After a failed test of the "correct" PRCL feedforward fit today (80444 and 80287), I tried again to fit a PRCL feedforward, this time using the good measurements Camilla took in 80444, using both the PRCL excitation and the correct PRCL preshaping.

There was some trial and error here, mainly due to the difficulty in matching the phase as well as the rising gain of the transfer function above 100 Hz, which you can see in this plot. This results in the best fit having some high Q pole just above 250 Hz, and of course the usual struggle to keep the gain low below 10 Hz. I found better results when I limited the gain between 1 and 10 Hz to a maximum of 0.5, and also reduced my fit range to go between 10 and 60 Hz, which is where the sensitivity improvement is required, and then limited the gain from 100-1000 Hz to 0.02. PRCL is limiting the noise up to 30 Hz, and far below th noise floor around 100 Hz, so it's probably ok to keep the PRCL noise the same or just slightly worse above 60 Hz (NB alog).

Finally, with Gabriele's help, I found a fit that worked! Here is the injection comparison with no feedforward and the new feedforward. I updated the DARM traces, so you can see that the PRCL contribution to DARM is slightly worse around 100 Hz, but much better up to 60 Hz, by up to a factor of 10.

The new filter is in FM7 (I have triple checked this!) and I have added it to the guardian and SDFed this in observe (screenshot).

I did some chopping of the PRCL FF to check the improvement in DARM, see plot. I think there is improvement between 20-30 Hz, but it's hard to tell.

I set out to rerun the noise budget injection of PRCL so we could remake the plots, but there was a lockloss while I was measuring (I think because people were on the floor?). I don't think the lockloss was from the NB injection.