World's first NOS 24/384 filterless DAC

[Telstar]

1)  a delt-sigma "24"/192 DAC
2) a true 24/192 Multibit DAC... with oversampling switched in but without upsampling

The NOS1 is a 24/192* multibit dac.

He has a sabre buffalo so if he got the time he can do the measurements for it.

[*actually accepting higher than 192k]

[manisandher]

OK Telstar, that's cool.

But I'm still just a little confused. Peter, are you simply proving that if you upsample, you don't need to oversample... and if you oversample, you don't need to upsample???

This fits with my (limited) understanding, because oversampling and upsampling are achieving the same result, but at different points, no? FWIW, I've never liked upsampling, whether done in hardware or software... BUT, I've only ever used an OS DAC, where IMHO upsampling simply isn't necessary.

But your thinking on getting the output stage right is facinating...

Mani.

[Telstar]

There's never been a clear definition of oversampling and upsampling, somebody refers to the bitrate (which does nothing and its trascurable), others to integer vs floating point resampling, the latter being sort of universally accepted.
ex. 4x upsampling as Peter said = 44,1k x4 = 176,4k
ex 4,something oversampling: 44,1k => 192k

Integer upsampling is the best way to do it, because it creates the interpolated points at fixed points in space (sorry for my poor explanation -i'm abroad and i dont have my reference links).

Long time ago (like 6 months or more) i had a long discussion with Peter about this subject. the conclusion was that if OS has to be done, be it integer of at least 4x (8x would be ideal according to Peter, but there are other issues that are not solvable at present times, so he is settling for 4x -in software).

If you scavenge on diyaudio and audio asylum you would find some of my posts about possible ways to use the computer as transport, about doing the resampling IN SOFTWARE and why it is better than any SRC available to use in electronic devices, including state of the art algorithms such as the ones used by Lavry, dCS, Meitner, chord and so on.
If we consider the FLEXIBILITY of the computer medium for any calculation done in the digital domain, the choice becomes obvious.

I'm very excited by Peter (and the digital engineer) progress of making all this theory a reality, and for a reasonable price. Yes, i'm one that cannot afford 6 figures and that i would be skeptical spending even if i could.
I understand his happiness regarding the output stage, because it is THE most important part in defining the sound of a DAC. Now, I'm hoping that MY amplifiers will be fast enought to follow it... (the tube folks wont be happy)

The new digital board that Peter is waiting will mostly improve on jitter only and will be more crucial for high-res material, so i believe that the sound that he is hearing now is 99% of what he'll experience with the new board with 16/44 tracks.

I'm not sure which amount of information i can post publically at this stage, so i'll stop here. I'm really looking at the final production unit of this dac and i have no doubts that it will be the best that can be made to be connected to a computer transport, short of programming a custom multibit converter.

[manisandher]

When I have some time, I'm going to dust off my old 'Sampling Theory' texts from university. I'm absolutely convinced that 4x oversampling (with no upsampling) should theoretically achieve the same result as 4x upsampling (with no oversampling). (Let's just stick to integers, because anything else just complicates things further.]

The fact that we've had oversampling for over 20 years and upsampling for only 10 or so is because, in IMHO, the marketing people hadn't realised that they could scam consumers with upsampling, knowing full well that their DACs/CD players were oversampling anyway and just didn't need upsampling. Actually, to my ears, upsampling just messes things up with an OS DAC.

Now, if you have a NOS DAC, I can totally understand why you would want to upsample. But you'll just get the same result as an oversampled DAC, surely?

I'd love to be shown wrong 

Mani.

[PeterSt]

Well, before getting to some answers, it may be a good thing if I try to explain from my views how I see both Upsampling and Oversampling; Indeed there is no strict definition for this (not that I read anywhere), but for sure two very different phenomena are going on, and it is a good thing if we all know what we're talking about, when.

Oversampling

This is the thing a sigma-delta can't do without. Now watch the part "over", which may tell us something like "overdone". Too much. More than we actually want. However, the sigma-delta (1 bit for any sigma-delta DAC and which may be a part of the whole functional bit structure) NEEDS to oversample, and afaik the least will be 64 times, and the common-most is 256 times (I think 512 times will meet the SSD standard, so any DAC (in a cabinet) which tells to support SSD, will be oversampling a 512 times (relative to 44.1 of course).
Never to forget : the more oversampling takes place, the more squares will become sines. At 64 times oversampling, for the higher frequencies this will be a fact; the larger the square wave (low frequency) the more oversampling it takes to round the square into a sine.

Upsampling

See this is a nice and easy "upsample" from 44.1 to e.g. 88.2, or maybe one or two steps further. Now understand the difference with "oversampling" as how I describe it : there's nothing "over". It goes to where we need or want, and it is not more destructive than what we require from it.
So yes, a 22050 Hz original full square, will become a 2 stepped square when 44.1 is upsampled to 88.2 (but it will be far from a sine yet).

Thus, it is the whole point that any not true 24 bit DAC, needs oversampling to operate, and while this process shifts the Nyquist frequency nicely as I described earier (like 256 times out of the audio band) it destroys the squares and thus harmonics. Besides that the heavy oversampling creates HF noise which must be filtered out again, but that is another matter. Otoh, this never allows the OS DAC to be "filterless", and those filters by itself destory sound again (roll off in the highs).

So Mani, if we'd "oversample" 4 times and compare that to "upsample" 4 times, of coure it is the same. But an Oversampling DAC is called like that because its physcis require the (heavy) oversampling as I described it, and it won't work with 4 times only.

Because an oversampling DAC just measures good because it *can't* have the false aliasing, one may tend to believe it is/sounds better. And of course, because of the inherently not present aliasing, it really is. But in the mean time all transients were turned into soft sines.

Now, what happens if we - preceeding thee DAC - upsample in software ? the only thing -besides that the software may use another "upsampling" (I say this on purpose here) algorithm than the OS DAC will use- is that we feed the DAC with e.g. 176.4, and now that will be oversampled a 256 times once more.
(note that there may be a difference here with SRC's involved, who for an NOS DAC may go from 176.4 to 192 instead of going from 44.1 to 192, once fed with 176.4 -> this is also related to the maximum sample rate the DAC can handle. Thus upsampling yourself in software may incur for different results in the SRC opposed to doing nothing in software (already because of the different algorithms used)).


Once an NOS DAC is used, we must realize that various "tricks" can be in order, like playing a 176.4 (or 192) file which may not need any treatment at all, because no aliasing will occur in the audible band. So, this time we don't need to upsample at all, and because it's NOS it wasn't oversampled also. This is IMO why it is so important that the DAC can inherently do 24/192 in NOS, assuming that more and more hirez albums will emerge, and NOS playing at that same rate without any conversions, will be theoretically the best one can think of, so better than 44.1 for that reason (no aliasing) alone. That you are listening to a higher resolution which may come to you as more refined ... well ... personally I have my doubts that *that* is audible. But the lacking distortion just is.
To this I can add that the virtues I heard before from hirez files (and which I heard since the NOS1 only), now seem to have disappeared again, because 44.1 now sounds as refined as the hirez files. Not because 4 x upsampling makes the sample steps more fine, but because the distortion is not there.

It is all rather complicated and nothing is real science as long as science seems to tell that OS is good. Example : we have 44.1 which is oversampled 256 times and we have a hirez file of 96 which is 256 times oversampled. We do realize that the resulting resolution is a matter ogf difference of 256 vs 257, right ? and that one of that 256/257 was a real higher not-fake step. Do we think we can hear that ?
Or another example : we have 44.1 vs. 96 and this time we use an NOS DAC. All the analogue stuff is so much slower that the "huge" steps of 44.1 can't be followed anyway, and the approx. 2 times smaller steps can .... 't be folowed equally (inthe time domain the steps are as large). So can we ever perceive a difference ? no, not with slow analogue. But then again, the 44.1 contains all the aliasing sh*t, while 96 does not. The chance is fairly high that we perceive *that* for a difference, right ?

But still no science ... just some reasoning.

Back to business : NOS 24/192 is nice for playing hirez material, but it is mandatory for Redbook !
At least that is what I say.

[PeterSt]

... filtering options will be by means of (player) software, and uploadable to the DAC in a later stage.

So, my understanding is that XXHE and the hardware will work together seamlessly. But will the DAC be able to work with other software players and/or digital inputs?

That depends ...

There will be options for more DACs (this can be in several arrangements, but I will tell about that later in a later stadium (weeks)), and this is meant for multichannel XOver.
Now, this will be done by an on board processor, but, the convolver files can be uploaded to the DAC. So, you make them how you want them, and the DAC will internally use them. And this is not what I was telling about yesterday of course ...

Yesterday I talked about the filter protecting the amplifiers ... if needed at all. Now, it is the idea that these filters will be applied in software (XXHighEnd), while the XOver "filters" are run by the DAC as I just explained. In either case, the both filter types will not be dealt with by the same instance.
Now, when the XOver is just not there, the on board processor can deal with the normal DAC filter, and the software does nothing in that case.

In either case all has a dynamic setup, because it can always be changed when you like.

Assuming that the software to upload the filter to the DAC is outside of XXHighEnd (and why not), the DAC will be running independently. However, if both filters are used at the same time, the software must take one part, and then the (XX) software is additionally needed to run the DAC properly.

I must be careful what I'm promising, because there is more I didn't talk about yet, and that "more" most probably will need software too, and *that* software can only be XXHighEnd or a redundant derival of it. I don't know about this yet, but I hope in a couple of weeks I do.

Peter

[PeterSt]

The point here is, it is not difficult at all to "create" detail, but usually this occurs in the higher frequencies, and most often it appears to be fake (harmonics !) afterwards.

... where the [delta-sigma] OS DAC can't do without the heavy oversampling because otherwise it can't operate, the NOS DAC can upsample to any rate we think is necessary if it can do that in the first place. And this is where 24/192 comes in as an important phenomenon, because the Good DAC just can't do that. So, now you also know why I did not show any picture of the Good DAC without all that distortion, because it just can't do the upsampling ... (mind you, some DAC chips can, but they are not 24 bits).

Peter, I would have loved to have seen the following included in your analysis:

1)  a delt-sigma "24"/192 DAC
2) a true 24/192 Multibit DAC... with oversampling switched in but without upsampling

My (limited) understanding is that there should be no difference between 2) and an upsampled NOS DAC.(If you'd like to borrow my D70, I'm sure we could arrange that. Is this a true 24 bit DAC?)

I hope this is sufficiently clear now, since my earlier post from a few hours back ?
But what you will see from the delta-sigma DAC (fully delta-sigma or not) is that it measures even better for the FFT graphs I showed. However, other graphs will show the roll off near 20KHz, which will not be there with the NOS DAC when measured properly (this is not even easy).
N.b.: My Buffalo is a kind of out of order right now, but then I have the Fireface. Note that this is not about how it sounds (like the FF doesn't "sound" at all) but how it behaves. I must see when I have some time (better : want to spend it), because in between the jobs I now rather want to finish the "analysis" stuff in XXHighEnd so I can at last upgrade again, in the mean time solving a few bugs which came around lately (and Mani, your 32 bit file support won't be in there yet ).

Peter

[PeterSt]

The new digital board that Peter is waiting will mostly improve on jitter only and will be more crucial for high-res material, so i believe that the sound that he is hearing now is 99% of what he'll experience with the new board with 16/44 tracks.

I'm not sure which amount of information i can post publically at this stage, so i'll stop here.

Haha, thanks. At this stage there are no secrets really, but not everything might come true. So, indeed it is good to stop "there".

Apart from better jitter specs (of which I don't think they are important at all at these already OK levels), the noise level will go down by 3dB at least, but possibly around 6dB. This is partly because of the now 4 layer board, and partly because of something else, yet to tell. But in the mean time something very different is going on, which is of great importance for the analogue stage. This actually is the reason that posting all those 60 screenshots was without real sense, because they will be overruled largely. Well, that is the expectation, when digital and analogue meet.

So many things are going on once that big hurdle has been taken. Now, the hurdle has been taken allright, but behind it may be the water pool. The reccie (good old rally term) showed all was dry though. But as you know it can always start raining unexpectedly ...

[manisandher]

Well, before getting to some answers, it may be a good thing if I try to explain from my views how I see both Upsampling and Oversampling; Indeed there is no strict definition for this (not that I read anywhere), but for sure two very different phenomena are going on, and it is a good thing if we all know what we're talking about, when.

Peter, thanks for taking the time to expound your thoughts - very helpful.

I'm not sure if I'm knowledgeable enough to agree or disagree, but one thing is for sure; I totally agree that a NOS 24/192 would be nice for playing 24/176.4 and 24/192 material - I don't see any need to manipulate the data in any way, shape or form at these resolutions.

As for 16/44.1, I've always felt that the 'old' multibit DACs/CD players with 8x oversampling were absolutely fine (with no upsampling required). But I can see that 4x upsampling with a NOS 24/192 would work nicely...

Mani.

[PeterSt]

That "8 x oversampling" would be the same as what I today call upsampling. So, with a multibit that can be done (at only 8 times) while a not multi-bit can only over do it.
What applied in the old days, still applies, weren't it that the old days chips (like 1541) just don't exist anymore, which for those 16 bit chips is even a good reason (at screaming for 24 bit these days).

So ... since the 24 bit chips also were killed, and aparantly the world thinks it needs to go single bit ... let's see what really is better (oh, I know hehe).

[Josef]

>I'm hoping that MY amplifiers will be fast enought to follow it... (the tube folks wont be happy)

Are you saying that 'tube folks' should not even try this DAC?

[PeterSt]

Hi Josef,

I'm not saying anything. In fact I'd say that a tube is super fast. This is different from possible "sluggyness" in the bass area, which IMO is unrelated to the speed I'm talking about in the DAC case. But maybe I don't know everything ...

[Telstar]

>I'm hoping that MY amplifiers will be fast enought to follow it... (the tube folks wont be happy)

Are you saying that 'tube folks' should not even try this DAC?

I'm afraid that some SET amplifiers could not be fast enough, yeah. But maybe i'm wrong.
My amp is a SET of jfet (the lovely firstwatt f3).

[leifchristensen]

is it possible to have "an executive summary" of the present status: I don´t have time (or patience) to read all these replies
thxs in advance
best
Leif

[PeterSt]

Hi Leif - This summarizes it a bit I think : http://www.phasure.com/index.php?topic=642.msg6497#msg6497

Now, a few days later, I can add that all your Tangerine Dreams, Beach Boys, Black Sabbaths and even Stealers Wheel sound as from today. But then without the compression of course. It is really weird ...

In one - two weeks I will present more definite results and all the measurements.

Peter