I like this...thanks.
Pictures always help....

Mark Skeels
Engineer
Competition Electronics, Inc.
TEL: 815-874-8001
FAX: 815-874-8181
www.competitionelectronics.com


RussellMc wrote:
>> So you can give it new names, but you're really just computing the
>> average of the last N samples.  (now, if there's a lot of difference
>> in samples, you might notice that recent samples have more "weight"
>> than older samples, but that's just a little detail...)
>>    =20
>
> I'll quibble gently.
> I know you know, and your comment further reflects the knowledge, but
> I think a better way of putting it would be something like:
>
> 'The "average" of ALL prior samples + the  latest sample, with the new
> sample being given a "weighting" of 1/Nth of the new "average" '.
>
> Note that "average" here does  NOT have the same meaning as 'average'
> usually does.
> Olin explicitly addressed this with his signature "do you really want
> to keep all N sample". I didn't think so".
> Readers are free to decide whether his "I didn't think so" is valid
> for them - and in many cases it will be - but by making this decision
> you have a MUCH different beast than an average over N samples.
>
> What this is, as has been noted, is a single pole exponential function.
>
> With a true N samples average you start dropping off whole prior
> samples N periods after they have happened. With the exponential
> function you drop off only an Nth of the effect of a sample N+1 ago.
>
> That they can be very different in response is easily seen by graphing
> some typical step responses. I just produced  some Excel graphs with
> varying N lengths and step functions of Y samples =3D 1 and all others =
=3D
> 0.
>
> A graph of only one such is attached.
> Blue curve is input step.
> Octarine curve is true average
> Yellow curve is exponential filter under discussion here.
>
> Do you think they are identical? I didn't think so :-).
>
> Whether this matters depends on application. Input function can make a
> large difference.
>
> I have used the exponential filter extensively in the past BUT it has
> such sluggish performance under some conditions when it is usually but
> not always useful that I sometimes add ad hoc rules to it to modify
> its behaviour. eg priming it with current sample value on input would
> be one possibility (depending on input function).
>
> The exponential function has the advantage of starting up properly
> from 1st sample whereas true averaging needs either a sliding length
> for the 1st N samples or some other handling of its boundary
> conditions. "Horses for courses".
>
>
>         Russell
>  =20
>
> ------------------------------------------------------------------------
>
--=20
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist
.