Sheila, Georgia, Craig Today we changed the phasing on REFL9 from 5.9 to 84.8 degrees. We've been investigating fast locklosses, probably caused by CARM. We noticed that if we moved the CARM servo board common offset around, we would win or lose arm power and see a change in the amount of DC power on REFL LF. (Pic 1) We also noticed that when powering up, we saw the offset in REFL9 I change by a lot, along with weird glitches in Q associated with servo board IN1 gain changes. (Pic 3) We went out to the racks, injected a line into CARM above the UGF of about 10 kHz, and checked the REFL9 I and Q monitors. We found that the phasing was terrible: there was far more signal in Q than I. We adjusted the phase, while measuring CARM and lowering the CARM IN1 gain from 3 dB to -9 dB to maintain our UGF. After adjusting the phase, we tested the CARM servo board common offset again and found the arm power and REFL LF to be relatively insensitive. (Pic 2) Sheila and Georgia are changing the guardian so that our digital CARM gain is lower since we have so much more optical gain now. I remeasured the CARM loop and posted the results in the pdf. The gains are in flux right now, for this plot CARM servo board IN1 gain = 1 dB, and Sum Node A IN2 gain = -15 dB at 5 watts of input power. We powered up from 5 watts to 15 watts, and I watched REFL9 I offset again (Pic 4). It still changed, but by about half as much as it did when we previously went from 2 watts to 5 watts in Pic 3. We are sitting much closer to the resonance now that we aren't dumping a factor of 5 of our optical gain.
We have tried to adjust the CARM gains correctly in the guardian to account for the increased sensor gain from our phasing. We have tested this in the adjust_power state, but not the CARM_TO_ANALOG state.
| 2W before rephasing | 5W before re-phasing | 5W after rephrasing | 2W after re-phasing (should be settings after CARM_TO_ANALOG now) | |
| SUM node A in2 gain | -3dB | -3dB | -15dB | -17dB |
| REFL IN1 gain | +11dB | +3dB | +1dB | 11dB |
| total | +8dB | 0dB | -14dB | -6dB |
The CARM_TO_ANALOG state could be re-written in a way that would be less confusing. We increased the sensing gain by 14dB with the re-phasing, so we have reduced the SUM node gain by 14 dB for each of the steps in the CARM_TO_ANALOG guardian.
After this we were stayed locked with 15W of input power and a PR gain of 45, for 20 minutes, then powered up to 20.5W of input power (PR gain dropped to 41) and lost lock after 5 minutes.
Georgia is now turning off CO2X and its chiller to prevent overflowing the bucket that is catching the leak.
It looks like the front-end world has crashed. Here's a screenshot of what I think are the relevant SDF differences for this aLOG such that we don't lose this new goodness.
Since nobody wrote, I summarize the story.
1. Some remember that, at some point in the (distant?) past, Q and I cables were swapped, i.e. Q output was used as I output instead of setting almost 90 degrees of phase in delay line. Everything was good except for cross-cabling.
2. Recently people found on the floor that the demod phase was off by almost 90 degrees.
3. At that time, Q and I were found to be NOT cross-cabled (I output went to I channel). Somebody should have disconnected them and put them back in a "correct" order.
4. The cables were swapped back again (Q output went to I channel) to make IFO happy.
Cross-cabling needs to be fixed before O3 to prevent this from happening again, but this is not a priority for now (it works).