[NBLUG/talk] excess-50

Mitch Patenaude mrp at sonic.net
Mon May 26 21:08:01 PDT 2003


> "Using excess-50 floating decimal numbers with
> 8 digits,...."
>
> What is "excess-50"?

I'm just guessing here, but I think it has to do with the way floating 
point numbers are encoded.
Each floating point number has 3 parts, the sign (usually one bit), the 
mantissa and the exponent

If we're using base 10, then a number written in scientific notation 
will look like
6.67*10^-11
The sign is positive, the mantissa is 6.67 and the exponent is -11.    
In order to represent a negative mantissa, you just set the sign bit.  
But most IEEE floating point standards don't do the same with the 
exponent.  Rather than encode the exponent as a signed number, it's 
encoded as an unsigned number, but a value is subtracted off before 
it's evaluated.  So, if that value was 50, then the exponent would be 
stored as 39 rather than -11.  (in real floating point, all this is 
done in binary).  I believe that this is what he's talking about as 
"excess"... it's the excess of the exponent that needs to be removed 
before you evaluate the exponent.

(I think this is done because typical users want more ability to 
represent very large numbers rather than very small, so this allows a 
bias in the range of possible exponents.)

   -- Mitch




More information about the talk mailing list