I agree with Pipeline that breaking this subject down so it's understandable can be tricky. Hopefully I'm not too full of shit here, but I'm going to try. (This is mostly for Tim).
16 and 24 bit systems behave a bit differently in regard to the noise factor at or near the LSB, or Least Significant Bit. Forget about noise for a second here. Noise is not an issue.
Digital noise comes from aliasing crap or something like that that occurs at and/or near the LSB. Imagine a very, very low volume, but distinctly audible clicking noise. It happens a lot when you start changing sample rates, which is another topic altogether. The way to get rid of it is a workaround solution at best, but a very effective one that still leaves you with a lot less noise than with tape hiss from the old Studer & MCI days. You add a very, very tiny amount of white noise to the signal. It's called dither. It masks the digital noise so's you can't notice it. It also blends in very well with practically any other sound, so you can't really effectively even hear this white noise.
Let's say you're listening to a cranked up Marshall amp with some hot chick banging out power chords on a Les Paul. She's not wearing a bra, and she's all excited. This digital noise thing would sound like a mouse fart in comparison, in 16-bit audio systems. 24-bit systems reduce that to about the level of an ant fart. This noise would be difficult to notice if you were completely focused on it and struggling to do so, provided that your monitor chain could actually reproduce it. You're likely to get much more noise in the system from other issues like cables, monitors, flourescent lights, HVAC systems and such.
Essentially, forget about noise floor.
<deep breath>
A bit is like a 2 position switch. A 0 or a 1. Binary. In PCM audio, this is how dynamics are determined, and it's divided into 6 dB chunks. It's easy to know how much range a 16 bit system has:
6 dB multiplied by 16 bits = 96 dB range.
The most significant bit has all the other bits to back it up. This is the audio range from -6 to 0 dBFS.
The least significant bit is by itself. The range of choices for dynamics are pretty simple here - either you have audio at ONE LEVEL ONLY, or no audio is present whatsoever. The 16 or 24 bit word length builds on it from there. Each time the word length adds a bit, the available number of splits to register a distinct dynamic level is doubled. 2 bits gives you 4 available choices and 12 dB range to be considered. The audio will sound like crap because there aren't enough variations to render it at an accurate level.
At 16 bits, the overall number of splits possible is simply 65,536. Half of that range lives in the area from -6 to 0 dBFS. As you keep moving down in 6 dB increments, the range keeps getting cut in half.
The main excuse that makes any kind of sense for printing hot levels in 16 bit systems is this "dynamic resolution" thing, or whatever.
The 0 VU = Line Level thing has not changed, ever, in decades. There is no standard for where your converters are calibrated to, as others have already said.
If 0VU = -12 dBFS, that's 12dB for transients. 12 dB is not a lot. It's wimpy. People calibrated their converters that way because of the dropoff of resolution, not the noise.
Between -6 and 0 dBFS, there are 32,768 "peg holes" if you will, to describe where the audio level is.
Between -12 and -6, it's half that much. 16,384.
-18 to -12, 8,192.
-24 to -18, 4,096.
It keeps crapping out.
Imagine that you just record one hit on a crash cymbal. It can easily fade by 40, 60 dB or more, and you can hear that quite distinctly when it's by itself. It loses its realism at lower levels because there aren't enough distinct spaces for the audio system (PCM digital) to describe where the volume is at.
In 24 bit systems, a level of around -48 dB has all of the full range in this regard as a 16 bit system at maximum.
Again, a bit is like a light switch. 2 choices, on or off.
2 <to the power of> 16 (bits) = 65,532
2 <to the power of> 24 (bits) = 16,777,216
All of a sudden, now all your cymbal fades and any sounds that end up in the quieter range of what you're recording can be captured much more accurately and realistically. You're almost down to the level of mouse fart again before things start to fall apart. So now there's no need to chump out on the headroom. More headroom is better - it's pretty easy to chew up that first 12 dB when you're recording something with a lot of transients.
Now consider that the individual recorded sounds of instruments are lower than what you want to end up with in a mix. Even if you record all the instruments at a very conservative level, say RMS levels at around -20 dB or so, once you start combining them for the mix, the signal strength of each individual sound will add to everything else. All the levels will "sum". Once you have 8 tracks or so in a mix, you could easily and very likely find that you have to start pulling track levels down to keep the overall mix from clipping. All of a sudden it doesn't seem to make very much sense to need to record stun/hot level tracks if you're only going to have to turn them down later (by a lot) anyway, when you can just as easily record them at much cleaner levels and still have to turn them down, but less. Even a mix should still have its own headroom, and won't be as loud as a commercial CD that's been rammed to the rails with compression and limiting. That's not the goal of a mix.
Myself, I'm guessing that the less you have to screw around with the volume of a recorded track in a digital mixing enviornment, the closer the final sound will be to what came out of the microphone in the first place. Regardless of the whole 16/24 "digital resolution" argument!
Again, if you run your preamps at line level and nothing clips (or even comes close!) things should be pretty decent. That's how it was all designed to work. Many pros advocate that printing even lower than that is just fine if not better under certain circumstances.
I'm sure there are other things to consider, but maybe this makes some kind of sense?
sl
