Topic: 32Bit Float recording - The Technical view

[TheJez]

I have to say that I trust it rather more than the response I got from "DeepSeek"

There is a lot of confusion and misperceptions and marketing nonsense about 32bfp recording on the internet. DeepSeek and it's western counterparts are only repeating what they've learned from the often misleading and incorrect content it finds on the web. Personally I'm a big fan of multi ADC 32bfp recording as I often have to deal with situations where it is unclear what a good record level would be upfront. So I like the fixed gain concept, and then accept the need for multi-ADC's to keep the noise down in the quiet parts.

[aaronji]

Since your DAW almost certainly upsamples to 32-bit for processing, there is no benefit to converting a single 24-bit ADC output to 32-bit floating point in the recorder. Just normalize to -1 dB (or -.1 dB or whatever) in the last processing step. Any rounding errors will be so minuscule as to be completely inaudible in virtually any recording situation. 

The dual (or more) ADC is the key to the whole thing, in my opinion.

[unidentified]

Since your DAW almost certainly upsamples to 32-bit for processing, there is no benefit to converting a single 24-bit ADC output to 32-bit floating point in the recorder. Just normalize to -1 dB (or -.1 dB or whatever) in the last processing step. Any rounding errors will be so minuscule as to be completely inaudible in virtually any recording situation. 

The dual (or more) ADC is the key to the whole thing, in my opinion.

If so, how can a device with only one ADC accurately claim to be a floating 32-bit device?

[aaronji]

^ It is easy to turn a 24-bit PCM file into a 32-bit floating point format. It won't make any difference, however, other than increasing the file size. It is advertising trickery...

[Dan33185]

As someone who has just gotten in to the whole 32 Bit float thing, I am wondering...are there any drawbacks to using a recorder with it? I get some people don't believe it makes a difference between 24 & 32 bit and they think it's all a scam, but are there any obvious negatives to using a 32 bit recorder as compared to 24 bit? Or at worst is there just no difference?

[grawk]

As someone who has just gotten in to the whole 32 Bit float thing, I am wondering...are there any drawbacks to using a recorder with it? I get some people don't believe it makes a difference between 24 & 32 bit and they think it's all a scam, but are there any obvious negatives to using a 32 bit recorder as compared to 24 bit? Or at worst is there just no difference?

For the recorders, the only concern I've seen is a mostly theoretical concern that noise levels can get high when the recorder needs to switch to the high gain ADC, because the overall analog noise floor stays the same, it's just amplified.   

[unidentified]

As someone who has just gotten in to the whole 32 Bit float thing, I am wondering...are there any drawbacks to using a recorder with it? I get some people don't believe it makes a difference between 24 & 32 bit and they think it's all a scam, but are there any obvious negatives to using a 32 bit recorder as compared to 24 bit? Or at worst is there just no difference?

Most importantly to me, it means I do not have to worry about properly predicting how to set the gain level in advance.  Set it up, turn it on, and then relax and enjoy the music.  As good as 24 bit but less worries.  Works for me.  YMMV

[unidentified]

^ It is easy to turn a 24-bit PCM file into a 32-bit floating point format. It won't make any difference, however, other than increasing the file size. It is advertising trickery...

So it negates the no-clipping advantage of actually recording in floating 32 bit?   

[unidentified]

As someone who has just gotten in to the whole 32 Bit float thing, I am wondering...are there any drawbacks to using a recorder with it? I get some people don't believe it makes a difference between 24 & 32 bit and they think it's all a scam, but are there any obvious negatives to using a 32 bit recorder as compared to 24 bit? Or at worst is there just no difference?

For the recorders, the only concern I've seen is a mostly theoretical concern that noise levels can get high when the recorder needs to switch to the high gain ADC, because the overall analog noise floor stays the same, it's just amplified.

Based on my extensive use of the F3, the word "theoretical" is the key word here.  Never heard any such noise in my many recordings with the F3

[aaronji]

So it negates the no-clipping advantage of actually recording in floating 32 bit?

The "magic" in the 32-bit float recorders is the multi-ADC setup. That's what allows them to not clip (unless the input clips), but, at the same time, keep the noise floor down. The 32-bit floating point side of it is really just a mathematically convenient way to implement the multi-ADC design.

[Gutbucket]

^That.

There are 3 steps:
1) Use of additional ADCs running in parallel with different levels of input gain to extend dynamic range.
2) Combining the outputs of the multiple ADCs into one data stream.
3) Storing that output.

You can have the first without the other two.  That is how secondary safety tracks have been implemented in some recorder for the last few decades. Those recorders stored both ADC outputs as separate files, one at a higher level than the other, which allowed the user to choose whichever one happened to fit the full dynamic range better than the other afterward.  Those recorders essentially operate no differently than modern multi-ADC recorders in regard to this first step.

The combining of the outputs is the magic sauce, the tricky part, and is the step that is related to the low-level noise modulation artifacts reported to occur in some cases.  In modern multi-ADC recorders this is done in real time while recording. But it can also be done afterward on the computer as long as the output of each ADC is stored separately by the recorder.  TS member TheJez has created a program that automatically merges the two separate files created by "safety track" recorders afterward.  That program and a more detailed description of what it does can be found in the SafetyTrackMerger (Windows + MacOS) thread- https://taperssection.com/index.php?topic=206443.msg2421571#msg2421571

Whether the merge is done in real time in the recorder or afterward on the computer, the combined output of the multiple ADCs is then stored in some file format.  That might be 24bit fixed-point, 32-bit floating-point, or whatever.  Some recorders which do the merge in real time allow you to choose the output format, others don't. 

The dynamic range of any one thing we are recording can fit in a 24bits fixed-point file.  But the full range of all the different things we might plug into a recorder doesn't quite.  Writing a 32-bit floating point file that has the capability of storing a ridiculously large range of level values fixes that, even if its not really much of a problem in actual use.  The real-word usable dynamic range of any analog input recorder is always going to be limited by the dynamic range capabilities of its analog input, which is usually smaller than what a 24bit file is capable of storing.  But a few ADCs working in parallel with different input gains can collectively exceed the dynamic range capability of that by just enough that the user needn't manually set input gain.. except they actually still do.. by manually switching between "line" and "mic" input sensitivities. 

That such mic/line switching is still required is one clear indication that 32-bit float recorders do not actually have the super huge dynamic range capability which is often advertised (only the file format itself has that).  They effectively have just enough more dynamic input range capability via their automated ADC switching to make them more convenient. We still need to make sure that we aren't overloading the input or not driving it strongly enough to avoid overloading or noise, but we can do that ahead of time when setting things up originally.

Beyond that, a 32-bit float file format has some advantages for some forms of processing.  But modern programs and DAWs always convert the input file format to some larger internal workspace format before doing the processing anyway.

So there are some advantages (even though the advantages frequently don't actually reflect what is being advertised) such as less or no need to worry about level setting, at least after you've done the due diligence to make sure your recording setup is not going to exceed the actual input dynamic range of the recorder.  And some disadvantages, such as having to store significantly larger files for the same length recording and sample rate, and those files not being able to be played in simple file players prior to doing some post processing and conversion back to fixed point wavs or lossless files.

[Ozpeter]

^ Excellent lecture on the subject, thanks!

It seems to me that 32 bit float with dual converters particularly suits devices such as the Zoom M2 and M3 "MicTrak" recorders which only offer their own built in mics as the input source.  This means the circuit designers can optimise the whole chain within the device on the basis of knowing in advance what signal levels the mic and its analog preamp will produce.  The M2 claims 135dB "maximum sound pressure input" which is - I think - about as loud as any device will handle.  And obviously the a/d converters will not distort if that SPL is encountered, nor I guess any preceding stages, as the mic output at that SPL input is a known figure to the designers.

Someone somewhere here did suggest that storing single a/d converter output in 32 bit float form could have the advantage that it would be useful to have that recorded format to hand when processing the audio within the device.  The humble H2e and H1e devices that do not claim (as far as I can see) dual a/d converters do have the ability to remix, normalise and convert to other formats within the device (if the user is desperate) and so perhaps that's the rationale - apart from marketing - in using 32 bit float from their (apparently) single converters.

I have not seen references to the use of single a/d converters in marketed "32 bit float" devices anywhere apart from these forums.  Have we discovered something?!

[Gutbucket]

It seems to me that 32 bit float with dual converters particularly suits devices such as the Zoom M2 and M3 "MicTrak" recorders which only offer their own built in mics as the input source.  This means the circuit designers can optimise the whole chain within the device on the basis of knowing in advance what signal levels the mic and its analog preamp will produce.  The M2 claims 135dB "maximum sound pressure input" which is - I think - about as loud as any device will handle.  And obviously the a/d converters will not distort if that SPL is encountered, nor I guess any preceding stages, as the mic output at that SPL input is a known figure to the designers.

Somewhat ironic that if the specifications of what is bing input is known to the the designers (such as is the case with the device's own built-in mics) they can easily setup internal gain staging to use multiple ADCs but then write the output from that to a 24bit fixed point file without requiring any level setting in the same way as a 32bit float file.  That content will fit.  It's dealing with a wider range of potential external inputs that gets somewhat trickier.

For folks doing production work on the cheap, a 32-bit float output may fit into their file management and post-production workflow a bit better, and that's an audience that is a focus for manufacturers, not music tapers!  Fortunately it works just fine for us too.  I balk a bit at the inefficiency of the unnecessarily increased storage size, the misleading advertising claims, and the potential loss of gain-staging skills by future recordists, but it pretty much works just fine so I shouldn't complain too much.  Just takes a bit of time and discussion to unearth the actual truth of it.

Use it if you have it and like it and make great tapes.

[Ozpeter]

I think that now 32 bit float is simply the norm in any newly released audio recorder - even wireless mics!   Without it many potential purchasers will regard the device as old tech.

[grawk]

Certainly for consumer gear.