elenna.capote@LIGO.ORG - posted 21:05, Tuesday 27 August 2024 - last comment - 16:31, Thursday 29 August 2024(79755)
New SRCL and MICH FF fit
I was able to new SRCL and MICH feedforward using the iterative method. Measurements were taken in alog 79693.
The current MICH FF is performing well except for below 30 Hz, so that's where I targeted in my fit. SRCL needed some improvement everywhere.
The new MICH filter is saved and loaded in FM8 and the new SRCL filter is saved and loaded in FM5. They are both labeled with the date '8-27-24'.
To compare how these filters are performing, I would use the templates Oli saved in the above alog, but first save the live trace as a reference to compare against.
I have not yet done the PRCL fit because I want to see how SRCL performs, and maybe redo the PRCL injections before trying the fit.
Here is some more information about the fits.
First, my MICH fit caused a lockloss this morning because I forgot to the check the phase on the filter. Sometimes, the phase gets flipped during the fitting. Usually, I compare the sign of the phase with the previous filter in foton and adjust accordingly, but I forgot to do it this time (rookie mistake). I have double checked the new SRCL filter phase and fixed the MICH phase.
Attached are two screenshots of the fittings. First, the MICH fitting compares the current filter, labeled as "reference 1" with the red trace which represents the new fit. The bottom right plot compares the fit residuals. You can see from this plot that the most improvement occurs at low frequency, with some small improvement at mid frequency. The SRCL fitting has many more traces, but compare the orange "current fit" trace with the trace labeled "best with less HF gain". This has more improvement almost everywhere. There is an increase in gain at high frequency, but it is less than an order of magnitude, so I think it's ok. The new fit also has reduced the high Q feature around 300 Hz that was potentially injecting noise. There is a factor 5-10 improvement between 10-50 Hz that will help the most.
These have now been tested, SRCL passes, MICH fails.
The SRCL screenshot compares Oli's SRCL measurement from four days ago with the "current" filter, and the new "trial" filter that I applied today. There is clear improvement everywhere (except for a small worsening between 100-200 Hz), so I think we should use this filter.
The MICH screenshot compares Oli's MICH measurement from four days ago with the "current" filter, along with "no FF" and "trial". The trial did what I promised, and reduced the coupling between 10-20 Hz, but clearly at the sacrifice of the noise everywhere else. I think we should stay with "current".
I am changing the guardian to select this new SRCL filter (FM5) in "lownoise length control" and the gain back to 1. There will be an SDF observing diff for SRCLFF1 that can be accepted by the operator.
Some thoughts about MICH FF:
In my opinion the hardest region to fit is between 10-20 Hz because of the presence of some high Q features, such as around 17 Hz. It would be worth considering what is causing those features- perhaps some bounce/roll notches. Do we still need those notches, or could they be briefly turned off during a feedforward injection? That might make the low frequency portion easier to fit and therefore easier to achieve good subtraction 10-30 Hz.
Looks like these are maybe BS M2 LOCK L FM10. We can try turning them off I suspect. Foton says they have a 2 second ramp, so should be okay to turn off just before the measurement (I'm not sure if we need them all the time, but maybe we do).
Today Sheila took another injection of PRCL for me so I could fit a new feedforward. The fit looked promising, however once it engaged it apparently caused oscillations everywhere, and I turned it off fast enough to avoid lockloss (thanks Corey and Ibrahim!). I checked the phase and gains beforehand, no high Q features, etc so I don't know what could be the issue.