I had previously raised a preliminary concern about narrow line noise couplings to h(t) from the NCal electronics (see LHO aLOG 53666). Robert had suggested that I can use the temporary magnetometer left on the NCal encoder electronics, close to the NCal mechanism itself (much closer than the VEA magnetometer), and compare coherences between h(t) and the temporary magnetometer with h(t) and the VEA magnetometer.

Using the timeline that Timesh has left us (LHO aLOG 53629), I first selected the necessary segments using the following segment database queries:
$ ligolw_segment_query_dqsegdb --segment-url=https://segments.ligo.org --query-segments --include-segments=H1:DMT-ANALYSIS_READY -s 'lalapps_tconvert Nov 26 2019 14:00 PST' -e 'lalapps_tconvert Dec 4 2019 23:39:00 utc' | ligolw_print -t segment -c start_time -c end_time -d ' ' > segments_ncal_on.txt
$ ligolw_segment_query_dqsegdb --segment-url=https://segments.ligo.org --query-segments --include-segments=H1:DMT-ANALYSIS_READY -s 'lalapps_tconvert Dec 4 2019 23:39:00 utc' -e 'lalapps_tconvert now' | ligolw_print -t segment -c start_time -c end_time -d ' ' > segments_ncal_off.txt
Then I computed 1800 s long FFTs, Hann windowed and 50% overlapping for H1:GDS-CALIB_STRAIN, H1:PEM-EX_MAG_VEA_FLOOR_QUAD_SUM_DQ, and H1:PEM-EX_ADC_0_11_OUT_DQ for the two different periods, whether the NCal was ON or OFF. These were computed and averaged using standard LALSuite tools lalapps_MakeSFTDAG and lalapps_spec_avg_long.

Finally, I used the coherenceFromSFTs.py script written by Greg and Kara to compute coherences over these two time periods for comparison.

Attached is the result from 10 Hz to 1000 Hz, plotted in 100 Hz sub-bands. Each page has 3 figures: left shows the ratio of whitened h(t) ASD (ON/OFF), middle shows the coherence between the temporary magnetometer and h(t), right shows the coherence between the VEA magnetometer and h(t). The whitened ASD is computed with a running median across 101 frequency bins of the time-averaged spectra. This is the basis for Fscans and effectively removes slow variations in the spectrum over different time periods. I'm only concerned in this study with narrow artifacts, so this is ok to do.

The ratio of normalized spectra show some line changes (excursions above 1 indicate worse lines), especially below 100 Hz, but we also see excursions below 1. I speculate that the lines have actually changed in frequency by more than 1 bin. I see this effect other times as well, so I don't believe this is coupled with the NCal being on or off. Most of the rest of the 100-1000 Hz spectra look unchanged.

Since I don't see very much difference between coherence plots, I feel a bit more confident that the NCal electronics are ok, but I don't think we can quite put this to rest yet until we move the temporary magnetometer to sit very close to the NCal power electronics. Robert suggested that we could move this at the next lock-loss opportunity, pending permission of course.

In any case, since the NCal is not essential IFO operations electronics, I'd suggest to leave them OFF and only turn on as-needed for measurements.