Since we've changed the ETM ring heaters, the Jitter cleaning needed to be retrained. Also, since jitter is the only one that's doing any cleaning right now, all the others also need to be retrained.
I used a time from over the weekend when the IFO had been online for ~9 hours, so plenty thermalized. With some suggestions from Gabriele, I retrained the Jitter subtraction and the Laser noise (high frequency) subtraction. I hope to try these out during the next time we have a short commissioning window (perhaps tomorrow afternoon, in coordination with LLO).
Attached are the offline results for Jitter and Laser noise.
Note to self, the Jitter noise is even better with the N_SOS=16 that Gabriele suggests, but implenting that would require a restart of the h1oaf model, so this is with N_SOS=8.
The offline subtraction of noises that come from the h1bos model is not quite working right now. This means that I cannot subtract the main LSC noises, MICH and SRCL, until that's figured out. I can (and should) train PRCL, since that one is on the h1oaf model, and not modulated. This may also be motivation to first do a linear subtraction of MICH and SRCL (in addition to PRCL), and let the bilinear h1bos model do less work.
Also, Jennie pointed out that there has recently been a big noise bump around 120 Hz, apparently since about May 27th. If you look at the offline Jitter cleaning, it's well witnessed by the jitter witnesses (IMC WFS A DC pit and yaw). Hopefully the online implementation of the cleaning does as well as the offline....we'll check tomorrow.