As noted in LHO aLOG 82394, I recommend starting improvements by adding an additional 16k downsampling filter. Attached is a figure showing a comparison of the current situation (red), just 2 16k filters (green), 2 16k filters and 1 65k filter (magenta), and 2 16k filters and 2 65k filters (black). The curve colors are the same as the original FOTON plots in LHO aLOG 82375. These curves are ratios of PSDs using different downsampling filters on the time series data that is manually downsampled by a factor of 32 (524k ==> 16k) versus PSD of the original 524k data (cut off with the maximum frequency of 8192 Hz).

The narrow instrumental artifacts are visible by eye in the 8 s FFTs of the DTT measurements, and are much more visible in the CW analyses that use longer coherence times. We can quantitatively assess the improvement by comparing the number of bins in this short analysis that are 1% in excess of the nominal ratio of 1:

Full band:
16k 65k bins above 1% = 14847
16k 16k bins above 1% = 6686
16k 16k 65k bins above 1% = 4892
16k 16k 65k 65k bins above 1% = 4809

0 - 2 kHz band:
16k 65k bins above 1% = 430
16k 16k bins above 1% = 21
16k 16k 65k bins above 1% = 0
16k 16k 65k 65k bins above 1% = 0

Note that even though the 2 kHz band has only a few hundred bins above 1%, these artifacts are strongly visible in run-averaged spectra that use longer coherent baselines, similar to all-sky CW searches.

One can see that the additional 16k filter does most of the clean up (we don't see a significant change by an extra 65k filter). By employing an additional filter, this should hopefully reduce or eliminate the impact of the narrow instrumental artifacts.