MP3 encoding relies on the psychoacoustic principle of “masking” – the idea that a tone played at a particular frequency will raise the threshold of hearing for frequencies on either side of it.
If you extend this idea to a complex signal, e.g music rather than a single tone, you’ll find that much of the replayed sound is masked (and therefore redundant as far as we are concerned – we can get rid of it without hearing a difference).
MP3 works by splitting the audio signal into a number of frequency bands and examining it to find out what parts will be inaudible due to masking (the encoder has a model of the human auditory system built into it). The bit-depth of each band is continuously varied (lower bit-depth = more noise) according to how heavily it will be masked at any one point. I.e. it might be possible to get away with encoding a band at 1-bit if the massive amount of resulting noise will be masked, saving 15 bits per sample.
I sometimes wonder what MP3s must sound like to a dog or other animal – pretty ropey I’d imagine.
That’s the basic premise of MP3 – I can’t remember the exact ins and outs of how it works.
As for an MP3 signal being mono below a certain frequency I don’t think that’s the case – I don’t see why it should be. Incidentally this is something that is often done by the mastering engineer – it was necessary to mono the bass when pressing music to vinyl, as loud, out-of-phase low frequency signals can cause the needle to jump out of the groove (the difference between the L and R signals (i.e. all stereo information) is represented by the vertical movement of the stylus).
