What is and what isn't 32bit float?

[Prime Mover]

I was under the impression that signal path in DP is pretty much purely 32bit floating point until it exits the master. i've never thought twice about throwing a trim plugin on, lowering it by -4dB, and then using the fader for doing automated volume changes. My thought is: if the trim plugin is set to -4dB, and the volume favor is set to +4dB, then the result would be an EXACTLY the same as it would be if you set both to 0. But now I'm not so sure, especially after seeing the "32bit float" dropdown option.

What exactly *IS* the "32bit float" option anyway, if recordings aren't made in it, and as I assume, most of the signal path is float, anyway?

[BKK-OZ]

I found this thread helped me understand the topic a bit more.

Also some good stuff here.

[Prime Mover]

Yeah, I read both those topics, and found some material from the manual (gee, why didn't I do that first?) It appears that contrary to some opinion, yes, you can record at 32bit float, and that's exactly what that option is for. And yes, the entire DP signal chain is 32bit float all the time (hooray, I haven't been screwing up my projects for the past 5 years!)

All I gotta say though is, why would ANYONE want to record at 32bit? I know I'm jumping on the bandwagon with this, but that just seems ridiculous. When a "silent" room has a noisefloor of 30dB, and you're neumann microphone has a 5dB noisefloor, what do you think you're going to pick up? I guess if you want to pick up the sound of a George vomiting 4 blocks away from inside your acoustically treated recording studio, I guess that's what it's there for.

[n2mpujack]

Yeah, I read both those topics, and found some material from the manual (gee, why didn't I do that first?) It appears that contrary to some opinion, yes, you can record at 32bit float, and that's exactly what that option is for. And yes, the entire DP signal chain is 32bit float all the time (hooray, I haven't been screwing up my projects for the past 5 years!)

All I gotta say though is, why would ANYONE want to record at 32bit? I know I'm jumping on the bandwagon with this, but that just seems ridiculous. When a "silent" room has a noisefloor of 30dB, and you're neumann microphone has a 5dB noisefloor, what do you think you're going to pick up? I guess if you want to pick up the sound of a George vomiting 4 blocks away from inside your acoustically treated recording studio, I guess that's what it's there for.

You're confusing bit depth and computer math calculations in the last paragraph. There's no such thing as a 32 bit a/d converter so you cannot possibly record anything analog at that bit depth. Generally when one sees the term 32 bit float that refers to computer math calculations which can be either floating or fixed point. That's not to say DP isn't doing an up-convert of the incoming signal to 32 bit bit depth though (which would properly be called over sampling).

[Shooshie]

Your audio input will be either 16 bit or 24 bit, but as soon as two tracks combine, or as soon as a plugin operates on the EQ or compression of a sample, you may end up with audio that begins to fill out that 32-bit format. The more you mix, the more you utilize 32 bits. When you bounce any of that to disk, DP has to truncate bits to squeeze 32 bits down to 24bits. That's where a 32 bit audio format may be helpful, but eventually, if anyone else is to hear it, someone is going to have to do the job of getting it down to 24 bits or more likely, 16 bits. That might as well be you so that you have some control over the outcome.

Don't worry about it too much. Every CD ever sold runs at 16 bits, so the bar is still set pretty low.

Shooshie

[Prime Mover]

You're confusing bit depth and computer math calculations in the last paragraph. There's no such thing as a 32 bit a/d converter so you cannot possibly record anything analog at that bit depth. Generally when one sees the term 32 bit float that refers to computer math calculations which can be either floating or fixed point. That's not to say DP isn't doing an up-convert of the incoming signal to 32 bit bit depth though (which would properly be called over sampling).

This is incorrect. According the the manual (pg 233), 32bit float mode *DOES* record at 32bits. The math calculations are universal, no matter whether you have 32bit float mode selected or not. Even if you're running in 16bit int mode, the internal engine is all 32bit float. That mode selector that we're talking about is recording mode, and yes, it does make 32bit audio files. Both the manual and Magic Dave have reiterated this.

[n2mpujack]

You're confusing bit depth and computer math calculations in the last paragraph. There's no such thing as a 32 bit a/d converter so you cannot possibly record anything analog at that bit depth. Generally when one sees the term 32 bit float that refers to computer math calculations which can be either floating or fixed point. That's not to say DP isn't doing an up-convert of the incoming signal to 32 bit bit depth though (which would properly be called over sampling).

This is incorrect. According the the manual (pg 233), 32bit float mode *DOES* record at 32bits. The math calculations are universal, no matter whether you have 32bit float mode selected or not. Even if you're running in 16bit int mode, the internal engine is all 32bit float. That mode selector that we're talking about is recording mode, and yes, it does make 32bit audio files. Both the manual and Magic Dave have reiterated this.

I'm not so sure because as I said there's no such thing as a 32 bit a/d converter. What I think DP is doing what I mentioned before - it's upsampling to 32 bit.

[Prime Mover]

That doesn't make any sense. If we know that DP is already processing everything at 32bit, then what is the point of upconversion? It would produce larger files with extra 0-bits for the sake of it processing it the same way it would process it anyway. I'm not convinced of your claim that there is no such thing as a 32bit ADC, all signs point to the fact that that's exactly what's going on.

Either way, I'm leaving it in 24bit mode, 32bits is a waste of space, IMO.

[Phil O]

I could be wrong on this, but it has always been my understanding that in 32 bit floating point, the mantissa uses 24 bits, and the other 8 bits are for sign and exponent. So a 24 bit converter would work nicely with this arrangement.

DP now has 32 bit floating point as a file format option. It's always had 32 bit floating point as it's internal engine. I've been using this option for recording lately, and although some might consider it overkill, I've been getting excellent results, so I ain't arguing with success. I don't really know if there's an audible difference, but I don't really care. This format has worked so well for me, that I'm gonna be using it for a while.

Phil

[David Polich]

That doesn't make any sense. If we know that DP is already processing everything at 32bit, then what is the point of upconversion? It would produce larger files with extra 0-bits for the sake of it processing it the same way it would process it anyway. I'm not convinced of your claim that there is no such thing as a 32bit ADC, all signs point to the fact that that's exactly what's going on.

Either way, I'm leaving it in 24bit mode, 32bits is a waste of space, IMO.

Just to clarify - there is no such thing as 32-bit audio converter.
Converters operate at fixed points. 32-bit is floating point.

If there was such a thing as a 32-bit audio converter, you'd see the
option to set its format to such. As far as I know, no such audio
converter exists at this time.

32-bit is a floating-point math process. Not a fixed-point audio
format.

[OldTimey]

while most A/D converters available to us work at a fixed bit-depth (usually 24-bit) some of these converters may have on board DSP that operate at 32, 64, or even higher bit-depths. If you use these processes destructively and record the result in DP, I imagine it'd be smarter to work at 32-bit so as to not truncate the result to 24-bit.

BAD: analog signal--->24bit (A/D)--->32/64bit (onboard DSP)--->24bit (digital performer)

BETTER: analog signal--->24bit (A/D)--->32/64bit (onboard DSP)--->32bit (digital performer)

I could be wrong but check this out from MOTU.com regarding the Compressor DSP included on the 828mk3:

The Leveler faithfully models the LA-2A using the on-board DSP with 32-bit floating point precision. Its controls match the front panel of the original LA-2A: Gain Reduction and Makeup Gain knobs, along with Limiter/Compressor buttons.

so i might be onto something here....

[Don T]

Hello,
I just wanted to add that some posters are intermixing the terms 32 bit and 32 bit floating point. These are two entirely different things that some readers may find confusing.

For clarification:
16 bit = 90db signal/noise (1 bit of noise)
24 bit = 138db s/n (possible but not probable)
32 bit = 186 db s/n
There was a 32 bit (fixed) option in Cool Edit back in the P2 / 32 bit pci slot days.
32 bit floating point = 1500db s/n



32 bit float is the standard on board math coprocessor function. It provides high resolution so the plugins are not truncating data and you don't have to worry about clipping except for non native plugs. The DAW has to establish a STD communication protocol to communicate with plugs not bundled natively. You get to choose (16/24). So even when you have a lot of dynamic range with 32 bit floating point, it still has to be bottlenecked back to the bit rate you select either within plugs or at the end when bouncing the file (& listening). And ultimately down to 16 bits for CD.

[David Polich]

while most A/D converters available to us work at a fixed bit-depth (usually 24-bit) some of these converters may have on board DSP that operate at 32, 64, or even higher bit-depths. If you use these processes destructively and record the result in DP, I imagine it'd be smarter to work at 32-bit so as to not truncate the result to 24-bit.

BAD: analog signal--->24bit (A/D)--->32/64bit (onboard DSP)--->24bit (digital performer)

BETTER: analog signal--->24bit (A/D)--->32/64bit (onboard DSP)--->32bit (digital performer)

so i might be onto something here....

Wow, this really is confusing to a lot of people.
Let's clarify again -

In DP, you are ALWAYS at 32-bit floating point. And that is with the
audio set to either 24-bit or 16-bit. The 32-bit floating point is a
MATH process. Not an audio process. Think of it this way - your 24-bit audio files are floating serenely inside the 32-bit floating-point
"sphere" that is DP. They are still 24-bit, will never be 32-bit or 16
or anything else other than 24-bit audio files. Think of them as little
pieces of plastic floating suspended in water. The water is the 32-bit
environment. The plastic bits are the audio files.

As we all know (or should know) - a DP project can contain both 16-bit
and 24-bit audio tracks. Now how would that work? Because once inside DP,
the bit-depth of the audio files is of no concern to DP. It is processing
everything using 32-bit floating point...mathematically that is. NOT as
32-bit audio.

Can you record everything at 32-bit? Yes, in DP you can. If you feel like
doing so, then by all means. Me, I'm staying at 24-bit which gives me one
less headache to think about, and it sounds perfectly fine to me.

[jloeb]

In DP, you are ALWAYS at 32-bit floating point. And that is with the
audio set to either 24-bit or 16-bit.

Unless you bounce, freeze or print a submix. Then, if your session rate is 24 or 16, you have dithered back down to whatever your session rate is.

If you have set the session rate to 32 bit fp, then your audio is printed to a full 32 bit fp resolution audio file, remaining at the bit depth at which DP processes audio as it passes through faders and fx chains.

The 32-bit floating point is a
MATH process. Not an audio process. Think of it this way - your 24-bit audio files are floating serenely inside the 32-bit floating-point
"sphere" that is DP. They are still 24-bit, will never be 32-bit or 16
or anything else other than 24-bit audio files. Think of them as little
pieces of plastic floating suspended in water.

This is true, until you print an audio file by bouncing, freezing, submixing or otherwise mixing down. Then, unless the session rate is 32fp, the audio has to come back down to earth.

So the simple solution is:
-Start with 24 bit source files.
-If you're going to print any files inside the session (freezes, submixes etc.), then do it at 32 bit fp session rate.
-When your mix is in final form, mix the final product down with dithering at whatever bit rate you need it to be.

[Don T]

Thanks David, your explanation is clearer than mine.