Dmitry Kiryashov wrote:

> There are more short version algorithm above ;) No additional cell
> required
>
> avg = (avg + 3*sample)/4 = ( ( avg + sample )/2 + sample )/2
>
>         MOVFW   SAMPLE
>         ADDWF   AVG,F
>         RRF     AVG,F
>         ADDWF   AVG,F
>         RRF     AVG,F
>
> 5 Cycles
>

Excellent observation Dmitry!

And in a personal message,
Bob Fehrenbach wrote:
>
> Hey Scott,
>
> >Have you considered an IIR (infinite impulse response) filter?
> >Specifically, I've had good results with filters of this type:
> >
> >   avg = (m*avg + n*sample) /(m+n)
>
>    Without additional 'round off' code, my experience has been that
>    these filters tend to oscillate when presented with constant data.
>    In some cases it was only one bit but a nuisance nevertheless.

Actually, they tend to round down such that in the example of my
previous message where m+n=4, the average will tend to be 1 to 3
counts too small. For example suppose that the current average is
4 and the input sample is 5. Then after an iteration,

 avg = (avg + 3*sample)/4
     =  (4 + 15)/4
     = 4

The fractional remainder (of 0.75) is lost.

Where as with Ralf's desire of taking the last 8 samples, a constant
input of '5' will eventually produce an average of '5'.

It's something certainly to keep in mind. If your signal jumps around
wildly then it is not a big deal. If you're trying to get a couple of
more bits out of an A/D converter by integrating samples, then this
is not the method you would choose.

Scott
--
"The problem with television is not the resolution."
                                 Hugh. F. Frohbach