Hi Matteo,
I made some suggestions via email, were you able to test those yet? In both of my suggestions, the attenuator would not be part of the cal, but would be inserted afterwards and compensated using one of the techniques below.
The problem with having the attenuator in during the cal is that you will drive the signal into the noise and the unit will not be able to make a good cal.
Here is the main contents from that email.
Port extensions would likely work, there is both controls for setting a typical path loss curve and a fixed DC offset (for narrow spans or coarse measurements). In this scenario the attenuator would be inserted post calibration.
Deembedding would be a more accurate approach. Do the cal without the attenuator. Then insert and measure the full 2 port s-params of just the attenuator. Save the measurement as an S2p file, and use that file for deembedding.
