Denafrips DACs might not actually be NOS?

[GoldenOne]

As an update, I'm currently talking with someone else who has an APx555 and Denafrips terminator plus with them currently. 

I've sent them my project sequence and they're running it through in NOS and in OS.

Similar, but different behaviour is going on with the termi plus. This is the "NOS" impulse:

And as with ares 2, getting the IR from the transfer function measurement instead of an actual IR test shows something resembling normal sinc.

Definitely not NOS either, but it seems they're running the DSP differently to the ares 2.

Full measurements coming soon

[The Computer Audiophile]

Would be so great if Denafrips would comment. 

[GoldenOne]

53 minutes ago, The Computer Audiophile said:

Would be so great if Denafrips would comment. 

Previously when I contacted denafrips they always told me to speak to vinshine.

I have spoken to Alvin at vinshine and he's been incredibly friendly and helpful.
Unfortunately Denafrips aren't actually telling him anything. They're saying it is nos and that they don't want to give any secrets away :/

[RedFuneral]

I've been following along with this discussion as a NOS diehard whose been playing with other gear as of late & thinking of making a return. I'm not really sure what to think of the whole situation, it seems on the face of it the Ares is being advertised as something it's not BUT we also have a long standing precedent now of Sigma-Delta DACs which include a NOS mode which can be used with standard Redbook content. How are these functioning? Are they likewise Linear Interpolation? If this is the case then it strikes me that Denafrips isn't doing any unusual and have simply labelled this mode by established norms.

 

I wonder because I recently ventured into experimenting with the NOS mode on my Gustard X16 DAC(Sabre ES9068AS chipset) and find that it achieves the effortless sound of NOS once the treble is properly restored with EQ(36db(!) down @ 20khz.) It likewise avoids the audible aliasing of your standard unfiltered NOS DAC(tested subjectively with frequency sweeps.) I'm quite curious what my DAC is doing and if the Ares or other DACs* might be doing something similar as what I'm hearing right now seems a clear 1up over the traditional way of doing things.

 

*I know AKM has something similar with their super-slow filters, to my ears those don't sound very good. They don't remind me of NOS R2R, that's why I'm shocked what I'm hearing from this Gustard.

[The Computer Audiophile]

Hi guys, I’ll have some news to report on Denafrips designs in the not too distant future. Hold tight. 

[GoldenOne]

1 hour ago, RedFuneral said:

I wonder because I recently ventured into experimenting with the NOS mode on my Gustard X16 DAC(Sabre ES9068AS chipset) and find that it achieves the effortless sound of NOS once the treble is properly restored with EQ(36db(!) down @ 20khz.) It likewise avoids the audible aliasing of your standard unfiltered NOS DAC(tested subjectively with frequency sweeps.) I'm quite curious what my DAC is doing and if the Ares or other DACs* might be doing something similar as what I'm hearing right now seems a clear 1up over the traditional way of doing things.

Gustard 'NOS' isn't NOS either. It's not possible to run a delta sigma DAC without OS.

This is the IR for the X26 pro in 'NOS'. Seems to be a very short, minphase sinc or something.


Also would strongly recommend against using 36dB of EQ :P, that's uh.....a little excessive. 
Personally I don't like this 'NOS' mode much.

Some other dacs like some of the teac and ifi dacs have a 'pseudo-NOS' option made possible by the architecture of the burr-brown chip it seems. Though it's not actually NOS all the way through. The top 6 bits are though.

Others like the ADI-2 and more recent AKM dacs have a 'super slow' filter which is basically a zero-order-hold OS filter. Still OS, but produces a result very similar to actual NOS.

https://www.audiosciencereview.com/forum/index.php?attachments/33251/

[Miska]

2 hours ago, GoldenOne said:

Some other dacs like some of the teac and ifi dacs have a 'pseudo-NOS' option made possible by the architecture of the burr-brown chip it seems. Though it's not actually NOS all the way through. The top 6 bits are though.

Others like the ADI-2 and more recent AKM dacs have a 'super slow' filter which is basically a zero-order-hold OS filter. Still OS, but produces a result very similar to actual NOS.

 

"NOS" in these cases means it bypasses the 8x digital filter, and just leaves the zero-order-hold OS in place. Same for TI and AKM.

 

DAC chips just don't have enough DSP power to run proper 256x or higher digital filters, so they typically have digital filter up to 352.8/384k rate and rest is just OS by copying same sample N times.

 

However, TI chips pass DSD inputs straight to analog FIR conversion stage and now AKM chips have similar feature feeding DSD data straight to their SCF conversion stage. So DSD inputs can be true NOS. This allows running proper 256x or higher digital filters and high quality modulators not constrained by the on-chip DSP.

 

 

[GoldenOne]

6 minutes ago, Miska said:

 

"NOS" in these cases means it bypasses the 8x digital filter, and just leaves the zero-order-hold OS in place. Same for TI and AKM.

 

DAC chips just don't have enough DSP power to run proper 256x or higher digital filters, so they typically have digital filter up to 352.8/384k rate and rest is just OS by copying same sample N times.

 

However, TI chips pass DSD inputs straight to analog FIR conversion stage and now AKM chips have similar feature feeding DSD data straight to their SCF conversion stage. So DSD inputs can be true NOS. This allows running proper 256x or higher digital filters and high quality modulators not constrained by the on-chip DSP.

 

 

The weird thing with the ti/bb chips though is that they show different results when you feed it 44.1khz content and use the 'nos/bitperfect' mode, vs using zero-order hold upsampling to 768khz etc and then feeding that to the dac. 

 

Not sure why

[Miska]

12 minutes ago, GoldenOne said:

The weird thing with the ti/bb chips though is that they show different results when you feed it 44.1khz content and use the 'nos/bitperfect' mode, vs using zero-order hold upsampling to 768khz etc and then feeding that to the dac. 

 

Not sure why

 

Because they are not designed to be used in that way. That "NOS/bitperfect" mode is "external digital filter mode". And they assume to be receiving 705.6/768k input instead of 44.1k. If you still send them 44.1k, they are running whole thing at 16x lower rate than designed.

 

[GoldenOne]

2 hours ago, Miska said:

 

Because they are not designed to be used in that way. That "NOS/bitperfect" mode is "external digital filter mode". And they assume to be receiving 705.6/768k input instead of 44.1k. If you still send them 44.1k, they are running whole thing at 16x lower rate than designed.

 

But if its doing zoh internally why does it behave differently when fed zoh externally? 

 

Is the modulator itself running slower?

[RedFuneral]

7 hours ago, GoldenOne said:

Gustard 'NOS' isn't NOS either. It's not possible to run a delta sigma DAC without OS.

This is the IR for the X26 pro in 'NOS'. Seems to be a very short, minphase sinc or something.


Also would strongly recommend against using 36dB of EQ :P, that's uh.....a little excessive. 
Personally I don't like this 'NOS' mode much.

Some other dacs like some of the teac and ifi dacs have a 'pseudo-NOS' option made possible by the architecture of the burr-brown chip it seems. Though it's not actually NOS all the way through. The top 6 bits are though.

Others like the ADI-2 and more recent AKM dacs have a 'super slow' filter which is basically a zero-order-hold OS filter. Still OS, but produces a result very similar to actual NOS.

https://www.audio “science” review/forum/index.php?attachments/33251/

I wasn't aware of the X26's NOS mode. Annoyingly I cannot find an impulse measurement of the X16 to confirm if it's the same. I only have the frequency response measurement I found here:

You can see the reason for my intense EQ(I intend to scale it back to taste, the negative gain required is at the extreme of what I can compensate for.) Do you know if the X26 is doing this in the same way as the X16? It seems to be an implementation developed by ESS & built into the chips but I couldn't find any documentation of its existence outside the 9068AS when I was searching last night.

 

I'm curious why the AKM super-slow approach doesn't sound NOS to my ears despite similar measured performance while the Gustard subjectively does. I don't have any experience to comment on other chips implementations.

 

Ok to check my understanding: Sigma-Delta DACs employ zero-order-hold OS(repeating samples, correct?) while the Denafrips is adding samples equidistant between those input? This would mean both methods have new samples however in the zoh method the samples are clones of those input while they're novel with linear interpolation?

 

As a follow-up question; are the polynomial filters in HQP similar in theory as to the linear interpolation in the Denafrips? I've been playing around with the trial of this software trying to find something comparable to the major sound quality improvement I hear running the X16 in it's NOS mode and reading the included documentation these sounded similar in theory: not NOS but following the NOS promise of preserving the impulse response.

[GoldenOne]

51 minutes ago, RedFuneral said:

I wasn't aware of the X26's NOS mode. Annoyingly I cannot find an impulse measurement of the X16 to confirm if it's the same. I only have the frequency response measurement I found here:

Just checked to make sure but yeah it's the same filter:

 

 

51 minutes ago, RedFuneral said:

You can see the reason for my intense EQ

It would be better to just use a filter which had a flatter treble response, rather than using this then trying to compensate with EQ I imagine.

For example adding a +24dB shelf in roon which looks like this (if you let me know your exact EQ settings I can replicate them and show you the resulting output and IR):

Just makes the FR look like this: 

IR:
 

 

51 minutes ago, RedFuneral said:

Ok to check my understanding: Sigma-Delta DACs employ zero-order-hold OS(repeating samples, correct?)

No, Delta sigma dacs will use any filter you choose, and will normally have the option of a few 'fast' 'slow' and linear/minphase filters etc.
Dac chips have stock ones but some manufacturers offer their own

 

 

51 minutes ago, RedFuneral said:

As a follow-up question; are the polynomial filters in HQP similar in theory as to the linear interpolation in the Denafrips?

No, polynomial uses cubic (spline) interpolation, whereas denafrips is just using linear interpolation, basically playing connect the dots.

 

 

51 minutes ago, RedFuneral said:

I've been playing around with the trial of this software trying to find something comparable to the major sound quality improvement I hear running the X16 in it's NOS mode

The most similar filter would probably be something like poly-sinc-short-mp I'd guess though @Miska is the one to ask. I'm not sure HQP has any filters that short as it kinda goes against the point of having HQP in the first place.

 

 

51 minutes ago, RedFuneral said:

not NOS but following the NOS promise of preserving the impulse response.

Well it's definitely not NOS, but it's not preserving NOS impulse response either. It's just a very very short filter by the looks of things.
 

[RedFuneral]

1 hour ago, GoldenOne said:

Just checked to make sure but yeah it's the same filter:

...

 

Well it's definitely not NOS, but it's not preserving NOS impulse response either. It's just a very very short filter by the looks of things.
 

Thanks, I appreciate your effort in measuring this. It seems safe to conclude that I'm not preferring the sound of this because it's somehow purer in the time domain or otherwise. I've actually owned this DAC since April and have been using the standard flat filters up until this past week. I like the DAC but really have only enjoyed music in short bursts, that changed when I found this 'one weird trick.' Subjectively it sounds like a major system upgrade where I'm not fatiguing while listening anymore. It sounds effortless and out of this I'm able to better focus on the music and follow along with it, vocals sound more real & fast music more legible. I've spent most of my freetime in the past 48hrs A/Bing this NOS mode vs alternatives(Standard filters, Windows upsampling, HQP) and my impression remains consistent that this mode is doing something unique.

 

Now I have to admit most of my music is of subpar recording quality(still it is lossless 16/44.1K) and over the past decade I've listened through real NOS DACs more often than not. Do you have any theories as to why I much prefer this sound? Could there be an additive coloration or distortion that's familiar to past gear I remember fondly?

 

I wasn't able to recreate this impression with any of the filters in HQP(they're similar to the normal filters built into the Gustard.) It also isn't the EQ as I've tested this with music without significant content >5khz which sounds identical with EQ toggled on/off. I likewise don't think it's just a preference for reduced treble as I just switched from an Asgard 3 amp to a Drop THX specifically to attain a brighter/faster treble.

[Miska]

10 hours ago, GoldenOne said:

But if its doing zoh internally why does it behave differently when fed zoh externally? 

 

Is the modulator itself running slower?

 

Yes, it is, because the entire thing is running slower. ZOH / S/H factor is static in that case. Or semi-static, since TI chips have three configurable ZOH / S/H factors, you just cannot exceed maximum output rates, but these are to accommodate 8x / 16x digital filter rates. Since their (TI and AKM) internal filter can go to max 8x while they support external filters that can go up to either 8x or 16x.

 

8 hours ago, GoldenOne said:

The most similar filter would probably be something like poly-sinc-short-mp I'd guess though @Miska is the one to ask. I'm not sure HQP has any filters that short as it kinda goes against the point of having HQP in the first place.

 

poly-sinc-short-mp is more like some DAC chip minimum-phase filter, but with the difference that it has >192 dB stop-band attenuation and thus reconstruction accuracy to 32-bit level.

 

polynomial "filters" (interpolators) are closest to "NOS" with least filtering efficiency, then a somewhat longer is minringFIR.

 

[Blake]

I do like the measurements and info provided here.  My hope is measurements and inquiry will continue to help advance the sonic performance of audio devices.  What I'd like to see down the road is further refinement in correlating measurements with the sound qualities perceived by the listener.  Anecdotally, it seems that in some instances better measurements don't necessarily correlate to subjective listener perceptions of a better sounding product.  What measurements really matter and which ones do not?  Or, what is an acceptable range for a given measurement, and beyond that measurement range, the measurements cease to matter in terms of perceived sonic performance? 

 

If a product is fairly consistently praised by a relatively larger sample size of users with divergent systems/rooms/hearing abilities and subjective preferences as sounding more like the "real thing" but some measurements of said product are less than ideal, why is that the case?

 

We know a lot, but there is quite a bit more to learn and for me the science of correlating product measurements to subjectively perceived sonic aspects (soundstage size, image specificity, tonal qualities, "body", etc) needs further exploration.

 

Of course, this is nothing new.  We've had products in the past like the cheap ODAC that was supposedly "perfection" due to measurements but in terms of sound quality very clearly it wasn't.  

 

I think having more reviewers with measuring equipment and collaboration among knowledgeable folks (thanks here to @GoldenOneand @Miska) will hopefully advance this part of our audio hobby so that in the future users could possibly use the various different published measurements to help them to narrow down choices of which products they should consider, while eliminating others, given that user's sonic preferences.

[The Computer Audiophile]

22 minutes ago, Blake said:

I do like the measurements and info provided here.  My hope is measurements and inquiry will continue to help advance the sonic performance of audio devices.  What I'd like to see down the road is further refinement in correlating measurements with the sound qualities perceived by the listener.  Anecdotally, it seems that in some instances better measurements don't necessarily correlate to subjective listener perceptions of a better sounding product.  What measurements really matter and which ones do not?  Or, what is an acceptable range for a given measurement, and beyond that measurement range, the measurements cease to matter in terms of perceived sonic performance? 

 

If a product is fairly consistently praised by a relatively larger sample size of users with divergent systems/rooms/hearing abilities and subjective preferences as sounding more like the "real thing" but some measurements of said product are less than ideal, why is that the case?

 

We know a lot, but there is quite a bit more to learn and for me the science of correlating product measurements to subjectively perceived sonic aspects (soundstage size, image specificity, tonal qualities, "body", etc) needs further exploration.

 

Of course, this is nothing new.  We've had products in the past like the cheap ODAC that was supposedly "perfection" due to measurements but in terms of sound quality very clearly it wasn't.  

 

I think having more reviewers with measuring equipment and collaboration among knowledgeable folks (thanks here to @GoldenOneand @Miska) will hopefully advance this part of our audio hobby so that in the future users could possibly use the various different published measurements to help them to narrow down choices of which products they should consider, while eliminating others, given that user's sonic preferences.

Well said. 

[luisma]

Does anyone knows if the Bricasti and the Dave have a NOS option and if this option is not linear interpolation? I might have miss it

 

[GoldenOne]

22 minutes ago, luisma said:

Does anyone knows if the Bricasti and the Dave have a NOS option and if this option is not linear interpolation? I might have miss it

 

The Dave and bricasti m1se are both delta sigma dacs so cannot run NOS

[barrows]

On 7/21/2021 at 5:00 PM, GoldenOne said:

The Dave and bricasti m1se are both delta sigma dacs so cannot run NOS

But one must also consider the Bricasti M21, which uses AD R2R DAC chips for its PCM R2R mode.  This DAC also features a discrete DSD converter section.  There is an onboard OSF section running on a fairly powerful AD SHARC DSP chip, which runs Bricasti's own oversampling filters (but I am not sure to what rate this operates).  I do not know if one can bypass the onboard OSF and send the input rate (PCM) directly to the R2R conversion section, or if by inputting a high enough PCM rate via HQPlayer one would be able to run the input straight to the R2R conversion stage.

My understanding is that via the discrete DSD converter, one gets true NOS DSD conversion-in any case, the Bricasti M3 I have here, which features the same discrete DSD conversion section, sounds glorious when fed DSD 256 via HQPlayer-Signature Rendu SEoptical-USB to DAC.  The M3 does nor have the R2R conversion option for PCM though, and just converts PCM via AD SDM chips (I only use this DAC via the DSD section myself).

[semente]

On 7/21/2021 at 4:28 PM, Blake said:

Anecdotally, it seems that in some instances better measurements don't necessarily correlate to subjective listener perceptions of a better sounding product.

The problem is that it depends on the listener, it's a matter of personal preference.

In other words, not everyone likes the same "presentation".

 

Ultimately each individual will have to go the extra mile and learn how to correlate his preferences with measurements. Easier said than done, particularly with electronics.

[numlog]

Measurements have some correlation to personal preference at transduce and maybe tube amp levels of distortion where it is directly audible, rarely at DAC or SS levels and it only tells small part of story if it does.

DACs and other electronics have been well optimised for basic measurements for decades, you are not hearing <-90dB distortions but you are also not listening to static test signals, we hear clearly audible distortions in DACs playing music because they are at clearly audible levels, but apparently they haven't found a way to measure these real world distortions.

 

Generally R2R dacs and tube amp measure worse than SS and DS and both have associations with being preferred subjectively, 

but correlation doesn't equal causation, it's largely the nature of these designs that lead to the perceived differences: minimal feedback, short signal paths, inherent linearity, discrete, Class A/lower efficiency and so on, almost opposite in nature to modern mainstream electronics.

 

For example when Holo Audio comes along with their R2R DACs with DS levels of distortion and Pass Labs with their low distortion Class A ZFB amplifiers trying to correlate measurements with sound quality doesn't work anymore,  it was a coincidental pattern not a rule.

 

[semente]

6 hours ago, numlog said:

Measurements have some correlation to personal preference at transduce and maybe tube amp levels of distortion where it is directly audible, rarely at DAC or SS levels and it only tells small part of story if it does.

DACs and other electronics have been well optimised for basic measurements for decades, you are not hearing <-90dB distortions but you are also not listening to static test signals, we hear clearly audible distortions in DACs playing music because they are at clearly audible levels, but apparently they haven't found a way to measure these real world distortions.

 

Generally R2R dacs and tube amp measure worse than SS and DS and both have associations with being preferred subjectively, 

but correlation doesn't equal causation, it's largely the nature of these designs that lead to the perceived differences: minimal feedback, short signal paths, inherent linearity, discrete, Class A/lower efficiency and so on, almost opposite in nature to modern mainstream electronics.

 

For example when Holo Audio comes along with their R2R DACs with DS levels of distortion and Pass Labs with their low distortion Class A ZFB amplifiers trying to correlate measurements with sound quality doesn't work anymore,  it was a coincidental pattern not a rule.

 

 

I do agree in general but not when you say that "Generally R2R dacs and tube amp measure worse than SS and DS and both have associations with being preferred subjectively". The fact is that some people do prefer them and others don't.

 

Also, maybe we can look at measurements of different topologies (minimal feedback, short signal paths, inherent linearity, discrete, Class A/lower efficiency) and try to determine whether or not the could be causing audible impact (which may be "positive" to some but "negative" to others).

And let's not forget that audiophiles can sometimes fall in love with a topology and this may impact/bias their perception...

 

I think that the first challenge comes from the difficulty in separating our personal preference from our ability to assess performance in an unbiased observationist manner. An increase in accuracy, which is difficult to determine through listening by itself, may end up sounding worse to some people.

[numlog]

1 hour ago, semente said:

 

I do agree in general but not when you say that "Generally R2R dacs and tube amp measure worse than SS and DS and both have associations with being preferred subjectively". The fact is that some people do prefer them and others don't.

 

It is a given that everyone will have a preference, I worded that poorly.

 

1 hour ago, semente said:

Also, maybe we can look at measurements of different topologies (minimal feedback, short signal paths, inherent linearity, discrete, Class A/lower efficiency) and try to determine whether or not the could be causing audible impact (which may be "positive" to some but "negative" to others).

And let's not forget that audiophiles can sometimes fall in love with a topology and this may impact/bias their perception...

 

All my experience leads me to believe R2R and ZFB have more potential accuracy than the alternatives, it was the audiophiles tales that lead me to explore them in the first place... often the tales exist for a reason.

 There are so many variables that a certain topology guarantees very little. Until there are advances made in measurements to better define 'accuracy' the best we can do is get as much personal experience as possible with different designs and topologies.

1 hour ago, semente said:

An increase in accuracy, which is difficult to determine through listening by itself, may end up sounding worse to some people.

I dont believe that increased accuracy will not sound better to anyone who is actually concerned with accuracy, it would be a preference for one aspect of accuracy (or inaccuracy) over another, tradeoffs like this exist in every piece of equipment.

If measurements are used to define accuracy I can agree.

 

 

 

 

 

[GoldenOne]

23 hours ago, barrows said:

My understanding is that via the discrete DSD converter, one gets true NOS DSD conversion-in any case

This should be the case for any DAC that can do native DSD conversion including AKM and Burr Brown chips (though not ESS).
Though calling it 'NOS' is true in that there is no further oversampling being applied, but also different as DSD is effectively what the result of delta sigma oversampling+Modulation looks like.
It's not comparable to NOS PCM if that's what you mean.
 

 

10 hours ago, numlog said:

Measurements have some correlation to personal preference at transduce and maybe tube amp levels of distortion where it is directly audible, rarely at DAC or SS levels and it only tells small part of story if it does.

DACs and other electronics have been well optimised for basic measurements for decades, you are not hearing <-90dB distortions but you are also not listening to static test signals, we hear clearly audible distortions in DACs playing music because they are at clearly audible levels, but apparently they haven't found a way to measure these real world distortions.

 

Generally R2R dacs and tube amp measure worse than SS and DS and both have associations with being preferred subjectively, 

but correlation doesn't equal causation, it's largely the nature of these designs that lead to the perceived differences: minimal feedback, short signal paths, inherent linearity, discrete, Class A/lower efficiency and so on, almost opposite in nature to modern mainstream electronics.

 

For example when Holo Audio comes along with their R2R DACs with DS levels of distortion and Pass Labs with their low distortion Class A ZFB amplifiers trying to correlate measurements with sound quality doesn't work anymore,  it was a coincidental pattern not a rule.

 

But why? Why are you quick to assume that those measured differences make no difference?
The products sound different do they not? And so therefore something must be different. The most obvious answer being..well...the differences measured :P

 

[numlog]

1 hour ago, GoldenOne said:

But why? Why are you quick to assume that those measured differences make no difference?
The products sound different do they not? And so therefore something must be different. The most obvious answer being..well...the differences measured :P

 

It is not a quick assumption, seeing how measurements relate to the listening experience has been something I've been exploring for a while... I would like to believe measurements tell all as it would get rid of all the guesswork and mystery.

 I do believe the differences or rather causes of differences could be measurable, if you analyse points earlier in the circuit rather than just the output... doing something like this would not be easy at all, and for results that still wont give you any clear answers.

 

There is so much going on in every part of a DAC circuit. As an example, from the AC outlet to a power input on a chip is just a single complex chain, at various points in this chain there are going to be ways it can be objectively improved further (larger/seperate transformers for lower impedance and reduced interference between supplies, larger capacitors for improved ripple filtering and lower impedance, RF filtering , better regulators etc. )  but if these improvements dont give an appreciable improvement in measurements of the output then in a measurement focused design it wont be given a second thought. I think this kind of thing is the case for a lot of commercial designs.