not typical Audio DAC-IC's for Top Class Audio Applications

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I have heard, any few manufacturers don't use the good known DAC IC types for their audio components like that one from regular audio brands (e. g. AKM, Burr Brown/TI, Analog Devices, ESStech (SABRE), Crystal/Cyrrus, Philips/NXP and Wolfson).
An example is this manufacturer:
http://www.audiomaat.nl/upload/documents/Mededelingen/Metrum octave hifi critic.pdf
Maybe there are much more in the meantime.

Fast industrial devices seems to be much more suitable than the usual good known ones in order to obtain best possible sonic results.

Is this really true?

What types from which semiconductor brands are here of interest?

Thank you for advices.
 
In what way? I just looked through the 768 and 9707 datasheets and they seem less informative than the typical audio DAC's datasheet. It's not, after all, AKM, Analog Devices, Burr Brown/TI, ESS, Cirrus/Crystal, NXP/Philips, or Wolfson's fault DIYers usually zero in on THD+N and DNR and ignore things like group delays or combined responses. Among the specs that do tend to show up in both types of datasheets SFDR does indicate contain some information THD+N doesn't, but the converse is also true. Similarly, A-weighted DNR doesn't say much about where a part's 1/f noise corner is but neither do noise densities at several MHz.
 
Glitch energy primarily as that's my primary hypothesis for what constitutes SQ in a multibit DAC. They show multitone performance, that's a much more music-like signal (crest factor) than a single tone. The variation of SFDR with frequency and with sample freq plots tell me quite a lot about the DAC's dynamic performance.
 
Yeah, unfortunately I don't have a good enough oscilloscope to be able to make meaningful measurements of how much switching noise rides out of the outputs. But an output buffer with a couple passive RC stages between the DAC and the op amp pins should do a decent job on whatever escapes the DAC. Cirrus sets up their eval boards this way and it's always seemed kind of strange to me other manufacturers don't.
 
Yeah I originally thought an RC set above the audio band would lick it. But then I listened... :p
Depends on what exactly one means by lick, I guess. A glitch is essentially an impulse and therefore is fairly flat in the frequency domain. As the the switching time for dynamic element matching has to be short compared to DAC's clock the impulse energy should have a bandwidth of tens to hundreds of MHz. RC filters cornered at ~50kHz reject the majority of this energy. So, if one's purpose in putting down a passive filter ahead of the output buffer's op amp is to prevent DAC switching transients from consuming the op amp's gain bandwidth product and slew rate, passive filtering does that well.

Such filtering is obviously not going to reject the in band portion of glitches. However, a corrolary of their broadband nature is THD, IMD, and DNR measurements all contain information about glitch energy. These are sometimes given in detail for an audio DAC's eval board, so one can have a pretty decent idea of in band behavior at sigma-delta part selection time. A more revealing test is DNR versus DC offset which, to my knowledge, is unfortunately published for only the ES9018 (slide 36) and WM8740 (figure 4---the "competitor's 117dB DAC" is likely the AD1853). DNR versus DC is primarily as a test for other issues but, to the extent glitch energy varies as function of output level, it captures that too (more here, though the talk starts rather slowly).

I'm curious as to what lead to the conclusion it's glitches you're hearing as opposed to other limitations of sigma delta DACs such non-peridodic steady state noise, nonlinear excess phase, or even just the impulse response of the antialiasing filter.
 
Last edited:
I think here we might be talking a bit at cross-purposes. I don't hear glitches on S-D - but then I gave up listening to S-D type DACs a couple of years ago. Glitch is one of the things that S-D is good at - but its soooo bad at other things that its just fallen by the wayside in my quest for the ultimate SQ.

So having come to the firm conclusion that multibit is the only way to get satisfying sound I've been working on reducing the glitches in that. You might find this thread I started in another place interesting - I'm working on a multibit DAC which will be public in both process and end result - Digital that sounds like analog. The no.1 aim is dynamics which I hear as the the main drawback of S-D DACs - although I have stopped listening to them, plenty of material I encounter has the characteristic sound as you'll note if you dig into the thread :)
 
Looked through the linked thread. I've been down the same path with the same filtering tradeoffs as to how one wrings the most out of 44.1 and where I ended up was pretty much where Cirrus landed with their parts. Non-audio NOS parts are interesting to me primarily as they permit oversampling and allow the sinc Nyquist zero to be pushed out of the audible band. There's a freedom to this in that one has full control over the antialiasing filter but no steep analog filter I could figure out how to synthesize hit the target phase flatness. So, in lieu of original material recorded at 96+, an arguably least bad tradeoff seems to be to oversample, use a slow roll filter to provide some image rejection, and let the rest of the images pass down the signal chain to the amp.

A logical endpoint of this line of thinking is to feed 2x or 4x oversampled data to a NOS part with the usual 50-75kHz third order Bessel or Butterworth kind of lowpass on the output buffer. As long as one's got good bit depth utilization within the DAC this should produce something resembling minimal ringing of the impulse response at the expense of minimal image rejection. The main difficulty with this is a lot of power amps don't have particularly high gain bandwidth products, making the images that pass through somewhat problematic. At the lower power end of the range it's not hard to build power amps with 100+MHz GBP but it gets rather more involved above a watt or so. It's also not well explored what goes on when drivers get ultrasonic input. But I've had a certain temptation to build this approach and at least try it out with IEMs to see what it sounds like.
 
Last edited:
the major problem of glitch of ordinary multibit DAC may be not harm of HF noise ,but distortion.
the binary weighted DACs have unequally glitches at each binary.
when upper bits switches ,the major glitch occurs.
when the glitch through the filter ,the glitch energy is integrated and join in pulse amplitude.
it deteriorate deferential linearlity in practical use, especially at MSB (= at bipoler zero).
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.