I am trying to estimate the IIP2 of a double balanced gilbert cell mixer.
To model the transconductor mismatches I am using dc voltage offsets at one of the gates.
I have also introduced mismatch in the load resistor.
I did a sweep QPSS analysis with two equal tones (f1 =  2.45 GHz, f2 = 2.455 GHz ) and LO = 2.443 GHz
When I plot the IP2 curves the slope of the f2 - f1 component is close to 3 instead of 2.
But when I plot the single ended output current of the mixer at f2 - f1 it shows the proper second order behaviour ( slope = 2 ).
Once I take differntial output the slope becomes close to 3.

What could be the reason for this strange behaviour?

thanks
Raja Prabhu