World's first NOS 24/384 filterless DAC

[Telstar]

At using the digital attenuation the next thing which happens is that I "throw out" those bloodily created additional bits at Arc Prediction, with the result of AP not doing what it intended. That is, when I have to attenuate digitally 24dB (which is very normal for my modest 27dB gaining amps), I've thrown out 4 bits again ...

What about using 32 bit floating point or 64 bit (the latter would require 64bit os probably but who cares? we are in 2011 already)?

Anyway, waiting for the 8ch version with XO.

[DannyD]

So, both the digital volume (as implemented within XXHighEnd) as AP are as good as they are, when each of them is used individually. But not both at the same time.

So Peter, for the best possible SQ are you recommending that we don't use digital volume and AP together?

[PeterSt]

At using the digital attenuation the next thing which happens is that I "throw out" those bloodily created additional bits at Arc Prediction, with the result of AP not doing what it intended. That is, when I have to attenuate digitally 24dB (which is very normal for my modest 27dB gaining amps), I've thrown out 4 bits again ...

What about using 32 bit floating point or 64 bit (the latter would require 64bit os probably but who cares? we are in 2011 already)?

Anyway, waiting for the 8ch version with XO.

Telstar,

The bits are "lost" in the end result. The calculations in between are unrelated (for both the volume and AP) ...

Peter

[Telstar]

Telstar,

The bits are "lost" in the end result. The calculations in between are unrelated (for both the volume and AP) ...

Peter

Are you sure? Because if i recall correctly this trick is used in some dac chips like the sabre, 44 bits for the volume attenuation, it firsts adss all zeroes and then reduce the "volume".

But you are the software wiz If you tried and it cant work, pity, we'll have to spend more to have it passive.

[PeterSt]

So, both the digital volume (as implemented within XXHighEnd) as AP are as good as they are, when each of them is used individually. But not both at the same time.

So Peter, for the best possible SQ are you recommending that we don't use digital volume and AP together?

Dan - No ... It is way more complicated ...

I have just 12 hours of measuring and listening behind me ... The digital volume keeps on winning it from any means of analogue. And I use Arc Prediction only because any other means is worse anyway to begin with (like doing nothing hence native 16/44.1). But ... this is the case of the Phasure NOS1 ... Remember, this has no other means of "filtering".

Let me first emphasize that I am nit picking as hell here, and if there's only the slightest degradation of sound it's a no go (this is for me, but also for "you"). But this is so detailed that sure not every track shows it, and a 73rd track may exhibit "shouting" where it shouldn't; I still won't accept it.

What I learned from these 12 hours (soldering in and out stuff dozens of times), is dat there's a huge placebo in measuring here ...
The stupid volume (at its best setup) measures BETTER.

At first I thought it was because of all the wires (which also work as antennas) that I couldn't get the "stereo" pot to work as good as before with the mono pots, but this was caused by setting up the lot separately for a direct output and the volume output which I didn't before (things feed back, oscillate, whatever). So, once I found that out, I could easily see that with volume it showed better than without. Hmm ...
Since this actually is only about a first harmonic determining the "HD level" with a difference of 10dB (!), it came to me that the whole volume thing just filters, and it coincidentally filters 2KHz better than its fundamental of 1KHz.
Besides that, there's a whole slope of "noise" which is at least 6dB higher before 21KHz than after it. I never payed attention to it much, because it measured better NET. So, the before mentioned 10dB was even relative to that 6dB higher noise floor ... And all together you can well say that with volume is 16dB worse than without, while it showed 10dB better. Now, this is not true, because without the volume that first harmonic is 10dB higher, but THIS is "natural" and just caused by the "filterless" DAC and Arc Prediction not being better than it is (but also not good by itself of course). However, think of it ... false harmonics being there at more or less natural points, is always something different from the whole lot being attenuated differently at different frequencies. And this causes "shouting" ...

All 'n all, yes, you would day that loosing bits necessary for "upsampling" should stay there once nicely created. But the analogue volume destroys way more. Also, this is ver well measureable, once you know what to look for.

I have been partly placeboed by those measurements in combination with nice theories plus listening (for a too short time !!) to tracks where apparently it doesn't matter much.

Please keep in mind : this is how things matter *here*, at actually listening to real live instruments, and it doesn't take even "a bit" to destroy that all.
Man, all this time ...

[DannyD]

Dan - No

... would have been good enough!

[PeterSt]

Dan - No 

I kind of need a "reasoning" for myself. It may appear to be wrong in the future again, but as long as I can't "tell" these things just by heart, for me it's good to read back later what I found in certain situations. Besides, I guess It's just honest. I make mistakes, and I want to know why. And I think there's not much wrong with sharing that. Ok, it will be confusing. But that's only because I "share" too soon.

Anyway, I can only hope it's appreciated. Some of these things are really difficult, and at some stage you may be able to help out. It's like XX ... how can I do this whole thing all by myself ...

[PeterSt]

Are you sure? Because if i recall correctly this trick is used in some dac chips like the sabre, 44 bits for the volume attenuation, it firsts adss all zeroes and then reduce the "volume".

Telstar,

This is not about "accuracy" or anything;
"Filtering" - which for its largest part is about upsampling, creates a 8 times higher resolution in the time (sample rate as well as the amplitute (bit depth) domain). So, if a 20KHz tone is represented in the 16/44.1 domain it is a square; Upsample that 8 times, and you'll have a 8-stepped square - which looks a bit like a sine again. Attenuate that 48dB and 8 bits are lost and a square is again the result. It doesn't matter anymore at which amplitude-precise levels the 8 steps were calculated, because only 2 steps are used in the "net" result (24-8 = 16 bits = a square at 20KHz).

But as "explained" per my earlier post, although this is true, it's not more important than the "noise" an analogue volume implies, which affects the whole spectrum.

But as everybody is using analogue volumes (or preamps) - and is rather satisfied with that, I'm afraid it is too hard to explain.

Maybe Gerard can jump in at explaining better what the difference is between "I never heard a cymbal" (well, that's what he told me in advance) and what he heard here. And FYI : those cymbals weren't here either when you were around ! I myself am hardly able to imagine the situation without, but it's a difference you most probably can't imagine. But *if* they are there, things become so critical ... It's not only cymbals which exhibit their higher frequency information of course; it's about everything. And you better have lumpy bassy sound instead, *if* it's not completely right ... Otherwise it only hurts ...

[manisandher]

... it came to me that the whole volume thing just filters...

This makes sense... and kind of concurs with the graphs you posted over at CA.

I'm still interested in the 'DAC is' setting though. Would setting this to '32 bits' allow for digital attenuation up to 48dB without loosing any of the extra resolution created by AP?

Mani.

[PeterSt]

Good point Mani. But I don't think so. But there's something else to take into account :

I think you know I have talked about the "balance" between the sample rate and the bit depth, and that this balance actually is just allright with 16/44.1. So, according to my "calculations" each other doubling of the sample rate requires one additional bit. So, 88.2 needs 17 bits, and never 24. Similarly, 176.4 needs 18 bits, and (as in my case here) 352.8 needs 19 bits. That's all, and the remainder of the available bits imply fake accuracy. If this is correct, and I think it is , there's 5 bits in 24 bit DACs available for attenuation without loosing anything. This is 30dB.
One thing, the 32 bit float calculation of Arc Prediction doesn't take this into account, and just uses the 32 bits, which right after that are decimated to 24 bits (not even dithered). This theoretically implies rounding errors again, because it should use 19 bits ( for 352.8 ) in the first place.

So ... I am thinking of a next version of AP which just does that, therewith creating a better "balance". Whether this ever will be audible I don't know, but the importance for now is the available headroom of 5 bits hence 30dB. IOW, I'm actually *not* throwing out those "bloodily" created bits, because their assumed accuracy is not supported by the sample rate (which is way too low).

I can understand that nobody can get this now (without the underlaying background), but I intend to proove this by graphs from measurements later.
On this matter, and its importance regarding that "necessary calibrating" I mentioned by means of analogue volume, I'm now kind of deliberately using a 200W amp into my 115dB sensitive speakers, implying that digital attenuation of around 30dB in order to have my acceptable SPL of around 90dB (unlike Leif Christensen who is trying to blow his roof off with 110dB). Well, completely nothing wrong with it, that I can notice.
But, in order to show that really nothing happens to the sound, I'm afraid I have to make that "balanced" AP version first. On the other hand, the way this should be done (correctly) almost definitely will lead to just another means of "AP" I have in my mind for longer; for me it is more "interesting" to just develop that, which officially shouldn't be called "Arc Prediction" because it won't do that. What then ?

"Balanced Interpolation" seems to cover it nicely. Yep. And it will replace Linear Interpolation, which really is worth nothing.

Peter

[manisandher]

'Balanced Interpolation' - Peter, have you ever worked for a marketing company? You'd be good!

Let me ask the question that we all want to ask (but perhaps are too 'afraid', knowing how many things you've got going on right now): when can we get our hands on a final production-version NOS1?

Meanwhile, have you given a prototype to anyone to try? I'm convinced that one of the reasons XX sounds so good is the continous feedback you've received from 'us' users every time you've implemented something new. I know that this is so much easier to do in SW than HW, but having a cadre of NOS1 'testers' before it goes into production might prove a smart move. I, for one, would be happy to pay the full price for a NOS1 and receive a prototype for now, until the production version is available...

Mani.

[christoffe]

Let me ask the question that we all want to ask (but perhaps are too 'afraid', knowing how many things you've got going on right now): when can we get our hands on a final production-version NOS1?

Mani.

Hi Mani,

as long as we are coming  with additional requests all the time (like a volume control, 8 channel outputs) Peter will never come to an end.
My proposal: a plain NOS 24/192 DAC as intended by Peter, and we will get it on 12-24-2010.

Joachim

[PeterSt]

Dear Mani and everyone,

Although nobody says it, I realize it becomes harder and harder to give a trustworthy feeling about this whole thing. Especially when a question like "when" pops up again, I hardly dare to answer anymore. But I will try another time :

The last time I told someone (per PM) anything about this, I said "next Monday the cabinet will be handed to the manufacturer for production". That was two weeks ago, and it again didn't happen. But it WILL happen now next Monday.

So, again I dove into the stupid volume thing, and again I came to the conclusion it just will not work. So, I now quit this definitely, although there's still the option with some internal trimpots to calibrate (oh boy, there he goes *again* ). But this will be unrelated to any external pot; it will not be there. Period. Instead though, you will have the best sound possible.

I must honestly say that at forcing myself to really have it all the best as possible, along the way I found all the bits and pieces of detail that at least influence measurement results, and I was able to bring that down to 0.002% THD+N @ 1KHz (unweighted) for the audio band, or 0.004% for the range up to 96KHz (which is half the sample rate my analyser can cope with, while actually the DAC runs at 384).
Of course, I could have done this with the cabinets in production, but without the volume "objective" I just wouldn't have dove into it anymore.
That I (by now) think it still can be better again must appear next week, when some high quality resistors arrive - one being in a "path" where good resistors should be, but are not at this moment. Changing them on the existing (100) boards is no big deal (maybe a day of work).

When I let produce the 100 cabinets right away, I will have them (painted and all) within 4 weeks. But this is dangerous, because things have changed compared to the proto I have here, and I'll feel more confident if first one other proto cabinet prooves all is right. This may add another week or maybe two.
When the final cabinets are here, we will be able to stuff some two per day with the further ready parts, test and ship them.

I really hope (but also assume) this was the last time I had to estimate first shipment dates.
Now you better prepare ... haha

Regards,
Peter

[Marcin_gps]

Will the cabinet be the same as for the prototype? (shape)

[PeterSt]

Yes.