Lisa, Kiwamu,
We think that the calibration has been wrong and therefore it lifted up the noise floor below 300 Hz. We need another set of eyes tomorrow to doublecheck our theory (we are currently too sleepy to do systematic investigation).
In short, we believe that the sign of ETMY L3 stage in CALCS (or somewhere else similar to it) is wrong which is exactly the same situation as what happened in this past July (28396).
Because I knew that our DARM open loop model is accurate and consistent with the recent measurement (29748), I made a calibration filter for DARM_IN1 which converts DARM_IN1 to DARM displacement as opposed to the use of both DARM_IN1 and DARM_OUT. Here is a comparison of the CALCS spectrum against the spectrum derived from DARM_IN1.
As shown in the plot, the CAL CS spectrum overestimates the noise floor below 300 Hz or so. This is exactly the same behavior as what we have experienced in this past July. In addition, we have noticed that the noise floor of CALCS changed as a function of the DARM control gain which should not happen in the calibration scheme used in CAL CS. Also, when we flipped the sign of the EY L3 stage in CAL CS, the leve of the noise floor became identical to the one derived by DARM_IN1 and also became insensitive to the control gain. This increased our confidence that the L3 sign was wrong in CALCS.
We are leaving the L3 stage gain flipped (DRIVEALIGN_GAIN -30 --> +30) for the night. If our theory is correct, we regain the binary range back to ~60 Mpc with this change.
Also, we took a DARM open loop and PCAL sweep measurement within the same lock stretch. We did not analyze them yet, but they are available at:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Measurements/PCAL/2016-09-21_H1_PCAL2DARMTF_4to1200Hz.xml
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Measurements/DARMOLGTFs/2016-09-21_H1_DARM_OLGTF_4to1200Hz.xml
Also, my code which generated the calibration filter for DARM_IN1, as well as, the generated filter are available at:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Scripts/ControlRoomCalib/CalibrateH1DARM_IN1.m
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Scripts/ControlRoomCalib/DARM_IN1_calib.txt
I don't know what was decided this morning, but somehow we ended up with the wrong actuator sign in the front-end calibration again. So I reinstated the sign flip in the drivealign calibration filter.
After we fixed the sign flip on the night, the Pcal to CAL-CS transfer function looked indeed correct. See the attatched screenshot below.
The transfer function from Pcal to CAL-CS is almost unity in magnitude and zero in phase which double-confirms that the sign flip was the right action. The dtt template was updated with the latest DELTAL_EXTERNAL whitening filter, known delays.
The dtt file can be found at:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Measurements/PCAL/2016-09-21_H1_PCAL2DARMTF_4to1200Hz.xml
The matlab script which produced the calibration filter can be found at:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Scripts/ControlRoomCalib/H1_pcal2darm_correction.m
Finally the calibration filter in ascii formt is available at:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Scripts/ControlRoomCalib/pcal2darm_calib.txt
By the way, in the script, I needed to add an additional minus sign in order to get the resulting phase close to zero rather than close to 180 deg. I feel like this is one of theose things we had found during O1, but I can't recall what introduced a minus sign. In addition, the calibration filter currently does not include the effect of the supernyquist poles of OMC DCPDs which introduces a phase delay at high frequencies.
The open loop transfer function(s) can be now analyzed by a copy of Evan G's code,
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/PreER10/H1/Scripts/DARMOLGTFs/compareDARMOLGTFs.m
The attached are the resulting figures from the code. I have not adjusted kappas or examined each piece of the DARM loop yet, but the model showed a good agreement with the measurement-- no crazy bug so far in the model.