I have taken fresh LSC feedforward measurements, after the IFO had been at 35W for about 2 hours, and the SR3 heater had been at 4W for at least 2.5 hours. This used about 30 min of commissioning time.
Note to self for the future: All of the FF_to_DARM templates should give identical results, since the feedforward signals go through the same path to DARM (via differential ETM output matrix elements), so make sure that the frequencies that have coherence are good enough in one template for all 3 LSCs, then we only have to take one version of this measurement, rather than the 3 I took today.
Since we haven't had a PRCL FF measurement before, I copied templates from MICH and SRCL FF, and then used the PRCL excitations from Sheila's noise buget folder. I also set the PRCL FF elements of the LSC output matrix to match the MICH and SRCL values of ETMX=+1, ETMY=-1. I'll try to find a moment to fix the screen, but the FF output matrix screen doesn't display the PRCL column of the matrix, so these are currently hidden.
Data is saved in the usual place: /opt/rtcds/userapps/release/lsc/h1/scripts/feedforward/ , and it's all checked into the svn. I did update the references in all of the templates, since I can't remember how old they are, but they're pretty old.
I spent some time figuring out the ipython notebooks for MICH and SRCL feedforward fitting, then reorganized them, removed a lot of unused code, and added a bunch of comments, so that hopefully it is a little more user friendly.
One of the big changes I made was iterating on the fits. I let iirrational run and get a good fit, then I apply that filter to the measured feedforward transfer function, and redo the fitting. I repeat this several times until the residual between the measured TF and the fit is very nearly unity in the band that we care about. Note that, as Danny and others had found, I do have to fiddle with the frequencies, coherence threshold, and fit order for each of my rounds of fitting, so this is not yet an automated process. Perhaps this is something that we can ask the CSWG to look into making more automated by iterating over different fitting parameters and then keeping the one with the lowest residual.
The first two attached figures are the results of my fits for MICH and SRCL. The blue dots are the measured feedforward transfer functions, and the colors are the fitted filters or residuals after different rounds of fitting. Orange is after the first fit, then green, red, purple, and brown. For MICH I am using the brown result, and for SRCL am using the purple.
Unfortunately, these didn't seem to help a whole lot. The third figure is DARM and coherence of DARM with MICH and SRCL with blue being earlier in the lock and red after I changed both MICH and SRCL to the new filters. I do wonder a bit if part of the reason that it doesn't look so great in these long averages is that the ASC is oscillating and the IFO is moving around with a ~2 minute period. For just a few averages, it seems like there is a bit of an improvement in DARM around 30 Hz, but it goes away after a few averages.