On Tue, 9 May 2000, Robert A. LaBudde wrote:

> At 12:24 PM 5/9/00 -0400, Andy wrote:
> >I think what you want is a median filter.  Take multiple samples, sort them,
> >then pick the one in the middle of the list.  Your spurious readings are
> >extrema, right?
>
> A more efficient, but just as stable, filter is a 10% or 20% trimmed mean:
>
> E.g., collect a set of fixed size (e.g., 6-10), drop the high and low
> result and average the remainder. With a set of size 8-10, you could also
> drop the next highest and next lowest measurements.
>
> The effective loss of information is only about 1/2 the number of points
> dropped.
>
> Obviously filtering of this type will cut the bandwidth by an equivalent
> amount.


Here's an implementation that I put on my web page for Bob Blick:

http://www.dattalo.com/technical/software/pic/medfilt.html


It does the sorting, median filtering, and averaging. It works with 12-bit data
samples and conserves memory by packing two samples into three bytes.

Scott