Topic: Recording in 24-Bit vs. 16-Bit

[wforwumbo]

Just to chime in, the reason DAWs use 32 bit (some are even moving to 64-bit capabilities) has less to do with dynamic range and more to do with overall numerical precision. This is noticeable when we start manipulating audio - speaking in general hand waves, a basic bell/peaking filter will perform better if the bit depth is larger.

From a dynamic range perspective, it can certainly help with overflows when gain staging, but that's secondary to overall sonic performance of manipulating the audio. Keep in mind, the vast majority of DAW users are distinct from us tapers in that they're producing full records.

[Gutbucket]

Now, yes, if the noise floor of the signal chain is boosted by the same amount, you're right  that it will be louder than the noise I heard in (3). But surely the total noise level will be a combination of both noise floors, and not just the highest one? This is usually pretty irrelevant because we users do reasonable gain staging, so the signal chain's noise floor is always much, much louder than the 24 bit noise floor. But if your recorder always needs to record the quietest sounds at something silly like -140 dB (to make sure a sudden, deafening noise will not clip), then suddenly both noise floors will be pretty similar, and adding one to the other will result in a noticeably louder noise once boosted!

Yes the noise floor is cumulative. If both noise-floors are close enough in level, the resulting noise floor will increase. But the cumulative noise floor of the signal chain is likely to be higher than the dither in the least-significant-bits of a 24bit file by more than a little.  And both will be considerably lower than the acoustic noise floor of any of the environments in which we are recording, including church mouse quiet, purpose build classical halls.  The real world result is the same - the acoustic noise floor of the venue almost always dominates the noise floor of our live music recordings, followed by the self-noise of the microphones.

The virtues of a multiple ADC design and 32bit recording needn't necessarily be linked, and I can see no benefit for myself or for other tapers in recording to 32bit, while I do see additional burdens.  That the files cannot be used as easily prior to conversion is something of an inconvenience, but the big thing is that the file size is 33% larger than 24bit. I record at 48kHz rather than 96kHz because it requires 50% less storage space and provides little to no loss of fidelity to my ear.   As for the processing argument, once imported to editing software the bit depth used in the calculation space is automatically increased to 32bit [edit- or more as wforwumbo mentions] and oversampling can be used were its beneficial, which some plugins do automatically.

I do realize that if I were using a multi-ADC recorder to record 24bit files using the same methodology as 32bit I'd need to normalize afterward, same as with 32bit, which works against the "used easily prior to conversion" aspect, but that's a burden I am willing to accept in exchange for the practical benefits of increased headroom and the ability to not concern myself with level setting at unfamiliar concerts.  All future recordings requiring 1/3 more storage space is not a burden I want to accept.

Based on all that it seems to me multi-ADC recorders writing 24bit files would be the most appropriate tools for live concert recording, as long as the manufacturers provide the capability of using their recorders in that way.

If anyone comes across a link to bench tests of multi-ADC recorders capable of saving to both or either 32bit and 24bit file formats, which specifically compare the results of both output formats, please post here or PM me.  Will need to confirm that the recorder uses its ADC in multi-ADC mode when recording a 24bit file.  Such a bench test comparison could answer this question definitively, at least for the recorder under test.  I do recall the SoundDevices representative posting here a few years ago when multi-ADC/32bit-recording was added to the Mixpres, confirming that they apply dither to 16bit output but not to 24bit.  Maybe that's because they determined the noise-floor of the signal chain up to that point is always higher, making the addition of dither unnecessary.  Testing for aliasing should reveal that.

[DSatz]

I hope this doesn't come as a blow to anyone's ego, but the data format used in 32-bit audio recording was defined by the IEEE back in the 1980s for other purposes, back before it was common for digitized audio to be processed on computers. There were no readily available hardware audio interfaces for that purpose, certainly not at the consumer level, and personal computers were scarcely powerful enough to be worth using for processing audio anyway. Who wants to wait minutes each time just to hear the effect of a simple editing decision? Plus hard disk space back then was limited and very expensive; in the mid-80s a hard drive with enough space to hold the contents of one audio CD would have cost something like $2,000 in today's dollars.

But over time the IEEE floating-point formats (plural--the one used in audio is one of a set) became an ISO standard as well, and Intel and other CPU manufacturers supported it in the form of native instructions of a peculiar sort: As you ran a program, if it contained those instructions, their execution would be farmed out to a physically separate "numeric coprocessor" that your main CPU could talk to IF you had paid the extra bucks for one. The Intel 80286 CPU in your IBM PC AT would talk to an (optional) 80287 numeric co-processor; an i387 co-processor was available for the i386. The story with the 486 is murkier and beside the (fixed or floating) point. If, however, like most customers other than science labs, you didn't splurge for the co-processor and wanted to run software that relied on having one, those instructions could be emulated in library routines. Those were of course much, much slower than dedicated hardware. On the other hand the internal coordination of the CPU with the co-processor also carried a certain amount of overhead. So ever since the Pentium hit the fan, numeric coprocessing capability has been integrated into all CPUs that you would ever find in a PC of any kind.

If a DAW operator is mixing many stems, trying out different EQ, reverb, levels and panning settings on each (some of which may be changing dynamically under software control), they don't want to have to wait minutes or hours for each proposed adjustment to "render" before they can hear what it will sound like. But they also want to leave the original stems unchanged so that they can revert any changes they've made, or revisit a decision later on. The more operations can be done in real time, the more flexible the software becomes, so on a dedicated workstation this can demand very large amounts of processing with multiple CPUs running in parallel. It makes anyone who remembers the original IBM PC with its 4.77 MHz 8088 (not even an 8086!) want to weep.

[aaronji]

^^ The ADCs in these recorders are all 24-bit, so it might actually be easier to record in 24-bit mode from the computational perspective. I think all current SD recorders are spec'd to 32-bit ADC precision, even though their top line models (Scorpio/888/833) don't even have a 32-bit floating point recording option. If you're interested in the math, read SD's patent. It mostly focuses on how they combine the outputs from the multiple ADCs. Pretty interesting, but not exactly beach reading...

[noahbickart]

Few amateur astronomers would spend five times as much for telescopes based on the supposed properties of their exterior paint; therefore such telescopes aren't generally offered to them. Few ham radio operators would imagine that signals sent through a transceiver carrying the insignia of a long-dead brand will reach farther than one with identical measured performance but no such brand; therefore such transceivers aren't generally offered to them. In consumer audio, however, we're totally swamped with such things. Even pro audio has bent in that direction over the time I've been in it. It's profitable to cater to the credulous; significant opportunities are lost if a manufacturer fails to do so.

^^^^^^^^^^^^
This is among the most important truths of audio.