(re) Review of Sound Devices MixPre-3 II Multichannel Recorder Review - Advanced mode

[andy3d]

1st generation UCX, 1st generation MixPre3
Unfortunately my old UC is no longer able to play clean sweep tone, so could not add it to the comparison.

 

[Rja4000]

J'ai public une version raccourcie de cet essai, en français, sur Audiofanzine.

I published a shorten version of this review in french on Audiofanzine.

 

[Scott_Williams]

Nice review! The lights was really cool.

 

[Bergante]

That's another subject. But dynamic range is also a key benefit, IMO.
DPA claims >124dB dynamic range, as an example.

That depends on the intended application. The DPA microphones have a very high maximum SPL which makes them awesome for close miking of musical instruments. But for other applications it is not that critical!

 

[signalpath]

Sound Devicies marketing department keeps their exact way of doing this secret, using just fuzzy words.

Great review.

Sound Devices was recently acquired by a private equity group known for dubious marketing.

I'm not clear why they call a two-path ADC "floating point". Floating point is a mathematical construct, not a signal processing technique. Moreover, maybe they forgot that "32-bit float" math cannot guarantee bit accuracy beyond the 24th bit. Zoom uses the same phrasing. It's a poor, confusing precedent.

Cirrus Logic has been publishing a slew of patents for this multi-ADC process. They call it "multi-path" processing, and that makes a lot of sense. It also makes it easy to distinguish between legacy "single-path" and emerging "multi-path" processes.

IMO, any audio process that uses more than one dynamic processing path should be called "multi-path".

(history: over the decades, multi-path audio circuits have been called many things: ranging, gain-ranging, auto-ranging, multi-range, multi-path, heuristic, floating, parallel, cascaded, adaptive, etc... of all the choices, "floating point" is probably the worst).

 

[chelgrian]

Great review.

Sound Devices was recently acquired by a private equity group known for dubious marketing.

I'm not clear why they call a two-path ADC "floating point". Floating point is a mathematical construct, not a signal processing technique. Moreover, maybe they forgot that "32-bit float" math cannot guarantee bit accuracy beyond the 24th bit. Zoom uses the same phrasing. It's a poor, confusing precedent.

Cirrus Logic has been publishing a slew of patents for this multi-ADC process. They call it "multi-path" processing, and that makes a lot of sense. It also makes it easy to distinguish between legacy "single-path" and emerging "multi-path" processes.

IMO, any audio process that uses more than one dynamic processing path should be called "multi-path".

(history: over the decades, multi-path audio circuits have been called many things: ranging, gain-ranging, auto-ranging, multi-range, multi-path, heuristic, floating, parallel, cascaded, adaptive, etc... of all the choices, "floating point" is probably the worst).

It's because they record in 32 bit float format, which has been annoying to process for years for example Logic Pro only added support in version 10.8 in November 2023.

The dual path ADC is just a means to produce a signal with sufficient headroom before clipping that there is any point in capturing in 32 bit float in the first place rather than 24 bit integer. You really aren't going to miss the loss of single bit resolution at the extremes where it happens.

Quite a lot of pro audio gear processes internally in 48 or even 64 bit formats.

 

[Ifrit]

for example Logic Pro only added support in version 10.8 in November 2023.

Shows you how far behind Apple is in this respect. 32-bit float processing has been in DAWs for at least 20+ years, same for recording files in that format.

 

[signalpath]

It's because they record in 32 bit float format, which has been annoying to process for years for example Logic Pro only added support in version 10.8 in November 2023.

The dual path ADC is just a means to produce a signal with sufficient headroom before clipping that there is any point in capturing in 32 bit float in the first place rather than 24 bit integer. You really aren't going to miss the loss of single bit resolution at the extremes where it happens.

Quite a lot of pro audio gear processes internally in 48 or even 64 bit formats.

Maybe I'm misreading their literature, but they seem to refer to their 2-path ADC process as "floating point" (the dual-ADC process, not the recording format). This makes it even more confusing. A 32-float format has nothing to do with a multi-path ADC topology. It's apples and oranges.

A20-Mini 32-Bit Float Recording - Sound Devices

The A20-Mini wireless microphone transmitter is the industry’s first and only transmitter which features built-in 32-bit floating point recording. This is very significant: with 32-bit float recorded files, audio gain decisions can be fully made after recording. This may sound provocative to an...

www.sounddevices.com

Virtually all DAWs today are designed on 32-float engines. Some (Magix, etc) have 64-float master process, with 32-float sub-processes. Only two DAWs I'm aware of are currently true 64-float: Reaper and the Steinberg suite. As multi-path processing (ADCs, DACs, power amps, etc.) become a thing, and our systemic noise floor drops by one, even two, orders of magnitude, we will need bit-accurate DAWs beyond the 24th bit.

 

[chelgrian]

Shows you how far behind Apple is in this respect. 32-bit float processing has been in DAWs for at least 20+ years, same for recording files in that format.

While various other DAWs have technically supported 32 bit float as an input and output format for a decade in fact there are large numbers of bugs with things like meters reading in 24 bit and some output paths clipping if you try and output anything beyond a full range 24 bit signal.

Everything, including Logic, has done internal processing at higher resolution for yonks but we are talking about the I/O format. You always want to process at a higher resolution than your I/O format and Logic historically processed using 32 bit float internally. I expect that they wanted to use 64 bit internally for dealing with 32 bit float as an I/O format and they only got round to doing that recently.

 

[Ifrit]

Yet again, most of the common DAWs have been doing 64 bit float internal processing for more than 10 years now. Although I remember one of the Logic fans bragging that point about 7-8 years ago, claiming that’s the only DAW doing that, which wasn’t true.

 

[chelgrian]

Yet again, most of the common DAWs have been doing 64 bit float internal processing for more than 10 years now. Although I remember one of the Logic fans bragging that point about 7-8 years ago, claiming that’s the only DAW doing that, which wasn’t true.

You know you could give the Apple bashing a rest it's deeply tedious.

 

[Bergante]

Great review.

Sound Devices was recently acquired by a private equity group known for dubious marketing.

Audiotronix? They own several high profile pro audio manufacturers such as Digico, Allen&Heath, Calrec…

Which dubious marketing do you mean?

Vitec has played dirty recently, but not Audiotronix as far as I know.

 

[signalpath]

Audiotronix?
Which dubious marketing do you mean?

They have a new mixer which uses the identical name and font as our classic recording channel, the Origin. When I notified their marketing VP, he said, "oh, just a coincidence".

Ha. What's that old saying? I guess we should be flattered.

.

 

[Bergante]

Ouch I stand corrected. I didn’t know that!

 

[Ifrit]

You know you could give the Apple bashing a rest it's deeply tedious

Apple fanboy gets irritated by the facts? What else is new... I don't care what platform to use, and use Apple and Windows all the time; but facts are hard to ignore.

 

[Rja4000]

I just found this illustration of the way this works

NB: This is coming from Tascam's portal, where they explain how their 32 bits float Tascam Portacapture serie works.
So let's give back to Caesar what is Caesar's ;-)

 

[Bergante]

There are several possible approaches. Zaxcom for instance, has a patented system called NeverClip based on a similar idea. They don't actually produce a 32 bit float file but the result is roughly equivalent, preamp gain is not so critical.

 

[FieldRecordist]

SD patented their technology some years ago: https://patents.justia.com/patent/9654134.

 

[Rja4000]

SD patented their technology some years ago: https://patents.justia.com/patent/9654134.

Thanks !
I found the same with the drawings:

US9654134B2 - High dynamic range analog-to-digital conversion with selective regression based data repair - Google Patents

A multi-stage analog-to-digital conversion method and system use window functions and translation to match high gain frames of data to target frames of data. The technique selects window data packets for the output stream based the stage of data having the highest gain satisfying selection...

patents.google.com

"FIG. 2 is a simplified block diagram of a two stage version of an array based high dynamic-range ADC that includes additional details relating to window translation and selection in accordance with another exemplary embodiment of the present invention."