32Bit Float recording - The Technical view

[Rairun]

I usually use a Zoom F3 paired with CA-11 cards, and after running some tests, I found that the microphones are the bottleneck in terms of self-noise. Whether I plug them straight into the F3 (with a battery box + adapter) or add a pre-amp to the chain with a lot more clean gain, the level difference between a test signal and the gear's self-noise remains the same.

I also ran the same test with a pair of Clippys, which are way more sensitive and have less self-noise than the CA-11s. With Clippys>Battery box>Adapter>F3, the difference between the test signal and the gear's self noise was about 4dB greater than the result I got with the CA-11s>Battery box>Adapter>F3 rig - that is, changing the mics increased the signal to noise ratio by 4dB. With the Clippys>Pre-amp>Adapter>F3 rig, the difference was 5dB. This means that adding loads of gain between the Clippys and the Zoom F3 only increased the signal to noise ratio by 1dB.

My conclusion is that unless your microphones have EXTREMELY low self-noise, it's pointless to worry about the Zoom F3's gain staging.

[Niels]

I was checking spec sheets for a couple of Zoom 32bit float recorders and notice repeated use of the following wording (this taken from the F3):

Equivalent input noise: −127 dBu or less (IHF-A) when waveform magnification is ×1024 with 150 Ω input (My empasis)

Apologies if this is a basic question or have been covered before, but why would the input noise be affected by the waveform magnification?

I seem to recall early discussions and confusion about the F3, and came out with the impression that the waveform magnification setting on the display was immaterial and would not affect the quality of an 32bit recording when normalised later.

What is it I am not understanding here?

[voltronic]

I was checking spec sheets for a couple of Zoom 32bit float recorders and notice repeated use of the following wording (this taken from the F3):

Equivalent input noise: −127 dBu or less (IHF-A) when waveform magnification is ×1024 with 150 Ω input (My empasis)

Apologies if this is a basic question or have been covered before, but why would the input noise be affected by the waveform magnification?

I seem to recall early discussions and confusion about the F3, and came out with the impression that the waveform magnification setting on the display was immaterial and would not affect the quality of an 32bit recording when normalised later.

What is it I am not understanding here?

You're not misunderstanding. You actually understand the unit better than Zoom does, apparently. The F3 manual has some glaring inconsistencies and downright false or misleading statements. One of the biggest being the "Waveform Magnification" business, where they state at one point that changing that setting does not affect the recorded level. But, as anyone who has used the F3 can tell you, it absolutely does affect the recorded level. The thing is, that level is only being adjusted post-ADC, so it's just a digital fader rather than an input trim or analog gain setting. The analog preamp gain feeding the ADC on the F3 is a fixed, non-adjustable level. This is the case for all Zoom recorders when recording in 32-bit float format (the F6 and F8 give you a 24-bit fixed option in which case the analog preamp gain is adjustable).

The statement from the specs you quoted strongly implies that analog gain is adjustable, because you typically measure EIN with the preamp gain at max level. Since we know preamp gain is fixed and we know what "waveform magnification" actually does, the portion of the statement you bolded should not be there.

[Ozpeter]

That's a great clarification - but I wonder whether when they said it doesn't affect the recorded level, perhaps they were treating the massive dynamic range of the 32 bit float data as something not related to recorded level?  In other words, changing the 'magnification' is not really changing the level - same as when you look at an insect through a microscope, it's not actually getting a whole lot bigger?  In other words, the 'recorded level' is set unchangeably by the preamp, and the post ADC gain is simply changing how it is numerically expressed?

Show me a hair and I'll split it...

[TheJez]

But, as anyone who has used the F3 can tell you, it absolutely does affect the recorded level.

This was an excellent question by Niels, and a great answer by voltronic.
But (and please correct me if I'm wrong) isn't it so that on the F3, the waveform magnification setting at the moment of starting the recording determines the level of the recorded audio, but changing the waveform magnification while recording does not change the level of the recorded audio?
When selecting a 32bfp recorder a while ago this is how I understood the F3. I actually liked this behavior as it makes so much sense in a 32bfp recording device with fixed analog stage. The Tascam FR-AV2 which I eventually bought does change the recorded level when changing the digital gain during recording. So in practice, I never touch the gain setting while recording, as it doesn't add anything to the quality of the recording and can only result in extra work in post. On the F3, at least, one can modify the magnification to get a nice waveform on the display without changing the recorded level!

[Niels]

Thanks @voltronic for taking the time to explain.

I have a feeling of deja-vu.
It has probably been covered repeatedly in some of the sometimes very long ongoing threads, but it has only recently become relevant for me when I got one of those 32bit recorders.

[Gutbucket]

The entire range between −127 dBu at the bottom and 0dBuFS at the top fits easily within 144dB!

[TheJez]

The entire range between −127 dBu at the bottom and 0dBuFS at the top fits easily within 144dB!

True... but if you allow the user to digitally amplify (or magnify or whatever it is called) the sound, you might end up above the 0dB at the top... Then the enormous dynamic range provided by the 32bfp storage format becomes beneficial and takes a source of error away.

[Gutbucket]

If it ain't broke..

There is good reason for implementing a multiple ADC switching architecture.  Make the dynamic range of the recorder larger than that of the microphone. I welcome that. But there is no good reason for forcing 32bit storage in place of 24bit in combination with it.  24bit is already larger than the increased dynamic range the recorder provides by way of the new multiple-ADC architecture. In contrast to multiple ADC switching, I don't welcome the unnecessary and unwanted complication of forcing the 32-bit file format on users.

It's not a tall ask to set gain initially for whatever microphone is being used, based on the specs of the mic.  One never need touch the gain setting again until switching to a microphone with significantly different specs.  That supports the more technical users. Casual users wouldn't need to do either.  Good design choices empowers users rather than unnecessarily constraining them. 

/rant

[Ozpeter]

I'm quite convinced the 32 bit float thing is purely an advertising gimmick, when the device has a single ADC.   I recently asked DJI whether their wireless mics (three versions now) which trumpet recording 32 bit float audio internally have dual ADCs.  They confirmed what I expected, they don't.  Off the top of my head I don't recall seeing any wireless mics advertised with 32 bit float internal recording by any manufacturer claiming dual ADCs in advertising.  (But maybe some just don't bother to claim of course)  And there seem to be lots of them these days. 

The technical point of 32 bit float from single ADCs seems to be simply that the digital format is the same as many other devices now produce, so if you are working in post-production with 32 bit float data from multiple recording devices, it's all the same format, and you don't have to spend (possibly small amounts of) time reformatting 24 bit data as 32 bit float.

[morst]

From a gain-staging perspective, processing at 32f makes things hella easier.
I mix to 32f now and then save those files before applying processing (gain/limit) where I output to 24 bit.

Running with 32f as a default probably saves the manufacturers a lot of customer service calls.


I can imagine a workflow where audio gets passed from one app to another and it sure is better if it's not clipping on any one of those.

[grawk]

Storage is cheap, and 32 bit has benefits. If it's available I use it. Same for 96khz sample rate.

[Gutbucket]

Help me to understand what I'm missing.  I'm fully onboard with processing, transferring between applications, and whatever else at 32bit  ..but I've long had the ability to do that with 24 bit raw recorded files which I can simply import into the 32bit float workspace of a digital editor with no additional steps - no reformatting of the raw 24 bit input files necessary.  Same goes for 16 bit source files or the combination of the two.  Once in the editor, I can do whatever processing is required in 32bit float and set the output to whatever format I desire.  How does 32 bit input files ease workflow over 24 bit input files?  Is what I'm doing somehow unique to Samplitude? I don't think so, but maybe.  Additionally,  without opening them in the DAW I can preview and modify the raw 24 bit files easily - playing them, splitting / interleaving tracks etc.

Same is not true for 32bit raw files. Not as easy to do many of the simple tasks.  I'm not trying to be a contrarian here, I just don't understand the ease of workflow argument.  Help me to see the advantage after recording.

I'm quite convinced the 32 bit float thing is purely an advertising gimmick, when the device has a single ADC.   I recently asked DJI whether their wireless mics (three versions now) which trumpet recording 32 bit float audio internally have dual ADCs.  They confirmed what I expected, they don't.

To my understanding (as outlined above) this achieves nothing useful to the user. 

[Ozpeter]

Well... good questions.  Such benefits as there might be are subtle rather than dramatic / essential.  I am currently archiving DAT recordings dating from 1987 onwards.  Classical music live stuff.  60+ done, I have yet to encounter any clipping (perhaps excluding applause but the clipped transients in that are not really obvious).  And I haven't been thinking, wow this would have been so much better if the audio was 24 bit, let alone 32 bit float.  But I did have to be careful with levels at the time and I did have to rely on a lot of experience.  Especially with operatic sopranos...  capsule splitters, some of them. 

These days I don't do serious recording but if I did, I'd lean towards 32 bit float for insurance as it were, but not expecting any obvious audible difference. 

[Gutbucket]

^ Right. Those advantages are what a multiple-switching-ADC architecture achieves on the recording end.  I understand and fully approve of that part!