After Gabriele's 73546, Jenne and I looked LSC-DARN_IN1 before and after the feedforward was added. You can see the new FF (in particular SRCL) has more noise 5 to 9Hz, plot attached of only old vs new FF. The larger peaks at 0.05 and 0.1Hz are caused by increased microseism, they are not present on Oct 14th. No other noise down to 1e-4Hz.
We turned off the MICH FF only 20:07UTC - 20:21UTC and then the SRCL FF only 20:22UTC - 20:35UTC. Turned off both FF from 20:35UTC until 20:42UTC where we had a lockloss. Plot attached of LSC-DARN_IN1 in these configurations.
As part of thinking about why a little bit of excess motion around a few Hz can cause nonstationarity in the GW band, I had a cursory look at the filters in place for the ETM R0 tracking.
I have not yet re-looked to see why we made different choices for the angular vs length loops, but the top frame of the attached plot shows the magnitude of the control filters for ETMY's R0 tracking for pitch and length. The bottom frame shows the rough magnitude of the open loop gain of the length loop, using a filter that was in the filter bank called 'plant', and including the gain of 20 that is in the filter bank.
So, it looks like the angular loops (I'm assuming the yaw filters, which are named the same, are actually the same as the pitch filters) emphasize this region where we've got a bit of excess motion in DARM due to the LSC FF, but the length loop does not. We should give this some thought, but consider adding some emphasis to the length tracking filter to have the length tracking work up to ~5 Hz or so.
A parallel path forward is that we should also consider is changing the highpass in the SRCL FF filter bank to be more severe (thus sacrificing some phase and efficacy around 10 Hz, but hopefully an overall win above 20 Hz), however this implies doing the full remeasurement and refitting of the SRCL FF. We can work on this later this week, or whenever we next have some commissioning time.
Here's a design for a more aggressive high pass filter, at the price of 20 degrees of phase rotation at 10 Hz. I think we might need a retuned of the SRCLFF after implementing it. The new high pass is saved into FM10 of SRCLFF1 (not yet loaded).
The noise reinjection between 2 and 4 Hz should be about 4 times lower if we manage to retuned the SRCLFF filter without increaasing the low frequency gain.
