REW Fast Subband Adaptive Filtering (FSAF) measurement

[LionIT]

I just discovered this new measurement option in REW.

AV Nirvana Thread
REW Guide

I have not been able to study enough yet to understand what it is for and what advantages it can provide compared to the sweep.
It would seem that it can produce valid TD+N measurements for complex signals such as noise or music, therefore ideal for testing the real behavior of the speakers.

Has anyone fiddled with this?

 

[ppataki]

I will, today or tomorrow
Will share my findings here

 

[ppataki]

Wow! This FSAF measurement is brutal...it gave me some really crappy looking TD+N graphs...

This is my FR and phase curve using L+R channels (sub incl.) with Dirac Live and custom linear phase low shelf and high shelf curves

This is my corresponding distortion curve with a normal sweep measurement:

And this is the TD+N curve with an FSAF measurement (pink noise, 15 sec):

So which is the 'real one'? And why is the difference so huge? Can it be noise related?

This is my noise floor curve with sweep and FSAF measurements respectively:

mdat file is attached
Any comments would be welcome
Thank you

 

[ppataki]

Forgot to mention that I used my UMIK-2 mic

 

[Sokel]

I have been trying it for practice since it got out,it's a really good measurement.
Looking at the documentation though is not as trivial as one thinks and to get fair results the gear needed are not average,one needs good ones.
So...

 

[Sokel]

Have been testing for fun just to what it does electrically.
I made sure results are repeatable to the last dB but something must be wrong.

Tested DACs are shown at the labels,same conditions,same signal,same everything (and 20 years of difference between them)

Can someone please explain this to me:

Test setup:




E-MU 0204



Khadas Tone board



Both

Surprising,isn't it?

Unless I screwed up something one would expect the 110dB SINAD KTB to be way better than the 20 yo 102dB SINAD E-MU.
An explanation please,this can't be right.

(needless to say I verified them both through the usual normal measurements)

 

[ppataki]

I am happy I am not alone with finding FSAF measurement a bit funky

 

[LionIT]

Have been testing for fun just to what it does electrically.
I made sure results are repeatable to the last dB but something must be wrong.

Tested DACs are shown at the labels,same conditions,same signal,same everything (and 20 years of difference between them)

Can someone please explain this to me:

Test setup:

View attachment 413060

View attachment 413052
E-MU 0204


View attachment 413054
Khadas Tone board


View attachment 413058
Both

Surprising,isn't it?

Unless I screwed up something one would expect the 110dB SINAD KTB to be way better than the 20 yo 102dB SINAD E-MU.
An explanation please,this can't be right.

(needless to say I verified them both through the usual normal measurements)

For clarity, what are you measuring exactly?
Analog output of devices I guess, right?
What is the measurement bandwidth?

 

[Sokel]

For clarity, what are you measuring exactly? Analog output of devices I guess, right?
What is the measurement bandwidth?

Yes,analog output.
I guess the BW is limited by the filters I have set and are visible at the test setup photo.
For the specific measurements it's an L-R 48db/oct at 21kHz for the upper limit.

Edit:Same pattern emerges with pink noise or music as a signal,I tested that too.

 

[LionIT]

Some DACs may filter out ultrasounds badly and create demodulation in audio bandwidth, especially with complex signals.
This is independent of the source signal bandwith.
Anyway, I don't know what device you use to measure, but I doubt it has such a large bandwidth...
Maybe even knows the input impedance of the measuring device is useful for making considerations.

Apart from that, did you consider this?

FSAF processing is very sensitive to clock rate differences between replay and record. The best results are obtained when using the same audio interface for both output and input, so there is only one clock source. If the input and output devices are different REW can use timing markers before and after the measurement signal to apply clock rate correction. Even with rate correction applied the results are heavily affected by the stability of the clock sources.

Also, if you used digital attenuation to even out the levels, this could partly explain the difference.
Can you measure in dBr or % and without attenuation (if previously applied)?

 

[Sokel]

Some DACs may filter out ultrasounds badly and create demodulation in audio bandwidth, especially with complex signals.
This is independent of the source signal bandwith.
Anyway, I don't know what device you use to measure, but I doubt it has such a large bandwidth...
Maybe even knows the input impedance of the measuring device is useful for making considerations.

Apart from that, did you consider this?


Also, if you used digital attenuation to even out the levels, this could partly explain the difference.
Can you measure in dBr or % and without attenuation (if previously applied)?

For further clarity that's how they measure normally at 1kHz given the limitations of my silly setup:



Khadas tone board



E-MU 0204


I considered impedance as it's a decisive factor sometimes but I have manage to minimize the effects as you can see at the normal measurements.
Levels are not evened,both outputs happens to be extremely close.But I did considered the crest factor of the signal as reported by REW (next to the little chart) to make sure I don't clip them.
Scaling and units don't help either.
Lowering the BW to 20kHz had some effect but extremely minimal.

I admit that all this is way above my pay-grade.
I would really like someone with proper gear to test it as well.

(of course all that are repeatable,I have tried it more than 50 times and its ok down to fractions of dB)

 

[Sokel]

dBr scaling:



E-MU



Khadas Tone board.

 

[LionIT]

What device are you using to measure?
Khadas shows some clock drift with 1kHz testing, that according to the REW guide can contribute significantly to the calculation of the FSAF distortion.
You could still do a full band sweep to see the THD how it behaves, and even an IMD 19 + 20 test to understand if the FSAF TD is justified in some way ...

 

[Sokel]

What device are you using to measure?
Khadas shows some clock drift with 1kHz testing, that according to the REW guide can contribute significantly to the calculation of the FSAF distortion.
You could still do a full band sweep to see the THD how it behaves, and even an IMD 19 + 20 test to understand if the FSAS TD is justified in some way ...

The ADC is E-MU.It makes sense as you think about it,E-MU surely gets an advantage as it's a simple loopback to it.
Maybe that's the reason.

(I did use the timing markers but it seems it's not enough)

 

[Sokel]

What device are you using to measure?
Khadas shows some clock drift with 1kHz testing, that according to the REW guide can contribute significantly to the calculation of the FSAF distortion.
You could still do a full band sweep to see the THD how it behaves, and even an IMD 19 + 20 test to understand if the FSAF TD is justified in some way ...

OK,maybe I found something,and it has to do with levels.
As shown before KTB is not the nicest perfomer out there when lowering levels.
So,at -3dB:


KTB


E-MU

Now look at -14dB (sadly where the music bangs normally) :


KTB



E-MU

 

[LionIT]

OK,maybe I found something,and it has to do with levels.
As shown before KTB is not the nicest perfomer out there when lowering levels.
So,at -3dB:

View attachment 413079
KTB

View attachment 413083
E-MU

Now look at -14dB (sadly where the music bangs normally) :

View attachment 413081
KTB


View attachment 413082
E-MU

If I don't misinterpret, the devices are showing the same behavior.
When you go from -4dBFS to -14dBFS of fundamental, THD+N calculated by REW (in distortion panel) gets worse by 10dB about on both cases.
And in the case of Khadas it does not worsen at the increase in frequency, so the cause of that TD detected with FSAF should not be that.
Try with IMD 19+20 maybe...

 

[Sokel]

If I don't misinterpret, the devices are showing the same behavior.
When you go from -4dBFS to -14dBFS of fundamental, THD+N calculated by REW (in distortion panel) gets worse by 10dB about on both cases.
And in the case of Khadas it does not worsen at the increase in frequency, so the cause of that TD detected with FSAF should not be that.
Try with IMD 19+20 maybe...

IMD and every other metric is normal between them.
So,the clock?Let's stay there before we go crazy