16 bit vs 24 bit vs 35 bit vs 36 bit vs 64 bit DAC sampling

1. timing/jitter (read the blog at sonogy research.com).
2. noise on the ground of digital signals, that either pollutes the analog ground or impacts jitter (back to #1)
3. Filtering - the reconstruction part, both by shifting the frequencies so they are easier to filter (that's what over sampling or up sampling essentially achieves) or building better filters.  A filter can bee deep, or linear, but typically not both. So the bit I wrote about "only after the reconstruction filter" means these filters can be imperfect.  Can be?  Are.  I'll also point out that that @cleeds said "its math".  True enough, but there is an ocean of difference between theoretical math, and practical implementations. Analog is perfect in theory too.

4. part B of the filters is that there are some very small signals that need to be amplified or impedance transformed at the output of a DAC.  Both present good old analog amp challenges, not all that different from a moving coil cartridge.
5. never underestimate the impact of regular old audio blocking and tackling.  Power supplies. Amps.  grounding.  parts selection.  Do you know how many DACs depend on relatively pedestrian chip opamps and ok-but-not-awesome power supplies?  Most.  You know why?  Its easy and cheap compared to doing it from scratch.

Nyquist goes to Hz sampling (the vertical aspect in the time domain) and the step landings - horizontal bits - would become narrower with higher Hz sampling rate.

This is a bit confusingly worded. Harry Nyquist inferred from Shannon’s work (its basically a corollary) that sampling at **higher than** 2X the highest frequency allows the original waveform to be reconstructed perfectly. yes it is the X coordinate of what becomes a Cartesian graph, when viewed as PAM. (pulse amplitude modulation).

The caveats are:

1. higher than, not = 2X. At precisely 2X you can get the wrong answer. Its called an alias.  For example a 1 kHz 1V sine wave sampled at precisely 2X could land samples all at zero (silence) all at 1V (perfect recreation), all at 0.5V (off by half) or anywhere in the middle.
   
   
2. perfect reconstruction. Not in this world.
   
   
3. ...but we can get very close. Likely much closer than a) analog does (maybe 60-70 dB on a good day) and also b) likely distortion below the levels of our hearing
   
   
4. another issue is that one ought not measure distortions as "all created equal". Music theory tells us that, if pleasure is the desired result, this is not so. Dissonance and consonance are pretty well understood.
   
   
5. In the end arguing for "perfection" misses the point that no system is perfect, and vinyl is a long way from perfect, as is tape.

on the topic of bit depth used in recording nad mastering, there are many reasons why more than 16 bits are used.  And, today, almost ALL recording and mastering is done at 24 bits. Must be better for us right?
No.
reason #1 is so that if levels are off ( remember 0vU levels sets?) there is sufficient headroom to keep all 16 bits int he end product.  You can record too high and not clip, or too low and not lose bits on the low end.

#2 is so that in the inevitable (regrettable too) mixing, processing etc. the fidelity lost is below the least significant of the 16 bits. I won get into how digital level math is done but if you think about it it may become clear that you have both Least common denominator (speaking loosely) issues and truncation issues.

So it allows for sloppy production.  OK, tha's harsh, btu not too far off.  It also allows for multi-track mix-down. Yuk again, but reality especially in rock/grunge/pop/hip-hop/rap.

Want to hear what properly level-set and unmixed 16 bits can do? Listen to old Verve or Mercury Living Presence.  Old 3 channel analog recordings. Careful levels sets. Loving attention to detail. Minimal processing. Superb.  16 bits

The actual true limit in bits is 20 for digital , the recast are extra bits that fillin a guesstimate , there are plenty of high tech white papers from Mojoaudio , Schiit Audio that explain in-depth 
dontget caught up in  that it’s engineering  quality makes much more of a difference and dac type many prefer the older Multibit 
Dac chip , which-are a form of a R2R  but on one  big chip 
and werel laser trimmed , like Burr Brown 1704, or Analog devises 1955, for example. Tom dacs have 2-3 linear power supplies 
and excellent filtering and pre and post regulation , noise is jitter 
and the better dacs have all this and a bunch of other parts in the engineered design , but it isnot cheap, also many quality dacs 
are now starting to implement a seperate streamer module 
that is also filtered and has linear power supplies.
high quality cost $$.

to several who have noted (correctly) that 20 bits is all that can be truly used ( and I'd argue more like 17), but wonder, "why 32, 35, 64, (finger to mouth 1 million)?
Before going on i have also seen "but rate" and "but depth" used.   Let's be clear: its bit depth, although that impacts bit rate. Yet in lossless transmission, its not even close to linear since most of those bits are zero.  Basically is 2 to the Nth power dynamic range. 16 bits means 2^16th ~ 96,000:1. 24 bits means 16.7Million:1. 

Some good and bad reasons.

1. marketing. But i honestly think this is not as central as maybe imaged by the cynics
2. More bits - say 24 gives huge benefits in the studio.  levels sets can be off.  Mixing can be done and when you lose a but or two, its not big deal.
3. With enough bits, we can perform  volume control and DSP int he digital domain with minimal if any losses - and all the math truncations are below the level of audibility.  I personally think this is a huge advantage especially going forward since we cna right many wrongs without all the processing distortion we suffer today.

Remember that the analog processing before the digital rendition does not have infinite signal to noise,and remember that a DAC is an analog chip, subject to noise and distortion.  Even 24 bits is WAY beyond this threshold.
G