[Rob Watts]

@Rob Watts is there mileage in using a hypothetical new WTA3 from 256FS to 2048FS? Or, changing WTA2 so that it outputs 2048FS?

The original introduction of WTA2 in DAVE was a big deal and you've talked over the years about the possibility that WTA2 could benefit from more taps. I'm just curious if you've had the opportunity to play with a larger scope than simply increasing the taps in WTA2.

It seems that everywhere you go looking for possible sound quality factors you get big surprises. Apart from pure tap-count (whether in WTA1 or WTA 2) is there a danger that you're running out of surprises left to unearth?

About a year ago I looked into that very issue with new filters. There have been some small benefits, but as you get higher in frequency the benefits of WTA diminish rapidly. It seems that the ear/brain is sensitive to the area of the transient error; as you double the sample rate the area of the transient error falls by a factor of four, so it becomes less of an issue the higher in frequency you go. The exception to this is if you get aliasing (then of course the error is lower in frequency increasing the area) but that's another situation, but is vitaly important for sample rate conversion. More on this subject in the future...

Thanks for posting those exemplary readings Rob. I wonder do you have readings for when the Pre-amp is engaged? I just wonder what effect (or not) the Pre has on the measurements Edit: Regarding noise floor

So DAC mode and pre-amp mode are actually identical from performance POV - it's just in DAC mode the volume is simply fixed. The measurement was taken in pre-amp mode.

Regarding the potential future ADC product Rob, is it you plan to incorporate a couple of 48v phantom powers in the unit for Condenser mics? I obviously understand plans can change but it would be very useful imo.

For ADCs for studio use absolutely it will come with switchable phantom power.

 

[Jawed]

About a year ago I looked into that very issue with new filters. There have been some small benefits, but as you get higher in frequency the benefits of WTA diminish rapidly. It seems that the ear/brain is sensitive to the area of the transient error; as you double the sample rate the area of the transient error falls by a factor of four, so it becomes less of an issue the higher in frequency you go. The exception to this is if you get aliasing (then of course the error is lower in frequency increasing the area) but that's another situation, but is vitaly important for sample rate conversion. More on this subject in the future...

1M taps was, as far as I can tell, originally a huge surprise for you in terms of the sound quality uplift it offered. Along with that was the increasing magnitude of the improvement in sound quality as you went from 0.25M to 0.5M to 1M taps. So that seems to represent an inverse relationship to the "area of the transient error" that you've described.

So I'm wondering if the WTA1 tap-length improves sound quality seemingly exponentially because the recording was itself so badly aliased. So the worse the original ADC and aliasing, the greater the benefit that 1M taps offers?

Is there a way to systematically identify the quantity/quality of aliasing that's in a recording? I'm just wondering if it's possible to find a relationship between the "effect" of M Scaler and the aliasing errors of an original recording.

I've also been thinking about a "Remastered by DAVINA" labelling model: a lot of recordings that are out there might have been originated in "HD", e.g. 96KHz or 192KHz and have been transferred to CD with poor equipment. They could be remastered for CD and/or "Redbook" streaming services such as Amazon or Spotify simply by using DAVINA to correctly filter/decimate. So we might find that existing HD recordings can get a significant benefit from DAVINA even for consumers who do not have M Scaler or any Chord DAC.

In Stereophile's review of M Scaler,

https://www.stereophile.com/content...-m-scaler-upsampling-digital-processor-page-2

John Atkinson wrote:

As coincidence would have it, midway through the time I had the DAVE and M Scaler in my system, I was preparing the CD master for Translations, an album of works by modern Latvian composer Eriks Ešenvalds, performed by the Portland State Chamber Choir directed by Ethan Sperry and produced by Erick Lichte. We had recorded the sessions at 24/96, so I was downsampling the hi-rez data with my dCS 972, and again I was using the DAVE to audition the various options. Once I decided on the optimal strategy, but before I sent the master to Naxos, I tried upsampling the CD files with the M Scaler to see how they compared with the hi-rez originals.

Isn't it truly bizarre that someone who "understands hi-fi and digital" made the recording at 96KHz instead of 88.2KHz, forcing a non-integer rate conversion during preparation of the master for CD?

At the very least I would imagine that DAVINA in this situation would provide the operator with no options: there is only one way to do this correctly given the processing power of the convertor.

 

[Kentajalli]

I've also been thinking about a "Remastered by DAVINA" labelling model: a lot of recordings that are out there might have been originated in "HD", e.g. 96KHz or 192KHz and have been transferred to CD with poor equipment. They could be remastered for CD and/or "Redbook" streaming services such as Amazon or Spotify simply by using DAVINA to correctly filter/decimate. So we might find that existing HD recordings can get a significant benefit from DAVINA even for consumers who do not have M Scaler or any Chord DAC.

These days there are plenty of outlets one can get Hi-Res audio in original 24/96 or even 24/192, so is there much mileage in down-converting by any equipment??

John Atkinson wrote:
Isn't it truly bizarre that someone who "understands hi-fi and digital" made the recording at 96KHz instead of 88.2KHz, forcing a non-integer rate conversion during preparation of the master for CD?

24/88 is not as popular on recording ADCs , though I am no expert. 24/96 and 24/192 seem to be the norm for some reason.
He should have used DXD!

 

[Rob Watts]

1M taps was, as far as I can tell, originally a huge surprise for you in terms of the sound quality uplift it offered. Along with that was the increasing magnitude of the improvement in sound quality as you went from 0.25M to 0.5M to 1M taps. So that seems to represent an inverse relationship to the "area of the transient error" that you've described.

So I'm wondering if the WTA1 tap-length improves sound quality seemingly exponentially because the recording was itself so badly aliased. So the worse the original ADC and aliasing, the greater the benefit that 1M taps offers?

Is there a way to systematically identify the quantity/quality of aliasing that's in a recording? I'm just wondering if it's possible to find a relationship between the "effect" of M Scaler and the aliasing errors of an original recording.

I've also been thinking about a "Remastered by DAVINA" labelling model: a lot of recordings that are out there might have been originated in "HD", e.g. 96KHz or 192KHz and have been transferred to CD with poor equipment. They could be remastered for CD and/or "Redbook" streaming services such as Amazon or Spotify simply by using DAVINA to correctly filter/decimate. So we might find that existing HD recordings can get a significant benefit from DAVINA even for consumers who do not have M Scaler or any Chord DAC.

In Stereophile's review of M Scaler,

https://www.stereophile.com/content...-m-scaler-upsampling-digital-processor-page-2

John Atkinson wrote:


Isn't it truly bizarre that someone who "understands hi-fi and digital" made the recording at 96KHz instead of 88.2KHz, forcing a non-integer rate conversion during preparation of the master for CD?

At the very least I would imagine that DAVINA in this situation would provide the operator with no options: there is only one way to do this correctly given the processing power of the convertor.

So the area of the error relates to the sample rates - so a 768kHz signal would have 256 times less transient error innately, expressed in area, than a 48kHz signal. So as you go up in incoming sample rate, the errors improve markedly. This is why higher sample rates sound better, as the area of the transient errors are getting much smaller.

Now the problem of reconstruction and transient errors applies to perfectly sampled (i.e. no aliasing on the ADC) as well as poor recordings that have aliasing. The issue has nothing to do with ADC aliasing - that's another problem, and can only be solved completely within the ADC.

When I did my thousands of WTA listening tests, I optimised the algorithm with a number of very different tracks to listen to specific aspects of sound quality - and I have never found a different WTA sweet spot with different recordings, so it's all about the "simple" problem of reconstruction - ideal samples or flawed (aliased) samples have no bearing on the reconstruction problem.

 

[Rob Watts]

I ought to add, that listening to WTA filters above 16FS (different tap lengths, different oversampling) does have different sound quality improvements. With the M scaler, you get improvements in tempo, pitch reproduction, timbre variation and instrument separation and focus; above 16FS so far I have not had any of these improvements - it's more limited to clarity, soundstage and the perception of starting and stopping of notes.

 

[Jawed]

I ought to add, that listening to WTA filters above 16FS (different tap lengths, different oversampling) does have different sound quality improvements.

So, WTA1 should be treated as the highest priority, getting the majority of processing in a constrained-DSP scenario.

But it does also seem as if 1M taps have not been "fully unlocked" in DAVE. So that when exploring the quality of M Scaler, its ability to "recover" the maximum sound quality of a DAVINA 705.6KHz recording that's been filtered/decimated to 44.1KHz, is at least partially obscured by a "small" WTA2 in DAVE.

The original DAVINA 705.6KHz recording would also be bottlenecked by a "small" WTA2, I suppose, where WTA1 is not being used at all.

So in both cases an evaluation of "transparency", or accuracy with respect to the original analogue signal that was fed in to DAVINA, is hampered. It almost seems as if you have to go round in circles: if you improve WTA1 you might have to improve WTA2 to fully unlock its benefits, and then improve WTA1 some more, to then follow that with WTA2 improvements to unlock more, and so on...

Presumably a very large, e.g. 1M, tap length in WTA2 is extremely hard to achieve because there is no FPGA that you have access to that can run fast enough to keep up with 256FS. e.g. you might not be able to get past 64,000 taps for WTA2. That would at least reduce the number of steps as you circle around optimising WTA1...

With the M scaler, you get improvements in tempo, pitch reproduction, timbre variation and instrument separation and focus; above 16FS so far I have not had any of these improvements - it's more limited to clarity, soundstage and the perception of starting and stopping of notes.

This makes it seem as if WTA2 is important to minimise "continuous" errors in the DAC, almost like there's a level of noise inside the DAC. Or as if there's a quantisation-noise mechanism that's interacting with the internal math of the DAC.

Looking forward to finding out about the new heights you've achieved in your WTA2 journey!

 

[Solan]

Davina is the code word for the ADC project notionally intended for Dave.It will be the first product to use my Pulse Array ADC.

Just curious, a "few" years later: What is the state of the world when it comes to AD conversion these days? Has Davina been a fruitful project?

 

[Rob Watts]

Davina got stalled with a key part no longer being available, and the pressure of other projects. So I have come up with a better method that removed this part, will be testing the concept towards the end of the year.

 

[RustyGates]

I am personally itching for news on the Power Pulse Array stereo/mono power amps. I've been eyeing on moving from a headphone system to a full speaker set-up since I've got my first home.

 

[Rob Watts]

Oddly enough I finished a prototype PCB last week that tests one of the new concepts in power pulse array - so progress is being made, if somewhat glacial!

 

[RustyGates]

Oddly enough I finished a prototype PCB last week that tests one of the new concepts in power pulse array - so progress is being made, if somewhat glacial!

Great to hear! Still seems like its in the R&D phase so probably won't hear about it till at least 2023 I suppose. Plenty of other power amps to try out in the mean time.

 

[kennyb123]

@Rob Watts is it accurate that the Qutest has a Class A output stage? I don’t see this listed for the TT2 or DAVE and thought they each had the same fundamental design and didn’t have an traditional output stage found on other DACs.

 

[dac64]

@Rob Watts is it accurate that the Qutest has a Class A output stage? I don’t see this listed for the TT2 or DAVE and thought they each had the same fundamental design and didn’t have an traditional output stage found on other DACs.

in general, all audio equipment, other than poweramp, are all "class A" output. this is because it consume very little current.

 

[Rob Watts]

@Rob Watts is it accurate that the Qutest has a Class A output stage? I don’t see this listed for the TT2 or DAVE and thought they each had the same fundamental design and didn’t have an traditional output stage found on other DACs.

It shares the same OP stage as Mojo, a very high speed discrete design with a single global feedback path, and this is pure Class A with 300 ohm loads; since Qutest would never see a load less than 300 ohms (as it is intended to drive amplifier inputs - typically 10k ohms to 47k), so it is correct in describing it as Class A. So with other designs, they too would be Class A with a 300 ohm load or greater. With loads less than 300 and high OP voltage, the OP stage goes into Class AB mode; with Hugo2, TT2 and Dave this is not a problem (unlike conventional Class AB) as crossover distortion is eliminated by the second order analogue noise shaper topology. In simple terms this works by employing another amp that has the only job of making the OP stage linear; together with the first amp this allows the complete elimination of measurable crossover distortion (and all other OP stage related distortions). The OP stage amp is inside the global feedback path, which means that the topology is still simple from the direct signal path POV - this is very important as a simple signal path means better small signal accuracy, giving better perception of detail resolution and depth.

 

[The Jester]

So in theory at least the Qutest would have the most transparent output stage ?
in theory the less components in the signal chain, in practice though I’m guessing maybe not ?