Electronics > Metrology
How to create VNA calibration coefficients (delay, C0..C3) by measuring devices?
wb0gaz:
I'd like to "derive" (for lack of the correct term) cal coefficients for a set of "standards" (open, short, load of both genders, so six pieces) which currently have none (and weren't intended as calibration standards.)
Is this possible and practical (from a DIY perspective), and if so, how would I go about this? Detail follow:
I'm thinking start with a known cal kit (HP 85033 series), measure the devices on (HP 8753 series) VNA, then use the resulting measurements (presumably as S-parameter listings) to generate the delay (and in the case of opens, the polynomial fringing capacitance values C0..C3) that would become the open and short devices' calibration coefficients. At that point, with some added (and for my purposes, inconsequential) added uncertainty, I'd have a second cal kit "derived" from the first. Load standard(s) might be taken as "ideal", given that in the HP 8753 series cal kit specifications, that seems to be the assumption.
If this is possible and practical (as a DIY project at home, not involving an outside cal lab or otherwise), what is the process that would take the measurement data (of the new devices, from the 85033-calibrated VNA, possibly as S-parameters) and transform that data into the delay and (in the case of opens) fringing coefficients?
If this is not possible and practical, any ideas on what the impediment(s) would be?
Thanks for any comments, suggestions, or requests for clarification.
virtualparticles:
It can be done. Measure the standard on a golden VNA and save the S1p file. Measure the S11 delay to the short or the open. Half of this value is the one-way delay. Pull the data into Excel and create a calculated reflection based on the known delay and guessed values for C0 through C3 or L0 through L3. Use the Solver to optimize the coefficients for best fit. That is, you'll create a column which is the squared difference of the real part and the other the imaginary part. Add the sum of both error columns together to get a total error term and minimize it by changing the coefficients. Using Excel to solve multivariate problems is pretty cheesy, but I don't have time to mess around and it works. You may find that the delay term alone is close enough. The effects of the capacitance and inductance are quite small at moderate frequencies.
Keep in mind that negative coefficient values are allowed.
wb0gaz:
Thank you so much for the approach information!
I'm already using the derived cal kit just with delay (that was easy to determine) and a guess for the C0 parameter (the HP 8753 series assumes L0=L1=L2=L3=0.) The cal kit need only work to 3 GHz.
I'll get started per your outline as it looks like it will be an outstanding learning experience and provide good utility.
It's very likely I'll come back to ask for "what do I do now" next step (the first challenge will be to get S parameters out of the VNA and into a spreadsheet, as it has no floppy or other media and I don't have GPIB attached.)
Thanks again!
wb0gaz:
Hello Virtualparticles,
I'd like get your comments on my first steps towards deriving a cal kit.
The experimental test set-up is a HP 8753A VNA with 85044A test set. The "source" (reference) cal kit is an HP 85033C type. The APC-7 connector on the test set is adapted to 3.5mm/SMA female using a Wiltron APC-7/3.5mm female adapter. All connector surfaces were pre-cleaned using isopropyl and lint-free swabs. This is my "golden VNA" as you described in your initial response to my posting.
The "destination" (target) of the exercise will be a pair of SMA-male PCB ("vertical" type) connectors. The short "standard" has been shorted with a blob of solder on the pins. The open "standard" has had all pins filed flush and cleaned of metal particles as well as I could.
The 8753A was left at default "reset" settings except for the following: cal kit was selected as 3.5mm (the 85033C is one of the factory cal kits available.) The number of points was changed from 201 to 11 (mainly so that I could eventually transcribe a smaller amount of data when time comes to make up the initial spreadsheet.)
The 8753A was calibrated S11 using the 85033C standards (the open standards have small center contact extensions that are tedious to use, so that's the principal reason I'd like to derive a less-golden, but mechanically friendlier cal kit!). I then installed my target male cal component and used port extension to find the offset closest to a "dot" on the left (0-ohm) end of the smith chart (that's the photo attached "short-smith.jpg", showing the round-trip delay with the male component installed; the round-trip delay on this component differs from the 85033C male short/open components as expected.)
The information of interest at this point would be the round-trip delay with the short target standard installed; I'd expect that the same delay would apply to the open standard as the two male connectors used for the project are identical (until they were made into the short and open standards as described.)
With calibration and port extension unchanged, I then replaced the male target with it's female counterpart (that's the photo attached "open-smith.jpg", leaving the delay unchanged.) I then used copy->list to display the 11 test points as R/X components. The 8753A doesn't have a built-in option to display S parameters, but I believe I can convert the R/X components to their corresponding S parameters as needed.
I'd expect at this point that I want to measure the artifacts of fringing capacitance of the open standard, with the delay (to the mechanical open point) removed from the S-parameters that I end up curve-fitting C0..C3 coefficients.
My question is this - am I getting started in the right direction?
Thanks!
Dave
(edited to replace the three screen images as the original three were incorrect due to calibration procedure error.)
virtualparticles:
Yes, you're headed in the right direction. Keep in mind that the measured S11 delay is round-trip but the cal kit definition is one-way, so you'll need to divide by 2. At the end of the day, all you need is something that mathematically recreates the curve on the screen. The VNA will perform the calculation and compare its results to what it is measuring in order to create the correction vectors.
For more on how that calculation is done, see this article by yours truly:
https://www.mwrf.com/technologies/test-measurement/article/21152010/copper-mountain-technologies-one-port-vna-calibration-a-look-under-the-hood
Navigation
[0] Message Index
[#] Next page
Go to full version