New 28-bit DAC coming out.

[solderdude]

Add a 1x buffer to drive speakers and you have a nice 100W/4Ω speaker DAC-amp.

 

[popej]

The top DAC should be muted 4 times per period, did you ever see such a commutation with no glitches?

Now think of digital side. Top side can't be dithered. Low part wont get flat tops after reconstruction filter. I cant say for sure, that it is impossible to match both path with 20bits precision (author claim 120dB THD+N). I simply think, that it would be much easier to use 2 complete DACs with 20bit resolution and switch between paths. Or maybe smoothly pan between paths at low volume.

 

[PGAMiami]

A DAC may accept and internally mathematically process 32 bit data, but this does not equate to having 32 bits of actual resolution in the DAC’s analog output

 

[Rja4000]

Ultimately, it's all about THD+N at low music levels -- where we are most sensitive to atmospherics and imaging. And this is a key advantage of multi-path. When we cross-fade to the top of the low-path (roughly -45dBFS), the low-path DAC IC begins converting at its most significant bit (bit-32) --- dramatically reducing low-level THD+N compared with legacy single-path architecture at the same level. This is immediately audible, and easily testable on an AP.

Hi John

Thanks for showing up here.

I have to admit I have a hard time to understand how, in practice, that could be audible.
In other words, what practical case would lead to an audible improvement, even in a studio.

For an ADC, benefits are easier to foresee.
I personally tried to measure some multipath ADC (Sounddevices Mixpre 6 II) and implementation and results were not very convincing, to say the least. So there is room for some dramatic improvement.

Is the benefit from the DAC mostly expected for intermediate repeated conversion purpose ?
Or just for monitoring low level moments in a huge dynamic range material ?

More directly interesting for us here:
Is that the area where you think current DACs are lacking today ?
And you say it's easy to measure with an AP.
What measurement would you suggest ?
And what reproducible way to ear this effect, if audible ?

 

[signalpath]

I think it is a bad idea due to the obvious source of a high-order nonlinearity. Look, the top DAC is 7 bits(I hope it is just a typo, I see no reason to cut 25 bits of AK4490 or similar DAC), if it is so, try to imagine how will it work with -35dbfs sine. The top DAC should be muted 4 times per period, did you ever see such a commutation with no glitches? It should work smoother with two DACs both 32 bits....

Thanks for the discussion. Yes, both DAC ICs are operating at their full bit range (otherwise, as you say, it would be a very bad idea). I wish I had time to engage in all of these comments, but we're slammed. Virtually every comment here is addressed in the AES Engineering paper (#21106) and the multiple published patents (US9871530, China CN107210752B, etc.).

The gate is best seen as a psychoacoustic processor, not a level follower. It's based on Louis Fielder's perceptual work at Dolby from the 1980s and early 1990's. Unlike IC range extenders, there are zero perceptual artifacts, though we measure an immediate (non-linear) drop in THD+N at the path junction. This is a good artifact, and a core reason for doing multi-path architecture. The AES paper is especially helpful by showing a 32-bit low-path DSP-DAC pad diagram (attached). Note, we used a bit diagram of 1-32 for clarity (vs. 0-31).

 

[signalpath]

Hi John

Thanks for showing up here.

I have to admit I have a hard time to understand how, in practice, that could be audible.
In other words, what practical case would lead to an audible improvement, even in a studio.

For an ADC, benefits are easier to foresee.
I personally tried to measure some multipath ADC (Sounddevices Mixpre 6 II) and implementation and results were not very convincing, to say the least. So there is room for some dramatic improvement.

Is the benefit from the DAC mostly expected for intermediate repeated conversion purpose ?
Or just for monitoring low level moments in a huge dynamic range material ?

More directly interesting for us here:
Is that the area where you think current DACs are lacking today ?
And you say it's easy to measure with an AP.
What measurement would you suggest ?
And what reproducible way to ear this effect, if audible ?

Hi RJA. A key benefit of multi-path D-to-A (compared with all single-path DAC architecture today) is lower THD+N at low perceptual levels, i.e., the "low-path."

 

[Rja4000]

Hi RJA. A key benefit of multi-path D-to-A (compared with all single-path DAC architecture today) is lower THD+N at low perceptual levels, i.e., the "low-path."

Thanks
So you mean adding a THD+N measurement at, say, -50dBFS would make sense ?

 

[mocenigo]

So science indicates audio benefits from 28 bits? Aren't there already 32/768 DACs out there anyhow?

They may be processing the data internally in the digital domain at a resolution of 32 bits, to increase precision, but once they start converting the samples to an analog signal, the actual resolution is about 21-22 bits, maybe 23.

 

[mocenigo]

But what if I want to get a nice system for my cats

So no deep bass, but a lot of ultrasonics?

 

[phofman]

IMO in I2S I/O codecs the 32bit length is used for simplification of the circuitry and software around. 24-bit format is not natural for 32/64bit computers.

 

[EERecordist]

Hi All. Thanks for the interesting discussion. The D-1 multi-path DAC is designed for the pro market. In recording, we stick microphones directly on snare drums and trumpets. These have around +155dB SPL peak level at the microphone. On the other end, the human ear can detect sound down to -8dB SPL. In recording, this is a real-world dynamic range of around 165dB, or around 27-bits.

Will everyone need 165dB dynamic range? No, of course not. We're creating a new professional standard, not a home standard. Most home listeners are fine with 100-110dB peaks, or even less with headphone applications (though significantly more if a large sub-woofer is used for explosions, hip hop, etc). The D-1 DAC is just the beginning. We seek to re-create every link in the audio signal path via multi-path architecture, including microphones and power amplifiers.

Ultimately, it's all about THD+N at low music levels -- where we are most sensitive to atmospherics and imaging. And this is a key advantage of multi-path. When we cross-fade to the top of the low-path (roughly -45dBFS), the low-path DAC IC begins converting at its most significant bit (bit-32) --- dramatically reducing low-level THD+N compared with legacy single-path architecture at the same level. This is immediately audible, and easily testable on an AP.

It's looking like production D-1 will achieve 28-bit performance, which is 40nVrms quiescent noise (broadband, unweighted) and +23dBu headroom, with zero ISOs. Output impedance is roughly 1.8 ohms. This performance is only achieved at the diff-bal XLR outputs. The RCA outputs are not specified. An AP cannot directly measure -146dBu broadband noise. Heroic measurement techniques are required to even get close to such a number. See Art Kay’s seminal papers on measuring ultra-low-noise.

The best multi-path analogy we've found is called HDR = High Dynamic Range Photography. HDR uses multiple exposures which are then intelligently combined to create a single image with much greater dynamic depth and detail than can be achieved with a single photo (multi-path v. single-path). Please read the AES paper (AES21106) for discussion on noise calcs, path continuity, linearity, etc..

Best with your project. I'm looking forward to lower noise microphones. There is a return of the trend of the ADC in the microphone with Schoeps. A long time ago I was active in our Society for Information Display chapter. We heard a talk by the Canadian company that pioneered HDR. They were bought by Dolby.

 

[rdytro]

So no deep bass, but a lot of ultrasonics?

they seem to hear (or maybe feel through all their sensory hairs) earthquake rumbling and audio bass long before I can.

 

[Gradius]

Indeed. We're moving systemic dynamic range from 120dB to 160dB.

Dogs, cats and dolphins will be happy I think.

 

[Kal Rubinson]

Dogs, cats and dolphins will be happy I think.

Or will they be angry?

 

[Bleib]

Or will they be angry?

Yeah. When they figure out the loudspeakers are no match

 

[Mnyb]

I'll just wait to see the audibility claims to be proven in some objective way , hopefully the design don't introduce some higher level artifact to get rid of inaudible ( but measurable ) problems somewhere else . A third party measurements for example what amir does is a good start .

I get the idea about a wide range ADC in a studio . But as a DAC in your home with reasonable control of system gain ? We will see , when someone tests it .

It's never the less an engineering achievement .

A way forward for home systems would be DAC's with higher output voltage and low gain amplifiers after that, if the signal is "born" above the noise to begin with .
If you assume 2 volt and tracks down don't we face some physics limits of just random thermal noise in passive components ?

 

[mocenigo]

they seem to hear (or maybe feel through all their sensory hairs) earthquake rumbling and audio bass long before I can.

This is definitely through other senses. Their ears are not so great with low frequencies.