Pep Covas, Evan Goetz, Ansel Neunzert, Brynley Pearlstone, Keith Riles

Summary:
We report on lines and comb investigations by the CW group that lead toward a version 1 lines and combs safe for cleaning in O2 H1 data. Run averaged spectra (covering all of O2 except a 1 month epoch, details below), provide a basis for data quality issues that can degrade CW/Stochastic searches. Any sharp spectral feature has a potential to contaminate and degrade astrophysical searches. We attempt to find all of the combs that are obviously non-astrophysical, and to determine any environmental influence for single lines.

If we can mitigate the most prevalent/egregious combs, this would go a long way towards improving the science that can be done with LIGO data. These combs are mostly at low-frequency (f < 100 Hz):

Comb spacing / offset
---------------------
1.0000 Hz / 0.0000 Hz
1.0000 Hz / 0.5000 Hz
1.0000 Hz / 0.9994 Hz
1.0000 Hz / 0.9987 Hz
0.9999862 Hz / 0.2503172 Hz
0.9878881 Hz / 0.0000 Hz

Notching or replacing data from the narrow bins only works as long as no other contamination is in nearby bins. Thus, it would be better to mitigate these combs in hardware.

A full list to the version 1 line/comb list can be found by following this link. Also through this link, one can find the un-vetted lines found in the H1 run-averaged spectrum.

Attached below are plots of the run-averaged spectrum in full, and in narrower frequency bands.

Details:
Using the C01 calibrated h(t) for O2 through July 2017 and C00 calibrated h(t) for Aug. 2017, we produce 7200 s long Short Fourier Transforms (SFTs), excluding CAT1 CBC vetoes. This mimics many CW searches and also provides good frequency resolution and coherence length so that most narrow spectral features that can impact CW searches can be found. We exclude the period from March 14, 2017 17:00:00 UTC to April 18, 2017 23:00:00 UTC to avoid the bad spectral contamination caused by the Pcal camera ethernet adapter (see aLOG 35640).

In the figures below, the noise-weighted run averaged spectra are plotted over the O2 run and notable features:
Figure 1, 20 Hz - 2 kHz; overall, better than O1
Figure 2: 20 Hz - 50 Hz; combs and lines galore (too many to list here)
Figure 3, 50 Hz - 100 Hz; more combs and lines but better than below 50 Hz, oddball narrow comb forest around 75 - 80 Hz
Figure 4, 100 Hz - 200 Hz; much better here than below 100 Hz, oddball narrow comb forest around 153 - 155 Hz
Figure 5, 200 Hz - 400 Hz;
Figure 6, 400 Hz - 1 kHz; violin harmonic regions are difficult/impossible to analyze (damping these modes may help immensely), oddball narrow comb forest around 970 Hz
Figure 7, 1 kHz - 2 kHz; violin harmonics problematic (again, damping will likely help), other oddball narrow comb forests around 1022 Hz and 1046 Hz

The procedure to find the lines and combs was mostly a divide-and-conquer manual inspection of the data. Narrow features were flagged, and combs could be identified by their common spacing. Some automated tools were helpful, but manual inspection was usually necessary. Combs are obviously non-astrophysical, so almost no vetting needs to be done, but the cause of these combs can remain elusive. 

Making the timing system LEDs stop blinking has reduced the amplitude of the 1 Hz comb with 0.5 Hz offset, the comb is not completely mitigated in O2 data. Further investigations are underway. Other 1 Hz combs also remain prevalent in the O2 data, probably GPS synced due to their narrowness. Coherence between the sites of these lines has not yet been established, but if GPS synced, then likely coherent (bad for any CW/Stochastic GW search).

Other lines that were found to have environmental cause in O1 and again appearing in O2 data are also in the version 1 line/comb list. Still other lines that are found to be correlated using the NoEMi tool on auxiliary channels are being vetted further using coherence tools. As further investigations proceed on other lines, we may have a version 2 of the lines/combs list.

Takeaway message: Mitigating the most prevalent of combs is going to have the biggest across-the-board impact on CW searches. These combs are listed in the summary section above.