The vertical mode of PR mirror (11.6 Hz) gets randomly excited during the lock acquisition; the lock can even become unstable on that (orthogonal) resonance because of variable and not predictable coupling of PRCL sensing to PR vertical motion.
A possible way to control the instability is to send a fraction of the longitudinal correction to the marionette vertical actuation. That sort of damping works for the arm test masses and is used to reduce a similar instability occurring on WI during the lock acquisition. Unfortunately the same strategy doesn't work with PR, because the coupling seems to have different sign at different steps of the lock acquisition.
For that reason we decided to implement a true local damper of PR mirror vertical mode, using as error signal the vertical LVDT which measures the relative motion of filter7 and its crossbar. F7 crossbar is part of the 3 bodies system (mirror-marionette-crossbar) for which the 11.6 Hz mode is one normal mode of resonance. The LVDT signal is sensitive enough to see the peak when the instability becomes large and a loop can be effective in reducing the Q of the mode, but we have to pay attention on the possible noise injected on the payload by a local control.
A measurement of F7 LVDT response to MAR_Y_CORR has been measured and fitted (fig 1). The fit requires a high quality factor, at least 1000. A narrow band loop around 11.6 Hz has been developed (fig 2, fig 3). Repeating the TF measurement with the loop closed, the Q appeared to be lowered down to 300 (fig 4).
The damper has been tested during a lock acquisition, when the 11.6 oscillation had reached a very high level, well visible everywhere (fig 5 - the probe Sa_PR_MAR_Y_CORR is the output of the damper, which was actually not sent to the actuator when the resonance was growing). The damping have been quite effective and fast, but during its action some low frequency motion of the payload was induced. This is visible on F7 LVDT and also on the quadrant signal used in loop for the beam control (ASC_PR_Y), saying that the action induced a small excess of PR vertical motion and beam angular fluctuation. The excess noise disappeared when the steady state was reached, so I would assume that the noise of the damper is negligible, but the lock did not last enough to check better. Anyway, given the effectiveness of the damper, I decided to reduce its gain by a factor of 3 and leave it permanently in operation. A long lock in CARM NULL is needed to see if this gain is enough to guarantee the stability at 11.6 Hz.
A possible improvement can come from a swap between the local sensor and the quadrant in CARM NULL: the SNR of the quadrant is quite better, but that signal is available only in CARM NULL, while the 11.6 Hz can affect the lock also in the previuos steps.