So here's a copy of a bit of an email I sent to RG, in case others may be interested, although some of you know this stuff already...
Not that I'm into the loudness war, I tried maximising on a track recently, stupid mistake, if you want it louder, mix it louder (is the result, which makes sense as screwing around with files (DSP) at PC bit depths destroys the signal (sorry for those who like DSP, and I agree it's very convenient etc, but it just does, it's simple physics/electronics), not so apparent in cheaper speakers, but through my QUADs is very apparent. Should see the jaws drop when I play the original .wav and compare to the (high quality) .mp3, it's chalk and shit... MP3 sounds horrible. But hey, if it's going on youtube or itunes the audio is crap anyway, but as Bob Katz says, you have to mix for that format (as well) as it's a big part of the audience these days and there are things you can do to to make a .mp3/acc sound how you want.
Not sure how much you know about how A/D convertors and computers work with .wav files etc ? so humour me.
Word based formats: (I'll rave about serial steaming (MP2, ACC) , "loss less (flac ?)" and Sony Super Audio formats another time, not to mention over sampling (fill in the gaps between the samples so can be D/Aed))
Each sample of the signal is turned into a digital "word" e.g. 48khz 24 bit means sample the real analogue signal 48,000 times per second and store the result in a 24 bit (binary) word
So loudest possible volume of sample is 24 ones (111111111111111111111111) and softest is all zeros (00000000000000000000).
CD qual is 44.1khz 16 bit So sample the signal 44100 times per second and store in result 16 bit word
So loudest poss is (1111111111111111) and softest is (0000000000000000) (obviously much less dynamic range than 24 bit as 8 bits less (bit depth, same idea for digital images BTW)).
Hopefully you're not glazing over and this is some use to you LOL
So when you get into maximising the vol, the idea is to take all the bits in a word and move them left (e.g. 0001 > 0010 > 0100 > 1000) while keeping faith with the other samples (gazillions of them) which you are also shifting left (by same amount ?? well not always...). So on a maximised CD you want your loudest sound to be (1111111111111111), What tends to get thrown away is all the peaks (like the first spikes on a snare hit) as they tend to be very full words but not for long. What you'll find is the level you recoded at, while looking a bit low (doesn't fill the screen with signal from top to bottom (like a maximised mix)) is preserving your peaks and dynamic range (or you'd have red lights everywhere). So you loose headroom, life and ambiance and it sounds compressed (coz it is, no peaks anymore (well significantly less)).
Remember I talked about 128 bit backplane for mastering decks ? (11111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111) = 1 word, so now you have some granularity and fu#cks the signal much less.