Excactly which frequency weighting is used to calculate the percieved loudness. The info I can find just says something about the midrange frequncies but are not very specific?
Can't tell you exactly, but to a first approximation it will be the A-weighted curve.
The reason that you don't get told much is partly because the exact algorithm is proprietary, but also because it's a good deal more complicated to calculate than just looking at a weighting curve. One of the things that makes it more complicated is that the curve varies at different levels (look up Fletcher-Munsen for more details). The actual calculation for percieved loudness looks far more like an Leq measurement - although I think that the response ballistics are a little different. What this means briefly is that you have to make an estimate of how much sound at different levels there is, and arrive at an overall loudness value from that. And as I've explained before, this is very much a compromise figure, because it can vary from one end of a track to the other.
It's bound to have a more meaningful value on tracks with a lower dynamic range. The wider the dynamic range, the better chance you stand of matching tracks by ear.