One reasons why the B4 56MHz I/Q noise is high comes from the small modulation witnessed by the B4 photodiode.
Figure 1 compares the 112MHz currently (purple) and 1 month ago at 13W (blue). The current 112MHz signal is about half of a month ago, when it should be double because the power has been doubled. So there is a factor ~4 of side-band missing. It is not clear to me how much is from a lower 56MHz modulation and how much from a lower sideband gain inside the PR cavity.
After using today LSC noise injections from ~14:30 UTC to update the MICH coupling transfer function and the mirror actuators gain, I have made a noise budget for the CLEAN time.
Figure 2 shows the noise budget, where most of the sensitivity curve is well explained. Note that the MICH/SSFS sensing noise is double counted here, as both the theoretical quantum/dark noise and the measured noise in the MICH/SSFS error point are added in quadrature.
Figure 3 shows the same without the MICH/SSFS error point projection, which removes the double counting, but also some of the features present in the SSFS error point
To be more clear Figure 4 shows the projection of quantum noise and Figure 5 of dark noise, where clearly B4 56MHz I/Q are dominant.
--------
Edit, after the alpha subtraction was updated
Figure 6, the total noise budget shows a factor 2 gap between the projection and sensitivity in the 20Hz - 60Hz region.
Figure 7, as shown already in the longitudinal noise budget entry, the PRCL projection is getting close to the sensitivity. However the MICH projections are different between the two analysis of the same data. It needs to be understood why that is the case.
-------
Edit 2, after checking the PRCL loop, and increasing the 8MHz modulation index by a factor 2 to match reality and get the right PRCL loop gain, the MICH noise projection looks consistent with the longitudinal noise budget, see figure 8. And the projection doesn't change between doing a linear and non-linear noise projection of MICH.
Figure 9, adding a ~1/f^3 mystery noise fills the gap in the projection between 20Hz and 50Hz.
I suspect the reason is the bad CMRF that is present since the shutdown.
Figure 1, shows the FMOD line coupling to DARM before and after the shutdown, the coupling is ~3 times higher now.
Figure 2, shows the noise budget during the PRCL noise injection. Note that the PRCL noise projection matches well the sensitivity curve, but that projection is not measured but modeled!
Figure 3, shows the noise budget if the CMRF is improved by a factor 100 in the model, the PRCL noise projection stops matching the observed noise curve.
So the observed PRCL coupling to DARM is passing through the SSFS loop. The PRCL loop injects noise on to the PR mirror, this PR motion creates a spurious signal on the B4 56MHz I signal used for the SSFS control. The SSFS loop imprints that noise on the laser frequency, which then couples into DARM due to poor CMRF.
