What is 32Bit Floating point

[vpunk]

I should probably know this already, but there's no such thing as a stupid question - right?

 

Could someone please tell me what 32 bit floating point means? Why is this so terrific?

[shivermetimbers]

32-bit floating-point mathematics provides enormous headroom for internal processing, making it nearly impossible to overload the signal — even during mixdown to 16-bit audio for CD release. In other words, Logic Pro 7 delivers excellence.

 

 

[vpunk]

So does anyone know how the 32bit works to do this?

Shivers?

[Ashermusic]

So does anyone know how the 32bit works to do this?

Shivers?

 

http://emusician.com/tutorials/emusic_daw_math/

 

And btw, yes there certainly is such a thing as a stupid question but your wasn't one

[David Nahmani]

OK very quickly:

 

In digital audio, you sample the value of an analog electric signal coming from the mic at regular intervals. Each sample is coded with bits. A certain number of bits gives you a discrete number of values you can code. Any analog value between those discrete numbers will have to be "quantized", or approximated to the closest digital value. The more bits you have, the less quantization errors you get, and the more faithful the digital representation of your analog signal.

 

So 16 bits is great, and 24 bits is awesome.

 

But still, if you're using a lot of processing, with 24 bits you could overload a plug-in or your summing bus for example. So to give you even more dynamic range, Logic's audio engine (and for that matter, most native audio engines out there) give you 32 bit floating point. So now you can use 24 bits to code the value of the signal, and you have 8 bits left to code the range of that value.

 

Now for a minute, let's say we're working with decimal numbers instead of bits, and let's say you work with 3 "bits" fixed point.

 

You can code 001, 002, ... up to 999. If you have 999 and one of the processing plug-in wants to amplify that signal by a factor of 10, then you'd have to code 9,990, which you can't. You're overloading.

 

Now let's say you work in 5 "bits" floating point. 3 "bits" for the precision, 2 "bits" for the resolution.

 

Now you would have:

 

999 (x01)

 

Multiply that by 10, and you get.

999 (x10)

 

You don't lose any resolution, and you don't overload.

 

OK this was really an oversimplification so there are a lot of things that a mathematician would say are wrong in this post, but I wanted to give you an idea of how it works.

[vpunk]

Thanks David!

Always on the ball.

 

Last summer I was recording a vocalist with bad control of her dynamics, and this caused serious clipiping - shouldn't the 32bit FP prevented this?

I was recording with a motu 828MK2.

[David Nahmani]

Last summer I was recording a vocalist with bad control of her dynamics, and this caused serious clipiping - shouldn't the 32bit FP prevented this?

No, the recording is done in 16 or 24 bit fixed point depending on your settings. What would prevent the clipping is to use 24 bits and turn down the gain on the mic preamp.

 

32 bit floating point is only for internal processing within Logic.

 

When recording audio, Logic has the stupidest job in the world: take the 0s and 1s coming from the converter, and place them in a computer file. That's it. Period. Any clipping occured in your hardware, at the input of the converter. Nothing Logic can do to prevent that!

[vpunk]

I see the light . . ..

 

Thanks again!