During an analysis of ITF angular stability, it has been observed that PR_TX low frequency residual rotation seen by AA signal can be well explained assuming that the source of noise is the actual rotation of NI. The AA loop imposes PR to follow NI rotation, which is large enough to be seen by NI and PR optical levers: the coherence between the two is almost 1 at 40 mHz and the TF is flat (fig 1). There is also coherence with all the other TMs of the arms: all together seem to have an excess noise in COMMpTX dof. This is evident comparing the OptLev signals in DIFFpTX and COMMpTX combinations (fig 2). The actual DIFFpTX rotation is forced to be very small (about 1 nRad) by the loop on B1p_QD, so the corresponding combination of the optical levers gives the level of noise of those sensors: we can see in figure that COMMpTX motion is much larger than the sensor noise.
The component of noise below 0.1 Hz is very likely due to the centering loop on the dither signal. The one used for COMMpTX is NI_MIR_Y_AA; in fig 3 we can see that it is noiser than the other three. We can also see a different shape below 40 mHz: it should be the suppression due to the COMMp slow loop. I would say from this plot that the gain is too high respect to what is really needed.
In fig 4 the coherence between two witnesses of this noise is shown: PR_TX is coherent with the west arm centering signals in the combination given by a rotation of the axis, while the combination given by the axis translation (soft mode) is not coherent. This is another confirmation that the excess noise affects COMMpTX.
The first thing to do is to reduce significantly the gain of COMMpTX slow loop and see what happens. I would do it both for TX and TY, even if a similar analysis for TY has not been done yet. The there should be the possibility to reduce a bit the noise using a different error signal and excluding the noisy one. The COMMp combination of NE, WI and WE seems to be the cleaner option.
