I was wondering also about how the cal compensation in REW works - whether or not this leads to an increased reading of the noise floor by ramping “apparent” input gain to compensate for loss of frequency response? Then there is also the possibility of filter leakage at very low level from the D/A converter……
I'm not an REW expert, so just my interpretation so far of how it is working from using it, with rather limited reading...
I don't think it is doing anything as fancy -or potentially confusing- as that. It just subtracts one curve from the other, and the curves are just straight measurements.
The cal has to just reflect what REW finds digitally between what it sends out and what it gets back and take that as an adjustment curve to factor out any quirks of the interface itself. It's not a perfect system as it can't easily account for all factors such as the different electronic characteristics of the gear you might subsequently connect to the interface afterwards which might change it a little (as you can see in the different bass rolloff in my two cal curves for example). Other calibrations can be added to the measurements though, especially -when it comes to acoustics meaurements- the measurement mic's freq response. Probably just stating the obvious here. I think when it comes to acoustic measurements then effects due to small details of the interface, are, ahem, in my experience lost in the noise
I don't see how REW can make any assumptions about the frequency response - that would defeat the purpose of actually trying to measure it wouldn't it?!
Re the noise issue I ran into, I was slightly horrified at first how noisy the interface seemed to be at those frequencies, but I'm not used to working at those frequencies and sampling rates etc and I wonder how the interface would measure when set back to the humble 44.1 I normally work at. If I get time I'll try that.
The two 'cal' curves above do not look anywhere near as bad as the first-efforts picture I posted. I experimented with averaging over multiple sweeps but in the end - having discovered the 'sweep length control - just used a longer sweep which has the effect of removing the noise effect on the sweep signal (by 3dB per doubling of the sweep length IIRC) . This produced curves which looked smooth enough to work with except at the very end of the line, even the unpadded one. I then used the same longer sweep setting as well as using the 40dB pad when taking the readings with the pre-73 in the loop.
'Filter leakage' (I'm not really sure what that is exactly TBH) is the kind of thing I was suspecting as being an artefact from the DAC which might be the source of the noise, and hence I was wondering whether the behaviour might change at lower sample rates. ?