J. Kissel, J. Driggers More to come, but OMC Whitening Configuration change (using 1 whitening stage instead of 2 whitening stages and a low pass) is in testing now. Quick answer from PCAL to DELTAL EXTERNAL broaband TF shows ~1.0-1.5% frequency dependent addition systematic error with the new configuration. Magenta is nominal configuration (2 whitening stages and a low pass), and red is test configuration. Sweeping now...
I have accepted in SDF the differences that this caused for the digital and analog filters (see screenshots of sdf files).
I have also modified the ISC_LOCK and OMC_LOCK guardians such that when we next acquire lock, it should come back to this state of only 1 stage of analog filtering.
- ISC_LOCK I just removed the edges that were connected to the OMC_WHITENING state near the end of the acquisition sequence. Lownoise_length_control now goes directly to Reduce_rf9_modulations_depth. This prevents ISC_LOCK from asking the OMC guardian to add the 3rd stage of whitening. Nothing else happened in this state.
- OMC_LOCK I commented out the lines in OMC_LSC_on that would have turned on the 2nd stage of filtering. Note that in Prep_omc_scan all 3 stages of whitening are turned off, so the OMC should now only have 1 stage of analog filtering.
Calibration measurement suite sweeps are complete, and we are back in nominal low noise, but Robert has taken over the IFO for commissioning. We are remaining in this 1 stage of whitening only configuration. We have resolved that it is not prudent to change the compensation filters, and/or update the calibration model parameter set, and we'll therefore just live with this increased systematic error in the low latency h(t). Eventually, after the systematic error is well quantified, we will create a new model parameter set, and re-calibrate the data -- starting at the next observation ready segment today. I'll post the official time as to when we've gone back in to observation ready, so as to clearly define the first observation ready segment with this new change.
The first observation ready segment that includes the above described OMC whitening chassis filter configuration change started at Mar 16 2020 21:18:54 UTC (GPS time 1268428752)
Not that there is anything profound here, but I caught a glitch. This is the first one that I've caught to grab a screenshot of after our change this morning of the OMC DCPD whitening filter configuration.
Note to self: compare the _OUT_DQ of this glitch with other glitches in the old nominal config that I have screenshots of, to (try to) see if these glitches were attenuated due to the new analog configuration.
The summary pages are reporting a significantly reduced range, but I'm not totally sure why.
In the attachment, I've plotted ~an hour of data each for observing times this morning before we changed the whitening configuration, and for an observing segment after we made the change. This is H1:GDS-CALIB_STRAIN, so has time dependent correction factors applied (although, as Jeff points out earlier in this thread those won't fully account for the small frequency dependent change that we've acquired).
Blue is an old nominal (3 analog filters) time, starting at 16 Mar 2020 16:24:40 UTC. Orange is the new nominal (1 analog filter), starting at 16 Mar 2020 22:19:30 UTC. The darker lines are the 50th %-ile, and the shaded regions are the 5th and 95th %-iles. I can't tell much of a difference by eye. Also, when I calculate the range from these median spectra, I get less than a 0.5 Mpc difference. So, I'm not sure why the GDS range on the summary pages is so different now than it was this morning.
I've used measurements of this OMC whitening chassis from (Sunday!) 2019-03-03 to predict the systematic error that transpired as a result of this configuration change. See further discussion in "PART II" of G2000527. Making a *very* long story short, (a) the systematic error incurred on the final DCS C01 calibration amounts to (a frequency dependent, but at maximum) 0.5% and 0.25 deg error at 100 Hz and 50 Hz, respectively. (b) the error arises from - poor modeling of the super-Nyquist response of this chassis in the calibration pipeline, - previously poor understanding of the circuit resulting in a poorly informed fit of the measurements, and finally - the original data set only was taken down to 5 Hz, where the chassis (namely the first and third filters, which are the whitening stages) has response at 1 Hz and thus *any* fit (be it Stefan's original fit to update the compensation filters, or Lilli's fit for super-Nyquist poles) is fundamentally limited by the data. Attached is a comparison of - the ratio between the 2020-03-09 (pre-change) and 2020-03-16 (post-change) broad-band PCAL injections -- and thus the *measured* systematic error incurred by the change, and - the estimated systematic error derived from a re-fit to the original data. Again, for further explanation of how this modeled error estimate was derived, see "PART II" of G2000527. The script the accompanies the analysis for G2000527, from which all plots come, lives in /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/ fit_OMCDCPDWhiteningChassis_20190304_forG2000527.py
