(Correction on Oct/09: POP_A RF 45 Q ADC saturation, not I.)
I was looking at the lock loss at around Oct/04/2018 17:57 UTC with 15W, which Jenne told me "may" be the fast lock loss. In the attached plots the power in the arms is lost at about 406.3 sec mark.
Since I wanted to see what happened to which signal first, in the first plot I high-passed all relevant LSC signals by zpk([0;0],[70;70],1,"n"), and it's clear that it's MICH and specifically POP_A_RF45_Q_ERR that lost sensitivity at 405.8sec (left middle and left bottom) at first. The disaster didn't propagate to OMC, REFL, SRCL and PRCL until 406.2 sec.
Removing low-pass (2nd attachment), from POP_A_RF45_Q_ERR signal it's clear that POP_A_RF45_I POP_A_RF45_Q before rotation hit the ADC ceiling (ERR signal is post-rotation and post-dewhitening, but the rotation angle for POP_A is about 90 0 degrees).
Before losing lock, for about 400 seconds it was going between +-20k counts, which is marginal at best for post-dewhite signal, and when the disturbance becomes larger than usual it rails.
I don't have time to analyze other lock losses tonight but I'd say we have to reduce whitening gain by at least 12 db or so. It's unclear if this is the cause of so-called fast lock losses but we need to fix this anyway.
Craig and I reduced the POP_45 whitening gain from 30dB to 12dB, and added a gain of 7.943 to the digital filters. This seems to have helped a lot as had two locks of about 1.5hours at 10W tonight. We didn't try powering up higher, but it would be an easy to check to see if the problem is totally gone.
We didn't check the dark offsets yet because we adjusted this in lock. We should probably go around and add filters that undo the whitening gain to all of our PDs, so that we don't end up with these random gains in so many of them. That will require changign the input matrix through, so I will not do it now.
(Sheila, Keita)
With the whitening gain of 12dB.
- We set the dark offset for POP_A_RF45_I and Q (H1:LSC-POP_A_RF45_I_OFFSET and Q).
- Digital gains were reverted back from 8 to 1. (H1:LSC-POP_A_RF45_I_GAIN and Q).
- Made -12dB gain in FM4 of LSC_POP_A_RF45 filters, loaded it and enabled it.
This is a total of a factor of 32, and this was reallocated by Sheila to
- lscparams.gain['DRMI_MICH']['1F_45'] in lscparams.py, from 0.057 to 1.824
- lscparams.gain['DRMI_SRCL']['1F_45'] in lscparams.py, from 0.058 to 1.856
- ALIGNIFO.py line 575, ISC_library.intrix['%sARM'%arm, 'POP_A45I'] from -10 to -320
The idea is that FM4 (-12dB in this case) is the inverse of the whitening gain. Next time you change the whitening gain, you can just change -12dB filter without worrying about guardian.
