[Rana, Gabriele, Jenne]
We have now locked the PRMI using REFL55 I&Q for more than one minute!!!!!
This isn't really the most useful plot as is, but it was created using:
/opt/rtcds/caltech/c1/scripts/general/getdata C1:LSC-POP22_I_ERR_DQ C1:LSC-REFL55_I_ERR_DQ C1:LSC-REFL55_Q_ERR_DQ C1:LSC-MICH_IN1_DQ C1:LSC-MICH_OUT_DQ C1:LSC-PRCL_IN1_DQ C1:LSC-PRCL_OUT_DQ -d 80 -s 1049013520 -c
This is just one of several long lock stretches. If I can get the TRIG_MON channels to be saved, we can automatically (versus my by-hand search) find lock stretches and make this kind of plot. Although we want them saved in some raw format so we can zoom in on selected axes, I think. This might require some python-fu from Jamie, or learning of python-fu for Jenne.
The secret sauce:
* The big key was changing REFL55's phase. It was -4 when we looked at the I&Q signals, and minimized the PRCL information in the Q-phase. We were able to get short lock stretches with this. During these stretches, Rana changed the REFL55 phase until the lock sounded (audibly) quieter. The final phase we settled on was +26. As we changed the phase, the lock stretches got longer and longer.
* We also tweaked up the POP22 phase. It was close from our previous efforts of looking at non-locked time series, but we perfected it by minimizing the signal in the Q-phase during lock stretches. We also found that it drifted (according to this method) by ~5 degrees over ~half an hour (I don't remember the exact time between our phase tunings).
* POP22's low pass filters (both options, ELP10 and ELP50) must be OFF for any lock to be acquired. Turning on either filter prevents locking.
* Normalization helped a lot. Without normalization we weren't really able to catch any locks, certainly not of any significant length. (0.004, using POP22I, for both MICH and PRCL).
** Normalization: use POP22I for both MICH and PRCL, value = 0.004
** Input matrix: MICH with REFL55Q, value = 0.01; PRCL with REFL55I, value = 0.01 (we used the small number in the matrix so our servo gains weren't too tiny).
** POP22 lowpass filters OFF
** Analog whitening OFF for REFL55, POP22.
** Analog gain for REFL55 I&Q = 27 dB
** Analog gain for POP22 I&Q = 15 dB
** Output matrix: MICH with -1 to ITMX, +1 to ITMY. PRCL with +1 to PRM.
** Servo gains: PRCL = 0.75; MICH anywhere between -3 and -20. Best in the -8 to -15 range.
** Vio2 filters in ITMX, ITMY, PRM (all actuated-on mirrors) were OFF. (Still need to lower the Q on these so they don't ring).
** PRCL and MICH triggering on POP22I. The trigger-off was always 20, but the trigger-on changed throughout the night from ~170 to ~50. I think 130 was a trigger value for at least some of the long-time locks.
** Low frequency seismic was small (i.e. no anomalous 0.1 Hz - 1 Hz noise) during successful lock times. (Not to say it must be low, but it was low when we were able to lock for long stretches).
Things we had looked at and thought about throughout the evening:
* Oplev calibration. See elog 8391 and 8393. Optimized BS and PRM to reduce yaw angular motion.
* Actuators all functioning as expected. We checked transfer functions of MICH_OUT/MICH_IN1 for locking with different optics, to ensure that at high frequency the response was 1/f^2. Also, we locked MICH with (a) both ITMs, (b) BS, (c) ITMX and (d) ITMY. We locked the PR-ITMY half-cav with (a) PRM and (b) ITMY. We locked the PR-ITMX half-cav with (a) PRM and (b) ITMX. Thus, we conclude that all of the PRMI-related optics are functioning as expected.
* Realigned REFL55 beam onto PD. It was clipping a bit, so the DC power wasn't steady (when ITMs were misaligned, PRM aligned). After alignment, the DC power as seen on a 'scope was much smoother.
* Turning off the limiters for the MICH and PRCL control signals allowed us to hear a high-pitched whine. From looking at the time series, it's predominantly in MICH_OUT. Rana speculates that perhaps the normalization is causing the UGF to wander temporarily to an unstable place. For a time there was a high-Q peak between 500 and 600Hz, but reducing the gain (of MICH?) eliminated that. Then we heard several times, irrespective of gain setting, the ~400Hz broad peak (I say broad because I was able to see it on DTT looking at the error and control signals, and it spanned +/-100Hz).
Things to investigate:
* Is there a good reason that we should switch to triggering on POP110, rather than the current POP22? From Gabriele, Jamie and my Finesee/Mist modelling last week, without the arms, the 11MHz and 55MHz resonate at different PRC lengths. If this difference is very small, then we are fine, but if the difference is large, it could be causing trouble - we're trying to catch the lock at the linear part of the 55MHz signal, but if that does not coincide with the linear part of the 11MHz signal, we're doing the wrong thing.
* For the POP normalization, should we be using the amplitude or the power ( POP22 or sqrt(POP22) )? Why? Look at this with a modelling sweep and/or analytically.
* Look at different noise sources, potentially sensing noise, coil actuator noise,..... We should check these out, and make sure we're not limited by anything obvious.
* Make a "restore" medm screen, rather than restore script. IFO Configure restore script can pull in values from the screen (EPICS values). One screen per configuration.
* Get TRIG_MON signals saved, write script to search for triggered lock times (between given gps times), then plot interesting signals for just before lock, during lock, and until just after a lockloss.