Topic: 32Bit Float recording - The Technical view

[morst]

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. 

25% smaller file size is a minor benefit.

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?

25% larger file size is a minor drawback.

[Ozpeter]

Given the size of video files these days, even 32 bit float audio files are not that much of a big deal.  Or large deal.  And of course the colossal size of inexpensive memory cards is another factor.  I think the humble H2e can handle a 2TB card.  That equates to a 32 bit float stereo recording of... considerable length.

[Dan33185]

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. 

25% smaller file size is a minor benefit.

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?

25% larger file size is a minor drawback.

I don't worry about size too much, I compress my original files using WAV Pack for archiving. Those are then transferred to an external HD, which are fairly cheap these days.

[Ozpeter]

Tascam claim in a reply to my question to them on YouTube that their latest recorders must use dual a/d converters because that is the only way to get 32 bit float output.  This is in conflict with Zoom who state that at least one of their devices does use a single converter to output 32 bit float from certain inputs of the device.

I have fired off this question to Tascam support in the USA -

"I am investigating whether recent handheld recorders by various manufacturers do or do not use dual a/d converters for 32 bit float output.  This investigation began when another manufacturer stated that their 32 bit float device did not use dual converters on one of the inputs of their 32 bit float only device.  That caused me to check the published specs for other devices, and there is a clear pattern that lowest cost 32 bit float devices do not claim to use dual converters.  Publicity for your new 32 bit float DR-07XP device does not mention dual converters.  But when asked about this on YouTube your representative states that it does, as they say 32 bit float cannot be achieved with a single converter (I disagree - and so does another manufacturer).  Could you resolve this matter?  On other Tascam devices, dual converters are highlighted in publicity.  So if the DR-07XP does have dual converters, why not say so in the publicity and manual?   Thanks for any info you can supply."

I will post any response I get here in due course...

[datbrad]

Tascam doesn't seem to be mincing words with their 32bit primer: https://tascam.com//us/feature/32-bit_float#:~:text=In%20a%2032%2Dbit%20float,low%20gain%20ADC%20does%20not.

If dual 24bit ADCs is the way Tascam explains how they are able to produce a 32bit output, I would be shocked if there is a superior process they are choosing not to use. It makes zero sense from a product marketing perspective for a company as established as Tascam to broadcast that they aren't using a 32 bit standalone A/D unless such a thing doesn't exist. Just applying my own logic, I don't know this for a fact.

[GLouie]

Bad link for me, an extra backslash. This seems to work:
https://tascam.com/us/feature/32-bit_float#:~:text=In%20a%2032%2Dbit%20float,low%20gain%20ADC%20does%20not

I am curious whether the little Zoom flaw from the deleted YT video last year is now a non-issue for all manufacturers, or if it was ever a real problem.

I also wonder if the Sound Devices patent affects other manufacturers implementations of 32 bit float, or if SD is going to start lawsuits.

[Gutbucket]

^ They are lying to you.

This is demonstrably not true: "After you've recorded audio in 32-bit float format, any quiet sections of your recording can be increased in volume in your DAW or NLE software without adding any noticeable noise. And loud sections that would have clipped a 24-bit or 16-bit recorder can be reduced in level, without any digital distortion or clipping."

The analog input dynamic range capability of the recorder, which ranges from it's noise floor up to clipping, is determined by it's analog input capabilities, not the digital file format.  Anyone can record a signal which is low enough in level to encounter the recorders analog input noise floor, which will then become apparent when the level is raised in the DAW.  And as we have discussed here at TS, an overly hot input which exceeds the capabilities of the analog input will distort and clip.  Either of those things can be easily accomplished.  It is true that within those limits no gain adjustment is needed, and that's really great.  But the limits of the real-world analog input dynamic range capability of the recorder is orders of magnitude smaller than what they are inferring from the 32bit float format and are really not that different than what we already had by correctly setting input gain manually.  Not having to make that gain adjustment when recording (as long as the signal fits the analog input stage limits) is the real advantage.  Wish they could be honest about it.

Here's an easy indicator of marketing bullshitism. Any time you come across an illustration like the one below in marketing materials, stop reading:

[Gutbucket]

Really, it doesn't matter to us how many ADCs or what the magic sauce inside is.  What does matter to us are the limits of the real-world analog input dynamic range capability of the recorder.  I really wish they would honestly talk about that. I see a lot of smoke and mirrors talking around this essential point by manufacturers and find it shameful.

[TheJez]

Really, it doesn't matter many ADCs or what the magic sauce inside is. What does matter to us are the limits of the real-world analog input dynamic range capability of the recorder.

You're right that there's a lot of marketing-nonsense spread by the companies. However, I do believe that multi ADC (if implemented properly) can and will broaden the analog dynamic input range capability. A proper implementation will not consist of just two or more ADC's, but also two or more analog input paths, each designed to handle their specific part of the total device's analog dynamic range.
It would be great if the manufactorers would be more clear how each of their devices implement the 32bfp technology, so we can separate the 'true 32bfp devices' from those who are merely 32bfp because that's the format they use to store the recordings.

[voltronic]

Really, it doesn't matter many ADCs or what the magic sauce inside is. What does matter to us are the limits of the real-world analog input dynamic range capability of the recorder.

You're right that there's a lot of marketing-nonsense spread by the companies. However, I do believe that multi ADC (if implemented properly) can and will broaden the analog dynamic input range capability. A proper implementation will not consist of just two or more ADC's, but also two or more analog input paths, each designed to handle their specific part of the total device's analog dynamic range.
It would be great if the manufactorers would be more clear how each of their devices implement the 32bfp technology, so we can separate the 'true 32bfp devices' from those who are merely 32bfp because that's the format they use to store the recordings.

Responding to the bold statement: You can't increase the analog dynamic range with multi-ADC implementation. As Gutbucket explained, the analog input stage is the limiting factor no matter how many taps off of it go to however many ADCs. The devices with greater dynamic ranges are the ones that have outstanding analog stages at the front end. The analog circuitry is also one of the main places where the cheaper brands save money, and they distract you by telling you about the enormous digital dynamic range even though the analog input stage will never get you anywhere close to utilizing that enormously tall and deep digital bucket.

Case in point: The long-discontinued Sound Devices 788T has an input dynamic range of 123 dB, whereas the Zoom F8N PRO is 113 dB (both A-weighted). The 788T has a single ADC feeding a 24-bit fixed container whereas the F8N PRO has two auto-ranging ADCs feeding a much larger (numerically) 32-bit float container. Yet the old 788 gives you 10 dB better dynamic range, and that's because of its higher-quality input stage.

In recent years the spec sheets have gotten quite inconsistent with different manufacturers, making things difficult to compare.
- Zoom does not specify A/D dynamic range for any of their recent 'Essentials' recorders, but we know through other specs they likely reused the old, noisy H-series preamps so the dynamic range is going to be rather limited.
- Sound Devices specs the A/D dynamic range for the 888 at 120 dB, and for the MixPre-10 II at 142 dB minimum. I find it hard to believe that the much cheaper MP10II has dynamic range that much better than the 888, even with multiple ADCs.
- Tascam specs the Portacapture X8 A/D DR at only 113 dB, but the FR-AV2 at 132 dB or higher. But if you look farther down, the FR-AV2 has far higher input noise.

[Rairun]

When these devices first started coming out, I was under the mistaken impression that each signal path had its own preamp - because in my experience, the difference between a shitty preamp and cleaner one was very noticeable. Adding +40 dB with the internal preamp of a Zoom H1 sounds a lot more noisy than dialling down the internal gain and using a better external preamp with the same amount of gain. But if you only need something like +15 dB for the signal to reach a healthy level, then the performance difference between internal and external preamps will be much closer in absolute terms. You won't hear any preamp noise either way.

So I thought these devices worked like this: one signal path would have high preamp gain (say, +40dB) fed into an ADC that then offset this by -40dB when writing it to a 32-bit float container, and a different path with little to no preamp gain, fed into an ADC with little to no offset. Combine both, and you have the advantages of high clean preamp gain for the quiet parts without worrying about overloading the device with your peaks.

But as far as I know, this is not how these devices work. A Zoom F3 appears to have the preamp gain fixed, which initially didn't make sense to me, since I always heard people say it was a conservative amount of gain. But this architecture only makes sense to me if the gain is permanently set to high (say, +40 dB), with one of the AD converters being capable of withstanding a really hot signal. Does this sound right? This would allow the device to actually take advantage of the clean preamp (which a conservative amount of gain wouldn't permit!).

What I don't fully understand is: if this is the case, is a preamp necessary at all here? If you can tweak the sensitivity of the ADCs, can't you just make the one intended for quiet sounds really sensitive, skipping the preamp stage altogether? Or is that not possible because you'd bump against some noise floor that is inherently higher than that of a quality preamp?

[unidentified]

All I can say is that my F3's preamp sounds pretty darn quiet, even when I greatly boost the volume of very quiet passages in post

[Rairun]

All I can say is that my F3's preamp sounds pretty darn quiet, even when I greatly boost the volume of very quiet passages in post

Same here! I am just interesting in understanding exactly how it accomplishes this.

[datbrad]

This is the only true 32 bit fixed integer audio interface I can find. There are 8 more bits but it still clips signals above 0 dBFS just like 24 bit.
https://www.merging.com/products/interfaces/merging+anubis

To me the fact that cheap recorders and wireless mic receivers are all moving to 32 bit float tells me this is feature for convenience and that's it.

[grawk]

my sonosax is also 32bit integer, and has a DNR of 135db. The only analog gain stage available on this is a +20db selectable gain.  It can write safety tracks as well.