Alexa, Elli, Sheila, Rana, Evan
Today we worked some more to make the sqrt(TRX+TRY) handoff more robust.
Now we can pretty reliably complete the transition by hand. We are working on implementing the transition in the guardian.
We have tried to reduce the CARM offset while locked on sqrt(TRX+TRY), but we cannot seem to get beyond 2 or 3 times the single-arm power without blowing the lock. The next step is to go through the CARM reduction sequence more carefully and characterize the OLTF of the CARM loop in this state.
Gain redistribution
As discussed in LHO#16252 et seq., we suspect that the common-mode board has gain-dependent offsets. If this is true, it explains why the interferometer can be knocked out of lock while ramping down or turning off ALS COMM.
To boost the CARM loop's immunity to these offsets, we redistributed gains as follows:
-
The gain of the ALS COMM input on the common-mode board was increased from −4 dB to +16 dB (
LSC-REFL_SERVO_IN2GAIN). Correspondingly, the gain on the AO into the IMC board was reduced from +16 dB to −4 dB (IMC-REFL_SERVO_IN2GAIN). - The input matrix gain for the slow path out of the IMC board (going into the digital CARM control) was decreased from 1 to 0.1 (LSC-PD_DOF_MTRX_5_26).
-
The digital gain for sqrt(TRX+TRY) was increased by a factor of 10 (
LSC-REFLBIAS_GAINwas increased from 0.8 to 8).
Even with these changes, success is not guaranteed. The gain steps can be heard very clearly when listening to LSC-CARM_IN1, and turning off ALS COMM on the summing board will blow the lock about half the time. We can get better results by using the gain slider on the common-mode board, but we can still hear the gain steps.
Modified handoff procedure
-
In
LSC-REFLBIAS, we engage FM3 (1 Hz pole, 40 Hz zero) and FM9 (1.6 Hz pole, 40 Hz zero). The gain is 8 ct/ct. -
In
LSC-TR_CARM, we set the offset to −0.5 ct. - We engage the sqrt(TRX+TRY) path on the common-mode board.
-
We engage
LSC-REFL_DC_BIASwith a gain of 30 ct/ct. -
Once the buildup has stabilized, we turn on FM6 (0 Hz pole, 1 Hz zero). Then we turn on FM4, which undoes FM3. Then we ramp the gain of
LSC-REFL_DC_BIASto 50 ct/ct. - We ramp down ALS COMM using the common-mode board input gain slider. Then we turn off this input on the common-mode board.
A lockloss plot is attached for an unexplained lock loss during the CARM offset reduction after handing off to sqrt(TRX+TRY). Other lock loss times, all 2015-01-27 UTC: 03:56:53, 04:05:25, 05:36:16, 06:40:59, 07:56:02, 08:37:20. The last four are during CARM offset reduction after the handoff is complete.
For a sqrt(TRX+TRY) offset of −1.5 ct, the gain we need for REFL_DC_BIAS is 50 ct/ct. Our OLTF looks good here. However, when we reach an offset of −2, we seem to lose lock, even though there is no obvious nastiness in the CARM spectrum.
Mystery
Today, locking DRMI without arms was pretty painless. In contrast, DRMI+arms lock acquisition was very, very slow for most of the morning and afternoon. After about 5pm local, it became painless as well. This may have been correlated with our changing the trigger settings to be twice as high with arms (compared to no arms), since the POP18 buildup is twice as high. But we haven't investigated this systematically.
Last night Sheila requested DRMI_LOCKED on the ISC_LOCK guardian when she left for the night at 1am to che k if this configuration is stable.
DRMI with green in the arms and IR held off resonance stayed locked overnight, and the power slowly degraded untill 4.30am when the lock dropped. Cuardian relocked DRMI untill 6am when DRMI lock dropped and did not recover. This shows that this configuration is fairly stable.
After changing our ALS gain rampdown to be in the CM board rather than ahead of it, we never broke lock due to the rampdown. However, we only tried it in this new way twice.
The lock loss plot attached above was typical of our current lock loss mystery. After moving to a new CARM offset, the CARM error signal seems stable (no peaking in the spectrum) and the loop shape looks good. The lock breaks without any characteristic sound - just a sudden lock loss. Investigation of the 5 LSC error and control signals shows no instability in the 100 ms before lock loss. The lock loss happens several seconds after the CARM offset is done ramping. Since the signal from the Thorlabs Transmon PDs is only 2000-3000 counts, we think that they are not saturating.
No guess yet about what is happening. Any speculations on lock loss causes is welcome.
We struggled with the ETMX bounce mode all of today. It seems to have started ringing up around 3 AM last night (according to the ETMX OL 3-10 Hz BLRMS trend). We spent a couple of hours damping it around noon today, but then it slowly started growing again today. We've now installed a 60 dB stopband filter centered at 9.77 Hz in the OL loops to see if this will stop the ringups. We have also installed a resonant gain filter for 9.77 Hz in the CARM loop to reduce the arm power fluctuations during the offset reduction.
To better monitor the bounce mode, we set up the ETMX OL lockin screens to demod the OLPIT signal at 9.67 Hz. So now one can trend the 0.1 Hz beat note in the lowpassed output of this to see what the Bounce mode peak height is at all times. Probably should make a dedicated BLRMS for each suspension with an OL to monitor its bounce mode height.
I am taking a look at the lock losses Evan listed in this entry. All the correction signals seem good except the BS one. This is, for example, lock loss 4_05 UTC, with different zoom of the correction signals during CARM offset reduction. Probably a campaign of loop measurements for the vertex DOFs can help.
More lock loss science. As far as I can tell, 5:36 and 6:40 are similar to the previous lock loss 4:05, while in 7:55 looks like CARM is indeed the culprit..
(BS oplev seems OK)
Given the prior lock loss science by Lisa, I speculated that the BS oplev loops were doing something bad such as glitches. (Note that the people did not use the ASC loops last night so that the oplev damping loops on BS had been engaged all the time). Looking into the last five lock losses that Evan posted, I am concluding that the oplev damping loops were not glitching or disturbing the MICH loop. The attached are 60 seconds full time series of various BS oplev-related channels for the five lock losses. The 4:05 event is the only one which clearly showed DAC saturation and the rest of them did not saturate the BS DAC before the lock loss. The oplev sum sometimes shows a fast transient, but it happens right after each lock was lost -- indicating that the transient was caused by the motion on the oplev QPD as the BS was kicked and it is NOT initiating the lock loss.
(TRX seems always lower than TRY)
I don't know if this is related to the cause of the lock losses, but I found that TRX have been consistently lower than TRY by roughly 10 % regardless of how big the CARM offset was. It is unclear if this discrepancy is from an unintentional offset in the ALS diff operating point or some kind of calibration error in TRs. In any case, we should fix it in order to reduce DARM coupling in the TR_CARM signal path.
What is the CARM offset and offset reduction in physical units? (pm or Hz of the arm cavitites)
Peter, we can use Kiwamu's plot to convert this offset into physical units (alog 15389). An offset of -0.5cts gives about 800 pm. We were able to bring the carm offset to about 300 pm stabily.
