[sdiy] ADC - anti-alias filter

Sarah Thompson plodger at gmail.com
Fri Mar 24 19:32:40 CET 2017

This is a lot like the sensor signal conditioning stuff I used to handle in
my previous job. Basically, I'd never recommend sampling without filtering
upstream. It *might* work, but for the sake of a little bit of effort you
can go from *might* to *definitely will*.

One of the best reasons to do this is that even if you don't have much in
the way of signal near or above the f/2 Nyquist limit, you can pretty much
guarantee that you will have noise there, which will alias and increase the
noise floor of your signal. So the best option for something like this
would probably be to put a 1 or 2 opamp Sallen-Key filter between the
front-end gain stages and the ADC. It may also be possible to mess with the
gain stages themselves to limit their bandwidth -- done right this might be
enough rolloff on its own. I tended to work on the basis of calculating how
much noise and/or out-of-band signal would give me about half a bit's worth
of peak to peak voltage at the ADC, then figured out (usually with LTSpice)
the minimal amount of rolloff I needed to hit that spec.

The other thing I'd strongly recommend with that circuit is paying a lot of
attention to power supply noise -- this will probably bite harder than
anything else with a lot of gain being used. One of the less talked about
reasons for this is that opamps get their power supply rejection from their
ability to servo the output voltage to correct for power supply variation
-- this works best with low gain, so as you start to max out gain
capability this eats into PSRR.

On Fri, Mar 24, 2017 at 1:13 PM, Steve <sleepy_dog at gmx.de> wrote:

> Howdy @ DSP experts out there,
> I was wondering:
> if done by the book, if you're doing ADC conversions of an audio signal
> and like it to be clean, you put a nicely steep LPF before it that,
> immensely reducing anything >= fsample/2.
> So somebody told me recently, he wants to analyse an audio signal which
> pretty much only has frequencies up to 4kHz, maybe 6 but very quiet, and
> he's interested only in 1..2 kHz of that.
> Hence, he wants to omit the LPF to save parts & PCB space, sample at a
> rate like 15..25 kHz and use a digital filter and then decimate to 4 kHz or
> so.
> Is that really feasible?
> The thing is, the signal is picked up by an analog mic (piezo) with
> 2 opamps after another, each has an integrated PGA of up to 16x gain,
> because the signal get get really quiet.
> Now even if the sound source picked up and the frequency response curve of
> the piezo make sure that the mentioned upper limit holds.
> Could that stuff not pick up noise in some environments which introduce
> frequencies above 1/2 of his higher samplerate and ​hence pollute the
> spectrum?
> Is it *ever* a good idea to omit the analog filter before sampling?
> - Steve
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at synth-diy.org
> http://synth-diy.org/mailman/listinfo/synth-diy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://synth-diy.org/pipermail/synth-diy/attachments/20170324/fdb44e7e/attachment.htm>

More information about the Synth-diy mailing list