The allocation of low frequency locking correction to IP has been repeated for BS and PR. The test on BS was basically the same as the previous time and the result was equal: the high pass filter applied to the marionette correction had an effect, but not that much as expected, because there was also an unexpected increase of the global correction (fig 1). Anyway, this strategy produces an amplitude reduction of marionette correction, if one look at the data in time domain (fig 2).

With respect to the previous test, the only difference in configuration was the contemporary presence of a similar locking splitting on PR. The test on PR had a better result: the gain below 100 mHz due to the high pass on marionette was quite relevant, becuase the global correction was unchanged (fig 3). Also in this case there was an unpleasant worsening: the peak at 200 mHz. For this problem, we hope we could find a solution by improving the IP plant compensation. The amplitude reduction of marionette correction is already relevant (fig 4), but we think that the problem at 200 mHz should be fixed before finalizing the change.

The new strategy have been tested also on the arms. In this case, the allocation to IP needs to coexist with GIPC, which is an allocation of different kind. Having some trouble at the first attempt, we decided to better study the existing GIPC. We found that the mixing of local and global signal was formally not perfect. We also re-designed the plant compensation, profiting of a recent improvement of the suspension modellization. At the end, the better tuned configuration allowed to increase the blending frequency between correction and accelerometer, obtaining a visible reduction of correction at low frequency (fig 5). The new configuration have been left running: the improvement is visible also in time domain (fig 7, fig 8). The comparison with recent data has been done in equal condition of environmental noise (fig 6 the wind speed, fig 9 the correction on PR).

After the improvement of GIPC, the additional allocation has been tested again without success, but we found the reason in a mistaken implementation. Fixed it, the additional allocation (tested in LOCKED ARMS IR) worked fine, apart from the usual little problem around 200 mHz (fig 10). We are not yet ready to use the new strategy, also because its engagement cannot be done automatically at the lock acquisition, but it requires some additional logic.