32 bit audio interfaces/ ultra low noise floor

[Icarus88]

Hello

I wanted to ask your opinion on the new 32 bit interfaces that have been hitting the market. The idea is that these interfaces capture input with such a large dynamic range that gain staging becomes irrelevant. From what I understand this is achieved by capturing two signals (one at a low gain, the other at high gain) and then interpreting between them seamlessly in software. The idea is that you get a much larger dynamic range and can theoretically turn up the "input gain" in post.

An example of these interfaces would be the Zoom UAC-232.

Does anyone have first hand experience with this type of tech? I'm specifically interested in the possibility of ultra low noise floors, as I use high gain amp sims, where applying 80db of gain is typical, meaning even a SnR of 100db (typical for consumer grade interfaces) means you get an audible -20db noise floor in your output (necessitating a noise gate at the front of the signal chain).

This thread is about bass guitar rig.

Thanks

 

[welwynnick]

My immediate reaction to this is "not a chance!". The number of bits isn't a limitation at the current state of the art, though it's getting close. However, a budget audio interface is hardly going to break records. The EIN is -127dBu, so it's good, but doesn't move the goalposts.

Having said that, I saw Julian Krause's review of the UAC-232, and he did take some very good measurements, so have a look at this.

 

[DVDdoug]

Analog (and acoustic) noise is worse than 24-bit quantization noise. The digital side of things isn't really the problem.

The main advantage is on the "loud end" where you can go over 0dB without clipping. I expect floating-point interfaces to become standard "soon". But overall, I don't think there is more usable dynamic range. I don't know how much headroom the Zoom units have... There is some (unknown to me) voltage-limit on the loud-end, and noise on the quiet-end so you are nowhere near the digital dynamic range of 32-bit floating-point (which is virtually infinite for all practical audio purposes).

This thread is about bass guitar rig.

And... electric guitar and bass are notoriously noisy!

 

[phofman]

IMO it's a clever and inexpensive way to solve the range issue. Those devices have two (or maybe more?) standard ADC chips with resolution < 24 integer bits, each ADC preceeded with a fixed attenuator/preamp to cover the whole dynamic range in total. Then a simple logic merges the ADCs output streams into one float32 stream, taking into account the attenuation/amplification factor of the given ADC which has not been clipping yet (clipping is easily recognized from the stream, also some ADCs report the clipping via registers, some even much faster via hardware pins - e.g. the famous ES9822Pro).

Since the UAC2 specification supports float32 (and general UAC2 drivers in major OSes accept that format), it's simple to pass the data without having to develop expensive proprietary drivers.

An option would be to pass the multiple int24/int32 streams from each ADC separately and do the merging/DSP on the host side - but that could not make use of the fast cliping conveniently reported via the ADC pins).

 

[welwynnick]

Thanks. My initial scepticism was misplaced. I feel like I've learned something today.

 

[somebodyelse]

It's a similar idea to this patent from Analog Devices for a wide dynamic range microphone. Or merging multiple camera outputs for HDR video.

 

[Tell]

This sounds like the exact same thing as HDR in photography (no I'm not talking about cheap looking extreme tone mapped images), capture the image at different exposures settings and combine them into one with higher dynamic range than those single exposures alone. Though I'm not exactly sure how it works on the technical side when it comes to audio, but I bet it works just fine.

 

[AnalogSteph]

This sounds like the exact same thing as HDR in photography

Same idea, yes.

Composite ADCs have been around for at least 3 decades in the pro audio world. Quoted specs for DG's system that came into use almost exactly 30 years ago include

THD+n of -121dBFs with an input of 997Hz at -30dBFs and linearity errors within 1dB down to -135dBFs, together with a largely flat noise-spectrum.

(In this case that means that the "bottom" ADC would have a -103 dB THD+N at -12 dBFS. The CS5390 used was rated 110 dB(A) for DR and -100 dB THD+N so together with the quoted ability to record at 96 kHz - which it can't natively do - I suspect that they threw a few of them at the problem, and no doubt several Motorola 56k DSPs as well, similar to the later Pacific Microsonics Model Two. By 1998 you could buy devices with composite ADCs off the shelf if you have a few ten grand to spare, or a hundred grand for a mixing console.)

Artifact-free crossfading doesn't seem to be quite as easy as some might think, but it does seem to be doable.