If you really are trying to measure 45 dB return losses, you are going to have to spend considerable time learning about "error correction models".
In the old days, one would hook up a "moveable" 50 ohm load, and slide it to different positions. At each position you would measure S11 vs frequency. Then you used simple algorithms to compute the "center" of the S11 points at a single frequency, and that was your calibration vector.
In more modern times, you take a know reflective load, as in offset short circuits, or an E-cal unit, make the S11 measurements, and knowing what the calibration stadards actual data should be, you can calculate the correction factors. There are a number of patents held by Vahe Adamian (sp?) of ATN that explain the process.