Denafrips DACs might not actually be NOS?

[semente]

2 minutes ago, idiot_savant said:

@GoldenOne you are, of course correct, in that filters should have less effect the higher in sample rate you go, so that brings us back to what is the point of a NOS mode at lower sample rates?

do you happen to know anything about the Holo analogue reconstruction filter? I can’t see it doing much good unless matched with an OS filter?

 

your friendly neighbourhood idiot 

 

It doesn't seem to be doing much:

 

HoloAudio May, NOS mode, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan) into 100k ohms with data sampled at 44.1kHz (20dB/vertical div.).

https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements

[GoldenOne]

15 minutes ago, idiot_savant said:

@GoldenOne you are, of course correct, in that filters should have less effect the higher in sample rate you go, so that brings us back to what is the point of a NOS mode at lower sample rates?

do you happen to know anything about the Holo analogue reconstruction filter? I can’t see it doing much good unless matched with an OS filter?

 

your friendly neighbourhood idiot 

Not too much no. Though it's not intended to be a replacement for actual oversampling.

 

The may has internal oversampling if desired (and can also upsample to dsd and play back using its dedicated 1-bit converter) and also has 1.536mhz/dsd1024 input support so that you can do exceptionally high rate OS with software like hqplayer. 

 

Its got all the options so the user can use it however they choose. 

 

The NOS mode is indeed NOS and seems to just have a gentle analog filter to get rid of ridiculously high stuff (>50khz etc). Though I believe @Miskahas mentioned that the holo dacs may be doing something more advanced than a basic low pass filter. (could be wrong there but I remember reading a mention of that) 

[botrytis]

1 hour ago, semente said:

VW brought us the emissions-gate. And now we have the NOS-gate.

 

What a strange world.

 

Mercedes, Audi, BMW, GM, and Ford were all caught in the emissions-gate also - VW was the the tip......

[The Computer Audiophile]

1 hour ago, idiot_savant said:

MHO the problem here is that you have a DAC that ( as it stands ) you can set it up to be really quite poor in frequency response terms, then as you increase sample rates via eg HQPlayer it suddenly sounds much better, when it could have been perfectly fine already

But using HQP you can take this DAC way farther than any internal filtering and over sampling could do. The horsepower in any DAC is just too weak to do what HQP does. That’s why people like NOS type DACs. 

[semente]

7 hours ago, The Computer Audiophile said:

But using HQP you can take this DAC way farther than any internal filtering and over sampling could do. The horsepower in any DAC is just too weak to do what HQP does. That’s why people like NOS type DACs. 

 

Some people like NOS DACs with Redbook too, often driven by CD transports. It's easy to forget that files are not the only format, perhaps not even the most common.

Several NOS DACs are limited to 16-bit and 44.1/48kHz. Philips' entry-level TDA1543 D/A chip is a favourite for its low price and if I'm not mistaken ease of implementation.

I tried NOS with CD for a while back in the day, with a Shigaraki transport+DAC combo.

[semente]

7 hours ago, botrytis said:

 

Mercedes, Audi, BMW, GM, and Ford were all caught in the emissions-gate also - VW was the the tip......

 

I know that.

It was supposed to be a joke, but if you and @The Computer Audiophile didn't get it then maybe I should have added a cheeky smiley face.

[idiot_savant]

@The Computer Audiophile I actually think that @etane makes a good point here - you're talking about very heavily OS *system*, yet there are many references to "NOS". You agree (thankfully) that NOS at redbook rates is not a good idea, yet many people will say that is what NOS *is* - no processing, what's on the CD is presented "purely" ( which is, IMHO not ideal ). Does it say in the user manual or advertising that you need to buy a PC and 3rd party software to optimise it? 

 

I think we're moving steadily off-topic, but...

 

As for HQPlayer taking this DAC much farther, what do you mean? You can move it farther from the speakers? Better frequency response? Lower noise floor? Less distortion? As for horsepower, how much is enough? If you're talking about reconstruction filters, a modern FPGA will have "more than enough" to comfortably exceed any realistic requirements. For example, a Xilinx Artix7-200T, as used by e.g. the Chord MScaler according to the Xilinx datasheet can do 9.2GMAC/s sustained, a Mola Mola Tambaqui has something like 2.7GMAC/s, and because they doesn't have to run e.g. Windows, OSX or Linux, then you don't run into any of the classic problems of actually extracting performance from CPUs ( cache misses, task switches, IO ). Is there a penalty or benefit for moving higher-rate audio *into* the DAC rather than doing it internally? Is there a penalty for having a highly clocked CPU in the same location? Is the Denafrips OS implementation not ideal?

 

I'm not saying any approaches are necessarily better than other ones, and I'm certainly not talking about particular products, but as you yourself said, it's not black and white - bigger number and more processing does not guarantee better measurable performance or sound quality.

 

The job of the reconstruction filter is a holistic one - in a particular architecture, the digital OS and analogue filtering *should* be closely aligned and complement each other - there is, for example not much point in oversampling to very high rates if the analogue filtering cuts in early - you are literally just adding switching noise, and vice versa if the OS filter is gentler than the analogue filter is designed for, you can easily run into slew-rate limiting, things becoming unstable etc. Also, some companies may also choose to tweak their digital filter to compensate for deficiencies in the analogue filter ( e.g. the analogue filter is a bit droopy, so a bit of DSP can give a peak appropriately ) - this is obviously problematic if you separate the two.

 

Is an R2R NOS DAC still "pure" if you have a PC doing OS & noise shaping to help linearise it? 

 

your friendly neighbourhood idiot

[manueljenkin]

I have personally found ess dacs when paired with a decent analog section sound better stock than with Hqplayer. Hqplayer filters sounded worse, and also lots of pc noise pushed into the dac compared to other players.

 

There's an elegant way around everything. Digital filtering is just for cases where you want to compensate for certain things in analog section, mainly delta Sigma DACs. Even there, you would have to play around with many things, including proper pairing of noise shapers to match to the analog section (and of course the modern chips are mostly just delta Sigma by name, internally far more complicated, 5 or 6 bit structures so the design of matching filter is very different). It's seldom a case of slap a generic FIR filter!!

 

There's also analysis on holo Spring in NOS mode:  https://dastereo.ru/t/fizika-i-lirika-r2r-matriczy/3916/114

[numlog]

There has been recent discussion on diyaudio about why might people prefer NOS (real 44.1kHz NOS that is), interesting stuff albeit a lot of it very technical and  hard to follow. One theory as I understand it is that a DAC will be able work better or more accurately at a lower sample rate as it has less work, but at the same time a DAC can only be as accurate as the input which at 44.1kHz is not a lot, it's all a set of compromises.

This NOS mode behaviour, if common to all the denafrips range, has virtually no difference compared to its OS mode, if you aren't externally oversampling to same rate as the interpolator then the DAC will be performing it's OS, . What rate is DAC running at internally? Only denafrips know. Could be 1.5Mhz, which isnt widely supported, or could be even higher, maybe it runs at multiple of input sample rate so internal OS in unavoidable.

The advantage of a real NOS DAC is it will remain internally NOS at any sample rate, 88.2k up to 1.5M, so you have you full external control.

Personally, I have experimented with adding OS to NOS DACs with HQP/XXHE and found lower sample rates to be more favourable in terms of sound quality.

Pure NOS 44.1kHz was also favourable in some circumstances, as GoldenOne says, it very much depends.

 

 

 

 

 

[PeterSt]

12 hours ago, GoldenOne said:

It'll be interesting to try nos on the phasure NOS1a if I can actually get it working. At the moment it's an H shaped paperweight

 

An H shaped paperweight which will byte you with its legs if you dare use it without reconstruction in advance of it.

You may like to play test signals at 16/44.1 though to prove its NOS-being. Note that XXHE will not be able to do that natively because it will always upsample 2x, never mind it's 16/44.1 still (it will just be 2x of the same).

 

12 hours ago, GoldenOne said:

I think technically though if upsampling to 705.6/768mhz then the dac filter itself makes very little difference regardless of design (so long as its not doing something completely crazy).

 

I would not count on that too much. :-) However, do notice your own context, which responds to other posts. Without context you are completely correct. In-context, upsampling to 705.6 without filtering will give very poor THD on to an NOS DAC. With proper filtering (and the same 16x upsampling) will be way way better. You have a real NOS/Filterless (R2R) at hand, so go ahead ...

 

 

[PeterSt]

1 minute ago, numlog said:

The advantage of a real NOS DAC is it will remain internally NOS at any sample rate, 88.2k up to 1.5M, so you have you full external control.

 

Exactly. And THE reason why the NOS1 was created. It came along with playback software ...

So Yes, I created the filtering first. The genuine NOS allowed me to judge it best.

Another thing - just saying - the NOS1 "chain" was made such, that at each sample rate it behaves / should behave electrically 100% the same. See the "2x of the same" I just mentioned for 16/44.1. That's a similar thing.

[PeterSt]

18 hours ago, idiot_savant said:

My take on this isn’t actually about a “gotcha”, denafrips are evil or rubbish - it’s just that there is so much speculation about how something may be better even though you can’t measure it, that I’d hope people would pick up on something that is clearly being measured. 

 

Hey Idiot, it is good to have you back ! I hope you stay for a while. Us morons seriously could use some idiot, once in a while.

[PeterSt]

From the Ares II manual :

 

 

At first I thought this was Chinese odd language. But this kept on bothering me, from where ever I read it :

 

 

This THD+N can not be mentioned without telling whether this is in OS Mode or in NOS mode. Plus, in NOS mode the figure would be way too good because it should be 0.04%.

 

Back to the first screenshot and the Chinese language :

 

NOS, as the name suggested, does not over-sampling to digital input data.

 

For me this is just too much emphasis on (the superfluous) "Input". But next up is:

- An FPGA which could do anything - mind you, this is way beyond the "input" as such (in our NOS1 situation it is right in front of the D/A);

- Nothing tells anywhere that this is about filterless, when it is about NOS (mentioned).

- It even is so that the filter responses are mentioned without telling that this can't work in NOS mode. However, it would be so that the filters can not be selected in NOS mode (dunno where I read that, but I guess @GoldenOne can confirm easily.

 

But now about this one:

 

 

It may again be about how to decompose this Chinese language, but what I (like to !) read is that there's analogue filtering in there. And otherwise I may not be knowing what an analogue "filter" as such is doing there for decoding DSD.

Anyway ...

 

A NOS DAC is only a genuine NOS DAC when /filterless is mentioned as well. I mean, nobody wants a NOS DAC with filtering behind it. Also not in advance of it. Thus, NOS/Filterless.

So as How I read it, the Ares II could be a NOS DAC all right, but it is not filterless ?

 

Which of course is different from people seeing linear interpolation in it. But this should be easy to check for literally that. IIRC Linear interpolation just does not want to work well above 11.05 ... (or a bit above that - I forgot).

Otherwise the FPGA could be doing things (wrongly). ... My own design of the NOS2 (which never saw real light) contained very complicated stuff around PCM and DSD at the same time, as the FPGA turned PCM into DSD-like behavior. I have no idea how it would have measured, but surely far from normal.

 

 

[The Computer Audiophile]

2 hours ago, idiot_savant said:

@The Computer Audiophile I actually think that @etane makes a good point here - you're talking about very heavily OS *system*, yet there are many references to "NOS". You agree (thankfully) that NOS at redbook rates is not a good idea, yet many people will say that is what NOS *is* - no processing, what's on the CD is presented "purely" ( which is, IMHO not ideal ). Does it say in the user manual or advertising that you need to buy a PC and 3rd party software to optimise it? 

 

I think we're moving steadily off-topic, but...

 

As for HQPlayer taking this DAC much farther, what do you mean? You can move it farther from the speakers? Better frequency response? Lower noise floor? Less distortion? As for horsepower, how much is enough? If you're talking about reconstruction filters, a modern FPGA will have "more than enough" to comfortably exceed any realistic requirements. For example, a Xilinx Artix7-200T, as used by e.g. the Chord MScaler according to the Xilinx datasheet can do 9.2GMAC/s sustained, a Mola Mola Tambaqui has something like 2.7GMAC/s, and because they doesn't have to run e.g. Windows, OSX or Linux, then you don't run into any of the classic problems of actually extracting performance from CPUs ( cache misses, task switches, IO ). Is there a penalty or benefit for moving higher-rate audio *into* the DAC rather than doing it internally? Is there a penalty for having a highly clocked CPU in the same location? Is the Denafrips OS implementation not ideal?

 

I'm not saying any approaches are necessarily better than other ones, and I'm certainly not talking about particular products, but as you yourself said, it's not black and white - bigger number and more processing does not guarantee better measurable performance or sound quality.

 

The job of the reconstruction filter is a holistic one - in a particular architecture, the digital OS and analogue filtering *should* be closely aligned and complement each other - there is, for example not much point in oversampling to very high rates if the analogue filtering cuts in early - you are literally just adding switching noise, and vice versa if the OS filter is gentler than the analogue filter is designed for, you can easily run into slew-rate limiting, things becoming unstable etc. Also, some companies may also choose to tweak their digital filter to compensate for deficiencies in the analogue filter ( e.g. the analogue filter is a bit droopy, so a bit of DSP can give a peak appropriately ) - this is obviously problematic if you separate the two.

 

Is an R2R NOS DAC still "pure" if you have a PC doing OS & noise shaping to help linearise it? 

 

your friendly neighbourhood idiot

@Miska is far more capable of talking about his product HQPlayer and the concepts than I am. 

[PeterSt]

One last thing which could also be important:

 

Someone hinted at some calibration which could be going on. Well, I don't know how much the Chinese steal from each other (or from me, for example), but the manual explicitly tells about the audio data being read "into the memory of the FPGA".

Why ?!?

Well, if you provided the FPGA with some pre-calibrated "data" for the way (waayyyy) too rough resistors in there (0.01%) then calibration (a mapping) would sure help. And I think one of the Holo versions really does this. So some Chinese will know about this.

 

And from their website:

 

The cost of high precision resistors is expensive, each resistor must be stringently measured and matched by machine with human intervention to cross-check.

 

I say: BS. Nice story, but BS. Matching them, by machine or otherwise, is useless because at one degree temperature change, all is a mess again. I can show you my attempt with uncountable bins, all in vain when proceeding with it a next day. Of they think they calibrated them. This is how the real-time calibration (of the Holo ?) does work. 

I must add that only when you know what you are doing, you can see the mess. So, measuring a 0.01% resistor at two digits of precision (make 3 digits of it, logically), is useless when you know that for a 24 bit DAC you'd need 0.000022% of precision (I do this from the top of my head and there could be a 0 too many).

[semente]

3 hours ago, idiot_savant said:

Is an R2R NOS DAC still "pure" if you have a PC doing OS & noise shaping to help linearise it?

 

If you are performing OS & noise shaping using software and a computer then ideally your DAC should be just a D/AC (no internal SRC, no filtering).

[idiot_savant]

@The Computer Audiophile - like I said, at this point, anyway, I'm not especially interested in details about implementations - I'm sure HQPlayer is very nice, my point is how *any* OS that is performed "outside of the box" *may* not be optimised for the analogue filter "inside the box", and my point remains that of semantic one - there seem to be multiple definitions of what NOS means to different people.

 

@PeterSt thanks for the welcome back, hopefully I didn't bore you too much last time with all the graphs! Point of pedantry - OS/NOS/filtering won't strictly speaking affect THD - The H stands for Harmonics - so e.g. a 6kHz tone will generate HD products @ 12kHz, 18kHz, 24kHz.... NOS artifacts will be at Nyquist +/- signal - so a 6kHz tone will generate "image" products @ 38.1kHz, 50.1kHz, 82.2kHz, 94.2kHz and so on for 44.1k samples, which will typically be much, much bigger than THD in amplitude terms.

 

@manueljenkin IMHO a *reconstruction* filter is required for redbook material, and this is in practical terms IMHO normally best implemented in a combined digital and analogue filter

 

@numlog - do you have a link to this discussion? 

 

your friendly neighbourhood idiot

 

 

 

[PeterSt]

34 minutes ago, idiot_savant said:

Point of pedantry - OS/NOS/filtering won't strictly speaking affect THD - The H stands for Harmonics - so e.g. a 6kHz tone will generate HD products @ 12kHz, 18kHz, 24kHz.... NOS artifacts will be at Nyquist +/- signal - so a 6kHz tone will generate "image" products @ 38.1kHz, 50.1kHz, 82.2kHz, 94.2kHz and so on for 44.1k samples, which will typically be much, much bigger than THD in amplitude terms.

 

16/44.1

 

With Filtering (16x) :

 

 

 

 

Without filtering (still 16x):

 

 

And exactly the same as native 16/44.1 which should be 0.04 (I told about that in the topic from 60 minutes or so ago).

 

So I am confident we talk passed each other ?

 

Edit: this is all @1KHz.

[idiot_savant]

@PeterSt so is this NOS doing a zero-order hold ( so sample-hold ) or a zero stuff? If it's a sample-hold, you might be clipping slightly, so if you reduce the level by say 3dB and see what's happening?

 

your friendly neighbourhood idiot

[PeterSt]

Here's one I did not even ever look at myself ...

 

This is 24 bits 44.1 filtered (16x):

 

 

And this is 24 bits, not filtered (16 x) :

 

 

I don't know what I had to expect from this, but not that.

[The Computer Audiophile]

35 minutes ago, idiot_savant said:

I'm not especially interested in details about implementations

That's understood. 

 

You listed tons of stuff that really shows you haven't looked much into the world of over sampling outside of DACs and I attempted to get someone involved to drop more information into this thread. Suggesting that current audio hardware is fully capable of all that's needed is a bit strange to me, but alas, if you aren't interested that's OK to.

[PeterSt]

4 minutes ago, idiot_savant said:

you might be clipping slightly, so if you reduce the level by say 3dB and see what's happening?

 

It's already at -3dBFS.

 

4 minutes ago, idiot_savant said:

is this NOS doing a zero-order hold ( so sample-hold ) or a zero stuff?

 

Sample-hold. Well, it's PCM1704.

[numlog]

@idiot_savant Full discussion can be found here

[PeterSt]

For further reference, this is what this DAC makes of 24/96 filtered 16x:

 

 

(we have this unfinished one coincidentally on the test bench, and later this will show as 0.00059% or so).

This is still better than "untreated" 24/96. With 24/192 there is no difference to be shown (plus it shows the same as the above plot).

[ecwl]

I have to admit I’m very confused by this discussion.

Granted, we have no idea what Denafrips Ares is doing in it’s implementation of NOS R2R playback.

But I see lots of people getting upset over NOS R2R not being truly R2R. But I’ve mentioned in other posts before that many DSD playback is not truly DSD either. And it seems that nobody gets very upset about that and vigorously defend it as “normal”.

To clarify, almost all “true” DSD playback is actually DSD with shift register with multiple elements. So technically, you’re not hearing “the original DSD” signal because of the multiple elements with shift register. And people tell me that’s normal and that’s still true DSD.

Let’s imagine Denafrips NOS also uses a shift register system for playback with say 16 R2R DACs all shifting at 16fs for 44.1kHz so you’re getting 16 R2R DACs all playing NOS except they’re all shifted by 705.6kHz = 1.4ms. To me, that’s still NOS (albeit with a type of filtering of some sort, which you’ll always need with NOS DACs, usually with an analog filter)

So how exactly is that different than a 2.822MHz DSD with 32-element shift register playback? Are we going to have people start demanding that their DACs be 1-element DSD DACs or else, their DSD DACs aren’t pure?

 

That said, maybe my understanding of NOS R2R DACs and DSD DACS are too elementary so perhaps somebody with more insights can explain to me the difference?