Keita and Daniel pointed out that one way to check if all of our range loss is due to subtractable jitter noise or something else, is to compare cleaned data from before the earthquake to after.
In the attached plot, the blue trace labeled "Orig" is the raw data from the 15th of July, before we (accidentally) switched to using only one DCPD. The red-orange trace labeled "CleanPostEQ" is from that same 15 July time, but has been cleaned. The yellow trace labeled "CleanPreEQ" is cleaned data from the 5th of July, just before the big Montana earthquake hit us.
The ratios in the bottom portion of the plot are the 2 cleaned spectra compared to the raw post-EQ blue spectrum. Because of this, the % improvements in the title of the plot are misleading.
The conclusion is that we definitely have some noise below 90Hz or so that is not witnessed by the IMC WFS, the PSL bullseye, the ASC Hard loops, or the vertex length loops.
I attach here another set of spectra, where in each figure the blue is cleaned data from before the earthquake (6 July 2017 04:00:00 UTC), the red is cleaned data from after the earthquake (21 Juy 2017 14:00:00 UTC), and yellow is the difference between those 2 spectra. These traces are the same on all 3 figures.
The other two traces, purple and green, are lines overlaid to help decide what kind of slope our residual is. The figure with "f2" in the name has 1/f^2 lines, "f3" in the name has 1/f^3 lines, and "f4" in the name has 1/f^4 lines.
I'm not sure what would give us 1/f^3 noise, but it looks to me like that might be the best fit. It's definitely not 1/f^4.
