During the unplugging I mistakenly cut the seed for the fibered amplifier, so triggering the safety mode. The amplifier will be switched back on soon. A further detail: no tuning of the PZT gain was done.

Also, a couple of pictures to give more details: second attachement is the IMC rampeauto board with the switch set to 1/f, third shows the unplugged cable from the amplifier to the 14MHz EOM on the laser bench.

Last night there were some lock perturbations due to the loop oscillation at about 30 kHz (see plot 1).
In order to attenuate this phenomenon, we reduced the gain of the PZT inside rampeauto till the annihilation of the peak (see plot 2 and zoom plot 3). 
To be noticed that there is little margin between keeping the lock and the peak exctinction.
We will let it in that state for the weekend.

Yesterday morning (Jan 18 ~9UTC) there was a period of glitches on the unconnected EOM correction that would have corresponded to fast unlocks (see figure 1). The IMC stayed locked.

Figure 2, 3, looking in time domain, it corresponds to a series of 10 glitches spaced by 100ms,

Figure 4 is another example. Again 10 glitches spaced by 100ms.

However this was the time of BPC offloading with picomotors. So it is unlikely to be the origin of fast unlocks in normal times.

In order to let ISC people locking the arms correctly, we recovered the normal state of EOM corr and IMC rampauto as before the intervention for the study of fast unlock (fig. 1). We put back the correct gain of PZT and the switch on 1/f⁴  and finally we replugged EOM corr signal.

To be noted that during the latter, unfortunately we cut again the seed for the fibered amplifier, putting it in safety mode. It has been immediately restarted.

During the night of the 16/01/21, while the correction of the EOM were unplugged, the IMC unlocked several times (see plot 1).
Those unlocks looks like fast unlocks (see plot 2 3 4). So it seems that the glitches on EOM_CORR are triggered by IMC_REFL_I and not the contrary. Investigation are still ongoing.
To be noted (see plot 2 and 3), that there are some kind of glitches on ML_FREQ_I shortly before the unlock. However this canal have some glitchy/noisy periods that does not seem directly correlated to the unlocks (see plot 5).

Note that in the 3/4 cases I have looke at the fast unlocks are precedeed by precursors glitches on the error signal in the preceding seconds.

Figure 1 shows an example, the glitches seem to be always in the same sign on the error signal.

Figure 2 zooms into one of them, the duration is around 20us, so ~50kHz, so above the loop UGF which seems to be ~20kHz with EOM unplugged.

Figure 3 shows another case where the IMC actually stay locked (and unlocked 30s later). What is interesting is that in that case the glitches look like a very noisy PDH signal. First all of the excursions are positive, then both signs, and then all of them are negative. As if there was some resonance passing through.

Figure 4 is the unlock 30s later, first excursion are negative, then both signs and the system unlocks. As if the glitchy mode was passing back through resonance.

Figure 5 is another example, only a few glitches in that case, all positive in the other signal.

Figure 6 shows a zoom during the time when the glitches have both signs. Not sure what can be learned from it as it saturates in both signs all the time.

It would be interesting to understand the origin of these precursors glitches. Do they happen in normal times (EOM plugged in) as a precursor to fast unlocks? It may well not be the case as the first occurence may unlock the system straight away, as in that case the glitch is in loop. Could be a parasite mode in the IMC? A start of a mode hop in the laser? Something else...

Figure 7, 8, 9, 10, 11. These glitches are also visible in the PMC error signal and the PMC transmission. For positive IMC glitches, the PMC error signal first goes down and then up, for negative IMC glitches it first goes up and then down. Strangely for the PMC transmission AC signal, there is always a drop, and it is much longer than the glitch in the RF error signal. Could this longer time scale be due to an analog saturation or is it real?

A few more comments about this fast unlock like glitches.

Figure 1, 2, 3, The glitches (at least when they are not too frequent) correspond to an increase in the PMC transmissited power which is surprising. For the IMC the transmission drops and the reflection increases, which makes more sense, light frequency is mistuned and doesn't enter the IMC cavity as well.

Figure 4, 5 The time scale of the glitch is the same in the IMC and PMC RF signal, around 20us. In the PMC TRA AC channel the duration of the transient depends on the amplitude of the glitch, but that is probably just due to the saturation of the AC electronics.