Good luck!
The way to produce an error signal to measure and correct for finesse asymmetry: put a low frequency sine at f0 on frequency noise (f0 = a few 10s of Hz); detect it on dark fringe; demodulate at f0. The correction may be applicable to the tower temperature. The error signal would be a nice tower thermometer.
This solution should be applicable to Virgo, if needed, when we have the full resolution at low frequency, were the requirements for frequency stability should be the more stringent. There is no reason why we should not be able to do that in Virgo+, or Adv. Virgo, may be even with thermal compensation system for thermal lensing and parametric instability mitigation.
The initial idea, that I proposed many years ago as a change request (!), was to put some heating tapes around the bottom part of the tower, to measure the etalon effect, evaluate the transfer function and hope to close a loop. This may be difficult because the time constant will not be much less than a day, but we can first reduce the day/night fluctuations with some feedforward, in order to request less gain from the servo-loop.
When a CO2 laser becomes available, it will be possible to heat directly the mirrors and the feedback will be quick enough. If we are smart, this can be done in parallel with the compensation of the lensing effect, by controlling independently the average power (etalon effect) and the geometry (thermal lens)
An other way to understand the need for gain change:
The slope of error signal (in V/Hz), in rampeauto, has decreased by 2 dB. One wants to add electronically a signal (SSFS_Corr) with similar V/Hz. The way to recover the cross-over between prestab and SSFS (at 20 kHz or so) is thus to decrease SSFS gain by a similar amount of 2 dB.
The SSFS open loop gain is:
G_ssfs = F_CARM*F_rack*T_IMC/F_IMC
where F_CARM is the TF between a frequency line before PR and demodulated signal on Pr_B5_ACp, F_rack the SSFS rack TF, T_IMC a low pass filter (pole of IMC), and F_IMC the IMC optical TF (=> cross-over with prestab).
Gc_Alpha_Out = alpha * alpha_filter * Gc_MICH_z + beta * Gc_PRCL_z
In case of problems with the alpha or beta corrections, or questions about what is added to the DARM correction (an important signal), this channel is critical.
I have three suggestions:
1) increase the line DARM line amplitude by a factor of 2 until weather conditions are better (and horizon consistently > 3.5). I think the final line amplitudes are set in step 12.
2) modify the GainAdj flag to monitor also the MICH and PRCL gain servos
3) check the performance of the MICH and PRCL gain servos, and increase their line amplitudes if they are having trouble (both now 1.5e??, so change to 2.0??, for instance).
The GainAdj flag might currently use the max of the DARMline signal. I would suggest something more like the mean over 10 seconds, which should be less than 1.2 (for DARM, MICH and PRCL). This would allow for an occasional loss of coherence without making a red flag. A yellow flag condition could be set at a mean of 1.1.
So I think that unless we notice something different today, we can safely ignore this flag for the first minutes of step 12.
The second plot looks at TraMC and D1T_DCHF. TraMC goes quickly into the ADC noise, making it of little use, but at least one can see that D1T_DCHF appears to be an AC coupled version with a single zero at DC and a gain at 1Hz of about 100. One can also see that the spikes and discontinuitied in D1T_DCHF look quite odd... the experts have been contacted.
Sc_IB_TraMC = 2.86W/V*DPS_DC (I am not sure 2.86W/V is up todate but it should not be very different)
Bs_IMC_D1T_DCHF = 56*DPS_AC*50 (factor 50 is applied in the Pstab rack, 56 is the transimpedance ratio between DC and AC channels)
(moreover Bs_IMC_D1T_DCHF is AC coupled with 16Hz cut-off)
plot 1 , at 10Hz Bs_IMC_D1T_DCHF=8e-4V/sqrt(Hz)
that should make 8e-4/56/50*2.86=8e-7V/sqrt(Hz) for Sc_IB_TraMC, well below ADC noise
plot 2, Sc_IB_TraMC=1e-4V/sqrt(Hz) @1Hz that should be
1e-4V/sqrt(Hz) /2.86*56*50*1/16 = 6e-3 V/sqrt(Hz)
on Bs_IMC_D1T_DCHF at 1Hz (measured 7e-3instead)
At least, in this regime those signals are coherent
http://wwwcascina.virgo.infn.it/commissioning/weekly/2007/May2007/Cleva_Pstab_230507.ppt
The more resolved plot 2 shows that the "high noise phase" (this one ending at 05:03:30 UTC) is characterizd by frequent short bursts, while in the low noise phase these bursts are much less frequent. Plots 3 and 4 show various signals that were not clearly related to our mystery noise. With plot 5, I found that Bs_SL_D2_RDCHF and D4_DCHF showed broad band bursts at the same time as SSFS_Visu.
Plots 6 shows these signals and the SL_Icorr_DC signals during the lock in science mode at the time of the change between high and low noise states. Plots 7 shows the signals as the ITF and IMC unlock, and the IMC relocks. Plot 8 is a higher resolution plot of the IMC relock (with the ITF unlocked), which shows that the noise bursts are present after the IMC relock and are visible as soon as the Pstab loop is enabled.
Further investigation is certainly necessary, but my preliminary conclusion is that we have a significant noise source associated with the Pstab loop. Disabling the loop has been seen to have negative effects, so we will have to work harder than that.
1) removing 100kg which was rigidly attached to the Brewster
2) replacing 50kg, but sitting on a rubber pad
The original "hard" mount, which was aimed only at increasing the mass of the BW, was found not to be effective at reducing B1_ACp noise. The "soft" mount, on the other hand, was modestly effective in damping some mechanical resonances of the BW which were visible in B1_ACp (particularly 220Hz).
Though this should have no effect on the AC alignment signal, in principal, there was a significant change in the IBz error signal. The result was a very slow drift of the IB (slow because the UGF of this loop is less than 100uHz) and an offset on the zCorr signal of about 0.6V.
To reduce the offset, after the work on the IB card this morning, I moved the Sa_IB_Gain_y_offs to 1150um. I also increased the servo gain (on the IBzAA.flt line) from 1e-5 to 1e-4, such that the loop converges in less than 1 hour (the noise is still very low, see plots).
ADC4_4 : Up Coil
ADC4_5 : Down Coil
ADC4_6 : Left Coil
ADC4_7 : Right Coil
Slim modified the MC DSP card and for the time being, we can monitor only one coil at the same time. The probe name is Sc_MC_Coil_current and you can choose the coil you want to monitor by changing the corresponding gain in the DSP card.
We made a preliminary analysis when the MC was in High noise configuration (it should be said that the MC readout have no shaping filter). The first figure shows that what we measure is coherent with what we send when above ADC noise. The next four figures make the comparison between the measured coil current and the sent correction signal rescaled with driving factor, actuation factor and sensing factor. So we can say that ADC noise is around 5-6 uvolt/sqrt(Hrz) that is compatible with usual value.
The only clue comes form the PR seismometer... it has a nice line at our favorite frequency, 241 Hz. We tapped there, but got nothing special.
We had to stop and restart the locking process to recover..but we forgot to check the Gc elapsed time, which increased during this operation and the ITF unlocked again at step 10, during the NEO transition.
After that, the elapsed time seemed reasonable in await (64000), but at step 8 was too big, so we manually unlocked and we stopped and restarted the locking process a few times until the starting value was lower, around 62000.
After that we could reach science mode again.
As already reported, Matt added a protection in Alp, so that the locking sequence stops if the elapsed time is too high.
Of course it doesn‘t solve the problem, but at least reduces the wasted time,
which for us was at least 3 hours tonight...Nemo was not happy.
---------------------
The transition to B2_mix caused the unlock of the ITF twice today.
I increased the B2_mix gain to -0.0014, not clear if this helps.
(only 1 success after that (first plot)..ITF still locked.). To be checked.
Stability
----------
This night we tested some long term stability of the ITF.
The ITF remained locked for at least four hours twice (see second plot), and all the unlocks were due to "technical" problems.
Sensitivity
------------
The sensitivity is now worse than before below 70 Hz (third plot).
Looking at B1, the sensitivity was good on 16th May,
13h50 UTC, at the end of a lock, and bad at 17h31 UTC, after relocking at step 12 (fourth plot):
16/5 11:27 13:27 863350068 Science Mode 12 12 -
16/5 13:51 15:51 863358674 Adjusting Mode 12 12 -
16/5 14:14 16:14 863360074 Unlock! 12 12 173
16/5 15:56 17:56 863366203 Manual unlock 1 1 0
16/5 16:08 18:08 863366923 Manual unlock 1 1 0
16/5 16:16 18:16 863367413 Manual unlock 1 1 0
16/5 17:03 19:03 863370250 Unsuccessful locking attempt 12 11
16/5 17:30 19:30 863371842 Locked! 12 12
The noise budget shows that the noise goes through MICH (fifth plot).
Since we were sleeping at that time, we are quite confident not to be guilty :-)
The only operation reported in the log at that time is Calibration automation .
We checked if the alpha filtered was changed in between, but it doesn‘t look like it was. We don‘t find anything else to blame... but since this didn‘t change, what did?
Since the power after the MC is stabilized, one might wonder where these power fluctuations come from. Beam jitter is a possibility, but it would have to be jitter that the BMS does not see. We have also noted previously that the IMC quadrants AC and DC signals have some coherence with B1_ACp and SSFS_Corr.
The coherence of the IMC quadrant AC signals with SSFS_Corr (esp. IB_z, but also others) indicates that they are somewhat miscentered. We don‘t know what effect this has on our mystery noise, but one might hope that good centering (e.g., no coherence between SSFS_Corr and the IMC AC signals) would help to align the IMC. Fearing that moving motors on the injection bench will unlock the ITF, we leave this job to the next shift.
