Paralleling DACs

[DonH56]

Lower than what? As a statement it is true, but I strongly suspect if you lower the integrated phase noise from 1 ps rms to 0.1 ps rms you won't hear it. Same thing for "cascading clocks" if they mean interleaving; done right there is no harm and may be some benefit, but done wrong will cause problems. I do not know anything about Mutec and cannot comment on their claims. I can say I have seen many claims over the years that claim a scientific basis but instead use science to befuddle the masses. Like showing graphs with no units touting the benefits of cable M's much wider bandwidth when the actual scale is milli-dB and MHz so, while true, the audible benefit is zilch. Or conflating statements like "lower phase noise is better" when the competing product's phase noise was already inaudible. I worked on systems with 140 ~ 160 dB of dynamic range and clearly 160 dB is better than 140 dB but if it takes 10x or 100x the cost and power and the effective SNR of the input signal is only 100 dB what do I gain?

Just to throw out a "good" example, making a single 24-bit, 384 kS/s NOS DAC is rather a challenge, and so interleaving a couple or more lower-speed 24-bit DACs to get the desired resolution and speed makes sense. The clocks would have to be precisely aligned to prevent timing errors that introduce distortion, but the end result could certainly be better (or worse) than a single DAC.

And to reiterate, I do not know anything about Mutec and have no idea if they are providing credible proof, as if they or anyone should care what I determine is "credible". I'll try to follow up the links tonight; need to wrap up a few things and head out of town for a few hours this afternoon.

 

[fas42]

This is a classic "everything matters" scenario - we have a "tower of accuracy", being fed by a lousy PS and grounding setup - ummm, let's try to maintain a standard of precision in every aspect that could possibly contribute to maintaining the theoretical accuracy ... rather a simple concept, but that seems to get bowled over in the excitement to "play with another set of toys".

Because audio "doesn't matter" - it won't kill someone if it performs poorly, in most cases - sloppiness usually goes along for the ride. So long as it's sufficiently blingy, bits of it look adequately impressive to the those present - "it must be good enough".

OTOH, if one decides to get fussy about what the ears pick up then it turns out that one has to be fussy, not in getting impressive performance numbers, but making sure every tiny detail has been fully addressed.

 

[Palladium]

IMO that's an idea that is sound (what, puns?) only in purely romantic theoretical grounds but worthless in reality. Semiconductors are already so well fabbed that the any variance between one DAC chip to the next of the same is astronomically small, and on overall it's so much more likely to screw up in a multi-chip implementation than a bog standard single chip solution which defeats the whole purpose.

 

[DonH56]

Depends on the correction and architecture. Device matching, active and passive, is actually worse as process dimensions shrink, and the smaller dimensions also lead to lower breakdown so less voltage headroom and higher leakage. Not real analog-friendly. Delta-sigma designs incorporate a lot of correction circuits that may or may not be present in R2R and similar designs. If you need to achieve higher speed and resolution than available devices provide then interleaving makes sense and is widely used now so it is not some theoretical exercise. And with device matching perhaps 1% or a little lower if you take heroic measures, that is a major impact on performance if you need to interleave them. Remember 16-bit matching is about 0.0015% and that is much lower than device variation on-chip let alone chip-to-chip. If two 16-bit DACs are mismatched by say 0.1% then you have a glitch at around the 10-bit level. These errors can be compensated but it takes some careful design.

I suspect the biggest benefit of interleaving for audio is the reduction in correlated spurs, increasing the overall spurious-free dynamic range (SFDR). Assuming it is done rightly. And, not knowing the cost of audio DACs, perhaps it is cheaper to interleave a couple of 100 kS/s 24-bit DACs to achieve 192 kS/s. But I have not seen a lot of DACs that actually meet the theoretical ideals for SNR and SFDR even at lower rates.

That said the specs of audio DACs seem more than good enough to me to perform without audible degradation as-is. Dealing with PSRR, CMRR, and other elements of isolation of the device on the board, and implementing the output stages of the DAC (i.e. anti-imaging filter and buffer/driver stages), seem to me to be much of what differentiates good and bad audio DACs.

 

[fas42]

That said the specs of audio DACs seem more than good enough to me to perform without audible degradation as-is. Dealing with PSRR, CMRR, and other elements of isolation of the device on the board, and implementing the output stages of the DAC (i.e. anti-imaging filter and buffer/driver stages), seem to me to be much of what differentiates good and bad audio DACs.

Indeed this is the case. I've achieved my "competent playback" with DACs of all persuasions - a bog standard unit has "enough quality" to get the job done, if all the surrounding issues are taken care of.

Again, it's always, always the weakest link in the chain - find enough of these, sort them, and then the subjective quality is way beyond what the majority of audiphile rigs achieve.

 

[Thomas savage]

Indeed this is the case. I've achieved my "competent playback" with DACs of all persuasions - a bog standard unit has "enough quality" to get the job done, if all the surrounding issues are taken care of.

Again, it's always, always the weakest link in the chain - find enough of these, sort them, and then the subjective quality is way beyond what the majority of audiphile rigs achieve.

No one knows what you mean by “ surrounding issues” nor what you do to “ sort them”.

It’s all gobbledygook.

 

[RayDunzl]

That said the specs of audio DACs seem more than good enough to me to perform without audible degradation as-is. Dealing with PSRR, CMRR, and other elements of isolation of the device on the board, and implementing the output stages of the DAC (i.e. anti-imaging filter and buffer/driver stages), seem to me to be much of what differentiates good and bad audio DACs.

"Most people focus on the word "digital" and assume that all of the "magic" happens in the digital processing, but nothing could be further from the truth! A look inside most audio converters would show that about 90% of the components are analog! ... It is safe to say that 90% of the measurable performance differences are directly related to the differences in the analog processing inside these products. "

https://benchmarkmedia.com/blogs/ap...1191-inside-the-dac2-part-1-analog-processing

 

[DonH56]

By transistor count the vast majority is digital, but by component count on the board it is almost all analog... Glad to see someone else agrees with and backs up my off-the-cuff assessment, thanks for the quote Ray!

 

[Wombat]

There is no substantial technical reason for line-level analogue stages to have poor performance, either, even lower cost ones.

 

[Gaston]

Hello dear audio friends,
This DDDAC sounds fantastic, and stacking decks improves the sound every time. And when adding shunt regulators (4 per PCB), they become even very much better.
I had an expensive upgrade with 2 extra shunt loaded decks, thinking it would not have been worth the money.. but resulted in a night without sleep. I don't drink or do drugs, but the music had an emotional impact I remember from the days, long time ago, that I did..
So, I don't know or understand (or care) about the technical explanation that it would not be good.. Maybe you are right when you say so.. but then you definitely never heard this DDDAC playing!
I can truly recommend that you try.
And you can start off with only one deck, one motherboard and power supply.. Sounds good. Not too expensive..
I had that, then 2 decks, then 2 with shunt regulators, then 4 with shunt regulators (which was the sleepless night version). And Now I'm building a new 12 layered fully loaded with shunt regulators version.
Good to know is that Audio Creative now only sells PCB's with Tentlabs shunt regulators already on board. Saves a lot of money if you want to make one.
Enjoy!
Gaston

 

[Krusty09]

I remember from many years ago these guys were parralling dacs for years and they still might. I do remember back in the day hearing his nodded CD players thinking they sounded very nice.

http://aplhifi.com/products/digital/

 

[BDWoody]

Has anyone ever tried comparing DAC's by running the left channel from one and the right from the other simultaneously either to headphones or an amp? Obviously you'd have timing issues to straighten out, but I wonder what that would be like.
Probably not very useful, but might be interesting.

 

[Kane1972]

I’m sure my old Denon AV Amp used 8 DAC chips per channel in stereo mode and 2 per channel when doing 7.1. Is this completely a different thing?

 

[Lambda]

I know this is a old thread but the theory of this could easily be tested. (from someone owning a audio analyzer)
Send a mono signal to a stereo DAC and compare both channels to both channels bridged together.

I would Expect 1 to 2 dB more SNR

 

[JayGilb]

Has anyone ever tried comparing DAC's by running the left channel from one and the right from the other simultaneously either to headphones or an amp? Obviously you'd have timing issues to straighten out, but I wonder what that would be like.
Probably not very useful, but might be interesting.

I was pondering that also this morning and would be a fun experiment for both headphones and speakers.
A few samples of delay might give the audio a wider image space, sort of like double tracking instruments when recording.

 

[pkane]

I know this is a old thread but the theory of this could easily be tested. (from someone owning a audio analyzer)
Send a mono signal to a stereo DAC and compare both channels to both channels bridged together.

I would Expect 1 to 2 dB more SNR

RME ADI-2 Pro lets you combine stereo channels on the DAC or the ADC side (or both). The noise level is reduced by 3dB in the process, easily measured.