Topic: 32Bit Float recording - The Technical view

[rocksuitcase]

Thanks for updating us on this stuff. I do not own one of these decks yet, but this type of information will be informative when I do buy a new deck.

[morst]

I still think his explanation doesn't really say why writing 32 bit float from a single 24 bit float ADC is a good thing, although he does say that the ADC used is of a fundamentally higher quality than past devices.

He admits that it is merely a handy thing for technical reasons, chiefly post production.
If one were to take a 24-bit recording and load it into a 32 bit float workstation, would that be the same thing though?

[TheJez]

I still think his explanation doesn't really say why writing 32 bit float from a single 24 bit float ADC is a good thing, although he does say that the ADC used is of a fundamentally higher quality than past devices.

He admits that it is merely a handy thing for technical reasons, chiefly post production.
If one were to take a 24-bit recording and load it into a 32 bit float workstation, would that be the same thing though?

Yes, that's almost exactly the same thing, except that with a conventional 24bit recorder, the recording might be clipping if the analog gain was set too loud during recording.
It was kind of funny to see his example clipping guitar recording that was fortunately repairable because it was stored as 32bfp. What he didn't mention was that the only way to get a clipping recording with that recorder is by applying digital gain, as the output of the 24bit ADC will never exceed 0dB per definition!
So indeed, if the recorder allows digital gain, then it is possible to set the gain 'too loud' and make it clip. Only when stored as 32bfp, this clipping is repairable, which is nice.
The alternative would be to NOT allow digital gain at all. Just store the output of the ADC as 24bit. Then the user can never set the gain too loud and there is never any clipping to be fixed. It does mean that any level/balance correction must be done in post, while in the current situation the level/balance correction can be done during recording. I do understand that they opted for digital gain + storing as 32bfp, it seems like a sensible choice for a recorder with fixed analog gain, a single ADC and only built-in mics just because of this level/balance correction benefit. I'm sure some folks will appreciate that feature, although I personally couldn't care less.

[Ozpeter]

I think he's using the H1essential which has no gain control at all, neither digital nor analog.  They seem to set the output from the ADC to exceed zero in the 32 bit float file even if the limit of the max SPL of the mic hasn't been reached.  Perhaps the idea is that typical sound levels will stay below zero, and thus may not need fixing in editing - louder sounds which would have gone into the red in older devices without significant reduction in analog gain now go over 0dB in the 32 bit float files, needing to be fixed later.

[TheJez]

I think he's using the H1essential which has no gain control at all, neither digital nor analog.  They seem to set the output from the ADC to exceed zero in the 32 bit float file even if the limit of the max SPL of the mic hasn't been reached.

Wow, amazing. If that is indeed the case, then they not only have a fixed analog gain, but also a fixed digital gain! That's the only way to get samples above 0dB from a single ADC. (As said, the maximum output of a single fixed point ADC is 0dbFS per definition.) In that case they might as well have simply stored the 24bit output of the ADC. Then there is really no point at all to make this device 32bfp, apart from marketing (or e.g. production benefits by sharing the same design, chips and software components for multiple recorders)

[Gutbucket]

^ That.

As a 'closed system' which includes the entire signal chain from microphones to saved digital file, the designer can presumably set analog gain during the design stage such that the level out of the microphone input stage optimally matches the input sensitivity of the ADC.   Once that has been done, there is no real imperative to provide user controlled 'gain.. other than as a way of achieving better normalization for quiet recordings, yet that can be managed afterwards.  If the design of the recorder uses a single ADC per channel rather than being one which effectively switches between multiple ADCs as a way of increasing dynamic range, performance shouldn't differ regardless of the output being saved as 24bit fixed or 32bit float.  The data content of both should be identical. Only difference is the size of the resulting files and how the output file is handled after being transferred for editing. 

[Dan33185]

I won't pretend to understand the technical part of all this (in fact it makes my head spin), but all I know is when I record using my Tascam DR-40X at conservative levels for louder shows, I still get some clipping. When using the Zoom H4E, I never have. So in my personal experience, 32 Bit Float has made the recording process much more fun, not having to worry about setting levels just so, or monitoring them throughout the show.

[morst]

I won't pretend to understand the technical part of all this (in fact it makes my head spin), but all I know is when I record using my Tascam DR-40X at conservative levels for louder shows, I still get some clipping. When using the Zoom H4E, I never have. So in my personal experience, 32 Bit Float has made the recording process much more fun, not having to worry about setting levels just so, or monitoring them throughout the show.

It is possible to overload the input stage on any recorder with a line in.
Once overloaded, running with low "knob" levels will prevent digital "over level," but the signal is already flattened on peaks due to design limitations of the input stage.
(There is a max level that any input can accept, over which, distortion will occur.)


If you are overloading the built in mics on the DR-40X, then no amount of turning the levels down will stop the "brickwall" distortion.
Move farther away from loud sources if this is the case.

[Dan33185]

I won't pretend to understand the technical part of all this (in fact it makes my head spin), but all I know is when I record using my Tascam DR-40X at conservative levels for louder shows, I still get some clipping. When using the Zoom H4E, I never have. So in my personal experience, 32 Bit Float has made the recording process much more fun, not having to worry about setting levels just so, or monitoring them throughout the show.

It is possible to overload the input stage on any recorder with a line in.
Once overloaded, running with low "knob" levels will prevent digital "over level," but the signal is already flattened on peaks due to design limitations of the input stage.
(There is a max level that any input can accept, over which, distortion will occur.)


If you are overloading the built in mics on the DR-40X, then no amount of turning the levels down will stop the "brickwall" distortion.
Move farther away from loud sources if this is the case.

Like I said, it makes my head spin    I'm not using the built in's on either recorder, I use externals plugged in to the XLR ports

[morst]

Like I said, it makes my head spin    I'm not using the built in's on either recorder, I use externals plugged in to the XLR ports

OK, well, then this is really a topic for a different thread...
Something in your signal chain is maxing out before the analog>digital conversion.


I'm sure you'll be happier when you get that sorted.

[voltronic]

Curtis Judd just put out a 2-video series: '32-bit float recorders & differences in audio quality'.

The first video is a nice background on the tech, and the second gets into more specifics as to which specific products implement this tech well, and how those that don't do it well cut corners.

https://www.youtube.com/watch?v=mxWH8V5BaYk
https://www.youtube.com/watch?v=f_9TM9QZPdo

[EmRR]

I won't pretend to understand the technical part of all this (in fact it makes my head spin), but all I know is when I record using my Tascam DR-40X at conservative levels for louder shows, I still get some clipping. When using the Zoom H4E, I never have. So in my personal experience, 32 Bit Float has made the recording process much more fun, not having to worry about setting levels just so, or monitoring them throughout the show.

It is possible to overload the input stage on any recorder with a line in.
Once overloaded, running with low "knob" levels will prevent digital "over level," but the signal is already flattened on peaks due to design limitations of the input stage.
(There is a max level that any input can accept, over which, distortion will occur.)


If you are overloading the built in mics on the DR-40X, then no amount of turning the levels down will stop the "brickwall" distortion.
Move farther away from loud sources if this is the case.

Like I said, it makes my head spin    I'm not using the built in's on either recorder, I use externals plugged in to the XLR ports

Tascam DR-40X
Maximum Input Level -3dBV (MIC) / +20dBu (LINE)

Zoom H4E
Maximum Input Level MIC: +4 dBu (1.78 dBV) / LINE: +24 dBu  (21.78 dBV)

The Zoom has a max input level almost 5dB higher than the Tascam.  If you switch to LINE and it will still supply 48V (assuming 'externals' means condenser mics), you won't overload.  I ran a Sony Minidisc forever with DPA 4060's feeding it set to line level, and it was fine.   If the mics have pads, consider using them.   I don't know if the Tascam has a built in mic pad option; that would also solve it.  If you are using an external preamp with fixed gain, it may be too much for the Tascam and not the Zoom.   All of these things are gain staging considerations that will affect analog overload before signal ever gets to the AD converter stage.

I noted way back at the beginning of one of these discussions that the Zoom F8N/whatever has much lower input headroom than the Sound Devices 'whatever'.   The Zoom F8n (non 32 bit) at least has a pad that can be switched in for mic use.  Doesn't mean one is better or worse than the other, they are design decisions that balance different expectations/needs, so one has to adapt to those parameters. 

The bad assumption with these devices is that the headroom window is adequate for all uses.  Hot condenser mics at loud events (or up close to loud things) typically output line level signals rather than mic level signals.  Remember part of the market is people in the field recording bird calls, so they don't assume a LOW amount of gain.  The Rolls Royce device here would have adjustable analog gain with metering, and 32 bit float capture with metering, then you could really control the headroom relationships. I'm not aware of anyone making that device.  Yeah, it's somewhat redundant; ok, so give us a separate analog overload indicator along with the converter metering. 

[Gutbucket]

Seems that what's driving all this is the simplification of gain staging by the user in the recorder.  As mentioned, the need for proper gain staging doesn't actually go away, its just made less critical for a significant part of the typical range. That is until it exceeds that safety bubble range and once again becomes critical, like it always was.

I'm all for reducing potential failures, but I wish manufacturers would do so without obfuscating the details of gain-staging and instead make the monitoring of levels and the awareness of where the signal fits into the maximum dynamic range of the recorder more transparent to the user.  Don't need to dumb things down in order to make things easier when the system could instead be made more transparent and empowering..

The Rolls Royce device here would have adjustable analog gain with metering, and 32 bit float capture with metering, then you could really control the headroom relationships. I'm not aware of anyone making that device.  Yeah, it's somewhat redundant; ok, so give us a separate analog overload indicator along with the converter metering.

I'd like to see an overload indicator on the analog input stage, followed by a peak/VU meter with a scale that shows the entire effective dynamic range through the converter, from the recorder's noise-floor up to 0dBfs.  That's supposedly a range of about 142 dB through a current SD Mixpre recorder. The scale might be made non-linear to fit if necessary.  Primary role of the meter when recording is to indicate: Modulation (meter "dancing movement" indicating active signal); Peak level; VU level; and in this case noise floor as well.  Showing the entire range allows for seeing modulation at all input levels, and allows for determining where the current noise floor and peak level actually fall within the available input range.  User can also easily visually determine the approximate total dynamic range of whatever signal is being recorded.

Extra credit for including a metering mode that actively changes scale with content, starting fully "zoomed in" and automatically "zooming out" as necessary to keep whatever the current peak-hold value happens to be at the top of the meter scale and the current minimum noise-floor value at the bottom of the meter scale.  The meters will then show the full range of active modulation with maximum possible visual resolution.  Pushing the "peak reset" button resets the display to some comfortable maximum zoomed in range or whatever the current modulation range happens to be if if exceeds the max zoom default.  Display scale numbering across the top of the meter updates with each automated "zoom out" step as the uppermost and lowermost values indicated on the display change with each excursion past the previous peak and minimum value.

[Ozpeter]

I suppose that what they do is to cater for the vast majority of users' needs, and tough luck on the rest... but, having said that, has anyone actually encountered an issue with recording on a proper 32 bit float device, in terms of wishing that they could have increased the analog gain, or reduced it?  Genuine question, I'm not taking sides here (wouldn't dare... )

[unidentified]

No, I have not encountered any such issues. I'm perfectly happy with the 32-bit system in my three Zoom F3s. And I've been recording shows for more than 50 years, starting in the cassette era.