Do you measure 3% offset in your through standard or in a DUT Waveguide? Did you check if your DUT and connectors are still fine? If |S11|^2 + |S21|^2 > 1 it normally indicates radiation or ohmic losses (corrosion on your DUT?).

Best regards

Thank you very much for your answer, Martin.

In TRL calibration I measure Reflection, then Thru and, finally, Line.

Then I use this Line as a DUT and measure S-parameters. It is an empty piece of waveguide.

I intend insert a dielectric sample and measure S-parameters for calculation of epsilon. This first measurement was a test for accuracy.

All my parts are clean and fine.

Best regards,

Valery

Ok. I guess that if you measure the Thru after the calibration everything is fine? Are the S11 and S22 (S21 and S12) identical? I once had an amplifier fail on a Agilent PNA similar to your VNA (Source was switched between the ports). This resulted in an unleveled source on one port. Do you have a power meter to verify if everything is fine? If you have a time domain option you could also check if there are undesired reflections in your calibration setup.

I measure 201 points of S parameters from each side, total 402, see the picture attached. Now I found that the shift of S21 changes if I disassemble the scheme and then assemble again, though I had several times the same 3 % deviation. In my last measurements I had -0.21 % and +0.49 %. It looks like it depends on accuracy of assembling.

In any case I am very thankful to you for your interest.@