A dual conversion receiver has two frequency translation stages. The first mixer translates from RF to IF; the second (IF) mixer translates from IF to baseband.
This architecture is supposed to help avoid the problem of DC offset, caused by LO leakage to the mixer input followed by self-mixing of the leakage down to baseband.
Direct conversion receivers, which have only one frequency translation stage, suffer from DC offset because the LO leakage energy is immediately translated to baseband. The dual conversion architecture takes care of the LO leakage from the first mixer by ensuring that it appears at an offset of IF from the signal of interest.
But the LO leakage of the IF mixer, if any, is translated directly to baseband when it self-mixes. Why isn't the LO leakage from the second (IF) mixer equally as problematic as the LO leakage in the single mixer of a direct conversion receiver?
I've read one answer in this book:
DC offset due to self mixing is introduced when the IF LO signal leaks to the input of the IF mixer and then self mixes to produce a DC output in the analog baseband. This on-chip IF leakage is more deterministic than its direct conversion counterpart in which the LO can leak off-chip to the antenna and be amplified by the low noise amplifier (LNA) before self mixing occurs. Unlike a direct conversion receiver, RF LO self mixing in a dual conversion architecture creates a DC offset at IF and does not introduce any impairment in the analog baseband.
but it's not clear to me why the two mixers are different in this respect. Why does the first (or only) mixer suffer from off-chip leakage, but the second mixer does not?