No additional headroom at 32Bit?

I’ve been trying to make use of the additional headroom 32Bit float should provide when recording.

I use my Steinberg UR-24C, which is supposed to transform and transfer from its AD converter into Cubase DAW with 32Bit. I have set everything to 32-Bit, incl. the file record format and the internal precision. Yet, I do not get any additional headroom above 0db. It clips and if I lower the volume of the recorded 32-file, I get no clean signal, just the same hard cut, clipping signal at a lower volume.

Any idea what might be going wrong?

You do not record in 32bit float…
the UR-C series has 32bit AD converters (I’m not sure how they did that, the technical data is not like I would expect from a real 32bit interface)
but not 32bit float

:open_mouth: So the ad converter delivers 32 Bit integer? That’s a pity :frowning: Thanks for clearing this up.

a real 32 bit float converter would be much more expensive than an UR-24

Why precisely? It’s just about coding …

The UR series, as any interface in comparable price ranges, have at best 20 bits of usable data (actually rarely more than 18 bits), the rest is just electronic noise. The best ($$$) interfaces barely make it to 21 bits, the rest again is electronic noise.
It would be a (very!) long explanation involving laws of solid-state physics, but it’s very hard to do any better at normal ambiant temperatures, and running your device at cryogenic temperatures would be “impractical”… :wink:
So the 32-bit figure advertised by Steinberg is 100% marketing fluff (to stay polite). Running at 32-bit mode instead of 24-bit mode just gives you 8 more bits of useless noise… pure waste.
You do have some gain (about 2-3 useable bits) when using 24-bit mode instead of 16-bit mode, but no value whatsoever to move up to the 32-bit format. (for recording that is)

A few rare companies (ex. SD Mixpre) use a 32-bit float mode combined with 2 (or more) AD converters in parallel. The clever trick (one way to do it) is two have 1 ADC tuned for lower-level signals plus 1 ADC tuned for higher-level signals. An FPGA (typ.) then selects the data from the first one as long as it does not clip, then switches automatically to the other one when the signal gets hotter. The numerical conversion is done on the fly (typ. FPGA) transparent to the user, except a slight latency penalty… well… there is also a significant penalty on your wallet $$. (other DSP-based variants also possible, but basic idea is similar)

In any case, you still don’t have a true 32-bit dynamic resolution, but you get better headroom and immunity to clipping.
You don’t need a pricey interface. The best solution is still simply to lower your faders to stay at a clear distance from clipping.
By definition, you will always clip when reaching 0 dBFS, whatever the resolution used.

Sorry for the long explanation…
Short version:
If you measure your Christmas tree in millimeters instead of centimeters, you will have a better resolution/precision, but it will still be too large (clip) for your car, whatever the units… :wink:


Thank you very much for your effort. Actually, I’ve got a PhD, theoretical physics, so I am not too unfamiliar with solid-state physics. Yet, I am into theory, no engineer, so your explanation is very helpful.

Actually, I was really going for the additional headroom, the noise I don’t care about.

And I agree, the 32-Bit that Steinberg sells this device with is very misleading.

Hey, Steinberg guys. You really should write a warning message on the UR-C boxes: Attention! 32 bit integer = NO additional headroom whatsoever.

Some have speculated that the 32bit sampling has to do with accommodating the on-board DSP chip. Still not right for Steinberg to obfuscate the details, but that would explain why they went in the direction they did (and of course when marketing sees “32bit” they would instinctively turn it into a “feature”).