Prompted by the glitches in h(t) correlated with glitches in B8 I have looked into how the noise subtraction in h(t) are performing.
B8_DC (and B7_DC) are already subtracted in the online h(t) reconstruction, the correlation found between B8 and h(t) shows that the subtraction is not perfect, but it does improve the situation a lot, and attempting to do the subtraction offline with a completely independent code gave a worse result.
Hower in the process I have noticed, that subtraction done offline is better than the official online h(t) in between 90Hz and 150Hz. To make sure that this not just due to the bad weather data, I have looked also at July 26 00:00 UTC when the microseismic is very low, and the same issues is present.
Figure 1 shows Hrec_hoft_20000Hz (the online h(t) reconstruction with noise subtraction) in dark red, and the raw online h(t) reconstruction in light red. In dark blue is raw online h(t) with the noise subtraction done offline, and the other curves show the spectrum of the subtracted noises.
The dark blue curve (offline subtraction) is better than the dark red curve (online subtraction) between 90Hz and 150Hz. There is a jump at 90Hz in the online reconstruction, as at that frequency it switches from using MICH for the noise subtraction to using PRCL for the noise subtraction, as the online subtraction code is not able at the moment to use two channels if the noise between them is correlated. The offline subtraction shows that the MICH contribution (yellow) is larger than the PRCL contribution (green) up to 150Hz. This would suggest that moving in the online subtraction the switch from MICH to PRCL from 90Hz to 150Hz would improve the sensitivity, but this should be checked by experts, as the contribution of PRCL is computed here after subtracting the contribution from MICH, and the two are correlated, so the PRCL contribution is underestimated.
Figure 2 shows the ration between the offline h(t) noise subtraction and the online h(t) noise subtraction. The sensitivtiy is up to 10% better at 90-100Hz, but in terms of range the improvement is slightly less than 1Mpc.
The script used to make the figure is /users/mwas/ISC/PRCLMICHsubtraction_20190728/subtractB7hoft.m