32 bit float question
32 bit float question
Sometimes you hear folks over at the Sonar forum comment on 32 bit float as if it is a bad thing, what is the difference between 32 bit and 32 bit float, and is one better overall?
Let's take a 4 digit decimal example. 1489 is fixed point number. A 4 digit floating point version would be 1.49 * 10^3.
Note that the floating point version has lost one digit of accuracy. On the other hand it has a much wider range: the fixed point number can go up to 9999, the floating point version can go up to 9.99 * 10^9.
One isn't better than the other. What matters is the rounding error in the final mix, and it's not easy to say something about that in general terms.
Most audio software uses 32 bit floating point numbers. The biggest problem with 32 bit floating point numbers is that they make developers think accuracy isn't a problem, which isn't necessarily the case. Nowadays some programs use 64 bit floats, which makes it almost impossible to go wrong. Fixed point math is more difficult, the developer has to think about accuracy all the time, which in a way can be a good thing.
MultitrackStudio uses 32 bit fixed point numbers for most audio processing, but 64 bit numbers are used in critical places to avoid rounding errors. There are, for example, 64 bit fixed point numbers in all dynamic effects (Compressor, Deesser etc.) and the EQ. The Master Limiter uses some 64 bit floating point numbers.
Giel Bremmers
Note that the floating point version has lost one digit of accuracy. On the other hand it has a much wider range: the fixed point number can go up to 9999, the floating point version can go up to 9.99 * 10^9.
One isn't better than the other. What matters is the rounding error in the final mix, and it's not easy to say something about that in general terms.
Most audio software uses 32 bit floating point numbers. The biggest problem with 32 bit floating point numbers is that they make developers think accuracy isn't a problem, which isn't necessarily the case. Nowadays some programs use 64 bit floats, which makes it almost impossible to go wrong. Fixed point math is more difficult, the developer has to think about accuracy all the time, which in a way can be a good thing.
MultitrackStudio uses 32 bit fixed point numbers for most audio processing, but 64 bit numbers are used in critical places to avoid rounding errors. There are, for example, 64 bit fixed point numbers in all dynamic effects (Compressor, Deesser etc.) and the EQ. The Master Limiter uses some 64 bit floating point numbers.
Giel Bremmers