Could someone explain to me, in a simple-enough way, the real difference between the two, and exactly why everybody seems to think that R-2R/Multibit is so much "better?"
I feel that the R2R vs DS debate is intimately related to the current cacophony over sampling rates. Higher sampling rates are either better, worse, or don't make a difference: go figure. Monty Python and Dan Lavry say that 192 kHz sampling speeds are silly, and actively damaging for playback fidelity. Neil Young and even James from FiiO say that 192 kHz and 384 kHz are useful to improve sound quality in the audible band. While this is clearly something akin to the theater of the absurd, I suspect that they're all right. They're just talking about different things.
http://lavryengineering.com/pdfs/lavry-sampling-theory.pdf
http://people.xiph.org/~xiphmont/demo/neil-young.html
Lavry clearly talks about
multibit converters (and probably Monty, too), those
R-2R (aka
"ladder") converters that take the full samples provided by a lossless audio file and convert them at hardware level before passing the signal to the analog output phase, with
no oversampling (
NOS). This technology is somewhat rare nowadays (and these DAC chips are more expensive and require extreme manufacturing precision when getting into 19-20 bit territory), and has been dying a slow death before seeing somewhat of a revival lately. Some of the products using R2R converters are old Theta (e.g. DS Pro Generation V), and newer Schiit multibit (e.g. Yggdrasil) or Audio GD (e.g. Master 7 or DAC-19). R2R DAPs are even rarer, the few existing breeds being the HM801, Hifi E.T MA9 and... surprise... Tera Player.
While this may be a bit of a stretch, these are apparently the
only DACs/DAPs around which are
bit-perfect, and they're definitely fewer than 0.5% of the market. They use the original samples in
full depth (up to the hardware limitations, usually ranging from 17-18 bits up to about 21 bits max), as these are
high-bit, low-speed devices just like redbook PCM is a
high-bit, low-speed format: 16-bit/96 kHz is ALL they need for perfect retrieval of analog waveforms within the audible band, and 16-bit/44.1 kHz works very well, thank you. What's crucial here is that 16-bit/96 kHz R2R technology exists, is proven, doesn't require ever changing formats and complex digital processing, does much less aggressive filtering of the data stream/sound signal, is perfectly suited for 99.9% of available audio material (slow PCM), and the only R&D required is care in real-world device implementation.
Neil Young and James however clearly have in mind
Delta-Sigma D/A converters (e.g. the DACs in Pono or X7), low-cost 5-6 bit devices which use complex digital-feedback techniques and algorithms (e.g. noise-shaping) to approximate analog waveforms. If with R2R we are hearing the device, with DS we are hearing the
algorithm. DS is the direct successor of 1-bit D/A converters and are thus at heart
low-bit, ultra-high-speed devices that rely on
oversampling techniques (e.g. 2^8 = 256 times the sampling rate). This is probably why many report that the
low-bit, ultra-high-speed DSD format sounds so much better and more "natural" on them. This is also what makes them quite... err... awkward with
high-bit, low-speed PCM, because the two are fundamentally mismatched. This is probably why manufacturers like FiiO, Hifiman and AK are constantly seeking to support ever higher sampling rates, to infinity (which is in many ways irrelevant, as how many copies of 768 kHz PCM files do
you have?).
What's most disconcerting with DS converters, is that they are
dropping original samples: since they can't handle more than 5-6 bits natively, it really doesn't seem to matter if you pass them an 8-bit or a 24-bit file, as they will still drop all the bits until they get to a 5-6 bit sample, and proceed from there to the analog conversion. (Talk about wasted space in a world awash with 16bit PCM played back almost exclusively on DS devices.) In order to operate, they need to slim down the high-bit PCM file into a low-bit version. And since they use so few bits, they produce a boatload of noise (i.e. a very high noise threshold), which is why they need to use oversampling to increase the bitrate and then use aggressive noise-shaping techniques to shift the noise just outside the audible band, a process which can go wrong in many ways. Since DS are fundamentally low-bit devices, it appears that they need ever higher sampling speeds in PCM to do their best, which along with DSD are merely cumbersome attempts to work around the fundamental limitations of the technology. I'm no engineer, but overall it looks to me that all this is hardly suited for high-fidelity applications, not least because they're not bit-perfect. But hey, they currently occupy 99.5% of the market, including "high-fidelity" flagship territory.
Sometimes I get the feeling that Delta-Sigma is so attractive to manufacturers mostly for scoring marketing points: "high-resolution" support up to 768 kHz and 32 bits, DSD, DXD, DSD512... Compare this to the paltry "up to 24 bits and 96 kHz" that R2R manufacturers often whisper. But if we avoid the polemics, on a theoretical level I see two big issues with DS.
http://www.head-fi.org/t/319569/r2r-vs-s-d-dacs/15#post_11860338
http://www.diyaudio.com/forums/digital-source/15439-how-does-delta-sigma-dac-work.html#post179844
http://positive-feedback.com/Issue65/dac.htm
First, this is a
lossy conversion process that uses
decimation to partially drop data, and then proceeds to digitally
synthesize high-bit performance. And after 15 years of R&D, the digital "glare" reported from the very beginning is still being reported today with top-flight DS DACs (when they playing back slow PCM). From my (approximate) understanding of DS technology, first they lose data from the original samples by downconverting each sample to 5-6 bits, then DS oversamples the data by some 2^8 or 2^10 (256 to 1024 times the sampling speed of 44100) to add an additional 8-10 bits of performance, then proceeds to using complex and obscure digital-feedback noise-shaping techniques to clean up the gargantuan levels of noise introduced by the earlier process and "sweep them under the rug" by shifting them just above 20 kHZ to where dogs and cats can hear it but not humans (so forget of using DS technology to test classical music on animals), thus adding a further ~8 bit performance. For instance ES9018 is rumoured to work at 6 bit + 8 bits + 8 bits, for a marketed headline 22.5 bits of effective performance. BTW, we should forever forget about "32 bit" DACs as neither technology is currently capable of handling it, each plateauing neatly around 21-22 bits, and apparently no standard digital audio file contains more than 24 bits worth of data...
http://www.metrum-acoustics.com/HexEN.html
http://www.metrum-acoustics.com/Design%20Philosophy%20Metrum%20Acoustics.pdf
My other theoretical concern also relates to oversampling, but from a different perspective. The Metrum document above, from another believer in R2R, focuses precisely on why so many DACs sound so differently from live performances. Ultimately this really is what we're all here for: recreate a live performance in our living room. It talks about "realism", "natural" sounding, purity of the sound, absence of digital artefacts (“rubbish”) and the like. They point among other things to
FIR filters used in DS oversampling technology, the absence of FIR filters in old R2R chipsets like TDA1541, and
the ear acting as a naturally sharp filter ("like a band-pass filter") meaning that adding one such filter in the audio playback device would be an overkill.
"Because our hearing naturally functions as a strong filter, our brains tend to interpret the signal from the NOS
DAC as if it has passed through a FIR-filter. This is due to the limited bandwidth of our hearing. Looking at
the picture on the top of this page, we can wonder how the eventual picture will look if another equivalent
filter is added by our hearing. It is well-documented by both musicians and authorities in the field of audio,
that especially percussion instruments suffer from this effect. It is therefore not unfounded when NOS DACs
are claimed to sound the most natural of all the alternatives. Because at the same time the testresults for all
NOS DACs fall short, the question can be raised wether the correct tests are being done to accurately gauge
their quality. All measurements are, after all, performed without the benefit of any filter."
Apparently while FIR filters (> 4th order) make oversampling techniques possible in the first place, it also plays as a redundant filter given our ears' natural filtering (> 6th order). This might go some way to explain the various reports of "artificial" or "digital" or "unnatural" or "lifeless" sound from DS devices with PCM, as opposed to "realistic" and "natural" sound from R2R products.
They go on with an interesting discussion of jitter testing, and conclude like this:
"NOS DACS have been gaining in popularity for the past few years, mostly based on listening reviews.
Especially people who regulary experience live music, appear to have a strong preference for this type of
DAC. As Kusunoki had mentioned in his article, it is primarily the behaviour in the time domain which gives
oversampling DACs their “unnatural” quality. This shows in the way that percussion instruments sound too
lacklustre and a sort of “excessive detailing”, which causes certain instruments to lose their timbre and
“warmth”. The question wether we should follow our ears or the results of tests remains on the table.
The development of digital audiosystems has not reached its zenith yet and we will certainly be confronted
with new developments in the future. Certain is, that due to High Definition recordings the need for
oversampling and sharp filters has lessened. How to approach the massive variety of CD’s, with their low
sampling rate of 44.1 kHz remains the question. To oversample or not to Oversample? Not oversampling
seems to be the preference of musicians and audio-professionals, despite their “limitations”."
For a related discussion on R2R vs DS see this thread:
http://www.head-fi.org/t/785488/the-new-r-2r-dap-thread