The Problem with 192/32bit Recordings

The Problem with 192/32bit Recordings

Edit: this not only applies to 192/32 recordings, but additionally to 24 bit multi ADC devices recorded at 192 like the Sony D100 or certain Sonosax recorders 

 One of the things we refer back to most on the field recording slack is issues recording at 192/32bit

If you've ever run into this issue on 32bit recordings, you might have noticed weird bursts of noise above 50khz 

Pictures via Aaron Brown

How does 32bit recording work?

In a single ADC 24bit recorder the pipeline works kinda like this:

1. The microphone creates a very small electric signal that's fed into the recorder
   
2. That signal is adjusted by a potentiometer (your gain knob)
3. That signal is conditioned by a preamp
4. That amplified signal is converted into a digital signal by an ADC (Analog Digital Converter)

A 32bit recorder works like this (using the mixpre series as an example)

1. The microphone creates a small electrical signal fed to the recorder
2. That is amplified through 3 preamps
3. The 3 amplified signals are converted to a digital signal
4. A "data buffer" breaks each of those 3 signals into millisecond bite sized chunks
5. The lowest and middle gain settings are processed and a processor calculates a vector of those two signals. In layman's terms, it's making a synthesized version of the waveform between those two signals. The same processes is repeated between the middle an high gain signals
6. A processor now looks at the three original amplified signals, and the two synthesized signals and determines which of those signals are clean, and which are distorted, and which has the least preamp noise.
7. The processor takes the cleanest signal from one of the three amplifiers, digitally amplifies to match where your gain knob is set, then writes it to disk.
   

So effectively the recorder is constantly looking at three source signals and switching between the signals to generate a "perfect" waveform before it's digitally amplified back to the desired level. Any distortion that's created in the 32bit digital file can be lowered back down without clipping since the file contains metadata saying what the waveform should look like once it's lowered. 

Imagine if you had a picture on your computer that was too big to see on your screen, just because it's off your screen doesn't mean that information is lost. Once you shrink the picture it will fit on your screen. The same logic applies to the audio file, it's too big for your speaker, so once you shrink it down you can hear the whole sound.

Why is 32bit a problem in 192khz recordings?

The hypothesis the field recording slack settled on is if you're recording and two of the three signals distort, the recorder is forced to pick the lowest gain stage, which then needs to be boosted up significantly. Since the amplifier doesn't generate noise evenly across the signal, there's a large burst of hiss generated by switching to the lowest signal. 

An additional hypothesis is that the processor that determines when the signal clips doesn't look at ultrasonic material evenly with things in the human hearing range. 

So our observations have been loud sources like percussive sounds and sources with lots of ultrasonic material that's louder than what's in the human hearing rage tends to have artifacts.

One last warning

The problem isn't just recording in 32bit, it's the structure of how the 32bit recorders are designed. After some thorough testing by the field recording slack, we were observing the same problems even when switching back to 24bit recording on the same device! This same problem is also observable in other 24bit devices with multiple ADCs. Eg. Sony PCM D100

What are my options?

Until the problems are fixed, we'd recommend using 24bit recorders like the sound devices 700 series, Zoom F8, or another recorder from the previous generation of recorders. In the future, digital microphones might also be a solution to this problem, but that's a topic for another article.

Another thing to note, is the Zoom F8's duel channel record feature does not create the "ultrasonic tearing" so if you're worried about issues with 32bit noise but want to record a safety channel, this is a viable solution.


____

For those interested, below are diagrams of the mixpre series patent diagrams with layman explanations of how it works.

Additional examples of added noise supplied by Dmitry Chernov
Zoom F6 with Nevaton MC59H Shotgun Mic

1_Below threshold before switch up

2_Above threshold after switch up

3_Above threshold before switch down

4_Below threshold after switch down