wrote in message
...
"As long as filtering is necessary, it might
as well be a narrow as the widest signal of interest and as sharp as
possible so long as it's convenient and doesn't distort the signal too
much."

"As sharp as possible" and "doesn't distort the signal too much" are somewhat
conflicting goals: In general, the steeper the skirts of a filter, the more
group delay variation you get there at the edges (hence, Butterworth has less
group delay variation than Chebyshev which has less than Elliptic). Now, you
can certainly account for this by widening the passband a bit and then perhaps
using even steeper skirts, or you can compensate for it digitally if you can
characterize it, but the main point here is that it does get rather complex --
hence the trend to have somewhat "looser" analog filters (and thus low group
delay variation) and then do whatever you want digitally.