Who is sampling more than 24-bits? 24 is really about 20.

Don't guess anyone is actually A/D'ing > 24 bits, but once inside digital hardware 32bit (and>) seem common.
What do you gain in 24 bit samples given dynamic range of 16? Noise floor?
And sonic improvements from sampling over 44.1 imply issues with the noise-shaping algorithims in D/A if I understand Mr. Nyquist correctly.

I realize phase coherence is "probably" the big issue with filters for 44.1.

cathode_ray wrote: Don't guess anyone is actually A/D'ing > 24 bits, but once inside digital hardware 32bit (and>) seem common.

For processing, yes, but 24 bits is the current common standard for conversion. You can buy 28-bit converters, but it's by no means clear that any audible improvement using those expensive beasts comes from the additional bits or the necessary extra care taken to make a good design at 24-bits.

cathode_ray wrote: What do you gain in 24 bit samples given dynamic range of 16? Noise floor?

Noise floor is a function of the circuitry up to and including the A-D converter. Designers usually try harder with 24-bit over 16-bit or 20-bit gear. The biggest thing you gain in practice is headroom.

cathode_ray wrote: And sonic improvements from sampling over 44.1 imply issues with the noise-shaping algorithims in D/A if I understand Mr. Nyquist correctly.

The actual implementation of the D-A is independent of the sampling rate. An oversampled design uses interpolation and noise-shaping algorithms at whatever sampling rate is used.

cathode_ray wrote: I realize phase coherence is "probably" the big issue with filters for 44.1.

Why? If the D-A is implemented as an oversampled design, the reconstruction filters can be digital FIR with inherently linear phase, so no phase issues. Only a simple analog roll-off is needed to reject the images at the oversampled rate.

Remember what you are doing,
You are taking a picture of the wave form 44.1k times a second and using 16 bits to describe what it looks like. If you have 20, 24, 32 bits to describe it, you will be able to describe it in more detail. The same goes with sample rates, if the "picture" of the sample changes in between 2 samples that picture or detail will be lost and so will the sound. If you are sampling at 48-192k each time you go up you are taking pictures of it more often and in most cases giving you more detail into it.

That said I can go 32 bit and 32 bit float (floating point math) but have not tried it. I use the standard 24 bit. I have tried higher sampling rates and I do get a smoother more full sound most of the time. I think the default archival rate for the Smithsonian is 96k.

Mark

cathode_ray wrote: Don't guess anyone is actually A/D'ing > 24 bits, but once inside digital hardware 32bit (and>) seem common.

Those two are not even comparable. You are not talking about the same thing still. Sampling and mix engine processing are two different things. Your 24 bit signal remains a 24 bit.

I use 44.1 cos most of my output will be at 44.1
this makes life easier
(there may be reasons to go to higher rates but only if you know why you are doing it and can weight up the pros and cons)

I use 24 bit cos it gives more headroom during the capture (record) process
again makes life easier

stop worrying about the Maths
concerntrate on the song and Mic choice and positioning