----- Original Message ----- 
From: "Bruce Partridge" <bgpartri@ca.inter.net>
To: "Microcontroller discussion list - Public." <piclist@mit.edu>
Sent: Thursday, November 04, 2004 9:02 AM
Subject: RE: [PIC]fwd: Floating Point Underflows and Overflows


> The application doesn't nornally deal with this kind of dynamic range.  I
> will never have any numbers that have exponents outside the range of +4
> to -7.  Anything above +4 is an error, and anything below -7 is just zero.
>
> But the +35 value occurs very occasionally.  Its a decay equation, so the
> results stay around for a day or so.  In searching for possible reasons, I
> discovered that I could produce the result with that particular
> multiplication.  There are probably others.
>
> So my next question is whether I am actually creating those numbers.  The
> only way that I can imagine them being created is if I am doing 
> calculations
> on values that would just be zero in integer calculation.
>
> For example 7.520001 - 7.520000 could easily appear in my calculations.
> Maybe doing the subraction on the internal values produces very small
> numbers.  But I'm just guessing on this part.

Yes, this is a common problem with using floating point. You lose precision 
very quickly when subtracting to numbers that are very close to one another.

Perhaps all you need is something like this:

       diff = a - b;
       if ( diff < 1E-10 )
            diff = 0;

(note that the same thing can happen when adding number of similar absolute 
value and opposite signs)

Bob Ammerman
RAm Systems

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