why not floating pont?

We all know there are no devices in common use that produce floating point audio data but many of us recording to floating point format. Floating point is just the on-computer container for the data, it says nothing about the precision or format of the input arriving at the recording application.

In another place a new user was asking about recording from LPs. His recording application, Sound Forge, will not record in floating point, apparently because of the soundcard. The program does have a 32 bit floating point option, but it will not function. The message reported is

An error occurred while opening an audio device. An unsupported media type was requested.
USB Preamp (USB Audio Codec) does not support 32-bit floating point input.

He reports he can select "32 bit PCM" which my experience can only interpret as 32 bit integer.
If this just some strangeness of the application or could there be some real hardware limitation that prevents capture in floating point format?

This inquiry has nothing to do with the usefulness of any particular bit depth for LP captures.

I think this is just a "feature" of Sound Forge. The recording part checks the capabilities of the input device(s) when the program opens, and this is flagged as the capability of the recording.

This means that the recording device controls the format of the file you are recording, rather than SF itself.

It does exactly the same here - I can only record in 16 or 24-bit, although it will playback a floating point file. If I open a new file as 32-bit floating point, I can't actually record into it with anything.

You can also apparently cut and paste between different format files. I haven't tried this to see how it copes or what the effect is.

 PS. You can of course convert the file manually to FP by using the bit depth converter. It's a strange program....

He reports he can select "32 bit PCM" which my experience can only interpret as 32 bit integer.
If this just some strangeness of the application or could there be some real hardware limitation that prevents capture in floating point format?

All hardware, regardless of what it is, outputs in integer form. That's all it is physically capable of. If something like Audition requests 32-bit floating point format, then it's the driver that provides it, not the sound device. All that happens is that the integer part is converted to a normalized value, and the exponent is filled with zeros. It may look like 32-bit floating point is being recorded, but actually, it isn't. Well I suppose that in one form it is, but not one that has any advantage at the recording stage. If you record in 24-bit and convert to 32-bit floating point afterwards, the result is absolutely identical.

We all know there are no devices in common use that produce floating point audio data but

I know I've never seen one, but according to Pohlmann, in his fourth edition of Principles of Digital Audio, they exist, and they can offer certain advantages. His discussion of their functioning starts on page 105.

We all know there are no devices in common use that produce floating point audio data but

I know I've never seen one, but according to Pohlmann, in his fourth edition of Principles of Digital Audio, they exist, and they can offer certain advantages. His discussion of their functioning starts on page 105.

You've probably misunderstood it. Whilst it's technically possible to produce floating point signals mechanically, you can only convert them into that form from something that determines discrete levels in the first place - so no real advantage is obtained. It's also a lot of hard work - and that's why all the world's best converters simply don't do it, but rely sometimes on devices like the Texas TMS320C6711D to do it for them. Still a lot easier and cheaper to let the software do it, though...