Which DACs bypass digital filtering?

[the_bat]

Sorry, missed the 176/192 requirement. The AN’s only go to 96k. 

[bilboda]

Sonnet Digital Audio Morpheus up to 384k. Who needs DSD with a dac like this?

[Kal Rubinson]

1 hour ago, bilboda said:

Sonnet Digital Audio Morpheus up to 384k. Who needs DSD with a dac like this?

Those of us who play DSD files.

[bilboda]

I play DSD, downsampled to pcm with HQplayer....sounds better than the upsampling to DSD and native DSD that I had going on before. The dac just outresolves the AKM4497 based dac I used for upsampling.

 

[semente]

You know what they say, there's no accounting for taste.

[bilboda]

apparently not

[barrows]

12 minutes ago, bilboda said:

apparently not

This from Sonnet's description sounds like complete BS to me:

 

" The result of this process is an extremely high linearity, right down to -140 dB, which gives our products a realistic 24 bit dynamic range.  It is musical, honors the Non-oversampling principle and simultaneously incorporating the technological progress that has been made over the years."

 

Where are the actual measrement results for this DAC?

[bilboda]

lol, I dunno, not sure I value your doubt over the designers pedigree...and I am not sure that measurements at this extremely inaudible level will tell you how it sounds either...asr aficionado?

 

[barrows]

10 minutes ago, bilboda said:

lol, I dunno, not sure I value your doubt over the designers pedigree...and I am not sure that measurements at this extremely inaudible level will tell you how it sounds either...asr aficionado?

 

Haha, not at all.  But I am skeptical about any company which prints what is clearly inaccurate claims about the linearity of their product.  No DAC in the world actually achieves 24 bit resolution at its output.

As to how it sounds subjectively, yes, for sure, it could sound wonderful, it is just that false claims like that make me very uncomfortable. 

[bilboda]

We could go way off topic, dark matter?, fancy cables, optical network? No measurements there either, I'm sure, and who are u to say what's impossible? Suffice it to say that his dac meets  op's requirements and I think it's terrific.

[mansr]

4 minutes ago, barrows said:

Haha, not at all.  But I am skeptical about any company which prints what is clearly inaccurate claims about the linearity of their product.  No DAC in the world actually achieves 24 bit resolution at its output.

As to how it sounds subjectively, yes, for sure, it could sound wonderful, it is just that false claims like that make me very uncomfortable. 

It is possible in some DACs (though probably not that one), with lots of averaging, to recover linearity down to the 24th bit. The usable dynamic range isn't anywhere close to that, of course. And don't get me started on "the non-oversampling principle."

[barrows]

10 minutes ago, bilboda said:

We could go way off topic, dark matter?, fancy cables, optical network? No measurements there either, I'm sure, and who are u to say what's impossible? Suffice it to say that his dac meets  op's requirements and I think it's terrific.

There is no DAC in the world which has 24 linearity at its outputs, period.  Find the measurements which say different and I will eat my socks!

Like I said, that does not mean this DAC does not sound wonderful, but it does mean that the company is misrepresenting its actual performance on their website, which makes me wonder about them.

[Account Closed]

16 minutes ago, mansr said:

It is possible in some DACs (though probably not that one), with lots of averaging, to recover linearity down to the 24th bit. The usable dynamic range isn't anywhere close to that, of course. And don't get me started on "the non-oversampling principle."

The useful dynamic range is probably 20 bits or less. I think the best ever measured was 21 bits but I can't remember which DAC that was. It is very hard to get much better than 20 bits just because of the inherent noise of moving electrons around.

[barrows]

4 minutes ago, bobflood said:

The useful dynamic range is probably 20 bits or less. I think the best ever measured was 21 bits but I can't remember which DAC that was. It is very hard to get much better than 20 bits just because of the inherent noise of moving electrons around.

Yes, exactly.  Which is why it makes me cringe when this company claims 24 bit resolution.  

[fas42]

Interesting paper that describes the "state of the art", on the ADC side; which can always do a touch better than DACs, and what one has to deal with ... https://www.researchgate.net/profile/Nikolai_Beev/publication/325285614_Analog-to-digital_conversion_beyond_20_bits_Applications_architectures_state_of_the_art_limitations_and_future_prospects/links/5b680228299bf1b9303caac9/Analog-to-digital-conversion-beyond-20-bits-Applications-architectures-state-of-the-art-limitations-and-future-prospects.pdf?origin=publication_detail

[Jud]

2 hours ago, fas42 said:

Interesting paper that describes the "state of the art", on the ADC side; which can always do a touch better than DACs, and what one has to deal with ... https://www.researchgate.net/profile/Nikolai_Beev/publication/325285614_Analog-to-digital_conversion_beyond_20_bits_Applications_architectures_state_of_the_art_limitations_and_future_prospects/links/5b680228299bf1b9303caac9/Analog-to-digital-conversion-beyond-20-bits-Applications-architectures-state-of-the-art-limitations-and-future-prospects.pdf?origin=publication_detail

 

"Furthermore, even though ADCs with nominal resolution of 24 bits are prolific, few parts can actually achieve SNR-equivalent resolution beyond 24 bits." It's not mentioned in the paper whether any such are in use in the audio field.

 

The problems with getting to such low levels of resolution are pretty fundamental. First there's thermal noise; and even if you wanted to invent a cryogenic ADC or DAC, there's shot (quantum) noise.

[bilboda]

This dac splits the bits in 2, 12 bits going to 1 set of ladders 12 bits to the other. The least bits are raised up to avoid thermal noise, afterwards they are recombined with a logarithm in use to get 24 bits. Read the info for your self and draw you own conclusions. It's beyond me. I can understand the presentation but can't evaluate it, This technology was established at Metrum by Cees and improved on with the new design. I don't think he'd use a method that wasn't sound and didn't sound good too.

 

[Jud]

10 minutes ago, bilboda said:

12 bits going to 1 set of ladders 12 bits to the other. The least bits are raised up to avoid thermal noise, afterwards they are recombined with a logarithm in use to get 24 bits.

 

You mean algorithm rather than logarithm. The bits being described aren't actual dynamic range.

 

However, the specifications claim a -155dB "noise floor." At 6dB per bit that's almost 26 bits. What they mean by "noise floor" I don't know.

[opus101]

Quite likely they know nothing of FFT gain so they just read that figure off their FFT. It would equate to the noise in a rather narrow bandwidth, perhaps <1Hz. But without knowing parameters of the FFT (number of bins, windowing) its relatively a meaningless number.

[bilboda]

6 hours ago, opus101 said:

Quite likely they know nothing of FFT gain so they just read that figure off their FFT. It would equate to the noise in a rather narrow bandwidth, perhaps <1Hz. But without knowing parameters of the FFT (number of bins, windowing) its relatively a meaningless number.

I think it's safe to assume that I know nothing of FFT gain and that you know nothing of how the designer is handling it (or not) but I don't think it is safe to assume that the designer knows nothing of FFT gain after decades of work with dacs. A simple search finds a dozen well documented examples of the concept. Not really a foreign concept to someone who works in that field.

 

[mansr]

6 hours ago, opus101 said:

Quite likely they know nothing of FFT gain so they just read that figure off their FFT. It would equate to the noise in a rather narrow bandwidth, perhaps <1Hz. But without knowing parameters of the FFT (number of bins, windowing) its relatively a meaningless number.

Yes, they are probably taking an arbitrary FFT and quoting the level of the noise as read off the y axis. With correct scaling of the values, the unit for this figure is dB/Hz. The total noise level is obtained by integrating over the frequency band of interest. Misinterpreting FFT plots seems to be quite common.

[mansr]

10 minutes ago, bilboda said:

I think it's safe to assume that I know nothing of FFT gain and that you know nothing of how the designer is handling it (or not) but I don't think it is safe to assume that the designer knows nothing of FFT gain after decades of work with dacs. A simple search finds a dozen well documented examples of the concept. Not really a foreign concept to someone who works in that field.

What they know or not is only tangentially relevant. Their claim is only slightly more believable than if they'd said they made a perpetual motion machine. It cannot be correct. They are either ignorant of their error or deliberately deceptive.

[bilboda]

Showing the depth of your knowledge does nothing to explain or expose the depth of the designer's knowledge.. So probably better just to ask him unless you prefer pot shots in the dark.

 

[mansr]

2 minutes ago, bilboda said:

Showing the depth of your knowledge does nothing to explain or expose the depth of the designer's knowledge.. So probably better just to ask him unless you prefer pot shots in the dark.

As I said, the claim is demonstrably false regardless of his knowledge.

[bilboda]

Well, I view your statement as unsupported, likely submitted for controversy only and thus false,until proven otherwise.🤨