[x RELIC x]

I suspect that I'm making a lot of incorrect assumptions.

Actually, I have to apologize. I’ve made an incorrect assumption in Thorsten Loesch‘s approach as I haven’t read much about it before. He says he is actually using "true multibit" for the most significant ‘top’ 6bits in the PCM signal and the rest is single bit delta sigma. I have no idea what benefit this would bring from a measurement perspective. Seems, to me, to be over engineering the problem in the wrong direction, but I’m no where near qualified enough to back that opinion up, just how it reads to me.

 

[JWahl]

Actually, I have to apologize. I’ve made an incorrect assumption in Thorsten Loesch‘s approach as I haven’t read much about it before. He says he is actually using "true multibit" for the most significant ‘top’ 6bits in the PCM signal and the rest is single bit delta sigma.

No Worries. I think the point that Thorsten was trying to make, was that most modern DAC chips take this approach internally. Not that he created a new concept there. He uses off the shelf chips that use that partial multibit/multilevel design, like the PCM 1793. Very few single-bit audio DACs exist nowadays. And as I understand, R2R beyond 16 bits becomes extremely difficult to maintain linearity, because of resistor matching required. And of course, Rob takes his own unique and rather complex approach toward conversion.

Regardless, I enjoy reading the insights of Rob, Thorsten, and Mike Moffat (of Schiit). All three take different approaches to conversion, but stand out from the crowd in their own rights. Rob and Mike especially seem to be pushing the boundaries of digital design, albeit in different directions.

 

[dmance]

... I enjoy reading the insights of Rob, Thorsten, and Mike Moffat (of Schiit). All three take different approaches to conversion, but stand out from the crowd in their own rights. Rob and Mike especially seem to be pushing the boundaries of digital design, albeit in different directions.

Don't forget to follow Ted Smith of PSAudio, Andreas Koch of Playback Designs and Bruno Putzeys of MolaMola. I'm sure there are other giants but all are converging on the same thing.

 

[Rob Watts]

I am absolutely sure we are not converging on the same thing at all.... R2R sounds exactly as bad as it measures; some people just like the sound of distortion.

 

[x RELIC x]

No Worries. I think the point that Thorsten was trying to make, was that most modern DAC chips take this approach internally. Not that he created a new concept there. He uses off the shelf chips that use that partial multibit/multilevel design, like the PCM 1793. Very few single-bit audio DACs exist nowadays. And as I understand, R2R beyond 16 bits becomes extremely difficult to maintain linearity, because of resistor matching required. And of course, Rob takes his own unique and rather complex approach toward conversion.

Regardless, I enjoy reading the insights of Rob, Thorsten, and Mike Moffat (of Schiit). All three take different approaches to conversion, but stand out from the crowd in their own rights. Rob and Mike especially seem to be pushing the boundaries of digital design, albeit in different directions.

Here’s one interesting article that he explains why his approach is different (multibit Delta Sigma vs ‘true multibit’) from the usual DS suspects, stating a couple times that only he does ‘true multibit’ for 6 bits of PCM:

https://www.audiostream.com/content/qa-thorsten-loesch-amrifi

I agree regarding Mike and have respect for what he has been doing.

Anyway, this is becoming off topic in this thread, and different from Rob’s approach.

 

[JWahl]

I am absolutely sure we are not converging on the same thing at all.... R2R sounds exactly as bad as it measures; some people just like the sound of distortion.

Perhaps. And that's why I thank you for releasing the Qutest (which I hope to buy after it's released). Of course, I prefer to keep my (intentional) distortion in the amplifer, rather than the DAC. I enjoy a sprinkling of tasteful tube harmonics sometimes. I go back and forth between my Mojo only, and feeding it to my hybrid-tube amplifier. I can certainly hear the merits of running Mojo direct, but I ultimately prefer the "seasoning" that external amplification offers. Thanks for giving us choices with that as consumers.

 

[ray-dude]

The Davina ADC project - because I wanted 135dB dynamic range - uses 40 elements as the pulse array reference for the ADC.

That's pretty incredible Rob (and thank you for the explanation...very helpful and illuminating). An ADC with 32 bits(!), with 8 bits of headroom so you can set your levels to leverage the entire dynamic range without worrying about clipping, and minimizing the need for post processing/mastering. I can't wait to hear a piano or cello that is recorded with this monster.

 

[musickid]

looking forward to seeing davina and what that brings with it.

 

[jayz]

The more I read into this, I think Rob and Chord are onto something very very special here with WTA and MScaler technologies especially as it aligns well with DSP fundamentals.

I still believe however that the absolute best scenario is when we finally arrive at a situation where we are able to superimpose say 1 min of audio trace - original vs reproduced, using various tools to determine the delta and improve. Be it either a simulation of a virtual design or real world measurements of an actual device, it doesn't matter. Structured listening tests are important no doubt but the scalability and repeatability of measurement techniques (current/future) based on fundamentals of physics and acoustics, has the potential to be equally effective, or perhaps even better. Perhaps one day, we will have supercomputers on the cloud running super high speed simulations of mixed signal designs with constrained random stimulus feeding input signals to discover the very last few abberations left in designs. And perhaps machine learning will play a part in debug iterations e. g . generating random seeds or configurations such that you could leave a simulation running for a month and at the end, the system will suggest several possibilities of how the design may be improved. I think we are on the cusp of a key enabler for all this.

 

[Rob Watts]

The kind of test that you are talking about is under development, in that I will be able to put a measurement number on the transient error from the decimation > reconstruction process. This will be useful to explore the actual errors involved in reconstructing transients from sampling; and I hope it will provide correlation between measurements and observed sound quality.

Although I am confident that this test will show huge errors for conventional filters against WTA - and it will hopefully prove the case for the WTA method, in that the errors will be so large nobody could deny that it would be significant. But what I am actually worried about is the change from say the 164,000 taps of Dave to the 1M in a M scaler - we may find that the actual measured difference using this test is small. Then we are back to using listening tests to actually evaluate whether these small changes have audible consequences.

I will give you another example. When developing Dave I had distortion harmonics at -130dB - which was matching the state of the art for DAC's - but I always wanted it to be better. And I had built lots of prototypes, steadily reducing distortion by small but significant amounts. Now this was an ego thing; I had distortion I could not explain, nor reduce, and wanted to get rid of it; I had no thought it would actually make a difference SQ wise.

Anyway, I made a breakthrough and discovered the source of the residual distortion and fixed it, so harmonics were all below -150dB, and higher order harmonics were some 30 dB better - a huge change in measured performance in one stroke. So then I listened to it and was very surprised - the difference was easily audible, and not what I expected - it sounded much more transparent with better detail resolution. Moreover, subsequently with other projects I have seen similar things - when higher order harmonics are reduced it sounds brighter with better transparency - and that is odd, as one would expect it to sound warmer with the removal of high order harmonics.

But the point I am making here is that -130 dB harmonics are well below the threshold of audibility, and reducing it should not be audible; but it is nonetheless. And this is where we get into difficulties with the sound science posters; not that something exists technically (and proven by measurements), but that these very small errors are actually audible.

 

[514077]

The kind of test that you are talking about is under development, in that I will be able to put a measurement number on the transient error from the decimation > reconstruction process. This will be useful to explore the actual errors involved in reconstructing transients from sampling; and I hope it will provide correlation between measurements and observed sound quality.

Although I am confident that this test will show huge errors for conventional filters against WTA - and it will hopefully prove the case for the WTA method, in that the errors will be so large nobody could deny that it would be significant. But what I am actually worried about is the change from say the 164,000 taps of Dave to the 1M in a M scaler - we may find that the actual measured difference using this test is small. Then we are back to using listening tests to actually evaluate whether these small changes have audible consequences.

I will give you another example. When developing Dave I had distortion harmonics at -130dB - which was matching the state of the art for DAC's - but I always wanted it to be better. And I had built lots of prototypes, steadily reducing distortion by small but significant amounts. Now this was an ego thing; I had distortion I could not explain, nor reduce, and wanted to get rid of it; I had no thought it would actually make a difference SQ wise.

Anyway, I made a breakthrough and discovered the source of the residual distortion and fixed it, so harmonics were all below -150dB, and higher order harmonics were some 30 dB better - a huge change in measured performance in one stroke. So then I listened to it and was very surprised - the difference was easily audible, and not what I expected - it sounded much more transparent with better detail resolution. Moreover, subsequently with other projects I have seen similar things - when higher order harmonics are reduced it sounds brighter with better transparency - and that is odd, as one would expect it to sound warmer with the removal of high order harmonics.

But the point I am making here is that -130 dB harmonics are well below the threshold of audibility, and reducing it should not be audible; but it is nonetheless. And this is where we get into difficulties with the sound science posters; not that something exists technically (and proven by measurements), but that these very small errors are actually audible.

So, to loosely quote Frederick Forsyth: The best measurement is the human ear, MK. 1. N'est pas?

 

[dmance]

I don't think the general audiophile population gets the implications of Rob's tests. That is ...if they prove his point that human hearing is beyond all reasonable limits of known ear biology and our understanding of brain processing. If humans can reliably detect differences beyond 150dB then a graduate student, somewhere, has to do more research and explain what is going on. Either we're all blessed with a quantum computer in our heads or the simulation algorithm designers left this impossibility in as a clue for us...

 

[richard51]

I don't think the general audiophile population gets the implications of Rob's tests. That is ...if they prove his point that human hearing is beyond all reasonable limits of known ear biology and our understanding of brain processing. If humans can reliably detect differences beyond 150dB then a graduate student, somewhere, has to do more research and explain what is going on. Either we're all blessed with a quantum computer in our heads or the simulation algorithm designers left this impossibility in as a clue for us...

That is the more important part for me thanks to remind that....

 

[JWahl]

It is certainly a great challenge for those of who are science/engineering minded to attempt to correlate what we perceive with what we measure. I personally lie somewhere in the middle of subjectivism and objectivism.

I of course don't think that measurements lie, but that perhaps there are other metrics we may have not looked into. For instance, Rob's focus on noise floor modulation. I can also accept that, as a alluded to earlier, I may have subjective preferences toward specific harmonic spectra.

It could also be that extremely objectively accurate equipment exposes inherent flaws in the ADC and mastering process, which I know is something Rob is trying to address with the Davina.

But then sometimes I doubt myself, because in today's audiophile market, it becomes difficult to separate marketing talk from actual technical information without yet having the proper technical background to interpret in the right context.

For instance, I was poking and prodding earlier about pulse array because I know the name is a proprietary brand name, not a generic description and I wanted to better understand the idea as an early engineering student (I'm only in Calculus II and Physics I, so not to the good stuff yet). Rob was kind enough to explain it as his own unique variation on a thermometer coded dac with constant switching.

Who knows, perhaps one day when Rob decides to retire, he can do one of those online master classes for digital signals and systems geared toward engineering professionals. Like the one Hans Zimmer has for composition, or Gordon Ramsey for cooking.

 

[paul2qute]

Apologies, I should have started this some time ago!


But I have been hit with project deadlines, migraines and other things so my first article about listening tests will be coming shortly.

The reason for doing this blog is so I can talk in more detail about technical aspects - such as how a noise shaper works for example. Also I want to be able to talk about projects as they are under development, so you can get an idea how these things actually happen. I have in mind to talk about the digital power amp project, and in particular Davina, the ADC project, as this has a number of research questions I want to answer for myself.


Please feel free to make suggestions for topics to discuss.



Rob

Rob watts got his own thread to talk about more technical aspects and new projects coming up,I don't know anyone in hi-fi that has such commitment to Chords customers,what a lovely bloke