Jump to content

32 bit floating point bit depth recording


Tommy T

Recommended Posts

As far as I know there are no converters that convert directly to floating point. They all sample in integer format. So even if the software lets you record and save the data in floating point format it is being converted from integers. That seems to be the same thing most DAWS do anyway. 

The extra Dynamic Range of a floating point number would give you no benefit down at the ADC level. Getting 24 bits from an Audio ADC is pretty challenging as it is.  Even though the ADC chips list specs that seem to use all the bits those generally are done on a special test board with expensive clocking and other test equipment.

 

Link to comment
Share on other sites

Thank you for the response. You're right about the converters only sampling in integer format. At the same time, the reasoning for 32 bit floating point record bit depth is "precisely" because of the extra Dynamic Range benefit down at the ADC level.  32bit integer gives a max of 192db (really good obviously), but 32 floating points yields greater than 1500db. It is in this way the floating point sample with literally deliver a cleaner sound regardless of the ADC simply because of the massive decrease in subsonic noise. Of course you can't hear it ( just like 50khz of vinyl records) unless your ears are trained, but you can definitely hear and feel it. Thanks again.

https://www.sounddevices.com/how-is-a-32-bit-float-file-recorded/

How Is 32-Bit Float Different From 32-Bit Integer?

32-bit float signals should not be confused with 32-bit fixed point. A 32-bit fixed point signal can accommodate considerable dynamic range, roughly 192 dB. While 32-bit integer offers a much larger dynamic range than 24-bit audio’s theoretical 144 dB, it pales when compared to the 1500 dB available to 32-bit float signals. If, for instance, processing adds 90 dB of gain to the 142 dB of dynamic range coming in, then 232 dB would be needed to capture this signal with no loss in headroom or noise. 32-bit float can do this, and 32-bit integer cannot.

Link to comment
Share on other sites

8 hours ago, Tommy T said:

I would love to have a 32 bit floating point recording bit depth option available in the audio data menu to accompany the 16, 24, 32, and 64 bit recording depths like Cubase and others. Icing on the cake.

Bit Depth vs. Internal Processing.png

I've experiment with this and it was awesome! Love the dynamic range and headroom, but I kinda miss the distortion with 24bit. It helps to shape the sound for me. Though this will be an advantage to CbB - your listeners wont know the difference. 

I say add it for those who want it, but don't need it. For me 24bit - 48000Khz sample rate brings the candy. 

Link to comment
Share on other sites

On 12/5/2020 at 6:52 PM, Tommy T said:

Thank you for the response. You're right about the converters only sampling in integer format. At the same time, the reasoning for 32 bit floating point record bit depth is "precisely" because of the extra Dynamic Range benefit down at the ADC level.  32bit integer gives a max of 192db (really good obviously), but 32 floating points yields greater than 1500db. It is in this way the floating point sample with literally deliver a cleaner sound regardless of the ADC simply because of the massive decrease in subsonic noise. Of course you can't hear it ( just like 50khz of vinyl records) unless your ears are trained, but you can definitely hear and feel it. Thanks again.

https://www.sounddevices.com/how-is-a-32-bit-float-file-recorded/

How Is 32-Bit Float Different From 32-Bit Integer?

32-bit float signals should not be confused with 32-bit fixed point. A 32-bit fixed point signal can accommodate considerable dynamic range, roughly 192 dB. While 32-bit integer offers a much larger dynamic range than 24-bit audio’s theoretical 144 dB, it pales when compared to the 1500 dB available to 32-bit float signals. If, for instance, processing adds 90 dB of gain to the 142 dB of dynamic range coming in, then 232 dB would be needed to capture this signal with no loss in headroom or noise. 32-bit float can do this, and 32-bit integer cannot.

Keep in mind that, without a converter that can handle those, your end result will just be truncated or converted anyways. This is similar to how some people record on 64 bit but don't have 64 bit converters. Sure, you end with a 64 bit wave file, but the extra 32 bits are filled with zeroes. It would be same thing as taking a 32 bit file and upsampling it to 64 bits. As per dynamic range...Humans become permanently deaf when exposed to signals louder than 150 dB.

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...