[Craig, Jenne]
On Wednesday I took measurements for iterative update filters for the LSC feedforward. SRCL and MICH don't really look like their ideal update filters are very far from unity at frequecies we care about, so I'm not going to fit and implement them today. However PRCL has some updating that it could use. Note for the future, slightly stronger excitations may help to get better, easier-to-fit data.
In the attached figure, the blue points are the measured transfer function to fit, which will become the update filter in series with the current feedforward filter. Note that in this plot, data points that have poor coherence are set to 1, so the low and high frequency portions of this plot look funny. The fit is only considering data points with high coherence, so those points are ignored. The green trace is the fitted filter that will be put into Foton. If our original FF filters were perfect, these update filters would be unity everywhere. You can see that my fit doesn't match the measurement very well below ~15 Hz, but I'm going to accept that in hopes that we make improvements at higher frequencies where we're actually detecting GWs - hopefully this will be easier to fit with better data next time we take these 'update' measurements.
At around 00:32 (+/- 1 min) on 9 Nov 2019 UTC I turned on this new update filter. We were out of observe for a few tens of seconds during the transition, then I accepted the SDF diff (FM4 in LSC-PRCLFF) and put us back in Observe. Seems to have perhaps improved the range a bit.
At around 00:49 I turned the new filter back off. Again I accepted the SDF diff, and we were back in Observe within a few tens of seconds.
At around 1:03 I turned the new filter back on, accepted SDF, and we were back in Observe within a few tens of seconds. I think we'll leave it on for the weekend to see how it goes longer term.
Craig will post a comment with some plots showing that perhaps we're making a bit of an improvement.
We did a on-off-on test with the PRCL iterative filter. Seems to have had no effect in the bucket. We definitely made things worse at ~10 Hz, likely because of a relatively poor TF fit. We have left this filter on because of the superevert that happened during this test. Does not seem to cause any additional noise in the bucket, so this should not matter for this weekend. There was a glitch during the 'off' time. I used median averaging to get the 'off' PSD. Unclear to me how close to the superevent the glitch was.
