I've produced new filters for use in the GDS calibration pipeline, primarily to get rid of spectral leakage seen, e.g., in this summary page.  The filters mentioned in LHO aLOG 45070 were inadequate, as the linked summary page shows.  The ASD plot below shows the performance of the new filters on data from the same day.  The new filters can be found in the calibration SVN at:

aligocalibration/trunk/Runs/PreER13/GDSFilters/H1GDS_1226769644.npz

They were produced from SVN revision 6052 and these parameters files:

• aligocalibration/trunk/Runs/PreER13/H1/params/modelparams_H1_2018-10-04.conf
• aligocalibration/trunk/Runs/PreER13/H1/params/H1_TDparams_1226769644.conf

The first parameters file above is based on aligocalibration/trunk/Runs/O3/H1/params/modelparams_H1_20181004.py.  I made this parameters file to be compatible with the old Matlab DARM model code, since the filter generation code in pyDARM is not yet ready.  Most of the measurement files were able to be read in, the exceptions being these files:

• aligocalibration/trunk/Common/pyDARM/H1aa.mat.  
• aligocalibration/trunk/Common/pyDARM/H1susdata.mat

These files were replaced by the ones used at the end of O2, referred to in aligocalibration/trunk/Runs/O2/H1/params/2017-01-24/modelparams_H1_2017-01-24.conf.

The primary differences between these filters and those of aLOG 45070 are:

• Longer inverse sensing filter (1.5 seconds vs 1.2 seconds).  The production DMT machines can handle this, but the testing machine cannot, so a similar 1.0-second filter is used on h1dmt3.
• Inverse sensing filter and actuation filters are split up into two filters of equal length: a highpass filter and a model filter (which includes a bit more high-passing).  Previously, we had convolved these two ro produce one, which mathematically should produce the same result.  However, since we are trying to attenuate low-frequency noise by many orders of magnitude with rather long filters (~10000 taps), double-precision digital errors add up, and this seems to have a greater impact on a single long filter versus two shorter filters.  This more closely matches what was done in G1800371.  (The recent changes are due in part to the port of all Matlab codes to python.)
• Both the model filter and the high-pass filter are windowed with a Kaiser window with Beta = 9.  We had previously used a Tukey window for the model filter and a Blackman window for the high-pass filter.

The first three plots attached show comparisons of the filters to the frequency-domain models.  The forth plot shows ASDs of GDS-CALIB_STRAIN and the front-end's CAL-DELTAL_EXTERNAL_DQ.  The last plot shows a transfer function comparing GDS-CALIB_STRAIN to the front-end's CAL-DELTAL_EXTERNAL_DQ.

The primary and redundant GDS calibration pipelines were restated around GPS 1226777593 to pick up these new filters.  They will likely be restarted again later today to pick up gstlal-calibration-1.2.5.