Displaying report 1-1 of 1.
Reports until 19:24, Monday 19 April 2021
H1 CAL (CAL, ISC)
jeffrey.kissel@LIGO.ORG - posted 19:24, Monday 19 April 2021 - last comment - 17:01, Friday 09 September 2022(58656)
New 2021-04-17 DARM Model Parameter File for new Improved Sensing Function
J. Kissel

This is a follow up on updating the sensing function part of calibration more carefully and completely than last Friday's "you could just" suggestion in LHO aLOG 58629.

Again, the exciting new results that among other things replacing ITMY has improved the DARM loop's sensing function by
    - increasing the cavity pole frequency,
    - increasing the optical gain
    - little sign of SRC detuning (though this statement is not strongly supported by data yet, just hopeful)

I'm starting to run through the T1800469 checklist of things to do in order to update the calibration accordingly. Thus far, I'm only considering an update to the sensing function. I've not processed, nor do I plan to update just yet, any stages of the actuation function.

I've created a new (py)DARM model parameter file:
    ^/trunk/Runs/O3/H1/params/modelparams_H1_20210417.py

It's got the following changes and why:
    (1) New ITM = higher optical gain and cavity pole frequency, and new foton export file
    (2) No evidence yet for detuning = zeroed out the detuned spring frequency
    (3) The power normalization is different for DC readout = New DARM_ERR/OMC_DCPD_SUM transfer coefficient
    (4) higher cavity pole = new file for export of foton representation of inverse sensing function
    (5) A long careful study of the OMC whitening chassis frequency response = new high frequency pole frequencies for OMC whitening chassis

Other than the following "small" levels of sensing function or actuation function systematic error, 
    (a) potentially new actuator strengths (they've not been updated since 2019-04-03)
    (b) that we *want* ETMX PUM coil drivers to be in state 3, but we are oscillating between being able to use State 1, and being stuck in State 2
    (c) we're still having trouble with violin modes, so we're using the "1st stage whitening and 2nd stage lowpass ON" state of the OMC whitening chassis, instead of either of the two configurations we used of that chassis in O3.
I don't know of any other changes (say, e.g. to the control filters). This model should be much better that the late-O3 model we have now, with only the above 5 changes.

So, after creating the model, I 
Re-ran the MCMC with 
    ^/trunk/Runs/O3/H1/Scripts/FullIFOSensingTFs/process_sensingmeas_MCMC_model20210417_meas20210417.py

to get 
    Optical gain, H_c (ct/m)                 | 3.473e+06 (+3598,-3641) or (+0.1036%,-0.1049%)
    Cavity pole, f_cc (Hz)                   | 450.8 (+1.526,-1.502) or (+0.3386%,-0.3333%)
    Detuned SRC spring frequency, f_s (Hz)   | 4.277 (+0.4536,-0.5162) or (+10.61%,-12.07%)
    Detuned SRC spring quality factor, Q_s   | 31.47 (+74.72,-86.12) or (+42.11%,-36.54%)
    Residual time delay, tau_c (usec)        | 7.98 (+0.7229,-0.7048) or (+9.059%,-8.831%)
and self-consistent posteriors of uncertainty, 
    ^/trunk/Runs/O3/H1/Results/Uncertainty/
    2021-04-19_H1_20210417Model_fmin30Hz_PCALSYSERR_None_meas2021-04-17_sensingFunction_MCMC_posteriors.hdf5
in case we need them to create a 68% CI on the response function systematic error (doubt it, but, just in case).

Updated and loaded the following in to the CS-DARM_ERR filter bank:
    FM4: "NewITMY_NoD2N": zpk([450.8],[7000],1,"n")
    FM5: "NewITMY_Gain": gain(2.879e-7)                        == 1/3.473e6
(and copied the gain over to the CS-DARM_CFTD_ERR bank, in case we have time to ever resurrect that.)
and loaded coefficients -- so these are now available for use in FM4 and FM5.

The, though I did not *write* them, I used
    ^/trunk/Runs/O3/H1/Scripts/CALCS_FE/createEPICS_for_20210417.py
to create the file for the "EPICs records" (the realized values of various transfer functions and TF ratios from the DARM model at calibration line frequencies)
    ^/trunk/Runs/O3/H1/Results/CALCS_FE/epicsrecords_model-H1_20210417_created-20210419.txt

Also, I checked if, with this model update, whether we still have the right the relative delay between the front-end sensing and actuation path. Did so with the script
    ^/trunk/Runs/O3/H1/Scripts/CALCS_FE/compute_relativedelay_AvsC_20210417.py
    
which indicates that the delay at the DARM UGF is -7.589 clock cycles. But -- remember -- the front end delay block doesn't change between 7 and 8 (see LHO aLOG 48620). So, maybe it's time that we figure out why Keita's thiran filter design (see LHO aLOG 48008) didn't work, and/or at least fix the bug in the RING_BUFFER.c code that limits the delay cycles to 7.

And finally, for the eventual (if we do) update GDS, I exported the "NewITMY_NoD2N" filter
    ^/trunk/Runs/O3/H1/Measurements/Foton/2021-04-17_H1CALCS_InverseSensingFunction_Foton_NewITMY_NoD2N_tf.txt

So -- now I think we're ready for an update to the calibration.
To do so, I'll need to:
    - Have a fully functional, running IFO, in at least a good a noise state (and in the same switchable filter configurations) it was on Friday 4/17
    - Measure a broadband PCAL to DARM TF for my "before change," then
    - Switch over the CS-DARM_ERR filters from having only FM2&3 to having only FM4&5 on,
    - use the "pushtoepics" input on createEPICS_for_20210417.py to push the update the EPICs records in the front-end
    - Measure a broadband PCAL to DARM TF for my "after change"
    - Look at sensing function TDCFs to make sure they make sense.
and if folks are feeling ambitious (or we need to use GDS-CALIB_STRAIN instead of DELTAL_EXTERNAL), then we can 
Non-image files attached to this report
Comments related to this report
craig.cahillane@LIGO.ORG - 00:50, Tuesday 20 April 2021 (58658)ISC
Our new 451 Hz cavity pole indicates our SRC losses are far smaller than before, around a factor of 4 better.

In the plot is shown the SRC losses vs DARM pole according to Buonanno and Chen model including losses.

Our new "design", or no loss, DARM pole is 463 Hz with the new ITMY:
DARM pole = FSR/(2*pi) * log((1 - ri * rs)/(re * ri - re * rs ))
where 
ri = sqrt(1 - Ti), re = sqrt(1 - Te), and rs = sqrt(1 - Ts),
Ti = 1.5%, Te = 4 ppm, Ts = 32.4%.

Buonanno and Chen agrees with the design DARM pole for no SRC loss.
Increasing the SRC loss pulls the DARM pole down, reducing our bandwidth.

------------------------------------------------------------------------------------------------------------------
During O3, Hanford's ITMY was 1.42% reflective, while ITMX was 1.5% reflective.  
Our design DARM pole was 457 Hz, but we only achieved 411 Hz, indicative of 4.2% loss.

------------------------------------------------------------------------------------------------------------------
Hanford's DARM pole is now comparable with Livingston's.  
Livingston's ITMs are 1.48% reflective, so the model is slightly different.  Their ideal DARM pole is 456.5 Hz.
And our estimated losses are now comparable as well: Hanford 1.0% losses, Livingston's 0.6%

------------------------------------------------------------------------------------------------------------------
This is further proof that our corner is greatly cleaned up due to the ITMY replacement.
The impact of the ITMY point absorbers was extreme.
LHO's DARM pole has been lower than Livingston's since O1,
and we've long struggled with unintentional SRC detuning and strong thermal transients which rendered us unable to attain high power.


More work remains to verify the improved status of LHO due to the ITMY replacement.
Our unintentional SRC detuning problems and the "disappearing" 9 MHz power are potentially solved, but must be checked.
Others more expert in the unintentional frequency dependent squeezing will have to weigh in, and we won't know until the filter cavity is installed, but its possible that may have been mitigated as well.

------------------------------------------------------------------------------------------------------------------
The code that produced this plot is darm_pole_src_loss.py.
It numerically finds Buonanno and Chen's DARM pole given some SRC loss.
Non-image files attached to this comment
evan.goetz@LIGO.ORG - 11:19, Wednesday 21 April 2021 (58690)
The new pyDARM code is still under active development, but I wanted to report that the MCMC code is giving consistent results as the old version. Note that the old SVN pyDARM uses DTT ASCII exports and the new pyDARM relies on a python module dttxml to obtain transfer function values. There have been observed slight differences in the numerical ASCII values exported and dttxml numerical values. This may lead to slight differences in the MAP values, as well as statistical fluctuations from the MCMC method itself.
    Parameter                                | O3 SVN pyDARM MAP Value (uncertainty)           | git pyDARM for O4+ MAP Value
    Optical gain, H_c (ct/m)                 | 3.473e+06 (+3598,-3641) or (+0.1036%,-0.1049%)  | 3.473e+06
    Cavity pole, f_cc (Hz)                   | 450.8 (+1.526,-1.502) or (+0.3386%,-0.3333%)    | 451.5
    Detuned SRC spring frequency, f_s (Hz)   | 4.277 (+0.4536,-0.5162) or (+10.61%,-12.07%)    | 4.358
    Detuned SRC spring quality factor, Q_s   | 31.47 (+74.72,-86.12) or (+42.11%,-36.54%)      | 36.65 
    Residual time delay, tau_c (usec)        | 7.98 (+0.7229,-0.7048) or (+9.059%,-8.831%)     | 6.74
craig.cahillane@LIGO.ORG - 17:01, Friday 09 September 2022 (64936)
I mention for my alog above that I neglected the length of the SRC in that plot.
Usually, the length of the SRC is neglected for DARM models, like in Kiwamu's equation 12 for the design DARM pole, which I plot above for the lossless case.
But for the precision required for estimating SRC loss, the length of the SRC is non-negligible.  

Also, the way I estimated the DARM pole here may be a bit unusual compared to the usual single-pole fit method of estimation.
See alog 64934 for more on that systematic.
Displaying report 1-1 of 1.