32 bit floating point

Posted by: Mike-B on 22 February 2015

Anyone done any testing using 32 bit floating point with CD rips &/or .flac, .wav conversions?

Apart fro 50% bigger file size than 24 bit,  are there any audible SQ changes ??

Posted on: 22 February 2015 by bicela

Hello Mike, I... is just a test... ripping SACD to .iso then extract tracks and recode to what is possible for now to use: .wav 32bit @ 192kHz.

 

First try, sound amazing... But I have not direct comparison with the same recoded to 24bit or 16bit i.e.

 

In the next week I would do some test.

Posted on: 22 February 2015 by sjbabbey

Hi Mike,

 

Only heard of 32bit FP used to provide extra headroom for audio editing (e.g. audacity) prior to resampling (dithering?) down to 24bit. Naim streamers specs seem to have limit of 24bit 192kHz for streaming although the spec for NDac indicates it can convert 32bit digital input.

 

Interestingly the blurb for the new NAC-N 272 reads:

 

"DSD64 can be played via UPnP, electrical digital or USB inputs, and DSD playback is kept as pure as possible by stripping out the digital header information, leaving only the audio to be optimised for the NAC-N 272’s DAC by integer resampling it to 40-bit/705.6kHz using Naim software, then feeding it to the 24-bit DAC, bypassing the digital filter."

 

I'm no expert but perhaps the initial resampling to 40-bit is similarly to allow additional headroom prior to being fed to the internal DAC.

Posted on: 22 February 2015 by Loki

32bit floating point? I'm sure my Apple Mac LCIII had that in the early 90s 

Posted on: 23 February 2015 by Huge
Originally Posted by Loki:

32bit floating point? I'm sure my Apple Mac LCIII had that in the early 90s 

Unless you had the FPU upgrade, then it was a software emulation*, rather like DSD on Naim equipment.

 

N.B. that's not to say that the software emulations are flawed in any way!  Slower, yes, but still perfectly fast enough for audio use.

 

 

* Unless the versions distributed in Valhalla came already fitted with all the add-on parts, as surely befits the einherjar!

Posted on: 23 February 2015 by Mike-B

I've dug a bit (no pun) deeper & concluded 32 bit is only useful at the recording end.

Ripping to 32 does nothing other than take up 100% more files space than 16 bit  50% more than 24 bit.  It does not change the original data resolution.         

Posted on: 23 February 2015 by Huge

Mike,

 

If you convert native 32bit WAV you will actually loose resolution with 32bit IEEE Floating Point.

 

The IEEE 32bit float has a 23bit mantissa +1 sign bit; i.e. it's resolution is exactly 24bit.

 

H

Posted on: 23 February 2015 by bicela

Dear Mike and dear Huge, your last post are very interesting.