Randy Yates wrote in message ...
[re. non-linear DSP algorithms]
One could argue that such
algorithms should upsample, do their non-linear processing, and then
downsample,

Not if you want to waste processing power needlessly performing
operations that will have no audible effect.

I don't understand your statement. Non-linear processing will produce
harmonics of the same order as the non-linearity's highest
power-series expansion. Therefore some upsampling is required in
order to perform such DSP without aliasing. Downsampling is then
necessary to return to the playback rate. Level compression is an
example of a non-linear process.

Whatever we may feel about high sampling rates, they are here, and
devices will need to deal with them. If you want to decimate them all
to 48 kHz, that's fine. I'd rather avoid one more filter in the
processing chain --- it's just one more place to screw up. Another
issue that hasn't been addressed yet are the possible ameliorating
effects of higher sampling rates on processing errors and bugs.

--Andre