S-D Modulation is a lossy/approximation of the original values.

So is R-2R.

I don't think S-D is much more destructive than R-2R. S-D is a little more unintuitive perhaps.

R-2R appoximation will depend on error in resistor values, current sources and perhaps switching glitches (among perhaps other things). There will also be noise limitations. Errors and non-idealities can result in a non-monotonic output to code word curve, which is not good.

From what I've read, there are grades of performance with-in a single R-2R devices (consider the PCM1704U < PCM1704U-J < PCM1704U-K). This maybe due to process and dunno if device differentiation is due to cherry picking or what. There are papers here and there to deal with R-2R issues. Here is one which I have not completely read but after quickly going through it seems reasonable:

http://liu.diva-portal.org/smash/get/diva2:559515/FULLTEXT01.pdf

It is possible that some folks feel a well implemented R-2R sounds better than a well implemented D-S. But I fail to see how proper quantization noise modulation destroys much more data than resistor ladder non-idealities. Dynamic range on both well implemented technologies seem pretty close to me, even after going crazy with compensation on an R-2R implementation.

Saying that something is worse that something else cuz somebody else said so may fly on a casual conversation. It does not fly on a more professional and serious conversation (and believe me, that's an understatement). I wouldn't mind that much if I have not seen this parroted over and over again in several places even in the face of measurments not supporting this crazy data destruction brought forth by well implemented DS designs.

Why hell, aren't most reputed Audio Analyzers based on D-S DACs and ADCs?

I don't have a problem if someone likes this or that DAC more and it happens to have R-2R. But the claim of D-S lossyness and approximationness vs R-2R does need a bit of data to support it IMO.

The linearity problems with R2R is due to resistor matching and thermal/electrical issues and not related to mathematical/algorithm approximation like Sigma Delta.

They are not related, but they are both non-linear. That said, if I remember correctly, I can come up with a delta sigma noise transfer function that may mitigate the passband noise to a point where >  21-bits are possible save for perhaps other non-idealities in the conversion outside of "the math" path.

Here outlines the problem with Delta Modulation(Sigma Delta Modulation is a derivative/improvement of delta modulation, with additional feedback/comparator & oversampling to improve full scale/slope output):

The slope of the input signal is bounded by the bandwidth of the signal and the number of input bits, so the sky is not the limit. A proper implementation would take that into consideration. Furthermore, many TOTL delta sigmas are not 1-bit but multi-bit (in the feedback as well), which together with a sufficiently large set of accumulators, should result in no issues as described there.

I read somewhere R2R DAC/ADC only works well to a certain frequency(up to 3 Mhz or so) , anything above in MHz to GHz, then S-D/D-S DAC/ADC makes more sense.

This is incorrect. R-2R are used for applications that require faster conversion, but lower resolution such as Oscilloscopes. D-S works better with applications that require lower speed conversion but higher resolution.

As far as cost, I would argue that indeed it would be cheaper to make D-S designs if for the equivalent R-2R you have to cherry pick and if you need a ton of discretes to make it work very well.

https://en.wikipedia.org/wiki/Delta-sigma_modulation

The issue with Sigma Delta(especially high order sigma delta) is it's using a complex form of feedback mechanism to maintain it's stability. Most Music Signals will stay within stable conditions, but some forms of music instrument(those with rapidly changing loudness/tones) can expose the weakness of Sigma Delta like French Horns and Obode.

You can achieve high dynamic range/SNR with high order Sigma delta that looks good on measurements(i.e. ES9018S). But does it sound good/right(tonally/timing) to audiophiles is another thing all together.

You have evidence that a well regarded D-S IC will not stay within stable conditions with 24-bit audio signals?

Also, I'm more interested in a PCM4222 than an ES9018S. I like multi-bit DAC chips better. Seems to me it combines the best of both worlds.

I got tired of this discussion on the other forum! I'll see if it's a bit less evasive (and effusive) here.

It basically seems to come down to trying to determine what this elusive difference is, between the two techs, that makes many discerning listeners prefer one technology, over another. Remember, these discerning listeners have picked out the technology of the conversion process, as the main factor. (Although I've never seen anyone discount the importance of implementation.) Why do they do this?

My incredibly simplistic understanding is that the DS process uses lots of approximations, and processes them very quickly, making it more susceptible to any inaccuracies (perhaps caused by the limitations and imprecisions in the hardware).
Whereas R2R uses a much simpler and real time process that is less susceptible to any imprecisions caused by the hardware.

And for the record, I'll go for sound over measurement, every time.

This IMO, is not fine. DS AFAIK no more approximates stuff than the R-2R does. It just easier to understand R-2R than DS because R-2R is equivalent to a bunch of binary additions. D-S is not.

Furthermore, DS process things at an oversampled rate because it requires bandwidth to move stuff out of the audio band. Which in turn makes it good for relatively slow conversions, but not great for applications that require Gbps conversions.

Is jitter potentially a bigger problem for DS?


Where does audio rate, as a far as speed is concerned?