World's first NOS 24/384 filterless DAC

[PeterSt]

In between the lines ... the new setup is playing right now. And if anything, it's better again !
Another () part is to arrive later this week.

And then measuring and fine-tuning.

(Hans Theessink seemed a good one to start with)

[PeterSt]

Allright. No long stories this time, just graphs ...

Below applies to all graphs :

• Phasure NOS1 2 channel 8 x mono DAC, 4 layer board;
• Setup Single Ended (SE). Note that with Differential (Balanced) all figures will be 3dB better and around half the distortion percentage. So, worst case scenario is shown;
• The DAC running at 176400Hz, apart from the last two pictures, where the DAC runs at 192000Hz;
• All measurements include the normal (2m / 6') interlinks and connectors used. As before there's a volume pot in the chain;
• Each graph shows the worse harmonic distortion or aliasing peaks. IOW if there were worse outside the bands of the picture, I had said it;
• What you look at, is besides Total Harmonic Distortion + Noise (THD+N), the best sound, meaning : THD+N figures can be better, but sound will degrade.
• No filter is applied and frequency response is flat over the whole audio band within 0.1dB;
• All is unweighed, and all figures incorporate a bandpass of 22Hz - 22Khz (but see last few pictures as well);
• Impedance of the load is 100KOhm.

FFT 5 Hz

Is 5Hz important ? maybe not. But if it's in the "data" it shouldn't destroy other frequencies, and NOS DACs can do that !


FFT 16 Hz

16Hz for sure occurs in music. So it just should be right.


FFT 50 Hz

50Hz, and still looking good.

For your reference : this is how some Zanden looks like at 50Hz (courtesy Stereophile) :

Keep in mind though, this last picture has an output of -0dBFS, while my pictures have -40dBFS. So, it is about the difference in levels.
So, suppose you can see a -145dB in the NOS1 50Hz picture, this is actually -145 + 40 = 105dB down.
For the Zanden this is -58 + 0 = 58dB down.


FFT 200 Hz

Everything still over 98dB down.


FFT 1000 Hz

Here we have the worst case of all frequencies; 1000Hz is only just over 90dB down.


FFT 2000 Hz

Ah, 2000Hz and already looking better again. Note that to the right (oudside of the picture) nothing is "worse" than you see here. This counts for all, or otherwise I'd say it.


FFT 5000 Hz

Not much to say about this.


FFT 8000 Hz

Note that with a Balanced connection this will be just over 100dB down (as everything is).


FFT 10000 Hz

Keep in mind that the peaks you see at equal distances of the base tone (20000Hz vs. 10000Hz here) are harmonics. IOW, no spurs of aliasing (which are tones reflecting on the Nyquist mirror which would be 22050Hz for 44100 sample rate, but which mirror is here 88200 for the 176400 sample rate.


FFT 18000 Hz

This one is just for the excitement. So, *if* we'd be able to hear 18000Hz, then we wouldn't want distortion products in the more audible band (because of aliasing) coming along with it, right ?


FFT 24000 Hz

Aha, now it gets interesting (or not). This is 24000Hz and more than the normal Nyquist mirror. Well, let's not forget that if we run the DAC at 176400, we formally allow frequencies above 22050 Hz (in fact up to 88200Hz) ... may they be fake or not. Thus, technically this again shouldn't cause distortions of any kind.

FFT 44000 Hz

Whoops ... what happened ?!
Well, I told you, when things would get worse outside the picture, I'll show you and get it in;
What you see here is something which is close to "illegal"; I officially started playing 44000Hz which should be okay for the Nyquist mirror being at 96000Hz here (the DAC operates at 192000Hz now !), and what you see is that a mirror tone comes mightly close to the original (whatever is "mighty close" at 60000Hz vs. 44000Hz). It is 60dB down, way out of our hearing capabilities, but our amplifiers might be bothered by it !

But wait a minute ... would a 44000Hz tone be in the data - hence music ?
Yes, theoretically it can, because I am playing a 24/96 (or 24/192) here, and 96000/2 = 48000 which is always more than 44000, so it is legal; Is it in the data, then it will be processed by the DAC.

While this may be less good in the situation an amplifier gets disturbed by it, nothing is the matter when not, because no mirror products jump into the audible range. And as you can see, they don't even go pass the original tone (or otherwise everything is 98dB or more down).


FFT 94000 Hz

This one is merely for fun, and it proves that when the 48000 border is crossed while running at 192000Hz, mirror products occur in the audible range. It is still 60dB down, but mind you, this is audible !

Ok, not anything to worry about, because no "official" sound creating a 94000Hz tone at this level will exist. But just for understandings : when running 24/192 material, in theory (technically) it can be in there.


FFT 1000 Hz whole spectrumm

This is the 1000Hz from before, and as being the worse (see text there), I thought to show the whole spectrum up to 96KHz.
As promised, you can see that nowhere harmonic distortion is worst than you already saw from the zoomed picture.
The worst you see is at the right, being a mirror product from the 96000Hz mirror.


Well, concluded for now, I don't think you'll find an NOS measuring better than this one (and not to forget, all figures will be better again when Balanced connected). Not a 16/44.1, not a 18/96, not a 16/192 (they exist), and for sure not a 24/192. But I'll be happy to make a deep bow for any device showing better. And remember please : this is about sound in the first place. I think it is the most obvious that when NOS (ok, running at 176400 which is still NOS to the terms, but which is upsampling on the other side) can show these figures, OS (oversampling 64 times at least !) is nowhere.
This is what I wanted, and this now has been done.

However ...

When you got the grasp a bit from what is happening to which kind of distortions, *and* you incorporate that nature will expose frequencies to well above the normal Nyquist mirror (being 22050Hz for red book CD) -but - they can technically exist on 96Khz (or 88.2) material only, you should see that we're really out of trouble when the Nyquist mirror would be at 192000Hz.

Oh, I forgot to tell ... this DAC is a 384KHz DAC now ...


Maybe I am trying to be funny, but I think this is really needed to make the best job of it.
And to keep in mind : the DAC is not upsampling (no SRC in there anymore), and it is up to the software in front of it to make the best of this.

And yeah, I hear you first question : how to connect that 384Khz to the PC ?
Working on that ... working on that ...

Peter

[Gerard]

Man you really should get a reward for this!! 

[Telstar]

[AUDIODIDAKT]

Peter I'm totally speechles,

I can't even comprehend with the idea how that must sound.

QUOTE
And yeah, I hear you first question : how to connect that 384Khz to the PC ?
Working on that ... working on that ...

After all a phasure soundcard ??
"The Missing Link"

   

[AUDIODIDAKT]

Q about volume control:

No digital volume control possible during playback?

just found out is doesn't. so

[Telstar]

Q about volume control:

No digital volume control possible during playback?

just found out is doesn't. so

The DAC has a volume control done the right way.

[Ava12]

It will have a I2S input, for CDP or iPod?
Greetz Ava

[PeterSt]

Sure ...

[Per]

Hi Peter,

Wow - these graphs show impressive low THD+N even when measured at -20dB. (0dB should be even better measurewise - but I am not an engineer so my understanding is pure layman on this). Have you measured liniarity at low levels - for instance at -80dB which is how some Hi-Fi magazines often test CDs and DACs.

Thanks again for sharing your work and enthusiasm. Now we all anxious await to hear the news

"PHASURE DAC-1 is ready to ship - introductory price - $499.99"

Take care.

Best Regards,

Per Borgen

PS: Hmmm - I still wonder how you get that I2S signal out of the PC... ;-)

[Telstar]

PS: Hmmm - I still wonder how you get that I2S signal out of the PC... ;-)

With a modified soundcard, easier is one with via envy24 chipset. Easiest on esi juli@.

[PeterSt]

Hi Per !

Well, I must confess something ... and found that out just this morning ...

I never trusted that very low S/N data (which seems to be some -157dB at the pictures), and so never talked about that either - while I should have, because they are so unbelieveably low. But they are just too low to be real. And they aren't ...

I just found some stupid mistake in the analyser software which had an "Auto" checkbox ticked for auto-sensing the input level (which is not 2VRMS (the standard) in my case), but which only auto sensed the input level when UNticked. So, now the noise is shown at the proper level which elsewhere officially measures -130dB in this case (this case : the status of things, that changing in a daily basis for testing).

As you can see below this actually looks better, because no harmonics (HD) can be seen anymore (compare with the earlier shown 1000Hz picture), but this is only necause they are well down into the noise.
Nothing changed about the real noise of course, but we now know this is at -130dB and not at an impossible -157.

On a side note, something I didn't mention before, but which might be interesting by itself to some :
I may expect that a few people with high sensivity horn speakers float around in here (like mine are 115dB), and as you know it is always a problem to have it all enough hum and noise free, not to hear (mainly) the noise from a distance. This can be done allright, but I never encountered a situation (at others as well) that you could be with your head in the horn, and hear NOTHING. Well, with the now hopefully real figures shown below (but just the same with the unreal figures from before of course ) you can hear NOTHING. Just nothing. And to keep in mind : in my case this is with the main amps at full gain because I don't use a pre-amp.

About so good THD+N figures, I'm afraid they went worse because of the before wrong analyser setting. THD+N is measured against the noise (the +N), and when the analyaser thinks the noise is at -157dBFS (FS = digital input level), it measures the relative amplitude of -40dBFS (as in the picture below) against that perceived noise level. And if that perceived noise level is wrong, THD+N is also wrong ...
All means that (as it came out) where the noise was 28dB down too much, the THD+N figures were just over 5 times too good opposed to reality. This too doesn't mean that suddenly things are super bad, but for sure it isn't as shown before (below shown THD+N is an indication only, properly not correct -> see below too).

(0dB should be even better measurewise - but I am not an engineer so my understanding is pure layman on this)

At this moment (read : with the proper settings as how they are now) this is true. Before it was not, because of mis-interpretation by the analyser. The stupid thing is, as how it was before it should have been right; now it is (seems) not. This is related to the linearity for THD of the DAC at the specific input levels (digital output level to it), so I guess - at this moment - I am not sure how to interpret this correctly, or how to set the analyser's setting so that I see what I expect. For example, below picture shows an input to the DAC of -12dBFS, and at 0dBFS THD+N drops to 0.02%. This really shouldn't be so, unless at the attenuation the headroom of the 24 bits is not used *OR* it is used in the first place, while it really should not. So, relation to your quoted question above, for 16 bit source material, digital attenuation up to 48dB really won't affect sound quality (hence THD), while the analyser now shows it does (note that it is not XX playing here, it's the analyser's generator). So in the end something is not right now, and I can't see quickly how to fix it.

Peter

[PeterSt]

I think I know what is going on ...

I'm not sure how others do this, but how to measure THD when it is under the noise floor. Ok, it can be done (just look at the earlier pictures) but that is wrong, because obviously THD+Noise should measure the noise along with it. Well, this is exactly what the last picture does, and what to do about the fact that the harmonic distortion is just under that ...

It is not as simple as "allright, so THD+N is equal to N only";
If you imagine an amplifier behind this, and it gains by e.g. 24 dB, it will gain the fundamental, it will gain the harmonic distortion (now invisible) and it will also gain the noise. Because ot the latter, HD will remain inaudible, and noise comes first.
You could say that it can become quite noisy then, and theoretically spoken this is true, and nothing different from usual. This, however, depends on the initial output of the DAC (say, that fundamental) and if that is quite low to begin with, the output will be more noisier because of the additional gain needed to compensate for the lower output.

Although I don't know at this moment how to properly express the relation between output amplitude, HD which is under the noise floor and the noise itself, one thing I already know, and that is the noise being totally inaudible (with my current gain of 24dB creating an SPL of well over 90dB) and no music running. This by itself tells that with far of sufficient gain HD will remain inaudible just the same, but this isn't enough for me.

I'll be back on this later.
Peter

[PeterSt]

Hi all,

Before nobody is going to believe in this project anymore, it is alive as can be. And, as I am known to never give up on something which once came to my head as possible, this time too it seems to work out. At last ...

Many of you will know (mostly PM) that I'm hunting a Firewire connection. Actually this is right from the start, although at that time I though it could be done relatively easy. Well, as it turned out, not so (at all). And I really found all the reasons why no audiophile Firewire DACs exist. One reall really must me a complete fool to hunt this ... but since I am ...

Besides each and every potential option I investigated, one of them I started to work on from off December (2008), and only today I can anounce that one is going to work.
Think of phone calls, many many emails, attempts to let others do all the work, all seemed to end up in nothing, and the throughput time this takes is inmense (well, just 8 months at this moment !). And during that time there never has been a moment of not outstanding emails or answers, but one thing I know : If I had to start a next life, I sure wouldn't be an electrical engineer. What a world that is. To me it looks like the least efficient world existing on this globe. Emails without answers, one-per-day email conversations after months ending in nothing which could have ended in nothing on day 1, etc. etc. Nothing for me actually.
But I want this ...

So, after now two weeks of trying to communicate with the software department of a large chip manufacturer, and feeling that this would never end at making clear specs, and probably ending up with wrongly working stuff afterall, I now decided for that particular Firewire product to write the "settings" interface and driver myself. And yes, the nice thing of this product is that it comes with an SDK (System Development Kit) and DDK (Driver Development Kit), actually created to do the development, which in practice nobody does because it is a hell of a job. The good thing is, I already know that the things I want are possible, based on "them" saying they could do it (thus it can be done) and furthermore by buying ready products which are similar, virtually combine the features they have, and thus now have a good idea what the possibilities are to be combined into this one unique product : NOS Phasure1.

So, let's sum up some features this subject will bring, if all succeeds - which it of course will :

• Low incoming jitter
  
  For those keen on few jitter, the data connection (read : Firewire which is normal data, opposed to SPDIF which is audio) is right next to the DAC (think in terms of 1cm), and from there on it is I2S. Can't be better for incoming jitter, which (by now) is known to siple through right to the output, never mind what is done underway to prevent that.
  
  
• Accepts 24/192 over Firewire ... and USB.
  
  
• Accepts 24/384 over Firewire. Yeah yeah, cannot be. Well, watch it !
  
  Small remark : This will only be possible with XXHighEnd as playback sofware.
  
  
• Choice of clocking by the PC or by the DAC (Firewire part).
  
  
• Parallel to the above (thus actually unrelated), WordClock In, WordClock out.
  
  
• Galvanic separation with Firewire/USB from the PC.
  
  
• Automatic Sample Rate changes to-from all Sample Rates.
  
  
• Asynchronous connection for Firewire.
  
  Here a disclaimer is in order;
  These days rumours go that Firewire would be an asynchronous connection by itself. I don't think this is so, or anyway not of the "real asynchronous" some USB DACs these days can do. This is related to the earlier mentioned subject "Firewire can clock the data" (opposed to the PC), and on that matter it seems just to be already there. But, first of all this is not a common Firewire feature. For example, if my Fireface800 can do it, please let me know how - apart from some Master setting which IMHO does exactly nothing in this case (being the master over the PC). Anyway ...
  When this "Firewire can clock the data" is already just it (fully asynchronous) well, then it's there inherently. And, I can only learn this with the SDK at hand, which I don't have yet. But if it's not "it", I can create it myself, a bit depending on the possibilities of accessing the on board memory, which is needed for this. That too can only be seen when the SDK is at hand.
  When all is accessible as I now think it is, all is based on the good clock the DAC uses, and which deviates a handful PPM only. I won't go too technical, but with a very low deviation of the internal clock, the internal memory can serve as a buffer that may last an album or more, and no such thing as incoming jitter can exist. A PLL (jitter creating by itself) will not be there (is not needed) and the inherent jitter will be that few PPM I just mentioned.
  Theory for now, and I strive for making it practice.
  
  
• The best sound !
  
  This needs a small explanation of course;
  The explanation will come through as I intent, only to those knowing/using XXHighEnd, and the means I use to let it sound good;
  As many know/assume or expect, the way XXHighEnd does this is all the most indirect, and while the software outputs bitperfectly, it is just waiting what the rest of the chain (up to the D/A) will do with it. This time, the chain is mine, so to say. In other words, I will be able to apply the exact same as we are used from XXHighEnd, which undoubtedly will boost SQ further. You will not be able to recognize it or compare, unless improvements at this driver level come along, and you upgrade the driver or the firmware which again is a level closer to the DAC (that would be the Firewire chip).
  So yes, here is a crazy fool who will offer firmware upgrades for a DAC to improve its sound.
  
  And oh, before someone asks, since all is programmable in hardware to my (or our) creativity, it all may and up with filtering (like it now happens at the software side in XXHighEnd) right in there. I talked about this before I think, but then this was about theories and virtual processing hardware devices (like "some" DSP). Here, the DSP is right at my hands and the theories are now narrowed down to "program it". But, for later.
  
  

Although there will be a complete separate SPDIF input (and AES/EBU) this really will be all. Thus, no Toslink, because it doesn't fit the architecture (although it could be made) and will sound lousy anyway. At least that is what I think at this time.
I2S out as well as SPDIF out can be there, but I didn't decide about that yet.
Analogue out was and remains RCA + Balanced.

I have a disappointment too, and the word has to come out at some time anyway :
As you may have imagined, I only go for the very best, and if there is only one slight little degradation of sound opposed to the possibilities (which are sometimes theories to start with), I won't do it. And so it is :

Throughout the time I have been examining all the dozens of means the analogue stage can be comprised of, plus one.
That one is a finding of myself, and it is not official. No, it is officially wrong by the grace of no engineer understanding how it can work.
The number of times I went back and forth on one of the dozens of options mentioned, and "my own" is countless by now, but each and every time it sounds the best by far, and measures the best at the same time.
I found this one relatively long ago (must have been April), so you'd say "why bother further". Well, apart from that it is not official which for me is a good reason by itself, the DAC will have a very low output. Summarized I make use of the DAC's best operational level (read : where are its best distortion figures), combined with a fully passive approach without any downsides (no components holding back or squeezing anything).
Well, opposed to what is normal, in this case the output is 18dB less which may be too much for you. If you currently play at 12 o'clock or something, really nothing is the matter. But if you recognize to reach a quarter past often, you will find yourself without juice too soon.

That's it for now. Ok, maybe I should add that again the DAC won't be ready tomorrow. I can tell more after the first looks at the SDK and all, which also will be the time I can predict the price. And I assure you, it will be worth it !
Don't even think "tomorrow" to hear me back on this, because this world is as it is, and I don't expect it to change. I mean, it was day before yesterday I took the decision on buying the SDK/DDK (which is only another 10K, yahoo !) and although I asked in the morning how to proceed, I heard nothing yesterday, sent a reminder today (reminders are always necessary somehow), received an answer at 2pm that I had to provide a purchase order and a FedEx account number - which I don't have, asked about other means and whether that was necessary and ... heard nothing. Just as expected. So tomorrow morning a reminder is needed, and I'm sure another question will come from that, as I am sure it will take weeks before I receive something here. I don't know what's wrong with ALL of these guys. And you can't get mad at them.

So far,
Peter

[AUDIODIDAKT]

AMAZING,

Really like that clocking of memory thought, really smart Peter (You Crazy Fool ).
So just loading music data on the memory and start clocking from there?
And writing your own drivers in SDK/DDK (its that or waiting for aliens to land, I guess)
Btw 24/384 needs your own drivers anyway

 

Please take your time, you are already lightyears ahead of all out there.

Roy