Occasionally the transition to the earthquake mode has caused locklosses. It took me a little while to figure out what was causing it, but I think it's mostly due to the BRSs adding a lot of very low frequency signal to the sensor correction signal, plus insuffficient low frequency roll-off of the sensor correction filter we use just for the earthquake mode. I've made a different filter (first attached plot) that has more low frequency roll-off, but gets slightly worse subtraction around 50-80 mhz. The dotted blue line is the new filter, the red is the old Eq filter and the gold is the filter we use during non-eq times. The new filter still injects more low frequency motion into the ISI cps below 10 mhz than the nominal filter, but less than the old filter. The reason we want such a low frequency filter is so we can try to suppress differential motion in the earthquake band, which the nominal filter only amplifies.

The second plot is a comparison of the outputs of different sensor correction filters during a earthquake when the transition broke the lock. The purple trace is POP A, showing when we lost lock, the dashed green is the seismic state, we lost lock 14 seconds after the request was made by the operator. The other 3 traces are the outputs of the different sensor correction paths for the Y arm, which are different ground signals run through different filters. The blue trace is the eq path with the new eq filter, the red trace is the old eq filter. The gold trace is the nominal sensor correction filter and path.  I don't post it here but the low frequency motion (< 10mhz) was much smaller on the X arm at the time.

The reason we lost lock was the DARM offset created by ramping from the gold trace to the red trace, pushing all of the chambers ~10 microns in the y direction over 30 seconds. The new filter doesn't create such a large offset, hopefully this will lead to fewer lock losses transitioning to the earthquake state.