You are here

The "float/fixed" debate and more digital myths

For everything after the recording stage: hardware/software and how you use it.

Re: The "float/fixed" debate and more digital myths

Postby merlyn » Fri Jan 10, 2020 4:06 pm

We never actually hear a 32 bit float. A to Ds and D to As are 24 bit.

This ties into your question earlier about what happens to a 32 bit float signal at +20dBFS. That can only happen inside a DAW.

Sampled data (24 bit) -> internal processing (32 bit float) -> reconstruction (24 bit)

By applying a gain of 20dB to a track the internal bus could have a level of +20dBFS but if you send that to the master bus with the master fader at 0dB it's going to be a distorted mess. The master fader would have to be at -20dB.

The master fader is dividing by a constant, in this case 10. It brings the 32 bit float back into the -1 to +1 range. Tracks can go over, but the master bus can't because it is outputting to 24 bit.
merlyn
Regular
Posts: 149
Joined: Thu Nov 07, 2019 3:15 am

Re: The "float/fixed" debate and more digital myths

Postby Hugh Robjohns » Fri Jan 10, 2020 4:41 pm

merlyn wrote:We never actually hear a 32 bit float. A to Ds and D to As are 24 bit.

That was the case... but we now have Zoom and Sound Devices promoting their 32-bit floating-point A-D converters in their premium portable recorders. So I suspect it won't be that long before we have other interface and preamp manufacturers offering and boasting about the same capability... not because it's really needed (in most conventional recording situations) but just because it's a bigger number that will please the marketing team. :crazy:

And D-As that directly accept 32-bit float via a USB connection can only be a trade show or two away, too... :beamup:
User avatar
Hugh Robjohns
Moderator
Posts: 29055
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: The "float/fixed" debate and more digital myths

Postby merlyn » Fri Jan 10, 2020 7:00 pm

It's interesting that you brought that up. It was the buzz around 32 bit float recording that led me to investigate.

This led to the patent for Sound Device's technology. It uses three 24 bit A to Ds in parallel :

Image

There's a load of DSP involved to produce a 32 bit float.
merlyn
Regular
Posts: 149
Joined: Thu Nov 07, 2019 3:15 am

Re: The "float/fixed" debate and more digital myths

Postby Hugh Robjohns » Fri Jan 10, 2020 7:12 pm

They've dropped hints that their practical implementation involves more than three separate preamp and A-D paths, too. And yes, lots of number-crunching required to blend the various output data together sensibly! Seems to work very well, though...

H
User avatar
Hugh Robjohns
Moderator
Posts: 29055
Joined: Fri Jul 25, 2003 12:00 am
Location: Worcestershire, UK
Technical Editor, Sound On Sound

Re: The "float/fixed" debate and more digital myths

Postby merlyn » Sat Jan 11, 2020 3:00 pm

I think sanity will prevail and 32 bit float recording will remain a niche activity. For a field recorder it makes sense -- it can't clip.

But for a USB interface I can't see it taking off. The DSP must introduce latency. Early adopters would report that their 32 bit float interface was more expensive, sounds the same and has more latency. Hardly a glowing recommendation. :D

Of course we'll see and sanity has had a habit of not prevailing recently. :D
merlyn
Regular
Posts: 149
Joined: Thu Nov 07, 2019 3:15 am

Re: The "float/fixed" debate and more digital myths

Postby James Perrett » Sat Jan 11, 2020 6:06 pm

I was going to mention the ESS Sabre series of chips which directly output 32 bit floating point data until I did a bit of searching and realised that they no longer make a 32 bit ADC - only the DAC. I guess that this shows that the market is too small for a chip maker to build 32 bit ADC's.

http://www.esstech.com/index.php/en/pro ... onverters/
User avatar
James Perrett
Moderator
Posts: 9931
Joined: Mon Sep 10, 2001 12:00 am
Location: The wilds of Hampshire
JRP Music - Audio Mastering and Restoration. JRP Music Facebook Page

Re: The "float/fixed" debate and more digital myths

Postby Logarhythm » Sat Jan 11, 2020 6:13 pm

merlyn wrote: Early adopters would report that their 32 bit float interface was more expensive, sounds the same and has more latency. Hardly a glowing recommendation. :D
I fear that you underestimate the potential for confirmation bias, to the extent that I'd wager a not insignificant number of early adopters would report that their shiny and expensive new interfaces improved on <choice of nebulous adjectives like "openness", "space" etc.> and the slight latency penalty was a small price to pay ;)
Logarhythm
Frequent Poster
Posts: 1019
Joined: Fri Mar 30, 2012 12:00 am
Location: On an intergalactic voyage to collect fresh electrons

Re: The "float/fixed" debate and more digital myths

Postby merlyn » Sat Jan 11, 2020 6:25 pm

Thanks James.
ESS wrote:The SABRE32 Reference audio DAC’s 32-bit Hyperstreamä architecture can handle full 32-bit PCM data
That's a 32 bit integer chip. 32 bit integer would give a theoretical dynamic range of more than 180dB. This one claims a dynamic range of 135dB or roughly 22.5 bits.

The practical achievable dynamic range at the moment is around 120dB or 20 bits.

It's not possible to convert a 32 bit float to an analogue signal because say a value of 1.0 is 1 Volt, the voltage could get insanely high like 2^127 volts, which I don't think has ever existed on Earth. :D
merlyn
Regular
Posts: 149
Joined: Thu Nov 07, 2019 3:15 am

Re: The "float/fixed" debate and more digital myths

Postby merlyn » Sat Jan 11, 2020 7:02 pm

@Logarhythm It's going to be interesting to see how this pans out :D
merlyn
Regular
Posts: 149
Joined: Thu Nov 07, 2019 3:15 am

Re: The "float/fixed" debate and more digital myths

Postby James Perrett » Sat Jan 11, 2020 7:11 pm

merlyn wrote:Thanks James.
ESS wrote:The SABRE32 Reference audio DAC’s 32-bit Hyperstreamä architecture can handle full 32-bit PCM data
That's a 32 bit integer chip. 32 bit integer would give a theoretical dynamic range of more than 180dB. This one claims a dynamic range of 135dB or roughly 22.5 bits.

You're right - for some reason I'd seen 32 bit floats mentioned in the same place as these ESS chips but it appears that the chips themselves are integer based. There seems little point in using the 32 bit feature in this case although I guess it may make later processing very slightly simpler as 24 bit integers aren't a standard data type in most programming languages.
User avatar
James Perrett
Moderator
Posts: 9931
Joined: Mon Sep 10, 2001 12:00 am
Location: The wilds of Hampshire
JRP Music - Audio Mastering and Restoration. JRP Music Facebook Page

Re: The "float/fixed" debate and more digital myths

Postby Folderol » Sat Jan 11, 2020 8:14 pm

In the Linux ALSA drivers it's a bit more complicated than that (dunno about others). This recognises (and you have to support). 32bit normal integers, 24bit bit passed as 32 bit with the top 8 bits ignored and 24 bit passed as 32bit with the bottom 8 bits ignored, and of course float which is 32bit. I understand the reason is that always working 32 bit is far more efficient, but I don't know if it's the OS that decides this or the hardware itself.

Forgot to mention. 16bit is handled as interleaved, i.e. 32 bit paired values... and is a pig to manage.
User avatar
Folderol
Jedi Poster
Posts: 11691
Joined: Sat Nov 15, 2008 1:00 am
Location: The Mudway Towns, UK
Yes. I am that Linux nut.
Onwards and... err... sideways!

Previous