FSAF (Fast subband adaptive filtering) measurement

miniDSP and AV Nirvana
That looks good @JanRSmit

This is what a virtual audio cable (bypass audio interface, bypass amplifier) can look like-

1740800375940.png


If you tweak your settings (I can't recall my REW settings) you may be able to squeeze out a big more.

The main downside with a 800ms impulse time, calculation may be a bit slow. Not sure if REW is multi-threaded or multi-core.
 
Interesting, so i first set the input or output of the Babyface to higher values, that did not alter the output.
Then i increased to REW level, i went to 290mV (2x145mV), and i got a funny TD+N:
1+2 FSAF-RBFP-96k-FilterYes-Music-IR800-ASIO1024-REW-290mV-DIST.png

I ran this test twice, no difference. When play back the res on speakers of laptop, i only vaguely hear music sediments.
The Input window in the FSAF panel does not show an overload, never less than -10dBFS. The puls at start and at end is highest.

So i went back to 145mV , ran the test, then increased to 200 , then with steps of 10 to 300, and up to 280mV, the result is:
1+2 FSAF-RBFP-96k-FilterYes-Music-IR800-ASIO1024-REW-280mV-DIST.png


Changing Asio buffer size from 1024 to 2048 did not give a difference, or IR length of 1000 or 800 also no difference.
The Babyface is not overloaded at all.
WEIRD.
 
THis morning i tested again.
First with White Noise (WN) i started at 100mV and in steps of 100mV increase. At 300mV, when hit start it shows an error box:
1+2 FSAF-RBFP-48k-FilterNo-WN-IR1000-ASIO2048-300mV-Setting-ERROR.png

The thing that struck me is the 143 dB SPL, but in the panel i set "No mic"
Hit Yes and here a snapshot during running:
1+2 FSAF-RBFP-48k-FilterNo-WN-IR1000-ASIO2048-300mV-Setting-Sample-ERROR.png

Headroom is red, so apparently in error. Yet the result looks ok:
1+2 FSAF-RBFP-48k-FilterNo-WN-IR1000-ASIO2048-300mV-Setting-DIST-ERROR.png


Even at 400 mV, of course Rew signals error, but result looks ok.

So now with WN and a filter:
400mV:
1+2 FSAF-RBFP-48k-FilterYES-WN-IR1000-ASIO2048-400mV-DIST-ERROR.png

And at 300mV:
1+2 FSAF-RBFP-48k-FilterYES-WN-IR1000-ASIO2048-300mV-DIST-ERROR.png

It still looks ok, but the panel shows error:
1+2 FSAF-RBFP-48k-FilterYES-WN-IR1000-ASIO2048-300mV-Sample2-ERROR.png


So now with music:
1+2 FSAF-RBFP-48k-FilterYES-MUSIC-IR1000-ASIO2048-300mV-DIST.png

To get rid of the hump, i had to drop to 280mV:
1+2 FSAF-RBFP-48k-FilterYES-MUSIC-IR1000-ASIO2048-280mV-DIST.png

The settings for 280mV:
1+2 FSAF-RBFP-48k-FilterYES-MUSIC-IR1000-ASIO2048-280mV-Sample3.png


So for some reason, i do not understand, there is a limit at Level = ~280mV in my set up.
I thought a possible explanation is the relation to the SPL calibration setting , in my case it was 14.5mV for 94 dB. So i changed that to 14.5mV for 104dB. No improvement.

Whisth the operating window with Rew V5.40 Beta70 icm RMA Babyface Pro FS(ASio) for me is clear.
I still want to understand a bit more this upper limit, perhaps someone can help me understand?
 
Preferably, a setup shall be able to loop at every point, i.e.:
- at application s/w
- at the drivers
- at the audio interface on the digital level
- at the audio interface on the analog level
- at amplifier output
- at loudspeaker acoustic output -> mic

...and the gain plan shall be set so that full-scale digital output overflow happens first, and the rest of the circuitry remains in the linear mode (as good as it gets).

I learned the hard way that it is much easier said than done.
 
John Mulcahy said:
Because there is clipping somewhere along the path for higher levels. Stop over-driving the system or adjust gains to stop the clipping.
Click to expand...
Yes John, i understand it is clipping somewhere, that was my question, as there was no sign of clipping on my Babyface, so where is it clipping?
First the voltage level calibration with BabyfacePro settings out is 0dB, in is +4dB:
REW-Voltage-Calib-Result-200mV.jpg

REW-Babyface calibration is quite accurate down to low values like from a analog mic f.i. 14.5 mV @ 94dB.



I did some more testing
First of all the music info showed in the measurement panel upper right:
Music-Test-04-Setting-96k-RBFPmin12-Settings.png

It suggests to me that this has peaks above 0dB, is this dBFS?
If checking same music with Musicscope, it gives a different picture:
MusicScope-Analysis2L-038_01_stereo_FLAC_44k_16b.png

Also the level shown changes with sampling rate (roughly 3dB per doubling sampling rate), and this is also visible in the Distortion results.

It appears to me that the level check option does not check the level of the music sample.

So what i did is reduce the Output level of the Babyface from 0 to -6 -9 , -12, -15 without any recalibration in REW. Obviously this is reflected in the results, the level of the Fundamental drops with the same value.


So what indication will tell me where the overload is? Would that be the level of the Fundamental in the Distortion panel?

Anyhow i did a full reset in REW and start all over again.
 
JanRSmit said:
It suggests to me that this has peaks above 0dB, is this dBFS?
Click to expand...
No, it is just showing a power spectrum to give an indication of the spectral content.

JanRSmit said:
the level shown changes with sampling rate (roughly 3dB per doubling sampling rate)
Click to expand...
That's an artefact of resampling the source data to the measurement frequency.

JanRSmit said:
It appears to me that the level check option does not check the level of the music sample.
Click to expand...
Check levels uses a speaker cal signal (filtered pink noise centred at 1 kHz) at the current measurement level.

JanRSmit said:
So what indication will tell me where the overload is?
Click to expand...
When you start the measurement REW will check the output data at the current measurement level and warn if that looks like it will produce clipping. However, sequences of sample values can produce inter-sample levels that exceed the sample values significantly (by 4 dB or more) which may cause clipping artefacts within the DAC or in subsequent analog stages even though the samples themselves do not clip, REW cannot detect that. The best protection against that is to use a measurement level that leaves significant output headroom. One way to do that would be to increase the measurement level until REW shows its level warning then drop the level by 4-6 dB more than REW indicates is necessary to avoid clipping. Another way would be to use a level below that which shows the onset of distortion in the loopback test, as you have done.
 
@John Mulcahy would it be possible to apply a soundcard cal or input cal to the resulting residual audio files? I have found that applying a soundcard cal does change the resulting distortion spectrum shown in REW, however it makes no change to the resulting audio amplitude over frequency.
 
The input cal should be applied to the captured signal before the FSAF calculations, but I'm not doing that currently. I have omitted it because cal files typically lack phase data and usually have low and high frequency roll-offs. Compensating for the roll-offs boosts low and high frequency noise and low frequency compensation can also cause problems with the FSAF calculations due to the increased IR length.
 
Interesting, amplifier rolloff at the frequency extends is generally fairly small, I wouldn't expect it to have a large influence on the calculations.

I would love to be able to give it a try, as an option. My use case is a bit different, as I'm experimenting with the "current sense" measurement above. Since the measurement is purely electrical, noise level is already very low so I don't expect there to be a problem. I'd like to use the soundcard calibration as a means of compensation for inductance variances between drivers, to make as close to an "apples to apples" comparison as possible when listening. Perhaps my thinking is flawed, but the idea would be to load a measurement frequency response as the soundcard cal, then repeat the measurement to provide a normalized result.
 
John, i am currently away for a short holiday.
But while in train a short feedback after i did reset rew and redid level calibration from null. And did some tests.

The test results are more consistent, from what i noticed looking at levels (y axis in measurement panel) the level calib was way off, unclear why.

Will also just like dcibel do test with a 0.1ohm Rsense when back from my trip.

To be continued.
 
You must log in or register to reply here.
miniDSP and AV Nirvana

Popular tags

20th century fox 4k blu-ray 4k uhd 4k ultrahd action adventure animated animation bass blu-ray calibration comedy comics denon dirac dirac live disney dolby atmos drama fantasy hdmi 2.1 home theater horror kaleidescape klipsch lionsgate marantz movies onkyo paramount pioneer rew romance sci-fi scream factory shout factory sony stormaudio subwoofer svs terror thriller uhd ultrahd ultrahd 4k universal value electronics warner warner brothers well go usa
Back
Top