The main message:
This afternoon we repeated the switch of DARM to EY, but this time Jenne retrained the feedforward to reduce the SRCL noise which dominated the measurements in 64478. The 3rd attachment: the DARM noise improves when the EX ESD bias is 0V, with improvements visible from 30-95Hz.
Details:
- nominal configuration: August 16 23:08:05 UTC
- DARM switched to EY with LSC cleaning returned by Jenne: 23:54:11 UTC
- EX bias off: August 17 00:07:36 UTC
Attempt to do this swap while preserving the calibration:
In our first lock today I made some copies of the calibration templates for ETMX actuator measurements and changed them to EY actuator measurements. It seems that the scaling that we are using for the ESD is good as shown in the 1st attachment, (for DARM control on EX we use a gain in DRIVEALIGN of -35.7, for EY we use -30 and these gains are well matched as shown in the first screenshot). In this measurement the EY bias had the wrong sign, I have been using a negative bias on EY to swap DARM control but this measurement was taken with a positive bias.
For L2 I did a similar measurement, her we normally use a gain of 15 in the L2 lock filter on EY and a gain of 23 in the EX L2 Lock filter. The calibration excitation is injected after the lock filter bypassing this gain. The second attached screenshot shows that the two PUMs are well matched, and that the different gains in the L2 lock filter shouldn't be needed to match the PUMs.
For L1, I unlocked the IFO when I tried to use the same excitation for EY as is used for EX. I don't understand why that happened, but it might mean that there is some large difference between the UIMs on the two test masses. We decided to just repeat the swap with the same gains that I used on the weekend, with the violin mode damping off, because the interferometer was stable in that configuration.
Calibration and feedforward in this configuration
We switched to the EY DARM actuator with a gain of 15 in L2 lock, while we know that to preseve the DARM calibration and the feedforward we should have used a gain of 23. This means that our calibration is rather wrong. Craig caputred a braodband pcal excitation for this configuration (4th attachment) which is saved at /ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O3/H1/Measurements/FullIFOSensingTFs/2022-08-16_H1_PCALY2DARMTF_BB.xml
Jenne returned the LSC subtraction in this configuration, which is really nice because it was much faster than our previous by hand method, allowing us to do a relatively quick test in this configuration. I believe that she tuned the LSC cleaning and also saw that the ASC cleaning could use some cleaning, but she only applied the LSC update. The 5th attachment shows the DARM noise after the cleaning and the SRCL coherence which is dominating the noise from 20-40Hz even after the cleaning.
Spectra and coherences with PEM channels
The last three attachments show spectra and coherences with PEM channels. First is PEM_ADC-0_19, which is the ESD power supply monitor. The spectrum of this channel does change when DARM control is switched to EY. This means that it does witness the DARM drive being sent to the test mass, but there are also peaks in this spectrum which are unchanged by the switch of DARM control so this channel is also picking up some other noise which is coherent with DARM. It is interesting that the spectrum also changes when the bias is set to 0, we could repeat this to see if it is repeatable. The coherence of this channel with DARM does show some changes when DARM control is switched to EY, but many of the features go away completely when the bias is set to zero, again indicating that this channel is picking up both the DARM drive to the suspension and some other source of noise which is coupling to DARM through the bias.
SUS RACK Y specta don't change as much between the three configurations, but again the coherence is reduced when the EX bias is set to 0.
The EY UIM has one stage of analog low-pass filtering engaged, while the EX UIM has none, so that could explain the lockloss. At one point I changed both UIMs to run with one stage of filtering (64212), but the configuration on EX was undone, presumably by guardian or SDF.
This morning we switched the DARM actuator to EY again and did some bias tests on ETMX. We tried setting the bias from -450V (nominal) to +450V and to 0V.
We stayed at each bias for about 15 mins. The line couplings are mostly best with 0V ESD bias (eg 32 Hz, 71 Hz, 83 Hz, 95 Hz), one line looks better with -450V bias (35.8 Hz), and one line looks better with the bias at +450V (~61.5 Hz), these are shown on the right side of the 1st attachment. Edit: watching DARM some of these lines are breathing so some of these statements might not be true!
The broadband improvement was harder to see than previously, but there is some improvement seen in the DARM BLRMS. It might be good to do another slower bias sweep while DARM is on ETMY later tonight.
Overnight I tried to sweep the ETMX ESD bias while the squeezer was shuttered. My test stopped working halfway through (at about 7pm local) so I didn't get to sweep through the positive voltages. The boradband noise reduction was again not as clear as the times Sheila posted and I'm not sure why. The range and BLRMS do still show that 0V bias improves DARM. I tried to make the warmer/redder colours in the attach spectra closer to 0V, and the green traces at larger negative voltage.
The attached plot shows that when Jenne retuned the cleaning for the configuration where we locked on EY, H1:OAF-CLEAN_LSC_COEFF_1_1 went from it's normal value of 7.0229 to 24.816 . For the later tests on the 17th and 18th, the cleanng was set back to the normal EX cleaning, which probably explains why we didn't see the broadband improvement.