Assuming that the sensing matrix was correctly measured, from the filter shape it looked as if the UGF of WFSs were set to a mHz or two. This was because Alberto and Bram found that setting gain higher made things unstable.
Today I and Alberto reproduced the problem by simply setting the UGF higher: Apparently 0.54Hz PIT mode and 0.64 Hz YAW mode goes unstable because of high Q (attached, glue traces).
The PIT mode number looks like it's in good agreement with JeffK's SUS resonance plots, though for YAW 0.64 Hz in our data looks more like 0.6Hz in JeffK's.
Anyway, I made some notches (0.54Hz and 0.44 Hz for PIT, 0.63 Hz for YAW) and brought the gain up by a factor of 10 or so and they didn't oscillate (red). With this setting, the UGF for PIT should be about 20 mHz and for YAW it should be about 55mHz.
I haven't tried to push anything further.
BTW 2Hz on the plot is caused by Robert's injection.
I and Bram briefly checked if things looks sane by giving an offset step to each DOF (ETM/ITM, PIT/YAW) one by one.
Each time, the feedback eventually brought everything back to where it should be, so it's OK.
Due to ETM - ITM servo coupling (remember, the servo is supposed to be diagonalized based on POS/ANG rather than ETM/ITM), when you kick one mirror there is an initial kick to both of the mirrors and eventually everything gets back to normal. Or it's supposed to be so.
For ITM step, the time constant for both the ITM and ETM to go back to the original angle looked about right.
For ETM step, for some reason the bulk of ETM step itself is removed almost instantaneously, and then ITM moves with what seems to be a right time constant while ETM goes through the same slow thing but with lesser amplitude.
Maybe the sensing matrix measurement was still far from nice. But anyway it works and I don't intend to fix that immediately.
Caution!
Now WFS is with some useful oomph, which is good. However, this also means that, when the arm drops out of lock, WFS output might go big and the cavity might not be able to relock on its own.
Though there are limiters to the WFS output (30 counts for PIT and 50 counts for YAW, which was set somewhat arbitrarily), 50 counts is large enough to tilt the mirrors such that 00 mode is really small in the cavity.
If this happens, you need to reset the integrator of the WFSs by pressing an ill-named "Clear all IP" button from H2 OAT medm screen (see attached) that is available from the ISC section of the sitemap.
If somebody writes an auto relocker that polls H2:ALS-Y_REFL_B_PWR_OUTPUT, disables CM PDH servo, press "Clear all IP" button and then enables CM PDH servo again, that would be greatly appreciated.
