Gain Structure in Reader's Digest Version
I think that what trips a lot of people up is that both the "rules" (so to speak) and the metering are different on the analog side and on the digital side of the recording chain - and no matter how short or long one's recording chain is, if you're recording a microphone or an analog instrument to digital, it's *always* part analog and part digital.
On the analog side, the idea is to get best balance of high signal level and low noise. In other words, if you set your amplitude at any given "stage" or point in the chain too low, you'll be limiting your dynamic range and reducing your signal-to-noise ratio. If you push the gain too far at any given point, however, you're just boosting the gain on the amount of noise you have so far and also running the chance of overdriving the circuitry on the high side and introducing analog distortion (digital distorts on the high side by clipping, analog distorts by overdriving the circuitry.)
On the digital side, down stream from the analog side, noise level and dynamic range are no longer the main issues; you're stuck with whatever noise you're carrying over from the analog side and (unless you use an expander) the dynamic range of the recorded signal is also already set in stone by the analog signal. All we're really doing is making a digital recording of the analog signal as it is. Sure there will be digital distortion and artifacting, but as long as we don't clip, that will be equally true regardless of what recording level we choose on the digital side.
The only real issue that remains to be concerned with with setting digital recording levels is that we leave ourselves enough headroom on the top side of the signal so that a) we don't clip when we record, and b) we have enough headroom for the mixing process (when we mix tracks together, the overall signal amplitude will rise.)
This is one reason why "record as hot as possible without clipping" is not the best advice; it leaves us with no headroom to mix. Sure we can turn things down after the fact to make room, but that's just giving us extra work that we don't need, and wouldn't need to even worry about if we recorded quieter in the first place.
Now, as it turns out, the engineers and engineering committees who design the gear we use thought about all this and actually designed both the analog and digital sides to work fairly ideally if we follow a basic road map. That road map more or less follows the concept of "line level".
All "line level" means on the analog side is the nominal or average voltage of the signal at which the gear is designed to operate to it's design specifications. The audio engineering equivalent of a GPS navigation system for following this analog road map is the VU meter; 0VU indicates a voltage of exactly line level.
Now, the more experienced engineers know how to read specifications and use their critical listening skills to know how to deviate a bit off-road every once in a while in order to take advantage of some of their analog designs either to maximize their signal-to-noise ratio (which is what gain structure is mostly about), or to purposely introduce some analog noise or distortion effect for specific special effects purposes, but they still have internalized the idea that they still in general want to wind up pretty close where that road map is taking them.
The more inexperienced engineer who has a hard enough time staying between the ditches without crashing, let alone leaving the road on purpose, will typically want to keep their signal averaging somewhere around 0VU. This includes the signal level heading into the digital converter. As they get more experienced and learn the personalities of their analog gear better, they may purposely stray a bit more, but until then it's a pretty safe bet to keep things pretty much on the road.
This means that when it comes time to convert to digital, we are somewhat at the mercy of how the analog to digital converter is designed; i.e. just what digital level it is designed to convert a given analog signal level such as 0VU into. This is where the -18dBFS figure comes in; while not all converters are designed exactly the same, it can be said that, on average, converters will convert a voltage of 0VU into a digital signal of -18dBFS. Some converters will be a couple of dB higher, some a couple of dB lower, but in text posts like this, it's easier to just average it to -18dBFS.
Now, there may be times where the analog signal is so dynamic - it has transient peaks that are so far above the average signal level (and that we never see on a VU meter because they happen so fast), that maybe that 18dB of headroom between the average signal level in digital and the top limit of 0VU is not enough, and we see clipping. This is quite rare, but it's not impossible. If so, then we can just turn down the input level on the converter/interface until we have enough peak headroom so as not to clip on the digital meters. No problem.
But if our peaks fall well below 0dBFS, there is no reason or advantage to boosting the signal into the converter;, it provides us with no sonic advantage to do so - i.e. it won't sound better to do so. It only serves to boost the volume of the analog noise level being converted and recorded, while taking away potential headroom that we'll want for the mixing process anyway.
That's pretty much the long and the short of it in a nutshell. For an even more detailed description of all this stuff, head over to
www.independentrecording.net and click on the "Metering and Gain Structure" icon.
G.
