Mark Rages wrote:
> Hm, as long as you are making an exhaustive post, you should mention
> the drawbacks of na=EFve implementation of this.  When looking at the
> steady-state response, the output will never get closer than 2**N to
> the input., because when the difference is less than that, the result
> of the right shift is zero.

Only if you make the naive assumption that the values are integers, which I
definitely never said.  In fact, it is common to use fixed point for exactl=
y
this reason.

I didn't get into representation issues because that wasn't the point of th=
e
post.  It gets a bit more complicated when you cascade multiple filters,
which is a common thing to do (more filters provide a faster settling time
at the same random noise attenuation).  Basically you need a
additional -Log2(FF) bits for each filter pole if you want to be able to
truly reach the input value after the appropriate settling time.  However, =
a
careful analisys of the signal to noise ratio usually shows that some of
those bits are meaningless and can be skipped.

>  FILTW <-- FILTW + FF(NEW*M - FILTW), FILT=3DFILTW/M
>
> M needs to be bigger than 2**N.  Practically, M should also be a power
> of two.

Which is just a round about way of saying "fixed point".


********************************************************************
Embed Inc, Littleton Massachusetts, http://www.embedinc.com/products
(978) 742-9014.  Gold level PIC consultants since 2000.
--=20
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist
.