It seems like a lot of thigs are going on in this small frequency interval just after the Ncal lines. To better understand the structures at around 42 Hz, we made several high resolution ASDs across January and February. The attached GIF image and the associated (static) PDF file describe the evolution of the line and bump structures over this period. This is a summary with some observations:
Jan 15: no line at 42 Hz but some structures are already visible at 42.4 Hz, as well as a small peak at 42.4 Hz. What I initially thought were sidebands of the 42 Hz line (ref. #63340) seem in fact to be not.
Jan 17: Another small peak at 41.65 Hz.
Jan 18: Increased noise baseline and new structure between 41.7 and 41.9 Hz.
Jan 19: The structure that just appeared seems to have moved to around 42 Hz. This structure appears to drift until Jan 22, as visible in this spectrogram.
Jan 20:: Sharper peak at 42 Hz.
Jan 21: New peak at 41.8 Hz, and structure centered at 42.1 Hz.
Jan 22 3:00: Peak at 42 Hz. This seems to come before the installation of the new chiller happened on Jan 23: #63062. The previous structure has now drifted to 42.2 Hz, and is different than the 42.4 Hz structure already documented.
Jan 22 11:40: the peak at 42 Hz is not visible but the structure at 42.2 Hz is now larger.
Jan 25: The 42 Hz peak is clearly visible. The structure at 42.2 Hz is either disappeared or merged with that at 42.4 Hz. The noise baseline is low.
Jan 26: Increased noise baseline.
Jan 27: The peak at 41.8 Hz is back...
Jan 28: ...and gone.
Feb 6: Increased peak at 41.6 Hz. The structure at 42.4 Hz seems wider.
Feb 9 and 11:New structure at 42.6? Is that part of what constituted the 42.4 Hz bump?
Feb 16: Dtructures at 41.4, 41.6, 42.4 and 42.6 Hz, in addition to the peak slightly before 42 Hz.
Feb 20: 42 Hz line disappeared after the chiller replacement: #63338. Structures at 41.4 Hz, 41.6 Hz, 42.4 Hz and slightly before 42.6 Hz are still visible but are likely not related to the 42 Hz line: they are no sidebands.
Some high resolution BruCo analysis should be done to understand the origin of the residual structures. The argument --medianpsdestimation
, new to version 3.2, should provide handy to avoid the spectral estimations being spoiled by the 25-minute glitches. Additionally, this analysis proves that we need a tool that automatically performs periodic comparisons between spectra.