By which you mean ultrasonic frequencies (close to the Nyquist limit imposed by the
over-sampling) though. NOT audio frequencies.
--Of course, it's not "Nyquist limited," but rather signals within
20kHz (or whatever your baseband upper frequency limit is) of the
analog sampling frequency and its harmonics...the signals that will
alias into your passband.
If used strictly for audio frequencies, a front end filter truly isn't required. The
rather gentle 6dB/octave LPF most application circuits suggest is simply to filter
any out of band signals. The recommended values for the ADC I've been using give you
a 1st order LPF @ 75kHz. That assuredly isn't going to give filter match issues in
the audio band. And, as I said, it could be left out if you're confident of the
bandwidth of the incoming signal ( i.e it doesn't contain anything in the megahertz
region.
Graham
Well, I'm glad to see that you apparently agree that protection
against aliases, however you do it, is a good thing. Clearly, if my
input has only "audio" frequencies in it, I don't need an alias
protection filter if I sample directly at 44.1kSa/s. You may have the
luxury of dealing with systems in which there is no explicit need for
an alias protection filter; when I was designing with delta-sigma
"audio" converters, I didn't have that luxury. You may even have the
luxury of not having to consider the effects of a first-order filter
with 75kHz cutoff at frequencies below 20kHz, but I also did not have
that luxury. Fifteen degrees phase shift (at 20k) is huge when you're
reporting phase to millidegrees. Even the 40dB attenuation of that
filter by 7.5MHz is inadequate when you're dealing with systems whose
input you guarantee to your customers to be protected against aliases
to 100dB. It's nice to live in a world where you don't have to worry
about little details like that, but in my world, I have to pay
attention to such details.
Cheers,
Tom