32bit float and outboard Mic Pre's

[digiot]

Just wondering how using out board Mic Pre's would effect the benefits of 32bit float recordings. The main benefit for me to record at 32bit float on the Carbon would be the ability to not stress about input levels and clipping.

Up till now I usually record at 24bit. When I record live bands I run all mic's in to my outboard Mic pre's (a mixture of API and Neve style 500 series). Up to 16 total mic's depending on the band. The 500 series racks run in to my Apogee AD16X that goes ADAT in to my Carbon. I usually run at 48k but if I want to run at 96k I run 8 of the 500 series in to the Carbon DB25 inputs and run SMUX ADAT out of the AD16X for the remaining 8 500's.
I am aware that if I run at 32bit I lose the ADAT inputs (24bit only), but how will the outboard Mic pre's connected analog to the Carbon benefit from the extra head room? If I don't set levels properly on the Mic Pre's they will surely overload the pre's before it gets to the converter. So does the benefit of 32bit float recording only apply to the Carbon's build in Mic Pre's?

Maybe I am totally overthinking this but since I usually run sessions alone (no assistants or interns), not having to keep an eye on 16 Mic Pre's would be a godsend.

[its2loud]

32bit float will not save you from over driven analog mic Pre’s. It does nothing to protect the analog signal chain. It only protects against digital clipping once it hits the A/D converters. If it’s coming out of the analog hole of your mic pre with too much gain, you’re stuck with it as is.

[digiot]

Originally Posted by its2loud

32bit float will not save you from over driven analog mic Pre’s. It does nothing to protect the analog signal chain. It only protects against digital clipping once it hits the A/D converters. If it’s coming out of the analog hole of your mic pre with too much gain, you’re stuck with it as is.

Yeah that's pretty much what I thought. What about the built in Mic pre's on the Carbon though? Those are in front of the AD as well.

[its2loud]

Originally Posted by digiot

Yeah that's pretty much what I thought. What about the built in Mic pre's on the Carbon though? Those are in front of the AD as well.

https://cdn-www.avid.com/-/media/avi...et-english.pdf

[LDS]

Carbon uses 32-bit fixed point converters, not floating point. They will clip at 0dBFS just the same as any other fixed point converter.

[K Roche]

Originally Posted by LDS

Carbon uses 32-bit fixed point converters, not floating point. They will clip at 0dBFS just the same as any other fixed point converter.

OK I am guessing the confusion comes from the session file bit depth option being "32-bit float ." for processing
And Avid's info being a bit ambiguous saying "32/192 conversion" and "32-bit float precision audio streaming".

[Darryl Ramm]

I suspect the converters are "32-bit float" but like all converters the internal representation will be fixed number of bits (often 24 bits... which conveniently matches the 32-bit float sign+mantissa). But yes a 32-bit float convertor still clips at 0dBFS... unless the vendor has shifted the level down going into the converter (and then digitally shifted it back up after conversion) to give it more headroom. And if Avid did that I sure as heck would expect them to explain this, and for folks here to notice

My Sound Devices field recorder does level shifting and outputs 32-bit float... a nice feature for Sound Devices' intended market.

Some claims of "32-bit" converters made by some manufacturers (not Avid) can be annoying stupidity. Occasionally see this pop up with consumer products. 32 bits however you slice it, fixed or float, is far far below the thermal noise floor, a physical impossibility to be useful.

I don't know what Avid's goal is here. It could genuinely be that the particular 32-bit float converters have better performance than many others, or Avid likes them for other reasons... maybe ease or integrating into 32-bit float support in AVB, or maybe Avid was thinking of adding level shifting in future (might only be a firmware update), who knows. But they'll effectively have 20-ish to 23 bits ENOB... where good ADCs operate today. That ENOB metric... Effective Number of Bits is a nice shorthand to compare ADCs.

I expect Avid here do good engineering and these converters to be great, but I'd prefer if all vendors would provide more detailed measurements like THD+N, low level sine wave signal plots, etc to show they have the engineering is done really well. And some of that is just engineering show-off, may be unrelated to what people can hear in practice, but beautiful engineering, and lack of some common flaws, is nice to see.

Again as everybody else has said these will clip at 0dBFS (unless Avid hid some level shifting here, and they have not disclosed that).

[LDS]

Originally Posted by K Roche

OK I am guessing the confusion comes from the session file bit depth option being "32-bit float ." for processing
And Avid's info being a bit ambiguous saying "32/192 conversion" and "32-bit float precision audio streaming".

Totally. The marketing is probably deliberately ambiguous. More must be better, right?

ESS, Texas Instruments, Analog Devices, etc are all making 32-bit integer converter chips nowadays. A lot of audio interfaces use them. MOTU M4. RME ADI-2 Pro FS R. They are in Antelope stuff. They are in Apogee stuff. Only a select few manufacturers seem to be deciding to give the 32-bit aspect a lot of attention in marketing, probably because so much of the performance gains disappear below the analogue noise floor.

I do think they probably have some benefit on better quality audio, particularly when it comes to the precision of the digital aliasing filters. Enough to flip tables for? Absolutely not. More the aural equivalent of drinking diet Coca Cola with your fast food meal because you want to lose weight...

What Sound Devices is doing is quite different (and patented as far as I am aware). They are using a pair of ADC chips in parallel per input, each specifically set for low or high gain. From the output they derive the 32-bit floating point audio.

Even in electrical engineering it probably gets a little confusing as modern converter chips support mono-mode, a process of using multiple channels in a multichannel chip. It is designed to reduce noise and distortion rather than increase dynamic range about 0dbFS though.

[K Roche]

Originally Posted by LDS

Totally. The marketing is probably deliberately ambiguous. More must be better, right?

ESS, Texas Instruments, Analog Devices, etc are all making 32-bit integer converter chips nowadays. A lot of audio interfaces use them. MOTU M4. RME ADI-2 Pro FS R. They are in Antelope stuff. They are in Apogee stuff. Only a select few manufacturers seem to be deciding to give the 32-bit aspect a lot of attention in marketing, probably because so much of the performance gains disappear below the analogue noise floor.

I do think they probably have some benefit on better quality audio, particularly when it comes to the precision of the digital aliasing filters. Enough to flip tables for? Absolutely not. More the aural equivalent of drinking diet Coca Cola with your fast food meal because you want to lose weight...

What Sound Devices is doing is quite different (and patented as far as I am aware). They are using a pair of ADC chips in parallel per input, each specifically set for low or high gain. From the output they derive the 32-bit floating point audio.

Even in electrical engineering it probably gets a little confusing as modern converter chips support mono-mode, a process of using multiple channels in a multichannel chip. It is designed to reduce noise and distortion rather than increase dynamic range about 0dbFS though.

Interesting Avid says Carbon is using 4 ADC chips per channel (but I don't know the configuration series or parallel ?)

[Darryl Ramm]

Originally Posted by K Roche

Interesting Avid says Carbon is using 4 ADC chips per channel (but I don't the configuration series or parallel ?)

They would be parallel, it's to average the signal/reduce the random noise. Meaning they see the same input signal and then their digital outputs are averaged. With four converters this should give a 6dB reduction in uncorrelated noise.

BTW now I see that Avid claim 126 dB dynamic range for these the Equivalent Number of Bits is

ENOB = (+126 dB - 1.76)/6.02 = 20.6

So I was a few bits too eager there.