noise floor.

 

[_redfox]

If there is sufficient data before the captured sweep starts (about 1.5 seconds or so) REW will use that to determine the noise floor.

Ah, so it happens automatically?

 

[_redfox]

Well, the question remains unanswered: how do I compare my new measurements with existing database?
It's obvious that new window produce different results. I want my measurements to be consistent and comparable with others', which were done with previous versions...

 

[John Mulcahy]

Nothing stops you using older versions. If you are trying to measure THD with harmonics that are below the noise floor the results aren't meaningful, whichever version you use, but the new windowing keeps the noise floor level with frequency and below the old versions. If some harmonics are significant you could limit the THD calculation so only those are included, and/or look at the individual harmonics as well as THD to have a better feel for what is significant and what isn't.

 

[_redfox]

Nothing stops you using older versions. If you are trying to measure THD with harmonics that are below the noise floor the results aren't meaningful, whichever version you use, but the new windowing keeps the noise floor level with frequency and below the old versions. If some harmonics are significant you could limit the THD calculation so only those are included, and/or look at the individual harmonics as well as THD to have a better feel for what is significant and what isn't.

Okay, thanks. THD+N just looks so strange due to lower noise floor in the newer version)
Will there be a menu option to toggle between window modes?

 

[John Mulcahy]

No, what value could there be in generating noise-contaminated data?