J. Kissel

Processing some of the data from LHO aLOG 51738, in which we compared 
    - the current "nominal" configuration of the ETMX actuator used for the DARM loop (and thus modeled for in the calibration of that loop)
    - a slightly modified configuration with the PUM L2A filters re-engaged, and a UIM boost engaged (see LHO aLOGs 51717, 51709, 51604, and LHO aLOG 49825),
we see the results are as expected / hoped for:
    (1) The PUM L2A filters do NOT impact the frequency response of the PUM L to TST L transfer functions for the ETMX actuator (like the used to in the former topology). That means there's no change in estimate of the PUM actuation strength used in the DARM loop calibration model, so there's no need to update that *in* the model, and we can just add the "nominal" and "new" data sets to the collection of data that refines the uncertainty on unknown systematic error.

    (2) The UIM boost -- as long as it's accounted for in the loop model -- is perfectly fine to use. Similar to the PUM results, it does not change the frequency response of the UIM L to TST L transfer function, nor the resulting estimate of UIM actuation strength. As with the PUM data sets, we can just add the UIM sets in to the collection of data used to quantify the uncertainty on unknown systematic error (which, we know we have, because we're not including the UIM high-frequency dynamics, see FRS Ticket 4691)

    (3) The PUM L2A filters do NOT obviously appear to impact the optical response measurement beyond is current time-dependence -- whatever is happening at low-frequency in the response is still there and time-dependent. This confusing low frequency response has been recently, severely reduced in magnitude, phase, and time-dependence after implementing the (unrelated) digitally requested 100 ct SRCL offset, but there is still some effect left (see LHO aLOG 51592). So, even though one can see that the new config's data has the largest magnitude deviation, and "crispiest" phase wiggle -- I wouldn't yet blame any difference "for the worse" on the new PUM L2A. Further, it's still well below the systematic error we've had in the past without the digital SRCL offset, which we've already agreed we can handle by making sure there's a large uncertainty in unknown systematic error (see G1901479).

This means that we can proceed with creating a newly updated calibration model as planned.

To demonstrate (1), see the first attachment 
2019-09-04_H1_L2ADecoupUIMBoost_PUM_actuationMeasurement_allresiduals_MCMCInput.pdf.
    Here, I show the residuals between the future 2019-09-09 DARM loop calibration model's estimate of the actuation strength (with all known frequency dependence, like the 1/f^4 pendulum response), and 4 data sets:
        (1) the 2019-03-27 measurement in which we still had the former, incorrect topology for the PUM L2A decoupling which left an unaccounted for parasitic coupling between the A2L path because our spot position gains were too large (see LHO aLOG 47941 for measurement and G1901481 for postmortem model)
        (2) An example data set from O3, in which we've just turned OFF the PUM L2A filters, and thus removing the problem
        (3) The most recent 2019-09-04 data, still in this nominal configuration with no PUM L2A filters on (and no UIM boost)
        (4) The 2019-09-04 data *with* the PUM L2A filters ON under the new topology (and also with the UIM boost on).
Note, this "future 2019-09-09" model has exactly the same values for the actuation strengths as the 2019-04-16 model, which we've been using throughout O3.

I demonstrate (2) [and a bonus demo that there's no affect on the TST L to TST L transfer function either] in the second [and fourth] attachments,
2019-09-04_H1_L2ADecoupUIMBoost_UIM_actuationMeasurement_allresiduals_MCMCInput.pdf
2019-09-04_H1_L2ADecoupUIMBoost_TST_actuationMeasurement_allresiduals_MCMCInput.pdf
    It's the same collection of measurement dates, just the different stages of ETMX's transfer function. Same conclusions here -- all data sets look identical, save for a bit of expected time dependence in the the TST stage, for which we account in other ways.

I demonstrate (3) in the third attachment,
2019-09-04_H1_20190909Model_sensingFunction_referenceModel_vs_allMeasurements.pdf
 by comparing the last few weeks worth of sensing function measurements that have all been in the same configuration (important for the sensing response, in the "July" spot positions and with a 100 ct digital SRCL offset) -- except for the last measurement (in red) in which the PUM L2A decoupling and the UIM boost are ON.

Plots produced by the following:
Model parameter set: 
    ^/trunk/Runs/O3/H1/params/modelparams_H1_20190909.py rev 8365
pyDARM source: 
    ^/trunk/Common/pyDARM/src/sensing.py rev 8379
    ^/trunk/Common/pyDARM/src/actuation.py rev 8379
scripts: 
    ^/trunk/Runs/O3/H1/Scripts/FullIFOActuationTFs/process_actuationmeas_collection_20190904_L2ADecoupling.py
    ^/trunk/Runs/O3/H1/Scripts/FullIFOSensingTFs/process_sensingmeas_collection_20190904_with20190909model.py