Thanks for your reply and the website link!
Very much looking forward to your review of the D18 :)
A good DAC needs upsampling.
Many people think that a DAC is supposed to reconstruct the stepped 16 bit 44.1Khz waveform and then simply filter off the sharp edges so everything above half the sampling frequency is removed.
However this is wrong! This is not at all how the sampling theorem works.
Yet our NOS DACs work this way.
Here is how reconstructing the original sample waveform should work:
Note also that it isn't actually a "filter" and it doesn't "ring".
The thing that's refered to as ringing is actually caused by the low pass filter used when encoding the data.
A good computer oversampler reconstructs the digitally encoded waveform by far perfect enough (with inaudible THD+noise).
And I'll say it again, to use a "minimal phase oversampling filter" is nonsense. Linear phase high bandwidth oversampling is the only technically correct reconstruction of the waveform.
To not oversample as in a NOS dac with 44.1 Khz data as described in the top of this post will give very audible (and measurable) errors (which we often perceive as warm btw, since specific frequency bands in especially the trebble are much lower in volume because of this error)
Contrary to popular belief these errors are not only for certain frequencies, but also for amplitude / impuls behaviour (these are not seperable in the wave, they come from one and the same thing).
So the funny thing is where people always say that NOS DACs etc have better impuls behaviour it's actually the other way around! :)
You need correct high tones to make a correct amplitude curve.
To make clear in a short way the error of NOS DACs btw: take 44.1 Khz sampling frequency. Now play a 22.004999 Khz sine wave (this is slightly under the Nyquist frequency so it can be reproduced.)
Depending on where in the phase of the sine wave the sampling occurs, it can mean that actual samples can be recorded at either the very top and bottom of the sine wave form, or exactly at the 0dB crosspoints (so the samples record actually nothing! Silence!) And the sine wave then very slowly changes phase to the sampling frequency.
To simply output the bit values (as a NOS DAC does) and then simply filter above the Nyquist, gives ACTUAL SILENCE where the wave is in such phase that the samples record nothing.
This is a severe error. It goes like a riple all the way through the entire frequency band till the lowest frequencies. However the effect is much stronger in the treble than in the mids than in the bass. At the highest frequencies the error can be 100% (silence when there should be a full scale wave), at half the Nyquist the error is maximally -6dB or -3dB (can't remember out of the top of my head), and at again half that frequency it is very small (4 times as small perhaps, possibly even smaller, again don't remember the math and too lazy to look it up :)
Now the sampling theory works differently. It says one can reconstruct the original waveform by making the "slowest angle" waveform that runs through the connected dots (it can do so because it assumes the sample values represent only a waveform that has no content whatsoever above the Nyquist)
Therefore one CAN actually recreate the waveform in the example above that is just a fraction below the Nyquist. So no silence where there should be a wave, but we get the actual wave back (even though a large timespan of the samples were taken at the 0dB crossing. All the errors disappear. (this also helps make clear the ringing that must happen when we bandlimit (filter) the original wave during the analogue to digital conversion, If the frequency is high already, it can't have a fast volume envelope as the volume envelope modifies the waveforms angle speed, this means it makes part of the wave higher in frequency and this is removed when this leads to a higher frequency than the nyquist. So to limit frequency is to limit impulse, ringing is an integral part of bandlimiting, no way around it, and bandlimiting is an integral part of the sampling theorem. It's not bad, it's high in frequency I don't know if we can hear it, but a DAC sure as hell can't fix it should it need fixing without doing other things completely totally wrong.
Sorry I had to get that off my chest it had been bothering me for a long time what I read about these issues in popular hifi talk.
Now so we need upsampling to give an honest reproduction of the waveform.
44.1 vs upsampled to 96 will make a big difference. 44.1 192 is slightly better than 96, audible on a good system but subtle if I'm correct.
384 very slightly better than 192, don't know if it's audible.
(edit: btw you can also upsample from 44.1 to 88.2 or 176.4 or 352.8, it will be lighter on the computer and give less errors with not so great upsamplers)
As for computer upsampling vs an asynchronous sample rate converter onboard a DAC:
The asynchronous here does not mean a good thing (like it does in USB data transfer), it means a very bad thing. It means there is no "lock" between the original and upsampled frequencies, it has errors in upsampling not only in bit depth rounding of values but also in timing.
However, it still reconstructs the waveform better than if we were to not upsample at all (only it gives us new errors)
The computer doesn't have these errors and performs "synchronous" SRC where the 2 frequencies are locked (well of course it has some errors, long story to explain, but these are so low for a good upsampler they don't matter)
So computer is better than upsampler on DAC. It's as sample as that, computer does it better.
I do belief that there is some other reasons why DACs use upsamplers, there are some tricks with them to reduce or modify jitter or certain types of jitter, I don't know the specifics of this.
In the Sabre DAC itself there is already ASRC being done (in a completely different way that as described above, as it's a very high frequency DAC it needs to do this) and also jitter reduction.
If the sabre is fed with a low jitter bit perfect perfect signal after the computer upsampler, this is logically the best possible way to feed the SABRE.
This is possible now. Take for instance the Audiophileo2 which has a claimed jitter of only 2.6ps rms (compared to about 30ps rms jitter beeing fed in the Anedio D1 to the Sabre DAC after USB conversion and it's PLL's) and does 24 bit 192Khz.
This should make all the difference in the case of the D18, and it could possibly perform better then the Anedio D1 etc when used in this way.
Please try it out, I don't have my D18 yet. I think you could be going for words like "incredibly neutral and revealing" instead of "warm" etc when you do :)
EDIT: What I wrote is not correct. The Sabre DAC in the D18 does it's own "upsampling" so it does reconstruct the waveform correctly without additional user upsampling. The only difference with using computer upsampling will be the slightly increased quality of good computer upsampling.
Sorry for being stupid. Thought I'd edit the related posts for the people who brows this thread later.
Edited by slackman - 12/25/11 at 5:44am