Howdy,
There are a couple of things to consider.
The higher the headroom, the lower the noise floor- 24 bits gets you about 114 db of signal to noise, if I remember correctly, which is pretty good. 16 bits gets you about 90 db s/n, which is still pretty damn good, but back in the old days a good console with Dolby SR going to a good tape machine using 456 could get about 130 db s/n (all these numbers are subject to the vagaries of my memory). Now why would 114 db s/n be better than 90 db s/n if the human ear generally can't detect the difference? Well, start recording and playing back twenty-four plus tracks and the noise adds up.
Another issue to consider is fooling the human ear. Analogue is more noisy but authentic than digital, to be somewhat simplistic. Some people think that 16 bit CDs suck, that the resolution isn't good enough, that the human ear is not sufficiently fooled into thinking the sound is "real" (these explanations may be a little simplistic, generalized, and open to debate, but I offer them humbly). 24 bits does seem to be more "real" sounding, and 24/96 is apparently going to be the new standard.
Digital does need big headroom because if you hit the threshold you're hosed. Some folks set "zero" as -10 db to prevent this, and then you have to start thinking about noise...
Probably the most significant source of noise for home recordists/budget studios is the room they are recording in. Second would be mics, then preamps or rack gear or cables.
Hey, I hate to be super obvious, but... Did you ever have an athletic career? Doc

I hate to drag this out again but here goes one more time.

24 bit digital is capable of 144 dB S/N. But the analog path is limited to about 110 to 115dB in all but the most exceptional of cases. This is due to thermal noise.

The best any tape ever did anywhere was more like 90 to 96 dB with all the artifacts and tradeoffs of dolby NR. When tape was king the best analog circuits around generally did no more than 88 to 96 dB s/n (again with a few extreme exceptions).

Now to answer the headroom issue:
Most high quality converters are calibrated such that 0VU on your analog pre amp will equal -20 to -14 dBFS. That is 14 to 20 DB below full scale. If you have a Digidesign 192i/o or an Apogee AD8000 each with s/n close to 110 dB, this leaves you with between 90 and 96 dB S/N which is again about as good as tape ever got.

Increasing the drive level into the converters is OK if you know what you are doing, but it does require that you push your pre-amp harder. If 0VU = -18dBFS than you need to feed +12 VU to hit -6 dBFS. If your pre amp sounds better at +12 than 0VU then go for it! If not then find out where your pre amp sounds best and use that. Either way you will have lots of headroom and more S/n (signal-to-noise) than any tape recorder ever had (again with the exception of some extreme gear).

So bottom line is: Run the signal level that sounds best for the analog signal chain. The 24 bit adc will handle anything that sounds good.

And lastly, never listen to anyone that tells you to track hot in 24 bit so you use as many bits as possible. One has nothing to do with the other.

130dB S/N on tape? Not in this universe. Especially not on 456.

One thing to remember is to maximize your dynamic range. I am calibrated to -18. I am well above the noise floor.

Most people think that you need to cmpress the snot out of the signal, and get it right up to the last useable bit. This is the reason today's music sounds so bad.

If you have a low passage in a sound, which requires low volume, then you do not require as many bits of dynamic range. What may only require 2 or 4 bits, people squish the snot out of it, and use 23 bits to represent that 2 or 4. A complete waste and mis-use of the concept. You get no more "resolution," you are just taking up more bits.

Originally posted by Brian Bosworth:
In the world of 24 bit digital, where you only have roughly 16,777,216 individual volume levels over a span of a certain reference voltage, is there a point where you can "over-do" the headroom figure? I guess what I am trying to ask is, if you work in the digital domain, is there a ceiling value for headroom you can hit before it stops making a difference?
Thanks!

Brian,
I think I see where you are headed with this.. Really the question is, since dynamic rage is limited by the recording medium, what is the point of high headroom mic pre designs and the associated costs? Well, headroom is a good thing. You cannot have too much headroom in any audio application, EVER!!! The higher the headroom the louder you can operate a pre, bringing it above the noise floor. You can always attenuate at the output, in the mixing desk or in an EQ, compressor or limiter. Also, high headroom designs just sound better! They are bigger, have more girth and dimension. There is no scientific reason or justification for opting to use cheap mic pres. Once you have pre amped a mic signal, it's all over except the mopping up, so do it right in the first place. You only get one stab at it. Good question! Kurt

I went and looked. The Dolby website says 105 db s/n is about the best you can get w SR. Oops (I did say my numbers might be a little off...).
Numbers aside, Brian, do you understand why people are always fighting for more headroom and better s/n?
Doc

Originally posted by lowdbrent:
If you have a low passage in a sound, which requires low volume, then you do not require as many bits of dynamic range. What may only require 2 or 4 bits, people squish the snot out of it, and use 23 bits to represent that 2 or 4. A complete waste and mis-use of the concept. You get no more "resolution," you are just taking up more bits.

But really, anything less than 10 bits would be noticably quantized, even to the end user. Take the case of 4 bits, for example. If you were to record something at 4 bits and turn up the volume later, it means that each of the 44100 samples per second could be only represented by 16 positions - Quantization noise central and lots of unwanted harmonics.

Of course, the other extreme - compressing the crap out of everything during recording/mixdown to "maximise the use of bits", especially in 24-bit recording, is pretty retarded.

I have to applaud this thread for being very correct n it's info. (okay, except for the 130db s/n tape spec :) )

This is a very misunderstood aspect of digital that get abused on internet forums. The truth is that most all converters have greater dynamic range than any preamp feeding it. RME actually has an intersting read on their website about how preamp noise can be considered "self-dither." From here on, greater sampling rates will be more useful to audio quality than greater bit depth. (except in mixer architectures where the bit depth must always be larger than the data bit depth.)

> why people are always fighting for more headroom and better s/n?

It is easier than listening.

I find it hard to deliver a full 16 bits worth of analog to my CD burner. Assume room noise is 15 dB SPL. Anything the size of a string quartet or smaller isn't going to reach 111 dB SPL peaks (15dBSPL room plus 96dB S/N). My last gig, a women's choir, peaked 108 dB SPL in a room with noise of 20 dB SPL (often higher), or 88 dB analog S/N ignoring mike and preamp noise. Most of the pieces were much softer than that meter-banger: say 92 dB SPL peak. And the live audience would not sit still. Many of the tracks look like 50 dB S/N. They sound fine.

However in post production I'm sometimes applying 12-15dB gain to get the volume of the softer pieces up where people expect CDs to play. Also EQ and reverb. I'm learning that doing this in 16-bit math ends up "gritty". Now if I'm doing more than 6dB or so of fiddling, I drop the file to 32 bit, fiddle, and then go back to 16 for the burning.