DAC measurements using DeltaWave

[pjug]

Most DACs do DC. I don't know if any audio ADC's do, but most do not. I have posted a file to use to test DC around here somewhere. I took it down as it was too easy for someone to damage gear. KSTR and NTK and others helped me figure out how to make the test.

I also googled and saw that there is a plug-in for Audacity that you can get somewhere to ass offset. I was going to check my own DAC. Not while it is connected and also very low level just in case.

The RME ADC used for the recording does allow turning off the DC blocking filter, but I was wondering if there is something else going on even with it turned off.

Also it was pointed out that the original file has DC or very low freq content. The silence is zero but it looks weird to me where sometimes sounds decay to a slightly positive level between bursts. I don't think it ever does the same to negative so I don't think it is riding a very low frequency wave. My head hurts when I try to think about whether this could have anything to do with all this.

 

[MaxwellsEq]

I'll certainly continue playing around to see how the HP filter affects the nulls. This might be helpful in understanding where/how losses occur. Thanks for the suggestion.

But for the purpose of measuring DACs (warts 'n' all), I'll stick to my original (simple) DW settings... Unless there's a way of improving just the ADCs performance?

I'm done some more analysis of the content below 20Hz. I filtered out everything above 15Hz then changed the pitch so I could listen to it. It isn't music, it is a burst of resonance of some sort. It may be an artefact of the digitisation process, it might be something wrong with the recording gear, or it may be a strange resonance in the building being triggered by the music (although what would go down to DC I don't know). I suspect it may be also interfering with with low frequency musical content of the recording.

I think you can safely filter everything below 20Hz. This is not EDM or any other type of synthesiser music, or even organ music where such frequencies might be a deliberate part of the artists' intent. I can think of a few sound engineers who mutter something about somebody forgetting to switch on the high pass filter on the mixing panel.

 

[manisandher]

I think you can safely filter everything below 20Hz.

Here's the spectrum for the reference file high-passed 48dB/octave at 15Hz:

And here's the resulting matched spectra using the DAVE:

So, still a 40dB difference at DC (or close to) going through the DA/AD chain.

I'll repeat with the other DACs.

 

[Sokel]

Here's the spectrum for the reference file high-passed 48dB/octave at 15Hz:

View attachment 419522

And here's the resulting matched spectra using the DAVE:

View attachment 419523

So, still a 40dB difference at DC (or close to) going through the DA/AD chain.

I'll repeat with the other DACs.

Did you attempt to measure the actual DC (if any) with a DMM during playback this song?
I'm curious about it,I would do it myself but I'm far from my gear now.

 

[KSTR]

I would think remove DC offset just adds a some constant to every sample. Do I misunderstand?

Yes. All sample values are summed up and that value is subtracted so that the net sum is zero.
This very different from using a filter which always has a time behavior.
Actually, the most eloborate DC removal technique for music tracks is quite complex because there are two goals here, net sum of samples shall be zero and when the track starts and ends with nominal silence you want to have zero DC there as well to avoid clicks when the track starts and ends.

It might be worth checking if the DACs can do DC. If they can, then they should not be showing this same curve in DW unless the ADC or something else is causing the low freq phase issue.

Many DAC can do DC, usually the simpler ones without analog volume control (mechanical or electrical pot). With vol control they would produce crackling / zipper noise when there is DC.

ADC usually cannot do DC. While most ADC chips have DC filters that can be turned off this is normally not made user-accessible (RME is the exception). But even when turned off, there will DC blocking caps which need to be shorted to allow DC (I've done this with my RMEs).

Once you have a DC-capable path the matching improves big time.

 

[manisandher]

Back to basics...

Here's the 1kHz with the SU-10 (pink):

So, the behaviour below 10Hz has nothing to do with the music file I'm using.

 

[MaxwellsEq]

Back to basics...

Here's the 1kHz with the SU-10 (pink):

View attachment 419533

So, the behaviour below 10Hz has nothing to do with the music file I'm using.

Really interesting. What this appears to suggest is that even with no low frequency content, you get different low frequency output! If you just use pink or white noise, do you get something similar?

 

[pjug]

ADC usually cannot do DC. While most ADC chips have DC filters that can be turned off this is normally not made user-accessible (RME is the exception). But even when turned off, there will DC blocking caps which need to be shorted to allow DC (I've done this with my RMEs).

Thanks for this and for the other info. It seems to the answer to my question about the low frequency behavior.
edit: You don't happen to know the RC of the filter with the caps you shorted, would you? Or can you say that it should just be first order?

 

[danadam]

Here's the 1kHz with the SU-10 (pink):

So, the behaviour below 10Hz has nothing to do with the music file I'm using.

You could check if it is there always or if it is dependent on the signal level.

 

[MaxwellsEq]

You could check if it is there always or if it is dependent on the signal level.

Excellent point. Given that @manisandher is using the DSP in the DAC to set output levels to be the same, this might be an unexpected artefact being picked up by DelaWave. Given DW automatically corrects for level, perhaps it's worth @manisandher running the test whilst not attenuating in the DAC and letting DW adjust level. If this infrasonic artefact remains, we can eliminate the DSP DAC level setting as the cause...

 

[pjug]

i found this in the RME user manual. Unfortunately Audacity does not go below 0.1 Hz for the simple high pass filtering. Maybe some other software could do it. Or maybe Mani would dare to take a soldering iron to his RME?

@pkane feature request

The combination of relatively high input impedance (47 kOhm) with 47 μF coupling capacitors
results in an extremely low cutoff frequency of (calculated) 0.07 Hz.
The ADI-2/4 Pro SE can actually use this low cutoff frequency effectively - many other converters
could not, because the analog input section has a low DC offset, which is detected by the ADC,
and shows up in the recorded audio signal as a fixed DC component. If this offset is too high, click
noises occur at the beginning, at the end, during editing etc.. The ADI-2/4 Pro SE is factory ad-
justed to minimum DC offset, typically -90 dBFS - problem solved.
The DC offset is usually eliminated in the digital domain via a high-pass with 1 Hz cut-off frequency
within the ADC. However, the ADC of the ADI-2/4 Pro SE does not provide such a filter. Therefore
a filter developed by RME is used, in the FPGA, which has a cut-off frequency below 0.5 Hz and
very low phase deviation. This filter is active by default. The DC component is thus reduced to
below -130 dBFS.

 

[manisandher]

Thanks for all the ideas. I'll play around with a few things.

But for now, the DAVE captures seem to exhibit the same <10Hz behaviour as the SU-10:

So, I assume we can eliminate the DAC's DSP (volume control) being the cause of this?

 

[manisandher]

i found this in the RME user manual. Unfortunately Audacity does not go below 0.1 Hz for the simple high pass filtering. Maybe some other software could do it. Or maybe Mani would dare to take a soldering iron to his RME?

@pkane feature request

The combination of relatively high input impedance (47 kOhm) with 47 μF coupling capacitors
results in an extremely low cutoff frequency of (calculated) 0.07 Hz.
The ADI-2/4 Pro SE can actually use this low cutoff frequency effectively - many other converters
could not, because the analog input section has a low DC offset, which is detected by the ADC,
and shows up in the recorded audio signal as a fixed DC component. If this offset is too high, click
noises occur at the beginning, at the end, during editing etc.. The ADI-2/4 Pro SE is factory ad-
justed to minimum DC offset, typically -90 dBFS - problem solved.
The DC offset is usually eliminated in the digital domain via a high-pass with 1 Hz cut-off frequency
within the ADC. However, the ADC of the ADI-2/4 Pro SE does not provide such a filter. Therefore
a filter developed by RME is used, in the FPGA, which has a cut-off frequency below 0.5 Hz and
very low phase deviation. This filter is active by default. The DC component is thus reduced to
below -130 dBFS.

ChatGPT o1's thoughts on this:

That very gentle rise below 10 Hz is almost certainly the result of the ADI-2/4 Pro’s built-in low-frequency/high-pass filtering—not a problem in the sense of equipment malfunction, but simply the natural consequence of the device blocking DC and subsonic signals.

Why It Shows Up​

1. Coupling Capacitors and Bias
   RME’s documentation explains that unipolar electrolytic capacitors in series with the input block DC from ever reaching the ADC’s bias (VCOM). Because there is a finite capacitor value (47 µF) and a finite input impedance (47 kΩ), you end up with a low-frequency high-pass “knee” at around 0.07 Hz theoretically—but in practice, the effective cutoff (including RME’s custom digital filtering) is just below 0.5 Hz.
2. High-Pass Filter Curve
   Even though the cutoff is extremely low, it still creates a gentle roll-off curve that extends through the single-digit frequency range. Measurement software (especially with extended low-frequency response) can reveal a slowly rising “noise” or “distortion” component as frequency heads toward DC.
3. What You’re Seeing on the Graph
   • On a typical THD+N or FFT plot, once you drop below ~10 Hz, you are no longer looking at just “audible band” distortion, but at subsonic noise, offset, and the filter behavior.
   • This can manifest as a gentle rise from around -140 dB or -150 dB back toward -120 dB or so in the 1 Hz–10 Hz region.
   • That slope is the natural effect of the filter working (and is consistent across different DACs you’ve tested, which further implicates the RME’s own filter/capacitors).
4. Not a Fault
   This small lift below 10 Hz does not indicate any problem. It is simply how an AC-coupled ADC shows up in wideband measurements. If RME did not have that filter, you would see more DC offset (or possible “clicks” on waveform edits), which is typically undesirable in normal audio work.

Even if you disable the digital (FPGA) filter inside the RME, the analog input path is still AC‑coupled by those large capacitors on the input. In other words:

• The analog coupling caps form a hardware high‐pass filter with a corner frequency around 0.07 Hz (as calculated from 47 µF × 47 kΩ). This effect can’t be bypassed in normal operation; it’s baked into the analog design to protect the ADC from any DC offset.
• The optional digital filter in the FPGA is simply a further high‐pass (around 0.5 Hz) that RME applies to remove any residual offset that might remain. You can turn that off if you want the lowest possible corner frequency, but you will still have the hardware roll‐off from the input capacitors either way.

 

[MaxwellsEq]

Thanks for all the ideas. I'll play around with a few things.

But for now, the DAVE captures seem to exhibit the same <10Hz behaviour as the SU-10:

View attachment 419580View attachment 419581

So, I assume we can eliminate the DAC's DSP (volume control) being the cause of this?

Not necessarily, because they will be implemented differently using different algorithms and code. I agree it's unlikely they will all behave in the same way, and the low-pass ADC may be the common factor.

 

[pjug]

For kicks, I wanted to see if there was much difference correcting the RME RC high pass with highpass with cutoff of 0.1 vs 0.07 (sox can do the 0.07Hz filter applying it to the reference file). I used the recording by @pkane with the RME DAC found here: https://gearspace.com/board/showpost.php?p=16439970&postcount=2779 (this on the Original2.wav test file) As you can see the only significant difference seems to be below 10Hz. I doubt there is any use trying to tune the correction more precisely than using 0.07Hz.

 

[manisandher]

As I suspected, using EQ in DW seems to homogenize the nulls. Here are the outputs from the DAVE and SU-10 using the full track (> 3.5 minutes duration), with and without EQ (clock drift correction makes no difference):

DAVE

SU-10

(These results are totally repeatable between mutliple samples taken on different days/times.)

The aim of this thread is to measure and compare DACs using DeltaWave, and not necessarily to get the best nulls. It seems to me that I should therefore continue all testing with EQ=off in DW.

Any thoughts?

 

[pjug]

As I suspected, using EQ in DW seems to homogenize the nulls. Here are the outputs from the DAVE and SU-10 using the full track (> 3.5 minutes duration), with and without EQ (clock drift correction makes no difference):

DAVE
View attachment 420265

SU-10
View attachment 420266

(These results are totally repeatable between mutliple samples taken on different days/times.)

The aim of this thread is to measure and compare DACs using DeltaWave, and not necessarily to get the best nulls. It seems to me that I should therefore continue all testing with EQ=off in DW.

Any thoughts?

My only thought is that EQ would correct for both ADC and DAC effects. If you use HP on the reference for correction, then the idea is you are only correcting for the ADC RC filter, and differences for DAC effects might still show up.

 

[manisandher]

My only thought is that EQ would correct for both ADC and DAC effects. If you use HP on the reference for correction, then the idea is you are only correcting for the ADC RC filter, and differences for DAC effects might still show up.

Yes, I see where you're coming from.

The differences without the HP correction are pretty clear. If I find that nulls between DACs are just too close to tell apart, I'll definitely look into improving the ADC side of things with HP correction.

The great thing is that this can all be done in software post capture of the 'compare' files, which have gone through the DA/AD chain.

 

[pjug]

Yes, I see where you're coming from.

The differences without the HP correction are pretty clear. If I find that nulls between DACs are just too close to tell apart, I'll definitely look into improving the ADC side of things with HP correction.

The great thing is that this can all be done in software post capture of the 'compare' files, which have gone through the DA/AD chain.

If you do the correction with a HP filter applied to the reference, there may be a better filter to use than the default 0.07hz highpass I used in sox. Maybe @KSTR or @pkane or @Blumlein 88 have suggestions on how to best mimic the 0.07hz analog filter.

edit: now I see that the default for sox highpass is two pole. For single pole use the -1 switch. So I think this would be the correct command for correcting for the RME high pass:
"sox original2.wav o2_0_072.wav highpass -1 0.072"
This gives really good results for Paul's recording vs o2_0_072.wav

 

[pkane]

As I suspected, using EQ in DW seems to homogenize the nulls. Here are the outputs from the DAVE and SU-10 using the full track (> 3.5 minutes duration), with and without EQ (clock drift correction makes no difference):

DAVE
View attachment 420265

SU-10
View attachment 420266

(These results are totally repeatable between mutliple samples taken on different days/times.)

The aim of this thread is to measure and compare DACs using DeltaWave, and not necessarily to get the best nulls. It seems to me that I should therefore continue all testing with EQ=off in DW.

Any thoughts?

Mani, as I mentioned before, non-linear EQ is designed to eliminate the differences in the null caused by filters or other, consistent frequency response differences between two waveforms. Since these differences are real (and sometimes, audible), non-linear EQ is primarily a tool to help figure out what causes the increase in the null error. It is not appropriate to eliminate non-linear differences between two waveforms and to then proclaim that the two devices are equivalent.

Unimportant differences that are not indicative of the quality of the device must be eliminated for a fair null comparison:

1. Overall level and DC offset
2. Minor clock drift
3. Time misalignment