I disabled 1Hz LPF which was somehow present in the SUM (not NSUM) of the QPDs and WFS DCs that are used for normalization of PIT and YAW. These LPFs ruin the rejection of the intensity noise for QPDs and WFS DCs even at 1Hz.
Here's a list of the filter modules where the 1Hz LPF (all FM1) was disabled: H1:ASC-OMC_[AB]_SUM, H1:ASC-REFL_[AB]_DC_SUM, H1:ASC-AS_[AB]_DC_SUM, H1:ASC-AS_C_SUM, H1:ASC-POP_[AB]_SUM, H1:ASC-[XY]_TR_[AB]_SUM.
Note: Without LPF, when large clipping action is going on and SUM goes very small, I can imagine that PIT and YAW become somehow worse than with LPF. If that ever becomes a problem, enable 1Hz LPF.
Note 2: Green WFS DCs and QPDs don't have the BPF for SUM.
Turns out that disabling the LPF for PIT and YAW normalization has a side effect of making RIN channels 1 (one, unity), but only for ASC-OMC_[AB]. Others don't seem to have RIN channel.
I don't see this as an urgent problem, but if this RIN function is needed, we have to add a dedicated filter module just for RIN normalization purpose.
Also, this will make the sensitivity of ASC-OMC_RIN to the detchar safety injections go away, but this is just an unintended side effect of the change in the normalization. (See alog 74056.)
Well, turns out that H1:ASC-[XY]_TR_[AB]_SUM setting was reverted back right after I changed it, but the rest stood (1st attachment). I might turn the LPF for the TR sensors off later, but not now.
Anyway, 2nd attachment shows that disabling the LPF for OMC QPD SUM for normalization made a big reduction in AM to ASC leakage (2nd attachment right column, though only OMC_A is shown here). NSUM is calibrated to RIN. Current traces are from last night after the LPFs were disabled (Nov/08/2023 10:00:00 UTC), and references are before (Nov/07/2023 02:30:00 UTC), both in nominal low noise. As you can see, disabling LPF reduced the intensity noise coupling to OMC ASC by a large amount especially for f>10Hz. (I'm sure that the coupling was reduced at lower frequency too but it's hard to see the effect w/o injection.)
3rd and 4th attachment show similar plots for ASC-AS_[ABC], ASC-REFL and in-vac POP. I haven't done any deep analysis but there doesn't seem to be much improvement if any. In general, the larger the centering offset is, the bigger the intensity noise coupling is with LPF on, so it's understandable that the OMC qpds show the obvious improvements (because they have crazy centering offsets and the coupling was high with the LPF ON).