Eric Quintero, Rana
Summary:
Studying coherences between the ground and the LSC drives, we find that we can reduce the longitudinal forces to the mirrors in the 0.1-0.3 Hz band by a factor of ~2-5 in most cases, for both sites.
Details:
The recent success of the length to angle (L2A) feedforward at LLO led us to wonder about the source of so much low frequency motion. The SEI loop feedback and feedforward has been heavily tuned over the years to reduce much of the motion. By looking at the coherence and transfer function between the ground seismometers and the LSC*OUT signals (in a similar manner to what we did with the length to angle in May 2016), we are able to make an upper limit estimate on how much this can be reduced by implementing a global FF (just as we did in early 2010 for the S6 run using HEPI/PEPi).
We took 1 hour of data from 1000 UTC on March 12. The RMS of the ground motion in the 0.1-0.3 Hz band was 0.5 um/s, which is high, but only moderately high for the winter time. The summer time motion is more like 0.1 um/s.
The attached plots show:
upper plot: LSC control signal & LSC control signal after ideal subtraction
lower plot: top 5 signals used in the subtraction (in practice, using ~3 signals is enough to do the biggest part of the subtraction)
This subtraction is done on a bin-by-bin basis in the frequency domain, and as such, its a best case estimate. In reality, implementing a causal filter which avoids injecting too much noise at 3-20 Hz will degrade the subtraction performance somewhat. We are now working on making a frequency dependent weighting so as to make realizable filters.
In the attached channel list, you can see that all ground seismometers and tiltmeters were used.
- In some cases, the tiltmeters do seem to be useful for doing this FF, although it remains to be seen if their noise at > 3 Hz is too high to preclude using them.
- As we also see for LLO, much of the motion is coming in through the Z direction. This implies that the vertical to horizontal coupling in the ISI (perhaps due to imbalances in the sensor calibrations?) is the culprit. In any case, we can fix it by FF just to improve the performance of the interferometer in the near term and then address the root cause at a later time.
- These LSC_OUT signals are just the control signals at the output of the LSC sub-system, but they are not really equivalent to force or displacement, since the filters inside the suspensions distribute the signals among the upper suspension stages with different filter shapes and crossovers: the suspensions do not 'look' like a simple pendulum from the LSC side. Some complicated suspension model + digital filter forensics should be able to convert these plots into physical units so we could figure out how much the motion reduction would really be, but I expect that the bulk of the relative motion between the mirrors and chambers comes from 0.1-0.3 Hz, as does the angular motion that the WFS loops have to deal with. It would be very handy if there was a matlab function that could read in a simplified suspension model and the digital filters / gains for a specified GPS time and return the calibration to go from LSC counts to meters.
- We will need to do some minor model changes to be able to bring the seismo signals into the LVEA at 2 kHz and then to send out the ~10 Hz BW signals to the various ISIs. There's some discussion between Joe B. and Jenne to clarify precisely what and how many changes.
- For LHO only, I've also used the BRS channels: H1:ISI-GND_BRS_ETMX_RY_OUT_DQ & H1:ISI-GND_BRS_ETMY_RX_OUT_DQ, which look like they are the right ones. But, it seems that they do not contribute to any significant subtraction during this period. IF they were already being used at this time, then perhaps that just means that their FF filters are already well tuned.
I've attached about 10 hours of BRS-Y driftmon data around the time of this crash. It looks like this crash was caused by trying to use BRS-Y in a bad range. Even if the data looks temporarily smooth, BRS-Y should not be used if driftmon is below -15k counts.
The close time association with closing of the remote desktop terminal was likely just a coincidence. It is still advised to not login to the machine remotely when BRS-Y is being used for feedforward, unless really necessary.