Gabriele and I have fit some new filters for the LSC feedforward based on measurements taken recently in a thermalized interferometer. These fits look reasonable and we think they are ready for testing. I am hesitant to put them in the guardian because the last time we tried this we made the noise worse and also caused a lockloss because of some unexpected high frequency behavior. We don't think that will happen again, but I want to be able to supervise the test.
If someone enterprising wants to check it out, the new filters are FM9 of the MICH FF and SRCL FF bank. They should be engaged with a gain of 1. It would be nice to see how they perform at the start of the lock and after thermalization.
Corey tried out these new filters at the start of a recent lock. I ahve attached a set of plots made by Evan. It appears the noise in DARM from 10-40 Hz is worse, while there is some improvement 60-100 Hz. The improvement seems to come from the slight reduction in SRCL coherence. My understanding is that the reference traces are the old feedforward, while the live traces are the new feedforward.
The injections we used to fit the feedforward were done after 6 hours of lock, once the thermalization has settled. This could be why the feedforward seems to be worse early on. I'd like to see this feedforward tried one more time later on, maybe around hour 6 if possible. If the feedforward still worsens the noise, there is some other coupling present that we do not understand.
One more note: comparing the new SRCL FF fit to the current settings, it appears the biggest change is the gain. We could maintain the same SRCL FF filter, but try adjusting the overall gain by a small amount and see if that improves the subtraction.
Concerning the evolution of MICH coupling during thermalization. This plot looks at a 7 hour long lock some days ago, with the old FF filters. I used NonSENS to estimate the optimal TF to subtract MICH on top of the running LSC FF. So this measures the residual MICH coupling to DARM after the reduction due to the FF. It's clear that the residual coupling gets lower with thermalization: this is expected since those FF were tuned with a hot thermalized IFO.
The bottom line is that we cannot have a MICH FF filter that works both for the cold and hot IFO. The same is true for a NonSENS online subtraction