Jenne, Sheila, Georgia, Keita, Varun
The input matrix using REFL-9I, A and B inputs for INP1-P and CHARD-P added noise in the error signal making it difficult to blend and increase power. Therefore, we reverted back to the old input matrix. The goal was to reduce the coherence of CHARD-P with DARM which is upto 40 Hz. Here is what we tried today.
- The plot1 and plot2 show the measurement of CHARD-P OLG before chard blend (recorded at REDUCE_RF45) and after blend at (measured at Low Noise Length). Based on the measurement we have excess phase margin lost in blending from as summarized in the table below:
| Frequency (Hz) | Phase Margin before blend (deg) | Phase margin at LowNoiseLength (deg) | Expected Loss from Cutoff filters (deg) | Unaccounted Loss in Phase Margin (deg) |
| 3.5 | 50 | 25.2 | JLP25: 18.2 | 6.6 |
| 5 | 54.3 | -0.2 | JLP25: -27.2 | 26.9 |
- Further at the state of lownoise length we found the DARM noise to be coherent with CHARD-P upto 40 Hz, see figure.
- We decided to skip the blending and try implementing a stricter cutoff. We implemented a 18Hz cutoff filter, with CHARD-P gain of 1.4 (gain of 1 lead to a ring-up). With these settings we were able to reduce the CHARD-P coherence with DARM down to 20 Hz but lead to glitching in DARM.
Suspected culprit of the CHARD_P issues: Currently in the CHARD-P blend, we use the QPDs TR_X-A, TR_X-B and TR_Y-A. The QPD TR_Y-B is not used as there are some hardware issues which show a coupling between pitch and yaw. We suspect the absense of TR_Y-B in the input matrix is causing coupling between the Soft loops and CHARD-P. Keita will be testing the QPD tomorrow morning.
quick note: I undid an earlier guardian change from when we were trying an unblended CHARD_P: reverted the LOWNOISE_ASC CHARD_P gain from 1.4 to 1. This was in lscparams. I loaded ISC_LOCK.
Attached are screenshots that show the situation with the O3 input matrix for CHARD P, (REFL45) and the blends on. This is the configuration where we are stable with 90MPc range, and CAHRD P noise is contributing to DARM up to 45Hz.
The first attachment shows the coherence of DARM with CHARD extending up to 50Hz in the middle panel, and the bottom panel shows that this is caused by noise from the CHARD B (QPD) signals, rather than the REFL WFS noise (CHARD P A is refl WFS).
The next screenshot shows from the coherences that TR Y A is dominating the high frequency noise of the blended signal, and the coherence with DARM. The QPD spectra show that both Y QPDs are noisier than the X QPDs.
The top right panel shows power spectrum of the output of the blends. You can compare this to the CHARD A blend shown in the third attachment. At 30Hz, if we remove the blending the WFS signal would increase by 60dB, meaning that the power spectrum of the error signal would be about a factor of 10 higher than the blended error signal at 30Hz. Based on Varun's loop measurements above, we thought that removing this blend would reduce some cross coupling and make the loop shaping easier, allowing us to engage the more aggresive cutoff used in O3. Being able to use the more aggressive cut off would reduce the CHARD P controls noise by a factor of 100 at 30Hz. We came to the conclusion that removing the blends and engaging the cutoff would improve our CHARD noise at 30Hz by about a factor of 10, so this would be the easiest way to get a lower noise spectrum.
However, when we did engage the cutoff18 filter, (with the CHARD P gain reduced compared to our normal gain by a factor of 2), we saw gain peaking in the CHARD loop at 3.3 and 6 Hz, which caused saturations of the PUM DACs. We reverted to the gentler (JLP25) cutoff, and attempted to lower the loop gain to give us more room for the more aggresive cut off. When we did this there was a 2Hz osciallation in many ASC loops which rang up slowly.
The final attachment shows a comparison of the DARM spectrum from the 90Mpc configuration (CHARD blends JLP25) and the trial of engaging cutoff18 with the blends off and the gain reduced to 1.4. With the cutoff18 on the spectrum was nonstationary, the noise below 20Hz is much worse due to the increase noise from the WFS but we do see the benefit of the cutoff.
Possible next steps for this are to debug the problem with TR Y QPD B (Keita and Filiberto are looking at the hardware this morning), or to potentially improve the loop shape by removing cross couplings with other loops (checking if our input matrix choice is cross coupling CHARD and PRC2).
The point was made today that the spectra in the upper right corner for CHARD_P A and B OUT are tricky to compare when looking at their noise, since they have their blend filters already applied.
Attached is a screenshot of those blend signals, before we start using them for blending (so, the IFO is at Reduce RF 45). I'm still showing the OUT channel, but as shown on the left side of the screenshot, the only difference between the ins and outs is that the B has the 76dB gain applied that matches the signal strengths. The blending filters themselves are off.
It does look like the B channel, which is the TMS QPDs, have somewhat less noise below 30 Hz.
Both signals see the 6 Hz feature which PeterF and Anamaria were pointing out look suspiciously like a HAM feature.
