John Payson wrote:

> I think the best approach probably is to keep a buffer of the last few
> samples (e.g. six) and then continuously take the average of the middle
> four (i.e. sum all six and subtract the minimum and maximum).  This would
> weed out single extreme points while providing more smoothing than simply
> doing median-of-three.  Note that finding the minimum and maximum of a
> set of 8-bit numbers is very quick; for longer sizes things are a bit
> harder.

BINGO! This is what I tried to say yesterday (obviously not too well). I
used this type of algorithm for part of that DTMF decoder we continually
elude to. However, like the DTMF decoder, I can't divulge the
median/averaging combo filter. All I can say is that it works.