I like how this thread suddenly buzzes with science. ;-) Engineers doing some engineering, I guess.

Isn't it impossible to distinguish signal and noise on a per-sample basis, though? A sample has no context, only a frequency. By itself, it can be both - no?