I wrote: the nyquist/lindquist/someone-else-who-was-pissed sampling theorems are based on the possibility of mathematically extracting frequencies from digital information in a STEADY_STATE situation. That doesn't mean that a speaker will properly reproduce those frequencies. Consider the dynamics of energy transfer. A digital signal at near-1/2-sampling frequency will have two datum points. The transitiion between them will be dramatic! the possibilities of energy transfer will not be comparable to an analogue sinusoidal waveform. .... and i missed a bit or two. Consider the entropic uncertainty of a signal that has two-and-a-bit datums, against a sine wave. Start from zero, and go to such a waveform. Is it a constant-amplitude sine wave at frequency z? or a decaying sine at a frequency (z-at)? There's more, and it's to do with the limits of fourier and sampling theory. Say you have a wave at a frequency of z that's sampled according to nyquist theory. can you distinguish it from a wave of a frequency z - delta z? It can be done, but it takes a while, and a good few samples to do it. And a good analogue system will do it quicker. someone (hopefully not me, i haven't the time just now) can probably apply wavelet theory and get all this from steady-state theory, and tie it up in a nice package. -- Peter Fairbrother