Why do non-OS R2R DACs sounds better (only to some).
I was discussing the R2R vs delta-sigma approach of PCM DA conversion with a local knowledgeable guy who is audio designer himself (though more on the analog side, yet he has designed digital as well and has a lot of knowledge on the subject).
R2R is basically a natural, straightforward approach in itself. There's no feedback in the process, unlike delta-sigma for which the feedback is necessary means of getting the DA done. And feedback is generally associated with corrections, approximations and messing with the time domain (as with the feedback you always correct with the time delay relative to the signal you're correcting with). This basically renders technical problems and various noise generated in the DA process which depends on the input sequence. Delta-sigma is from the very beginning on the path of constant improvement of the DA process...because it requires constant improvement, because of its imperfection.
The thing is that constant tone and FFT tests don't reveal flaws in delta-sigma conversion. To reveal noise produced you need more complex tests which requires bit more complex test signal which will provoke what's going on.
One of such signals from what I could read is a square signal, which consists of various harmonics of the base frequency.
With R2R, this process is so simple that you don't have anything other than error from the resistor ladder due to resistors precision limits, which shows as HD (plus IMD which results from the THD). There's nothing more to provoke which wont show on simple tests, that's all.
As for NOS vs OS, as much as I have understood, NOS is the was which is least demanding for output filtering, as it is most tolerant for less steep filtering - it's even tolerant for a non-filtered approach. And it's the steep output filters which produce pre and post ringing. Which is nothing other than a time domain oscillation audible in a HF range, rendering an unnatural effect - nothing similar happens with the sound in the nature. My opinion is that brain is unable to handle it - since it's not even remotely close to natural phenomena, brain doesn't know how to filter it out from the impression he sends to the cortex.
Hearing mechanism aims to render information from the raw tonal data entering the ears (and reflecting in the ears). When unnatural tonal signal enters ears, brain cannot draw correct information from it because it's impossible - but it still attempts to do so. The end result of this attempted brain processing on unnatural signal is easily far from what was intentional when producing the record. And this is the reason why relatively small in scale time distortion such as ringing is can do much wrong to an impression.
NOS design can sacrifice the strength of filtering out the ultrasonic noise from a DA reconstruction to make the pre and post ringing much lower (or non existent), in an idea that ultrasonic content, its eventual audibility plus IM byproducst in the audible range, will do less audible damage than ringing time domain distortion, from the 'viewpoint' of a human hearing mechanism. It can be either completely non-filtered approach (no ringing but more noise) or filtering with lower steepness (less ringing than steep filters like brick wall but also more ultrasonic noise, though much less than with non-filtered approach).
With PCM HR signals of a higher sampling frequency, 88.2kHz or higher, there's a benefit since ultrasonic noise from DA reconstruction happens higher in the frequency (where attenuation from a less steep filter attenuation of a noise is stronger), so there's a clear benefit over a 44.1kHz signal DA. So for such signals 'mild filter' approach may benefit more than with non-filtered NOS.
DSD is a different story since DSS native conversion is different than PCM and basically it's again more straightforward than delta-sigma PCM conversion. So DA converters with direct DSD DA conversion path (no conversion to PCM prior to DA) can benefit in sound since DSD conversion doesn't require output filter at all, so no ringing as well. However typical cheap DAC converts DSD to PCM, then processes DA as with ony other PCM signal.
Perhaps the high frequency artifacts that modulate into the audio band mask additional information allowing the brain to concentrate on what it most wants to hear?
I suspect it's about less damage, comparatively. Artifacts themselves are noise which can further increase the IMD and it can't be beneficial to the sound. But the alternative delta-sigma deals with specific additional noise, plus ringing.
