If the cal is good the 50 Ohm load can be de-embedded, provided the amplifier is stable for the VNA set-up. That's the measurement part of it - the design is a different issue.
What is the sensitivity to an s22 cal error of δ? Assuming δ is real, then Z
o is off by a fraction 2δ. The corresponding error in voltage gain would be (1 + 2.δ)(1 - 2.δ.Z
o/R
L). If the s22 cal is good for 30 dB (δ=0.03) the error in voltage gain for R
L=250Ω is below 2%. A 30 dB on-wafer cal can be achieved out to 10 GHz even with an archaic SOLT substrate once the cables have warmed up.
M.G.Rajan
www.eecalc.com