I did a brute force search for correlation between the BNS range and almost all possible channels in trend.ffl, while considering only those samples for which the range was above zero. Among the winning channels are the demodulation phase mistuning of B2 (LSC_B2_8MHz/56MHz_DPHI) and some signals related to the MC Z position, see fig 2 and the attached list. To me, this hints at a dependency on the mistuning of the length of the IMC with respect to the modulation frequency (FMOD_ERR). Since the modulation frequency is still fixed, the length of the IMC should remain constant too. Looking at the LVDT of the top stage, there are jumps of about 100 um after unlocks, superimposed with a slow drift that is probably caused by the tides, which is not a real change of IMC length. In Fig 3, I tried to remove the slow drift and convert the length change in a frequency mistuning of the 56MHz sidebands, which gets as large as 50 Hz. Note that in the early days of initial Virgo, a mis-tuning of the modulation frequency of 4 Hz was enough to spoil the sensitivity!
Fig 4 shows the horizon vs this mistuning (note that I do not know where the real zero is, and the subtraction of the drift might not be perfect).
It should be very interesting to tune the demodulation frequency during a lock (assuming that the bug that prevented us earlier is resolved).
