HAM1 ASC FF was off from 1396381762 to 1396382969 (20 minutes starting 90 minutes into lock). Using this data I could retune the feed forward filters that subtract HAM1 motion from the ASC signals.
I based the attached jupyter code on an older version, and automated the subtraction training for all pitch and yaw ASC signals (CHARD, INP1, PRC2, DC1 and DC2).
Looking at the filters that are loaded now, we used to do the FF only for the pitch degrees of freedom. The subtraction seems to be doing something good in yaw too, so I suggest we upload and try the yaw feedforward filter.
The first plot shows the expected subtraction predicted by NonSENS. The second plot shows the absolute value of the transfer functions.
Finally, the attached text file contains the foton filter definitions for all the feedforward filter banks. I haven't upload those flter to foton, and not tried them yet. Before each filter there is the name of the filter bank where it should be loaded, for example H1:HPI-HAM1_TTL4C_FF_CART2ASCP_1_1
[Jennie, Jim, Gabriele]
We tried the new filter, and there's something wrong with them. Even with a gain of 0.5 (instead of 1) they make some of the ASC signals much worse, see attached plot.
We'll use the time we got today with FF off (starting 1396987833 and lastring 10 minutes) to debug the problem and retune the FF
In the attached plot: green = FF off, blue = nominal FF, red = new FF with half gain
Gabriele, Jim, Jennie W
I saved the new HAM FF filters in H1SEIPROC foton file. These will only be loaded in during our commissioning window tomorrow.
The code attached to this alog had a bug that produces the wrong filters. Attached an updated version, hopefully without any more bugs. Also attached the new text filw with all the filters.
The two plots show the expected subtarction and the absolute value of the filters.
Gabriele's new coefficients have been loaded into SEIPROC and are available, but haven't been turned on or tested yet.