Does a truly transparent ADC actually exist?

[MAB]

Yet no answer about how you matched levels. @JamesQuorn0

Yes, 'great unknowns' were mentioned as if there was a mysterious circle of confusion that prevents comparison, but matched levels not addressed.
That Quantex that I mentioned in response to the odd challenge earlier was used for digitizing the output of an a field ion microscope (my Avatar), and was in fact calibrated to a refence source, most of the experiment was calibration and level matching. The big challenge isn't the A2D, it's noise and filtering, and if doing comparisons matching.

Looks like we may be setting ourselves up for an analysis of noise differences between experiments, rather than comparisons of A2D. I hope not, but seems likely.

 

[Blumlein 88]

Far be it from me to defend vinyl (I find the whole thing a pain, to be honest - and listen digitally 90% of the time), but the 'technicalities you speak of' (presumably THD, SINAD, and so on) don't quite measure up with aural perception. There's a lot of research into this, for instance:

https://hifisonix.com/wordpress/wp-content/uploads/2017/11/Perceptual-Levels-of-distortion.pdf

It's also worth looking into Auditory Scene Analysis, studies displaying the impact of hypersonics using EEGs, even the impact of auditory subliminals. Point being, there's a lot of stuff one DOESN'T consciously hear that impacts overall listening pleasure. It's a fascinating subject - I attended a talk a couple of years back on the subject in Manchester and it was eye (well, ear) opening.

Digital is a technically superior product given the measurables at our disposal today, but does it 'sound' as good? Working on both mediums (tape mainly, as opposed to vinyl) daily for a living, I can tell you that digital audio is certainly more fatiguing (in captures, where mastering is a non-issue) and this is a commonly shared feeling from those working with both. I wish it wasn't so, and is in fact one of the reasons I started this thread in the first place! Clearly, this type of evidence is anecdotal and will remain so until the ability to measure these great unknowns is available to us.

Now, that'll put the cats amongst the pigeons!

The DS metric in the paper might be more useful than THD and IMD numbers as those require quite a bit of interpretation and aren't great for just picking one number. However, I see nothing in the paper that tells us if digital has any particular problems. A lot of the testing was clipping distortion, and the conclusion it is more audible while THD and IMD are not well correlated subjectively. Well DUH! This test used digital recordings. So presumably they didn't fine unclipped un-highly-distorted digital a problem in their research.

The other stuff about hypersonics, and subliminals hasn't replicated well.

 

[JamesQuorn0]

Yet no answer about how you matched levels. @JamesQuorn0

I've already stated that I'll document the entire process when we get the Lavry back. We A/B pretty much everything here, so we're well versed on that side. The Tascam was done by ear, the rest were matched using a DMM.

 

[JamesQuorn0]

The DS metric in the paper might be more useful than THD and IMD numbers as those require quite a bit of interpretation and aren't great for just picking one number. However, I see nothing in the paper that tells us if digital has any particular problems. A lot of the testing was clipping distortion, and the conclusion it is more audible while THD and IMD are not well correlated subjectively. Well DUH! This test used digital recordings. So presumably they didn't fine unclipped un-highly-distorted digital a problem in their research.

The other stuff about hypersonics, and subliminals hasn't replicated well.

I think you've missed the entire point of the paper, to be honest.

 

[Blumlein 88]

I've already stated that I'll document the entire process when we get the Lavry back. We A/B pretty much everything here, so we're well versed on that side. The Tascam was done by ear, the rest were matched using a DMM.

Why the mystery, just spill the beans.

 

[JamesQuorn0]

Why the mystery, just spill the beans.

Mystery? You want me to list out the entire process across multiple devices and multiple tests? To run the gamut through sending out test signals, getting readings and so on? Are you having a laugh?

 

[manisandher]

Mystery? You want me to list out the entire process across multiple devices and multiple tests? To run the gamut through sending out test signals, getting readings and so on? Are you having a laugh?

Welcome to ASR James .

Hang in there. I'm confident something interesting will come out of this thread... eventually.

Mani.

 

[Blumlein 88]

Mystery? You want me to list out the entire process across multiple devices and multiple tests? To run the gamut through sending out test signals, getting readings and so on? Are you having a laugh?

No just a general procedure you used. For instance with tape how do you match levels for music? One good way is to append a test tone to play and record digitally to go with the music that follows without changing anything. Then that can be used subsequently to match output. No need to do a step by step for each single thing you did. Is that so hard to describe?

 

[MAB]

Mystery? You want me to list out the entire process across multiple devices and multiple tests? To run the gamut through sending out test signals, getting readings and so on? Are you having a laugh?

Yes. Most of experimental verification is test setup. Most of scientific papers spend significant time discussing experiment setup methodology, often more time than results so that other experts can know and understand, and have information to allow trust or debate of the experimental setup, otherwise perpetual mystery ensues. The experiment should be described in enough detail that other investigators can reproduce exactly. New discoveries like Onnes finding superconductivity (for instance) elicit a massive debate about test setup, Onnes spent tons of time defending his labwork, even doubting his own observations. Onnes' doubt of results is typical of how science is approached. Contrast that with recent infamous and bogus studies on high-T superconductivity where the investigators alternated between outright refusal to incomplete descriptions of their test setup and methods, which caused massive and justifiable doubt that ended up with the results being found clearly wrong. An investigator who can't or won't describe every step is considered to not fully understand their measurements, or faking them. People who come from a non-science background don't get this, but we have enough problems without some amount of discipline. And what you intend to demonstrate will require discipline, and ability to rule out multiple sources of experimental error that have nothing to do with analog to digital conversion, but instead have to do with the experimental method. Fortunately you aren't trying to get room-T superconductivity to work, but instead are going back a few decades to repeat similar work that was done while A2D and related DSP were developed.

One more thing, nothing to do with ASR or science or A2D... I don't even know what null-hypothesis you are going to test. You opened a technical thread saying you were non-technical, which is fine! You plan to test something that textbook signal processing says is inaudible in any reasonable DSP processor, and practical application indicates other variables in the signal chain are going to be many orders of magnitude larger than the A2D conversion process, also fine. 4 pages in we are still in a circle of confusion, with jerks like me who actually make and measure D/A and A/D converters for a living trying to figure out what hypothesis you are testing, and what method you will use. I guess I am saying you shouldn't be surprised, and we likely won't let you off the hook on and lingering mystery.

Edit, removed a partial sentence that didn't make sense...

 

[mdsimon2]

Take a SOTA DAC, something like an SMSL SU-10. Take a SOTA ADC, something like an RME ADI-2 Pro. Sync the clocks (trivial if using spdif into SU-10).

Capture the analogue output of the SU-10 five times. Compare captures to original in DeltaWave. (DeltaWave will ensure pretty much perfect level-matching.)

Original-to-captures null to around -70dB.
Capture-to-captures null to around -110dB.

Draw your own conclusions.

Mani.

I am curious what conclusions you have drawn from your DeltaWave testing?

I just started playing around with DeltaWave yesterday and I am pretty convinced that differences in pkmetric between DAC / ADC combos can be attributed solely to magnitude / phase response. Distortion / IMD performance seems to have no impact. I'm not using high performance hardware (MOTU Ultralite Mk5 DAC to ADC) but I've been able to get substantial improvements by playing around with magnitude / phase response on the same hardware.

For consistency I've been using the Gearspace Original2 track. I used Audacity to record and Quicktime as a player.

I first tried running the DAC / ADC at 44 kHz and got a pkmetric of -63 dBFS. Looking at the delta of spectra and delta phase it was clear that the differences were mostly due to phase behavior from minimum phase filters as well as some magnitude response ripples. I also confirmed this with REW frequency response measurements.

I then tried upsampling from 44 to 192 kHz in CamillaDSP and recording at 192 kHz. This gave me much better results as the magnitude / phase response is flatter through the audio band, this time I got a pkmetric of -79 dBFS.

I then played around with impulse inversion in REW to generate a FIR filter to correct the DAC / ADC magnitude / phase response while still using upsampling to 192 kHz. I implemented this filter in CamillaDSP. This resulted in a pkmetric of -109 dB based on the flat magnitude and phase response. At this point I would attribute most of the difference to the resampling operation as all of the spectra delta is in the 20-22 kHz range.

For reference here are my DeltaWave settings, all default other than I unchecked "correct clock drift" as it was unnecessary as the ADC and DAC were using the same clock.

Michael

 

[manisandher]

I just started playing around with DeltaWave yesterday and I am pretty convinced that differences in pkmetric between DAC / ADC combos can be attributed solely to magnitude / phase response. Distortion / IMD performance seems to have no impact.

Surely, you could have saved yourself all this trouble just by using linear phase filters in the first place?

I am curious what conclusions you have drawn from your DeltaWave testing?

Well, that ADCs aren't "solved", as many here suggest.

Mani.

 

[manisandher]

I won't have much time over the weekend, so here's a couple of captures I took today. The setup was:

Roon (24/96) -> SOtM sMS200 streamer -> SMSL SU-10 DAC -> ADC (linear phase steep) -> Audacity (24/96)

Level-matching achieved by passing a 1kHz sine tone and adjusting output to within 0.1dB in Roon, using its 64-bit processing.

FIrstly, let's make sure the two setups look OK using @pkane 's Multitone software (not level-matched in Roon):

Both look reasonable to me.

And here are the captures:

RME ADI-2 Pro: https://drive.google.com/file/d/1nLID40lJIAEjFbAIIdzw-nDwAwIscGRn/view?usp=sharing
E1DA: https://drive.google.com/file/d/1bXBCkE5QG7_9h40N2PFXSz9ibJesHz56/view?usp=sharing

For those of you who don't use DeltaWave, here's the spectrum of the delta (RMS -115dB):

And yet, they sound different to me.

Mani.

 

[olieb]

And yet, they sound different to me.

How do you know?
You did a blind test? ABX?
I don't know about you but after knowing which is which and after seeing the graphs I could imagine all kinds of things and my assessment would be quite useless. I am pretty convinced that the difference is not audible and so it would be no surprise for anybody if I would not perceive any.

 

[mdsimon2]

Surely, you could have saved yourself all this trouble just by using linear phase filters in the first place?

This was a low effort first attempt. I was using an interface on my desk, connected to a computer on my desk, running already installed software (CamillaDSP, REW). The most "trouble" was digging out a Windows computer that I don't typically use and installing / figuring out how to use DeltaWave. Digging out DACs / ADCs connected to other equipment or stored away doesn't make sense when I am just learning the tools.

In the future I plan on testing DACs with adjustable filters and seeing how that changes the result as well as different ADCs. Also interested in seeing how much magnitude correction without phase adjustment improves the null.

I like to methodically experiment with things to gain understanding, rather than make claims with no backup.

Well, that ADCs aren't "solved", as many here suggest.

But you can "solve" them by correcting the slight frequency response deviations with EQ?

Michael

 

[LTig]

I won't have much time over the weekend, so here's a couple of captures I took today. The setup was:

Roon (24/96) -> SOtM sMS200 streamer -> SMSL SU-10 DAC -> ADC (linear phase steep) -> Audacity (24/96)

Level-matching achieved by passing a 1kHz sine tone and adjusting output to within 0.1dB in Roon, using its 64-bit processing.

FIrstly, let's make sure the two setups look OK using @pkane 's Multitone software (not level-matched in Roon):

View attachment 349837
View attachment 349838

Both look reasonable to me.
[..]
And yet, they sound different to me.

Of course they do. The E1DA capture is 0.4 dB louder than the ADI-2 Pro (see the H1 levels).

 

[manisandher]

Of course they do. The E1DA capture is 0.4 dB louder than the ADI-2 Pro (see the H1 levels).

You obviously missed: "FIrstly, let's make sure the two setups look OK using @pkane 's Multitone software (not level-matched in Roon):" These were just to test that there was nothing untoward going on in the setup.

If you'd bothered to download before posting, you'd see that the captures are level-matched to 0.1dB, as I stated.

Unbelievable.

 

[manisandher]

But you can "solve" them by correcting the slight frequency response deviations with EQ?

Where's the "slight frequency repsonse deviation" between the captures?

Mani.

 

[pkane]

You obviously missed: "FIrstly, let's make sure the two setups look OK using @pkane 's Multitone software (not level-matched in Roon):" These were just to test that there was nothing untoward going on in the setup.

If you'd bothered to download before posting, you'd see that the captures are level-matched to 0.1dB, as I stated.

Unbelievable.

The differences are really tiny, Mani. Are you saying you can hear the difference between these?

One thing I can suggest is that there may be some differences due to the handling of intersample overs. As you can see, True Peak is +2dB for one, and 0.8dBTP for the other:

 

[mdsimon2]

Where's the "slight frequency repsonse deviation" between the captures?

Mani.

First, I should credit @KSTR who explained that differences in null testing are the result of frequency response deviations (magnitude and phase) here -> https://www.audiosciencereview.com/...t-deserves-in-measurements.37107/post-1306236.

Which is why I asked you in in this thread -> https://www.audiosciencereview.com/...ng-the-performance-of-dacs.39379/post-1387866 to perform a basic frequency sweep in REW to better understand the differences in pkmetric you were seeing, unfortunately it doesn't look like you ever got around to attempting that, .

Here are the results of the REW frequency response sweeps for 44 kHz and 192 kHz without EQ and 192 kHz with EQ. You can also see these differences in the deltawave screenshots I posted previously and commented on as well.

For reference here is the frequency response of the filter that improves the pkmetric by 30 dB at 192 kHz.

Michael

 

[LTig]

You obviously missed: "FIrstly, let's make sure the two setups look OK using @pkane 's Multitone software (not level-matched in Roon):" These were just to test that there was nothing untoward going on in the setup.

Yep, I missed this. Shit happens ... Anyway it would make sense to run the Multitone software on the actual samples used for blind listening, so one can be sure that the levels are properly matched. Since Multitone does not run on my Linux PC I can't verify myself.