Stefan, Peter, Rich It was found that the overall DC gain, and the pole-zero location associated with the second filter module of the OMC DCPD Whitening chain, was somehow dependent on input signal level. A parametric study was done on the OMC DCPD whitening chain for different whitening filter combinations with and without DC offset. Notes detailing the parametric study are attached to this entry. There were two causes of these phenomena: 1. Gain Change - As the DC input to the OMC DCPD Whitening amplifier (D1002559 Chassis S1101627) is varied, the DC level causes saturation in some of the DC coupled gain stages. This saturation causes the normally high-impedance amplifier input to be fairly low due current flowing into the input protection diodes internal to each opamp stage. This lower input resistance in conjunction with the series resistance of the FET switches produces a voltage divider with a division ratio dependent on signal level. 2. Pole-Zero Change - The second filter module had an ECR (E1500252) performed that required a 1uF capacitor to be used. This type of filter function is supposed to be implemented with a plastic film dielectric capacitor, but a ceramic capacitor was used instead. The piezoelectric nature of ceramic capacitors can result in a capacitance that will change with applied voltage. This caused the pole-zero location to vary as a function of DC level. This capacitor was changed to the proper type and the problem went away. Data (attached) was taken for both OMC DCPD chains by injecting signals into the input of the OMC whitening chassis (DCPD Preamps disconnected), and taking the output differentially from the analog output of the whitening chassis with the ADC cable still attached (to account for any loading effects). This data was taken with and without a 5V DC offset to verify that there is no longer any effect on the signal path gain. We note that the above were all small effects, but significant in the world of 1% calibratin uncertainty. The DC gain reduction with a 5 V offset (similar to the offset in NLN operation) was 0.2 dB. The errors in the filter stage with offset were -- phase: 0.6-1 degree at 50 Hz; magnitude: 0.15 dB gain difference between 50 Hz and 200 Hz.
Rich verbally confirms that he and Peter used the measurement setup as described in LHO aLOG 47225 and D1900027, using the coil driver test box as a differential driver. They *did not* gather the test setup's transfer function, however, so this will need to be done in order to analyze the data (attached above, DCPD_Whitening_v2.zip) to the desired precision. Also note that details of the measurement files can be found on the last page of the DCPDWhiteningChanges.pdf attachment. Also -- I opened (and subsequently closed) FRS Ticket 12453 covering the problem of "changes in frequency response as a result of input signal level" that the solution in this aLOG fixes. Since this issue will likely impact LLO as well, I've opened up an integration issue, IIET Ticket 12454, such that the above mentioned change becomes an ECR to be implemented at LLO as well (assuming they have the same problem).
Updated the FM1,FM2,FM3 whitening compensation filters for PDA and PDB.
They are now matched at the +-0.15% level (see plot).
And we verified that they are no loner offset dependent.
Also, to be explicit: We disabled the 12dB, 6dB and 3dB whitening gain steps in both PDA and PDB. The 24dB step is still operational.
The new foton filters are:
PDA:
FM1: zpk([10.44],[0.996],0.9734,"n")
FM2: zpk([49.63],[497.5],-0.9995,"n")
FM3: zpk([10.372],[0.9865],1.002,"n")
PDB:
FM1: zpk([10.160],[0.969],0.975458,"n")
FM2: zpk([49.72],[497.7],-1.0004,"n")
FM3: zpk([10.467],[1.000],.9973,"n")
Using this filter, we injected pink noise into both DCPDs, and plotted the PD ratio in various configurations to verify the filter compensation. Specifically, plotted are:
A(FM1,FM2,FM3)/B(noFM)
A(FM1,FM2,FM3)/B(FM1,FM2)
A(FM1,FM2)/B(FM1,FM2)
A(FM1)/B(FM1)
A(FM1)/B(noFM)
A(noFM)/B(noFM)
A(FM1,FM2)/B(FM1)
B(FM1,FM2)/A(FM1) (opposite rat!)
The xml file with the data in the plot is in /ligo/home/controls/sballmer/20190303/DCPDfinal.xml
Finally, we still will have to re-match the photo diode light levels in lock (alog 47217).
All this was done to fix the problem identified in alog 47247.
Attached are plots of
- Transfer function wit- over-without offset. The insanity is indeed gone.
- PDA: FM1, FM1FM2, FM1FM2FM3 - all normalized with their compensation filter.
- PDB: FM1, FM1FM2, FM1FM2FM3 - all normalized with their compensation filter.
In short: the current compensation is good at the 0.02dB level. One could still improve it...
The plotting MATLAB code is here: /ligo/home/controls/sballmer/20190304/DCPD
I've moved the contents of Stefan's results from his home directory on the local file system into the calibration SVN (though I've not yet modified his plotting script for the new directories and/or to spit out the right file names).
I used the follow commands to do so:
Data:
cd /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/OMCWhiteningChassis/2019-03-04/
cp /ligo/home/controls/sballmer/20190304/DCPD/SRS00* ./
cp /ligo/home/controls/sballmer/20190304/DCPD/srs00* ./
cp /ligo/home/controls/sballmer/20190304/DCPD/47254_20190303213* ./
Script:
cd /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/
cp /ligo/home/controls/sballmer/20190304/DCPD/plotData.m plot_omcdcpdwhiteningmods_20190304.m
Results:
cd /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Results/OMCWhiteningChassis/
cp /ligo/home/controls/sballmer/20190304/DCPD/PD*.png ./
rename 's/PD/2019-03-04\_H1OMC\_WhiteningChassisTFs\_PD/' PD*.png
Thus the new script name to plot the data is:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/plot_omcdcpdwhiteningmods_20190304.m
which should be modified to analyze the following data:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/OMCWhiteningChassis/2019-03-04/
SRS00*.78D
srs00*.asc
using the notes in the file
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/OMCWhiteningChassis/2019-03-04/
47254_20190303213544_DCPDWhiteningChanges.pdf
and to produce plots and results with similar file tags as
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Results/OMCWhiteningChassis/
2019-03-04_H1OMC_WhiteningChassisTFs_PD*.png
(or the code should be re-written to use your favorite fitting code [maybe in python using IIRrational] and produce plots and results with similarly explicit file names with indications of the date, the interferometer, and WhiteningChassisTFs in the file name.)
Lilli S.,
First, I have renamed the data files in dir /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/OMCWhiteningChassis/2019-03-04/, so that they match what is described in note /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/OMCWhiteningChassis/2019-03-04/47254_20190303213544_DCPDWhiteningChanges.pdf
The test box has been factored out. (Test box data: /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Data/CoilDriverTestBox/S1900070/)
The fitting script is /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/fit_OMCDCPDWhiteningChassis_20190304.py
The results are in the table below. The fitting plots are in the attachments.
The low-freq poles/zeros roughly match Stefan's results. In O1 and O2 we used 1 stage of whitening (FM1), and the high freq poles were at [(14.34e3 + 14.60e3)/2, (18.68e3 + 18.57e3)/2, (98.94e3 + 100.45e3)/2] Hz. The new fitting for FM1 is consistent with O1 and O2.
Take PDA FM1FM2FM3 noDC as an example, manually build TF by removing the high-freq zeros, and plot the residual -> see diff.pdf
By ignoring the high-freq zeros, there's little impact on magnitude, but there's phase difference at higher freq (~<1 deg below 5k Hz).
| PDA | PDB | Comments | |
| FM1FM2FM3 (no DC) | PDA, Filters: FM1FM2FM3noDC Results: z:p:k [ 3.90426349e+06 3.90426349e+06 4.99892253e+02 1.08290122e+00 1.08040698e+00] [ 1.04233314e+01 3.26866242e+04 3.26866242e+04 1.13980014e+04 4.97078295e+01 1.04233314e+01] -8.68011786846 |
PDB, Filters: FM1FM2FM3noDC Results: z:p:k [ 4.64029327e+06 4.64029327e+06 5.00447159e+02 1.07523773e+00 1.07282309e+00] [ 1.03389750e+01 3.28642175e+04 3.28642175e+04 1.15193027e+04 1.03389750e+01 4.97939024e+01] -6.26981682358 |
Low-freq zeros near 1Hz, 1Hz, 500Hz, and poles near 10Hz, 10Hz, 50Hz (as expected) High-freq poles near 12K, 33K, 33K Hz. |
| FM1FM2FM3 (4.99V DC) | PDA, Filters: FM1FM2FM3DC Results: z:p:k [ 4.83659991e+06 4.83659991e+06 4.99682906e+02 1.08106936e+00 1.08299624e+00] [ 1.04231076e+01 3.28748478e+04 3.28748478e+04 1.13460627e+04 1.04231076e+01 4.96890494e+01] -5.6929144981 |
PDB, Filters: FM1FM2FM3DC Results: z:p:k [ 3.93271594e+06 3.93271594e+06 5.00449636e+02 1.07585358e+00 1.07328174e+00] [ 4.98000266e+01 3.28631506e+04 3.28631506e+04 1.15205406e+04 1.03375026e+01 1.03375026e+01] -8.72188264417 |
Low-freq zeros near 1Hz, 1Hz, 500Hz, and poles near 10Hz, 10Hz, 50Hz (as expected) High-freq poles near 12K, 33K, 33K Hz. |
| FM1FM2 (no DC) |
PDA, Filters: FM1FM2 Results: z:p:k |
PDB, Filters: FM1FM2 Results: z:p:k [ 2.11805525e+06 8.72055868e+05 5.00733633e+02 1.14755102e+00] [ 2.04506589e+04 1.24360067e+04 8.94998324e+04 4.97729997e+01 1.02056871e+01] -12.7328787001 |
Low-freq zeros near 1Hz, 500Hz, and poles near 10Hz, 50Hz (as expected) High-freq poles near 12K, 20K, 89K Hz. |
| FM1 (no DC) | PDA, Filters: FM1 Results: z:p:k [ 4.22938089e+06 4.22938089e+06 1.17507390e+00] [ 1.27384958e+04 1.93716076e+04 9.29113659e+04 1.04895576e+01] 13.3497796427 |
PDB, Filters: FM1 Results: z:p:k [ 4.84207436e+06 4.84207436e+06 1.14699930e+00] [ 1.29122349e+04 1.93696499e+04 9.46279852e+04 1.02067553e+01] 10.4974545401 |
Low-freq zeros near 1Hz and poles near 10Hz (as expected) High-freq poles near 13K, 19K, 93K Hz. |
| noFM (no DC) | PDA, Filters: noFM Results: z:p:k [ 1.26091781e+07 3.57048935e+07 2.05610666e-01] [ 1.28075383e+04 1.92624890e+04 1.84656469e+06 4.98134421e-02] 0.980629117874 |
PDB, Filters: noFM Results: z:p:k [ 2.82948542e+07 2.84161583e+07 2.05615042e-01] [ 1.29617754e+04 1.93273505e+04 2.08761329e+06 4.98133861e-02] 0.629314166113 |
High-freq poles near 13K, 19K Hz. Unphysical low-freq zero near 0.2Hz and pole near 0.05Hz. |
Concluding from the results of the fit above, I'll be updating the calibration model parameters to change from the previous measurement of the DCPD whitening chassis (before O2, when the nominal configuration was to only have the first stage of whitening -- "F1" or "FM1" -- on; see LHO aLOG 28087) from PDA PDB PDA PDB PDA PDB uncompOMCpoles_whitening = [(14.34e3 + 14.60e3)/2, (18.68e3 + 18.57e3)/2, (98.94e3 + 100.45e3)/2]; to the new fitted measurement results, and choosing the new nominal configuration, with all three filters on (the 1st and 3rd stages, which are whitening and the 2nd stage which is a low pass, i.e. "FM1FM2FM3" or "3 filt"), and using the 5 V_DC offset fit results because it's more representative of the nominal running conditions of the chassis (with the poles rounded to the nearest Hz) PDA PDB PDA PDB PDA PDB uncompOMCpoles_whitening = [(11.346e3 + 11.521e3)/2, (32.875e3 + 32.863e3)/2, (32.875e3 + 32.863e3)/2];
Lilli S.,
The residual contribution plot is attached. This was generated using /ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/fit_OMCDCPDWhiteningChassis_20190304-vsOriginal.py
(based on reference model modelparams_H1_20190118)
The results is averaged after using different poles/zeros in PDA and PDB, instead of using the same set of averaged poles/zeros in PDA and PDB.
Per Jeff's request that the original DCPD files be included, I am attaching the data extending down to 0.5Hz for the OMC DCPD Whitening chain as taken on the floor by signal injection into the front panel of the Split Whitening Chain via an SR785
Attached are two plots:
A) The comparison of TFs:
1. All low-freq z:p + high-freq p + out-of-band z from IIRrational fitting
2. Dead reckoned z:p = [500, 1, 1]:[50, 10, 10]
3. Low-freq z:p from Stefan's fitting in 47257
4. Low-freq z:p from Stefan's fitting in 47257 + high-freq p from IIRational
B) The comparison of contribution residuals, by using the latest H1 model 20190301
The scripts live in
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/fit_OMCDCPDWhiteningChassis_20190304-compareTF.py
/ligo/svncommon/CalSVN/aligocalibration/trunk/Common/Electronics/H1/Scripts/fit_OMCDCPDWhiteningChassis_20190304-compareRes.py