Since I found that the the AS sensitivity seems to be limited by circuit noise, I inserted a RF amplifier just after the AS RF output.
Now, the sensitivity is improved and limited by the dark noise of the PD.

(Note: I did not apply the open loop TF on this xml file.)
REF3: dark noise + circuit noise + WT filter noise + ADC noise
REF4: circuit noise + WT filter noise + ADC

With this situation, I injected the acoustic noise:

REF5, 6, 7: with acoustic excitation
no reference: without acoustic excitation

We could see the coherence only at the same frequencies, around 200 Hz as we saw before (elog).