• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

DAC measurements using DeltaWave

I had a thought in the gym this morning - without ChatGPT ;).

What if I apply a high-pass filter (6 dB/octave, minimum-phase at ~0.07 Hz) offline to my reference file, to create a new 'compensated reference file', Ref_c. I'll then use Ref_c for the input to the DA/AD chain, and as also as the reference in DeltaWave.

The idea is that in using Ref_c, the RME’s own coupling filter should have little or nothing to do <10Hz.

Thoughts?
Try it and see. The larger issue with the filter is not that it causes damage below 10Hz, but that it causes phase change over the entire bandwidth.
 
Try it and see. The larger issue with the filter is not that it causes damage below 10Hz, but that it causes phase change over the entire bandwidth.

Yes. I should have written: "The idea is that in using Ref_c, the RME’s own coupling filter should have little or nothing to do."

Will try over the coming weekend.
 
6 dB/octave, minimum-phase at ~0.07 Hz
Sorry - lost track of what's you are doing. Perhaps you could have a "what we've learned so far post"?

Anyway, why a minimum phase filter? Given you are doing it offline and the track is quite long, you don't need to worry about latency. Or are you concerned about "pre-ringing"?
 
I had a thought in the gym this morning - without ChatGPT ;).

What if I apply a high-pass filter (6 dB/octave, minimum-phase at ~0.07 Hz) offline to my reference file, to create a new 'compensated reference file', Ref_c. I'll then use Ref_c for the input to the DA/AD chain, and as also as the reference in DeltaWave.

The idea is that in using Ref_c, the RME’s own coupling filter should have little or nothing to do <10Hz.

Thoughts?
Then you would be comparing a 2nd order filter (recording of digital first order followed by analog first order) to a1st order filtered reference.
 
Then you would be comparing a 2nd order filter (recording of digital first order followed by analog first order) to a1st order filtered reference.

Hmm... I'll give it a go anyway as I'm curious to see what happens :).
 
To get a deep null at LF, we want to make the frequency response of magnitude and more importantly phase the same for the compare.
The recording contains the 1st order highpass at 0.07Hz of the ADC, therefore the reference used in DW should also contain the exact same highpass function (as exact as possible).
Everything else will reduce null depth from the phase shift.
And it's a good idea to apply a steep highpass at 20Hz on the source file first, before using it for the other processes.

I think we've been at that point already many posts ago.

And once again, always use DW's bandlimiting feature.
 
Hmm... I'll give it a go anyway as I'm curious to see what happens :).

Mani, you might like this new version :) https://app.box.com/s/h2u20jzre4bake02n5aj40kc5r2woaa0
(because it's the same version number, you'll need to uninstall previous v2.0.15 from Windows, and then install this one)

I fixed a couple of small bugs, including one you reported with IIR filter being listed in the results log instead of FIR.

The other issue was a scaling problem with the minimum filter calculation. Now that that's fixed, and the fractional corner frequency can be entered directly, you can try what you wanted to do all along: apply a minimum phase HP filter to the reference waveform. Here's the result (notice the RMS of the difference):

1736900835737.png

1736900470747.png

1736900408900.png


The relevant filter settings:
1736900583854.png
 
Revisiting my older ADI-2 Pro FS loopback recording. It appears that a minimum phase, low corner frequency HP filter benefits the RMS null there, as well. This is using this recording:

Original RMS of the difference was -54dB without the filter.

With a 0.4Hz MP HP filter applied to the reference, the difference becomes -71dB. You can see the difference in phase:

1737032536200.png



Phase difference without filter:
1737032637797.png

---------------
EDIT: to demonstrate that it is not just the amplitude of the frequency response that is responsible for the improvement in null RMS, here's exactly same amplitude-response filter applied for this comparison, but this time with linear phase:

1737036613560.png
 
Last edited:
@pkane for my understanding, it appears you are only applying the HPF to the reference and NOT the recording, is that correct?

Michael

Correct. The theory was that the RME devices have a DC filter, with a very low frequency cutoff, below 1Hz, that also acts as minimum phase filter. As the result, phase is altered in the recording towards minimum phase.

To try to match this, the suggestion was to apply a similar minimum phase, high-pass filter to the reference recording and see if this makes the two files match better. It appears that this is the case: when the reference recording is HP-filtered with a minimum phase filter with corner frequency below 1Hz, the RMS of the difference waveform becomes significantly smaller.

Again, I don't recommend manipulating the reference waveform for any real comparisons/measurements, but as an experiment, this seems to point to minimum phase being a significant influence on the null result. I've tested this with a few other devices, and there the MP filter had no significant influence on the RMS of the difference.
 
Gotcha, thanks for the info. This seems like another confirmation of @KSTR's findings regarding how a very low frequency HPF can limit how good the null can be. When I first saw the chart below, I made some digital loopbacks while applying very low first order HPFs in CamillaDSP, the nulls exactly matched this chart for 1 Hz (~55 dB) and 0.1 Hz (~75 dB).

1737040517863.png


Michael
 
Again, I don't recommend manipulating the reference waveform for any real comparisons/measurements...

Paul, if the compensating filter is matched to the ADC's filter (as closely as possible) and then applied to the reference file only, why would this not be a valid way of increasing the sensitivity of the comparisons/measurements?

Might a good validity test be that the compensating filter improves the comparisons/measurements for all DACs equally?
 
Paul, if the compensating filter is matched to the ADC's filter (as closely as possible) and then applied to the reference file only, why would this not be a valid way of increasing the sensitivity of the comparisons/measurements?
Because you don't know the exact amplitude or phase of the ADC filter and you can't measure it directly without other components and filters in the recording path.

Might a good validity test be that the compensating filter improves the comparisons/measurements for all DACs equally?
That can help, but I assume it'll vary with DACs to some degree. ADC filter can vary also between the same model ADC units due to component variations and/or tuning and adjustments.
 
Back
Top Bottom