The LSC feedforward, and in particular the MICH feedforward, has needed regular updating since we reduced the IFO input power to 60W in June. I wrote up a summary of the "saga" as of the start of August, see 72037. My main assumption was this: ignoring major IFO changes such as input power/TCS/DARM offset, the changes to the feedforward have to occur because we "uncover" more and more LSC coupling as we improve the low frequency sensitivity. To justify this idea, I referenced the fact that Gabriele and I (except for the first retuning on June 22 with the power reduction) have mainly been doing iterative retuning of the feedforward. Specifically, we run our "retuning" injection with the feedforward on so that the noise coupling we attempt to reduce is any residual noise coupling left over while our main feedforward runs. However, since that alog we have again needed to update the MICH feedforward multiple times without any corresponding improvement to low frequency sensitivity, finally prompting me to think that something here is wrong. However, it is not a significant change in the feedforward from time to time, but merely a few percent change that we iteratively improve against some baseline decent feedforward.

Gabriele and I have found evidence that the MICH coupling to DARM is changing because the ETMX test mass actuation strength has been changing from ESD charge accumulation. Jeff details the effect of this charge accumulation on the calibration in LHO:72416, and has some notes about seeing this effect in the past, effects on the DARM loop, etc.

This change in the test mass actuation can also change the coupling function for the LSC noise contribution. In particular, when we measure the coupling of MICH for the feedforward fitting, we measure two functions. One is the DARM [W] / MICH [N] coupling, and the other is the DARM [W] / DARM [N] coupling. The changing strength of the DARM actuation will effect the required strength of the feedforward actuation. In fact, looking at the filters we have tuned for the MICH feedforward in August, they all have the same shape, but different overall gains.

We make a measurement of the second function above by injecting from the MICHFF filter bank with the input off and the feedforward filter off (but a gain of 1), so we can capture whatever effect is upstream of the feedforward filter banks. Gabriele plotted all of these measurements that we have taken since June 22 and normalized them by the June 22 measurement. The result is shown in the first image attached. This plot shows that the DARM actuation is changing in the same direction over time. We also tracked the Kappa TST value, and noticed that is has been steadily increasing since June 22.

This effect is most visible in the MICH coupling, and we have needed to update the MICH feedforward iteratively more than we have the SRCL feedforward. We think this is because the subtraction of MICH is much more significant than the subtraction of SRCL, by at least a factor of 5. Looking at the implemented SRCL feedforward since June 22, a similar change in gain is evident. The bigger changes in the SRCL shape have been mostly at low frequency to reduce injection of excess SRCL actuation noise into DARM which worsens the DARM RMS.

We predict that as the test mass actuation strength changes, we will continue to need to update the feedforward to improve the subtraction of noise and reach our best possible sensitivity. The MICH feedforward was updated yesterday, Aug 30, and the calibration has been updated today, resetting Kappa TST to 1 (72594). We should track the Kappa TST value. If it becomes even a few percent different than 1, Gabriele and I imagine we will need to make another iterative update to the feedforward.

While we think the few percent change in actuation explains most of the few percent change in the MICH coupling, there could be other changing factors. We have been tracking possible alignment changes in the interferometer related to OM2 TSAMS changes, unexpected temperature changes, etc. Changing IFO alignment could also contribute to some of the changing LSC coupling we have witnessed (although the how of this process is less clear to me). If we manage to mitigate the charge accumulation on ETMX, we should continue to track the MICH coupling in case there are other effects that compromise the success of the noise subtraction.