Understanding VNA error terms and what calibration actually removes

Short write-up of what we found on the bench, in case it saves someone a week.

1. The twelve-term model for a two-port accounts for directivity, source match, load match, reflection and transmission tracking.
2. Calibration moves the reference plane and removes the systematic error, not the random error.
3. Uncalibrated drift over time is not removed; that is why you re-verify with a check standard.
4. Knowing which term dominates your accuracy tells you what to improve in the setup.

Open to alternative explanations.

Solid VNA traveler note — the twelve-term model names the systematic terms calibration can remove; it does not remove drift, connector repeatability, or a bad check-standard habit.

Editorial keepers we will pin next to every "the VNA said cal passed" claim:
1. Calibration moves the reference plane and subtracts systematic error (directivity, match, tracking); random error and post-cal drift stay.
2. Knowing which term dominates your uncertainty tells you what to improve (adapter stack / torque vs source match vs cable flex) before blaming the DUT.
3. Re-verify with a check standard (known attenuator / airline) at session end, not only at start — that is how you catch drift the error model never saw.
4. Freeze kit identity, torque, adapter stack, and the check-standard readback on the same sheet as any S-parameter claim you will ship or publish.

Open: which error term usually eats your uncertainty budget first at the band you care about — and do you forbid freezing DUT data without a post-cal check-standard?

— rfforum 编辑部 / Editorial Team

On the bench at X-band the term that usually eats my uncertainty first is effective directivity after a worn or loosely torqued load standard — residual |S11| on a good load creeps from <−40 dB toward −30 dB and then every reflection measurement lies.

Practical sequence I use after a SOLT:
1) Immediately measure a known check attenuator / airline; freeze those numbers on the traveler with kit ID, torque, and adapter stack.
2) Flex the test port cable once; if S21 phase moves more than your spec allows, re-dress or replace before touching the DUT.
3) Re-check the same standard at end of session. Drift that appears only then is temperature / connector, not something the twelve-term model removed.

I would forbid shipping DUT S-data without a post-session check-standard readback. Curious what residual directivity people tolerate before they retire a load standard.

Interesting direction, will keep watching. Please update if there's progress.