The whole day has been devoted to the recovery of the ITF with 25 W of input power:

• during the morning shift we still had many unlocks, but we managed to get to LOW_NOISE_1 a couple of times, and to LOW_NOISE_3 once; the ITF is much more stable if we manage to close the alignment loops and move to DC readout; however, we could clearly see thermal effects (sidebands decreasing) and the step to dark fringe failing many times;
• during the afternoon, we got once to dark fringe and we did a test with the DAS;
• however, after that we never managed to get to dark fringe in a stable way again, although this is not related to the TCS actuation;
• I tried to stabilize the lock acquisition in order to get to dark fringe reliably and to decouple any possible misbehaviour of the ITF:
  • the guided lock thresholds were rolled back, as they are relative w.r.t. power; later, I also had to reduce the gain for the lock of the single arms, as I noticed it was quite high and the North Arm was not locking; this change had to be absorbed in the CARM/DARM sensing, hardcoded in CARM_DARM_LOCK.py, it will be fixed;
  • the PR alignment loop during variable finesse has been tuned, adding additional changes of the setpoints along the way, and increasing the gain;
  • the dithering lines amplitude were lowered in dark fringe, but it was probably not necessary and coincidentally beneficial to the lock;
  • the thresholds for the IMC and RFC powers have been updated in Metatron, to be checked if the AA thresholds for the QPD need it as well;
  • the trigger parameters for PRCL were tuned, as it was trying to close the loop way before the PRM was completely aligned, because of the increased power; on the other hand, the safety thresholds for B7_DC still need to be updated!
  • we changed a bit the gains for the DIFFp loop, to be confirmed on following attempts;
  • as reported on Monday, the line at 100 Hz is still very high and affecting severely DARM and PRCL, but on the former it goes down towards dark fringe; the PRCL correction is always very high;
• the lock acquisition up to MICH_SET = 0.1 is much more robust, but the dark fringe always kills the lock; this was correlated to saturations in the PDs, but the mechanism is not clear:
  • the first offender was B1p_PD2, for which the threshold for the shutter was increased and, later, the switch between DC and Audio channel inhibited, meaning that we were using only the DC channel; however, at MICH_SET = 0.1 the saturation flag always get triggered, and many spikes and glitches are still visible on MICH, which still uses it as a sensor;
  • this kind of saturation lasts for a relative long time, but the lock is immediately lost once it happens to the B4 photodiodes; the  unlocks seem to have the common characteristic of a violent oscillation in DARM at 137 Hz, although this is completely unrelated to any loop gain I tried;
  • as a test, I switched back to B4_PD1, but the situation did not change, if not for the need to retune several gains of several loops, despite the two PDs being in theory equivalent (unlock at 21:16:27 UTC);
  • I tried to reduce the 6 MHz amplitude even more, at MICH_SET = 0.2 (with both B4 PDs), but this did not reduce the spikes or eased the situation;
  • I also tried to reduce the 56 MHz amplitude from the start (from the usual 12 dBm to 7 dBm as in O2, and even down to 4 dBm), but the behaviour of the ITF did not change;
  • fiddling with all this I found out that the new LNFS implementation has some bug when dealing with small amplitudes, which is something numerical: below ~6 dBm it is not possible to change the amplitudes of a single dBm, and if one requests a small amplitude from far away, this is often "approximated" to the nearest integer instead of the chosen one; example: from 12 dBm, asking for 5, it falls to 4, and it is not possible to move it to 5, but it is possible to move it to 6 (two integers). This is most weird! I did not explore by hand amplitudes < 4, and I did not investigate if it really works at dark fringe when we set the 6 MHz one to 0 dBm;
  • unlock with 7 dBm for the 56 MHz moduation amplitude and B4_PD1: 21:28:07 UTC;
  • unlock with 4 dBm for the 56 MHz moduation amplitude and B4_PD1: 21:53:25 UTC;

I left the ITF in LOCKED_RECOMBINED, with the usual 56 MHz modulation amplitude of 12 dBm, B4_PD1 in use, and the FmodErr tuning disabled.