question about LO leak and LO+2*IF leak in a transmit branch

hi,all.
Recently I designed a transmit branch,including:mixer+filter+driver 1+driver 2+
PA(use GaAS FET matched by cap and ind). Now comes a problem: when measure
the branch at the driver2's output,the LO leak and LO+2IF leak are both about -70dBc(the PA doesn't work). But when measure at PA's output the LO leak is about
55dBc and the LO+2IF leak is about 50dBc. The output of PA hasn't reached P1dB(that is: the linearity is still good),and the IM3 is also good. Then why the LO leak
and LO+2IF leak becomes so strong, it badly disturbs me for several days.

I wish to get help from you!
Best Regards
L.J.zheng

Try to verify the gain of the PA at LO frequency.
Second reason could be a bad layout.

What's your PA's transfer function? plotting that out maybe helpful.

I measured the PA, the gain of PA at RF and LO and LO+2*IF frequency are almost the same(difference<1.5dB). For the RF is in UHF band, and the IF is only 21.4Mhz.

In this case make sense to have higher LO level at the output of the PA.
If you fail a spec with this issue, have to filter the LO signal in the front of the PA.