I am doing some measurements on a frequency converters with a N5242A. Connected to the VNA, I have 20 inches of cables that connect to a switch matrix. At the output of the switch matrix I have approximately 10 feets of cable up to the frequency converters. The total insertion loss between the VNA and the frequency converter is approximately 4 dB at the RF frequency (1.6 GHz).

My calibration plane is at the input of the switch matrix just after the 20 inches of cables, I use an e-cal to do the calibration. I also measure the s-parameters of the switch matrix + 10 feet cables and I use the s2p files I get to de-embed the measurements of my UUT. I chose to do this because I have a high number of path between the analyzer and the frequency converters and it was not practical to calibrate the analyzer directly at the converter ports. I get some great results for all of my measurements (conversion gain, noise figure, IMD3, group delay ...), matching nearly exactly the results I get at the calibration plane. However, for the return loss measurement, I don't get good results.

I have made some tests with a 50 Ohm load. When I do the VNA calibration at the frequency converter ports, and I measure the 50 Ohm load there, I get a 50 dB return loss. If I de-embed the measurement like described above, I get something close to 35 dB. I don't understand why I get this result. The network I de-embed starts exactly at the VNA calibration plane and ends exactly at the measurement plane.

Any idea of what could be wrong with my setup?

My calibration plane is at the input of the switch matrix just after the 20 inches of cables, I use an e-cal to do the calibration. I also measure the s-parameters of the switch matrix + 10 feet cables and I use the s2p files I get to de-embed the measurements of my UUT. I chose to do this because I have a high number of path between the analyzer and the frequency converters and it was not practical to calibrate the analyzer directly at the converter ports. I get some great results for all of my measurements (conversion gain, noise figure, IMD3, group delay ...), matching nearly exactly the results I get at the calibration plane. However, for the return loss measurement, I don't get good results.

I have made some tests with a 50 Ohm load. When I do the VNA calibration at the frequency converter ports, and I measure the 50 Ohm load there, I get a 50 dB return loss. If I de-embed the measurement like described above, I get something close to 35 dB. I don't understand why I get this result. The network I de-embed starts exactly at the VNA calibration plane and ends exactly at the measurement plane.

Any idea of what could be wrong with my setup?