Dear All,
Well, I'm sorry this topic has now reached the stratosphere, but George
has raised some interesting points. The rest of you may go back to sleep,
since (as I said last time) we are now at the leading edge of technology.
Point 1: Yes, I agree it's an artificial test, in the sense that one sample
could not occur in a proper Nyquist-filtered analogue-to-digital converter.
But I was talking about digital-to-analogue converters at the time, and a
default standard test-procedure like I described seems to be evolving to
document the performance of D-to-As. But I fully accept that I ought to have
made it clear that I *include* an anti-aliasing filter in *both* A-to-Ds and
D-to-As; they are in practice (and often physically) inseparable, and their
overall performance should be documented (if only to show cases where
anti-aliasing has been omitted, as in some digital signal processes!)
Point 2: Sorry George, you are just plain wrong here. The widest uncorrupted
frequency range (let's pick a nominal frequency of 22.0499999kHz to show
what I mean) will inevitably mean a very considerable smearing along the
time axis. No filter, analogue or digital, can cut sounds between
22.0499999kHz and 22.05kHz without generating what looks like "ringing" on
(say) square waves. If "ringing" isn't synonymous with destruction of
transient response, I don't know what is! It is possible our moderator may
have something to say here. Analogue video signals must be "brick-walled"
without video artefaces like ringing, and I have often wondered how it is
done! But, for documenting the performance of A-to-D converters (or
D-to-As), *including* their anti-aliasing filters, I cannot think of a
better test than an impulse test.