Gain compression with top-level behavioral simulations

Hello everyone,

I have a "problem" concerning gain compression in TX top-level testbench considering overall system.

The TX budget is as below:

RF side -- external rf diplexer -- external PA -- external RF filter -- rf amp -- 2nd mixer -- external if filter -- 1st mixer -- if amp -- external if diplexer -- IF side

all RF blocks has 50 ohms ref. impedance and IF blocks 75 ohms impedance.

I am using behavioral models in verilog-a in each block. I am having compression gain in 2nd mixer (specified in the model as 7 dB and simulated 4.5 dB) and rfamp (specified 10 dB and simulated 8.5 dB). Also, external PA had simulated gain 2 dB more than specified.

I am trying to understand why it is happening because apparently blocks linearity is ok to avoid gain compression.

I thank if anyone has an idea about this.