At 12:23 PM 8/19/2005 -0400, you wrote:


>Of course, this is easily done with fixed point as well. The difference
>is people are more careful with fixed point. They consider the range of
>numbers, the end cases, and choose a fixed point position base on those
>results.
>
>With FP people tend to think "I can handle ANY range" and end up causing
>very bad things to happen.

Another thing-it's not hard to get into trouble in real situations using
the "small" FP representations (like 24 bits). If you're just doing m*x+b
type stuff, then the problem never crops up, but if you're evaluating
an ill-conditioned polynomial etc. it's pretty easy to lose enough
bits to cause problems with real-world 3-4 decimal digit numbers. This is
not "round-off error". Conceptually, what happens is that two almost-equal
numbers in a partial result are subtracted, leaving many fewer bits than
were there originally. If possible, a better algorithm is one answer, but
that's not always convenient (for example, it may consume a LOT more code
memory if you substitute many low-order cubic splines for one or two poly
fits).

The FP math will be at least as good as the fixed-point math if the
number of bits *in the mantissa* is at least as many as the bits in the
fixed-point numbers. The bits devoted to the exponent are essentially wasted
as far as this issue goes, since you handle the scaling manually during
programming.

Personally, I find myself using FP math more these days in instrumentation
and control firmware. Back in the old days, when every byte and every cycle
counted, it was most undesirable in most applications, but now it has its
place.

>Best regards,

Spehro Pefhany --"it's the network..."            "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com
->> Inexpensive test equipment & parts http://search.ebay.com/_W0QQsassZspeff


-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist