When we measure the A2L coefficients, we use high frequency lines (20-30 Hz), but the DHARD and CHARD loops have finite roll-off, and high coupling to DARM. For this reason when we inject a line on a single test mass, the loops are actually propagating that line everywhere (you can clearly see the line in DHARD).

I did a quick test on ITMX, before losing lock for unknown reasons.

I first measured the ITMX yaw to length coupling, using my script described in 44532. The measurement is done with a line at 30 Hz, averaging 60 seconds. I then added notches at 30 Hz on all CHARD and DHARD loops (using FM6, which replaces the old 40W or 50W filters). I then repeated the measurement.

Without the notch, the measurement predicts a Y2L gain of -0.798.
With the notch, the measurement predicts a Y2L gain of -0.822. A second measurement gives -0.829.

So the difference between notch on and off (0.024) is larger than the variability between measurements (0.007), and might set the limiting factor in out A2L measurement and cancellation. The measurement accuracy could be improved by longer averaging.

When the IFO is back locked stably, I'd like to carry out a round of A2L tuning with notches and see if this allows us to get a better decoupling. So far I managed to run once ITMX_YAW and ITMY_YAW and update the gains. Then the IFO lost lock again without any apparent reason.