Topic: 32Bit Float recording - The Technical view

[Gutbucket]

My bolding and snip edits below >

[snip..] A 32-bit recording system can't help you if your input stage is overloaded, or (conversely) if the gain of your input stage is set so low that the noise floor of that stage is high relative to the signals coming in. If the signals coming from your mikes exceed the headroom limit of the first circuit stage, you'll get a glorious, 32-bit float recording of a clipped signal. If the gain of the first stage is set so low that you have, whatever, say 20+ dB headroom that you're not using, then that's 20+ dB of dynamic range that you're discarding in order to preserve that headroom--and you'll get a glorious, 32-bit float recording of the noisy signal coming out of your first stage and going into your glorious, 32-bit A/D converters.[..snip]

That's all true of course.  Regarding the bolded part, the thing is that the acoustic noise floor of the venues in which most folks here at Taperssection are recording is almost always quite high.. significantly higher than the electrical noise floor of the entire signal chain, even in what are perceived to be the most quiet moments.  Yes, running input gain low enough such that the loudest peaks overload something else in the signal chain first other than the ADC (the mics themselves, or in my case the input stage of the F8 when its gain has been set to minimum and the limiter has not yet been engaged) is going to increase the electrical noise floor of the recording system. But if that electrical noise floor is buried well beneath the acoustic noise floor it just won't matter.

When run that way a 24bit fixed-point recording is the same as a 32bit floating-point recording in the practical sense - everything fits with extra dynamic range left at both the top and bottom of the available range.. except the resulting 24bit fixed-point wav files are smaller and more universally usable. Either way one needs to normalize the recording.. which s the significant behavioral change forced onto folks when switching to 32bit recording, as I see it.

I'd guess the maximum possible dynamic range in the situations in which most "tapers" are recording likely does not exceed 60dB or so, maybe 70 at most- determined by the acoustic noise-floor of the venue at the bottom and the highest peak at the top.  Its not difficult to set input gain to easily accommodate that, then simply not revisit it it again.. at least not until switching to microphones with significantly different sensitivity or low frequency response.  The significant break is not the new file format, but in no longer "operating the recorder the old way".

The "old way" was manually setting input gain so that there is just enough headroom up top to avoid overs.  That makes/made sense when:

• Avoiding the need to normalize the level of the resulting recording.
• Recording analog to tape.
• The dynamic range through the signal chain of of some small and inexpensive digital recorders digital recorders was not really large enough, nowhere near the total range available to 24bits.
• For some folks wrangling "the sound" of certain preamps by "running hot".

Now things have evolved to the point where even cheap recorders seem to achieve sufficient dynamic range to allow setting input gain once and then leaving it alone (for what most are recording here at TS).  After that, the results end up the same as recording to 32bit.  32bit removes the ability to set gain, but you still need to check it the first time to insure the recorder is not being overloaded! My initial experience with the Diety PR-2 is a cautionary example.

[Gutbucket]

..some backstory as to how I came to the above revelation-

When I was regularly attending and recording orchestral concerts in a few near world-class halls, from a position directly behind the conductor where the dynamics were extreme, I determined the gain setting needed to accommodate the full dynamic range of those events and then never needed revisit it.  The acoustic noise floor in those extremely well-isolated halls was significantly lower than anywhere else I'd record. It took some doing early on to determine correct level setting, and then to fully determine what established the noise floor of the recording, but I was able to dial in gain such that the noise floor of the recording was dominated by the acoustic noise of the room while thunderous fortississimo (f f f) passages and any overly energetic applause by immediately adjacent concert goers were fully accommodated.  I was able to do that using an external preamp into a digital recorder which when operated in 24bit mode realistically achieved a dynamic range of only few bits more than would have been available in 16bit mode. I had a second lower-gain preamp setting I'd switch to for PA-amplified material dominated by heavy subwoofer content or for recording in very close proximity to a drum kit where transient peaks are extreme  That's all it took - two gain settings to accommodate any real world live music performance.

When I later moved to using Zoom F8 in my open recording rig, that previous experience in correctly gain-staging the more dynamically limited small recording rig made switching to setting gain once and leaving it alone somewhat easier to adopt.  Honestly it took a few years to break the old habit of "normalizing via gain tweaks prior to recording, rather than afterward", and I admit I've not fully done so.  I'll still up the gain for quieter material partly for raw playback convenience and to get the meters showing good movement, even though I know the recording won't really benefit or suffer from doing so.

/end of thread transplant.  Thanks to all involved for the discussion.

[Ozpeter]

All of that useful transplant reminds me of how I used to work in the 16 bit days, making commissioned (ie not stealth) classical recordings.  My mic preamp (from pair of Sennheiser MKH series mics set up for MS) had a red line marker on the input level control, which covered everything, except for a few surprises...  What I came to realise was that everything was actually the same level (in broad brush terms) at the mic.  So if it was a symphony orchestra, the mic was a somewhat away from the source, or if it was a solo acoustic guitar, the mic pair was much closer, and so the level at the mic itself was similar.  But for recording amplified music, I suspect it's much less predictable, and if it's stealth, you can't guarantee being at the sweet spot in the room every time.

As for the digital side of things having a 1500dB dynamic range but not the analog side, thus raising the possibility of making excellent recordings of analog clipping, that's where a device like the Zoom M2 Mictrak offers some reassurance - the entire chain can handle 135dBA, from mics through preamp to dual ADC, which is threshold of pain level I believe.  So it's about as set-and-forget as anything in one easy to use piece.  But not very low profile of course!  So if that clips, then the PA crew are risking people's hearing.  Sue them!