In the past days Pierre and Yuhang noticed that the level of the LO to the fiter cavity mixer was not the one required by the mixer, so they amplified it. After this change we were not able to lock the cavity on the TEM00 peak anymore. We went back to the previous configuration but still we could not lock properly. It seemed that we were locking on a sidebands and, strangly enough, we could lock on the top of TEM00 only by switching off the rampeauto power supply. But of course in this configuration the gain was much lower and the lock was not stable.

We found that the problem was that the sign of the lock had changed (even after going back to the original configuration). 

Note that the sign inverter on the rampeauto doesn't change the error signal sign, the only way to change the sign is by shifting the demodulation signal sent to the mixer of 180 deg. We did it and the cavity could lock again.  

We also learnt that swiching off the reampeauto power supply somehow changes the sign of the correction that it provides.