OK, I've modified my Z measurement software for the MS420 to accept the B-WIC. The results are mostly good, with one remaining issue.
Here is a plot of three current sense R's. Each one is plotted twice; one with Open-Short-Load (OSL) compensation applied, and one without, so we can see what difference the OSL makes.
The formula (polar coordinates, HTBasic syntax) to apply to the measured S21 magnitude at each freq to get correct impedance value is:
RealS21 = 10^( (S21 - 26.956) / 20.) ! subtract the fixture's CH 1 divider from the MS420's dBr, and convert to a voltage ratio
Zmag_Ohms = ( ( 1. - RealS21) / RealS21) * 2.35 ! Series-thru impedance with 2.35 Ohm shunt R
It is worth noting that the phase term plays no part in these equations, and is carried through unchanged. All they're doing is scaling the magnitude vector.
OSL corrections are applied after this. When they are being measured, they are also processed by the above before being stored.
It should be noted; I do not have the standard short-load card for the B-WIC yet. One is on the way. I used a 3/4"-wide gold-plated shorting bar, and an 1/8W 0.1% 100 Ohm axial resistor for the OSL measurements. The difference between them and the factory-supplied standards could be impacting the results. I'll know in a few days when the standard one arrives.
Some observations on the results:
: The OSL correction is what makes it useful below 1 Ohm or so. Without it, the impedance accuracy is really poor down there.
: My OSL-corrected values run about 2% high. This could be because I didn't include the 50 Ohm source R of the generator in the divider value. And/or the 1% tolerance parts used in the fixture. Easy to correct for.
: Above a couple Ohms the accuracy becomes increasingly good enough to not use the OSL. This agrees with Agilent's findings with generic Series-Thru. At 10 Ohms the curves were identical up to the 10MHz plot limit.
: Overall, I would agree that, with OSL, 20mOhm is the lower limit for this topology. Below that, the trace gets too noisy to be useful. It is already too noisy at 25mOhm for my needs.
: These sweeps were done with a 10Hz IF bandwidth. To see if I could clean the trace up, I dropped the IF BW to 3Hz. The difference was minimal. Not enough to offset a tripling of sweep time. I then increased drive level by 10dB and dropped receiver sensitivity to match. No difference. The S21 short sweep is only 60dB down so this isn't a dynamic range problem.
: Below a couple Ohms, the Z curves with OSL are not accurate at high freqs. The 25m and 100m resistors are axial current sense types and should show more inductance than they do.
: The phase curves are not accurate above approx. 100kHz , with or without OSL, going capacitive while the Z was going inductive. The native sweeps on the analyzer showed it too, so it's not a math issue; it's coming out of the fixture that way. I did not run a "thru" sweep on the analyzer before doing these, the curves without OSL show a ~ -7ยบ phase mismatch between the channels. This is not in the analyzer; it is coming out of the fixture. OSL corrects the mismatch but the high-freq error remains. Perhaps Omicron applies an inductive constant correction factor?
I'll withhold further comment until the short-load pcb arrives.