KVR Audio

morelia · Post by **morelia** » Fri Jul 23, 2004 9:03 pm

Can anyone of you geniuses tell me what "floating point" refers to when used as in "32 bit floating point", In laymans term if that is possible

Just curious.

Thanks

groovy · Post by **groovy** » Fri Jul 23, 2004 9:20 pm

Floating point number is a number with a decimal point where the point can move, a number can be represented as fractional parts. That is the number can be 1.00001, or 2.55555, or 95522041.1456.

morelia · Post by **morelia** » Fri Jul 23, 2004 10:32 pm

Thanks. I am not sure that I understand completely but maybe I never will without going into it more deeply.

tee boy · Post by **tee boy** » Fri Jul 23, 2004 11:33 pm

As i understood it, floating point is a method which deals with fractions while fixed point deals only with whole numbers. Fixed point represents numbers in binary, and can often use the LSB (or is it MSB, i forget) to represent whether the number is negative or positive. The most common method i believe in fixed point is a similar system called two's compliment. This also allows for negative numbers but i wont go into it now. You with me? Ok then, floating point deals with numbers in a totally different way. Rather than simply representing the decimal number in binary, it represents the number in scientific notation - for example 1.01 X 10. I couldnt tell you which method is better, but Im sure both have advantages and disadvantages. Like Im pretty sure that fixed point is less prone to errors. What i can tell you is that 32bit floating is better than 24bit fixed for audio (or at least according to a dsp guy i know). Everytime i mention PT he hits me with a loads of dsp jargon explaining why Nuendo is inherently better than PT because of this very point.

But please dont quote me on any of this, cuz i am NO dsp guru. I had a class on this stuff at uni, and i remember very little (it was first thing on a Monday

).

Spe3D · Post by **Spe3D** » Sat Jul 24, 2004 12:01 am

http://www.research.microsoft.com/~holl ... float.html

http://www.psc.edu/general/software/pac ... /ieee.html

http://nssdc.gsfc.nasa.gov/nssdc/format ... 2-Bit.html

VicDiesel · Post by **VicDiesel** » Sat Jul 24, 2004 3:03 am

groovy wrote:Floating point number is a number with a decimal point where the point can move, a number can be represented as fractional parts. That is the number can be 1.00001, or 2.55555, or 95522041.1456.

In the most literal sense, yes. But that's not very illustrative of how floating point is actually implemented.

In practice, a floating point number is a number of the form "0.something times 2-to-the-something". With 32 bits, the first something takes up 20 or so bits (the mantissa), one bit for sign, and the remaining 11 for the second something (the exponent). Woops: also a sign bit for the exponent.

That's it.

The good thing is that you can reprensent both very large and very small numbers, from 2-to-the-max-exponent to 2-to-the-min-exponent.

The bad thing is that those numbers are not regularly spaced, which leads to all sorts of interesting discussions about noise floors and such.

V.

Floating Point