LSC feedback-related peaks have been visible in DARM at 32Hz and 40Hz. it was demonstrated that they could be removed by notch filters in MICH servo.
These peaks were quite distinctive in the MICH/PRCL/SRCL error signals and therefore it was suggested that the these servos contaminated the DARM at these freqs. This peak has also been visible in the control spectrum of the 45 MHz modulator (i.e. the noise level before the EOM drive stabilization). Since the stabilized out-of-loop measurement of the driver noise did not have these features, the source of the peak are still unknown.
Since the UGF of the MICH is ~8 Hz, I tried to add notch filters to MICH to see if we could remove a part of the noise to DARM. The following notch filter was registered to FM9 of MICH2 filter bank with the name "not32/40".
New Filter: MICH2 FM9 not32/40 notch(32.3,35,50)notch(32.7,200,10)notch(40.05,200,20)
(FM9 originally had the following filters DelMe2 zpk([0.5;0.5],[500;500],1, "n")
I am not sure where it came from. ALOG didn't have the description about this "DelMe2".)
Attachment 1: The notch filters consumes ~1 deg of phase mergin at 8 Hz. This filter could just be turned on during a lock.
Attachment 2 shows the ASDs of the MICH/PRCL/SRCL feedback signal with and without the filter. The peaks were removed from MICH, and luckily they disappered from PRCL too. This means that there is a coupling from MICH to PRCL. There is also a peak in SRCL at 32Hz, but it was not coming from MICH.
Attachment 3 shows the effect on DARM. The lower plot is (raw) DARM, and the arrows indicate the removal of the peaks. The upper plot shows DARM after noise subtraction, where the improvement is still visible (particularly at 32 Hz).
Attachment 4 shows the coherences between DARM and MICH/PRCL/SRCL without (upper) and with (lower) the notch filters. Now there is no large coherence between 30Hz and 45Hz except for the one at 41Hz, which is not MICH-related.
This filter was registered in FM9 of MICH2 filter bank. It was turned off, and is not yet implemented in Guardian. I will talk to other commissioners next week about it.
This MICH2 filter change was registered in guardian (and loaded).
- LOWNOISE_LENGTH_CONTROL section of ISC_LOCK.py was edited.
/opt/rtcds/userapps/release/isc/h1/guardian$ diff ISC_LOCK.py.orig ISC_LOCK.py
5342a5343,5345
> #engage 32Hz and 40Hz notches in MICH filter to reject injecting these peaks in MICH and PRCL. See LHO ALOG 68050. Koji Arai 20Mar2023
> ezca.get_LIGOFilter('LSC-MICH2').turn_on('FM9')
>
The code succesfully ran. Committed to svn.
Tagging DetChar. The CW group should be particularly excited about this change, and maybe could diagnose a "before" vs. "after."
It does not appear this notch is having much effect. The 32 Hz line is still present, and it is most coherent with MICH. https://ldas-jobs.ligo-wa.caltech.edu/~gabriele.vajente/bruco_CAL_1364130018/LSC-MICH_IN1_DQ.html
I tried adding the same 32 Hz notches to the MICH ASC loops, based on seeing the significant contribution at 32 Hz from the various ASC loops, alog 68305. It doesn't seem like the notches had much effect at all on the peak height in the loops or on darm. I took a quick low resolution comparison and got excited because I thought I saw improvement, but the high res spectrum showed no change.
In both plots the blue reference trace was taken before I turned on the MICH ASC notches, and red live trace is with the notches on. The LSC MICH notch was on for both times.
It may have actually been that the 32 Hz line height was changing over time as well, so I was fooled by the low resolution spectrum because it was already decreasing, and then when I did the high res the peak increased again.