Worlds Greatest DAC and what it does differently

[davide256]

9 hours ago, beerandmusic said:

 

This effort has nothing to do with music in itself, it has do with a project of mine to decide if i want to consider a PCM dac.

My current belief is that I can be very content with PCM, whereas I used to believe that i would not ever purchase anything that did not support DSD...so you just have to have patience to see where I am going.  My guess (but i may be wrong) is that most audio engineers already accept PCM is more than sufficient and capable of playing the full audio spectrum equal to quad DSD "if done properly".  I still do not understand even the smallest fraction of what most here do, but i "think" i know more than I did "yesterday".  Anyway, to me, and I started the thread, it is not pointless...it is a starting point for a debate to see if there can be a consensus.  If there can't be a starting point, there is no reason to dig deeper.

I have found that DSD upsampling matters with CD quality choral pieces; there is normally a fuzziness associated with massed voices that up sampling removes, allows inner

counterpoint singing to be heard clearly vs covered up. I've heard this also with better PCM gear but the price tag becomes unaffordably expensive.

 

 

[davide256]

18 minutes ago, barrows said:

My feeling is that there is some "magic" with DSD 256 (and perhaps above).  But of course this is not really "magic" at all.  High sample rates, including high rates of PCM-like quality allow for relatively simple actual conversion paths.  This is how most Sigma Delta style DACs work.  Tthese high rates also allow for even discrete (non chip based) converters, which give designers additional freedom (if they are clever enough) to develop a unique converter topology which may offer advantages over standard chips.

For chips, when one looks at, say, ESS-it oversamples to a very high rate (similar to DSD rates) and 5, 6, or 7 bits, and then converts to analog from this very high rate.  dCS uses a similar approach, but they do the final conversion into analog using a discrete converter.  Chord, again uses a similar approach, again using a (similar to dCS) discrete final converter stage.

Jussi's (AKA miska) DSC-1 DAC design, now adopted in various different iterations by Holo Audio, Denafrips, T+A, among others, is a DSD based discrete converter stage designed to be fed by a very high rate single bit signal, generally DSD 256 and above.

Mola Mola, in the excellent Tambaqui DAC, uses a variation on the discrete converter theme as well, with the final converter stage being a multi-element switch and resistor set up, similar to DSC-1.  Same with Playback Designs, and I suppose Emm Labs...

I am listening as I type to Bricasti M3 DAC, which uses a different, discrete converter stage design of their own, essentially also based on a high speed switch and filter, for single bit DSD input only.

 

Most of these different DACs sound different, but I have experience now with 4 different single bit based DACs, all of which have used some variation on a discrete converter stage.  All of them have seemed to share, in differing degrees, a quality of conversion which sounds more natural, and easy going to me.  I do not mean that they are rolled off, or "soft" sounding, but that they sound more like music, and less like an electronic representation thereof.

 

There are a lot of really good DACs out there these days, but for me, I do seem to prefer the DAC which use discrete conversion approaches, with high rate DSD input.  This can be done on a chip as well, as with AKM DAC chips' "direct DSD" path option.  But, when a designer goes discrete, they have control of the performance, and they may be able to do better than what is available on a generic DAC chip (or not, as the case may be).  

Are there source side benefits? I'm told the Mojo doesn't really do DSD, converts to PCM but using Stylus/Roon sounds cleaner to me with DSD 128/256 upsampling than max PCM rate.

[davide256]

14 minutes ago, barrows said:

I am not sure what you are asking....

Chord DACs, except DAVE, all convert incoming DSD to PCM 705.6 before then upsampling to very high (MHz rates) at, as I recall 5 bits.  the this data goes to their discrete conversion stage.  I would not send dSD rates to a Chord DAC, as it will be down sampled to PCM 705.6 anyway.  If you want to oversample in software with a Chord DAC (again excepting DAVE), I would send PCM 705.6 to the DAC, as this will bypass some of the DACs onboard processing.  Note that this what the mScaler does...

The question is

 

assuming  adequate processor power and memory with competent software, which will be less impacted  as USB source by issues in the source hardware and software,

1 bit xMhz DSD output or 16/24 bit 705khz PCM output?

[davide256]

1 hour ago, beerandmusic said:

 

what is an example cd i can test what you are speaking about?  Can you share a cd, track and time that you are talking about? 

John Williams soundtrack for the movie "Rosewood", 2nd track "Look down Lord". Massed voices for the entire track have a fuzz to the sound without upsampling

and an interspersed baritone counterpoint becomes vague and hard to distinguish without up sampling

[davide256]

On 6/12/2020 at 5:45 PM, Miska said:

 

It could be inside the DAC, or outside of the DAC. I'm doing it in player software.

 

And yes, I'm talking about analog output of the DAC too. Modern DACs heavily rely on DSP algorithms, to the extent that DSP algorithms largely define base performance of the DAC.

 

Example; here you can see Holo Audio Spring 2 DAC running with 1 kHz -120 dBFS input, TPDF dithered to 24-bit, at 1.4112M sampling rate:

 

And here the same, but data noise-shaped to 20-bit to match linear range of the R2R ladder:

 

You can see that the second one is at least 20 dB (10x) cleaner! In fact it is much more than 10x cleaner, while having actually more dynamic range, but limited by analog noise floor.

 

Another view, here is same DAC running sweep with built-in oversampling (8x) to 352.8k:

 

You can see that the first image around 352.8k is at -60 dB meaning about 10-bit equivalent reconstruction accuracy. And actually four image pair bands visible. At 192k sampling rate it would be much worse.

 

Here's the same, but at 1.4112M sampling rate with oversampling done in player software:

 

You can see there there are no images at least clearly visible at multiples of 1.4112M sampling rate. The difference here is more than 40 dB (100x)!

 

And here's the same again at DSD512:

 

Here you can see that within about 100 dB range, 1.4112M PCM through R2R section and DSD through the DSD section produce similar image levels on this DAC, so we can reach at least decent RedBook reconstruction.

 

In all cases, source file is 44.1k sampling rate. We could also look into narrower spectrum to see differences in there.

 

 

This is just a very brief example what DSP can do to operate the conversion section to perform well.

 

Does the width of the blue band determine imaging stability?  seems to me the wider the band, the greater the deviance in volume reproduction  for overtones.