I would suggest checking the IOP SDF files to make sure this is OFF by default.  I am pretty sure that is true at LLO, but I am not positive

Jamie, Anamaria

We ran into a problem when I had commented out a mode and then once I hit load on the violin guardian, it went red. Strangely I had commented out an ITMX mode and it was complaining about an ETMY mode, but we never figured that out.
The way the guardian is written, it gets its mode info in the main, not in the run. So the state has to be rerun to get the new mode numbers.

The settings (gain, FMs, etc) however are reread in the run state, so it's ok to change those and hit load.

The solution is to set the guardian to manual, rerun the current state (so main updates its mode numbers) and then set it back to auto.

Overall calibration sign is incorrect for H1:CAL-DELTAL_EXTERNAL_DQ.

In the attached, left half shows the Pcal_X_RX_PD_OUT_DQ to DELTAL_EXTERNAL_DQ transfer function at three PCALY calline frequencies. DELTAL_EXTERNAL_DQ is calibrated in meters using H1DARMFOM dtt template, but I removed two 1Hz poles from PCAL_X_RX_PD so that basically it becomes the scaled power (positive means more power).

See how the phase of the TF is close to -180 deg at all three callines. This means that the overall sign of DELTAL channel is wrong. During O2, the same measurement showed tha both L1 and H1 were at ~0deg, which is the indication of correct sign (LLO alog 35346). This measurement is quite conclusive and that's the reason why we used this for LIGO and VIRGO calibration sign review in O2.

Incorrect sign is not because Craig did something, it has been like that for quite some time.

Let me explain the idea behind the measurement.

Let the Pcal power be P and the actuation function of Pcal be A such that the change in the Y arm length is

dY=AP.

LIGO-VIRGO sign convention is

dL=dX-dY

where positive dX or dY means longer X or Y arm.

The measured transfer function is

dL/P = (-dY)/P = -AP/P = -A.

Since Pcal pushes EY toward ERMY, A is positive at DC (i.e more power makes Y arm longer), but at frequencies much larger than the highest pendular resonance the test mass response is that of a free mass, so A is negative at e.g. ~36Hz and ~312Hz.

Therefore, if the sign of calibration is correct, power to DELTAL transfer function  -A is positive, i.e. the phase is zero, not -180 deg.

Now, PCAL TX (transmitter) and RX (receiver) power channels are calibrated in a funny unit where the complex suspension response is embedded in the RX channel calibration itself (but only in part) as a calibration filter. That filter is shown in the foton (attached rightmost). Important thing is that DC phase as well as the phase for f>30Hz or so are both basically zero degree (phase=-3.2deg at 36Hz). In this sense, for f>30Hz, RX_PD_OUT is just a scaled version of the power. We know that the channel goes positive when there is light on the PD, goes zero when no light, so positive signal means more power.

RX_PD_OUT(f<0.01Hz or f>15Hz) = C*P

dL/RX_PD_OUT = dL/P/C where C is a positive number.

The measured quantity is just dL/P divided by a positive number.

Therefore, if the sign of calibration is correct, phase of RX_PD_OUT to DELTAL transfer function is zero, not -180 deg.

I could in principle do the same analysis using the channel without any calibration, e.g. CAL-PCALY_RX_PD_ADC_IN but this is not DQ channel so I cannot look back, which is inconvenient.

We're trying to get help from JeffK (who is not at the site) and JoeB to learn how to correct the sign in a calibration-mode-friendly way.

Great work, team!

We also measured the relative sign of EX L1 and L2 using ~16Hz callines on Wednesday.

L1 cal line is at 15.1Hz and L2 cal line at 16.7Hz. Since they're right next to each other, it's really easy to make a relative phase comparison of this path. By comparing this measurement with the calibration model we can establish if the relative sign of L1 and L2 actuation model is correct. Our conclusion was that it used to be correct, got incorrect when Craig flipped the L1 sign. That's one of the reasons why we decided to change the L1 sign back.

Let the actuation function from H1:SUS-ETMX_L1_DRIVEALIGN_L_IN to the displacement of the mass be A1, and from L2 to the displacement be A2.

15.1Hz is close enough to 16.7Hz, so DARM OLTF G at 15.1Hz is almost the same as at 16.7Hz. The same thing could be said for the sensing function C too. Because of this, the callines will show up as an error signal as

error=A1*L1calline*C/(1+G)+A2*L2calline*C/(1+G)=(A1*L1calline+A2*L2calline)*C/(1+G).

Transfer functions from L1 and L2 calline to the error signal are

TF1=error/L1calline = A1*C/(1+G)

TF2=error/L2calline = A2*C/(1+G).

If we make the ratio of the two, we'll get the ratio of the actuation function regardless of the sensing and OLTF.

TF1/TF2=A1/A2.

Getting back to the actual measurement (attached left), phase of the Measured transfer function from L1 to DCPD-SUM at 15.1Hz was 58.9 deg, -173.6deg for L2.

phase(TF1/TF2)=phase(A1/A2)=58.9-(-173.6)=232.5deg (measured).

You can use any signal in the sensing path as an error signal, in this measurement I'm using DCPD_SUM. I'm using H1:SUS-ETMX_L1_CAL_LINE_OUT_DQ etc. and L1calline and L2calline, which are directly added to H1:SUS-ETMX_L1_DRIVEALIGN_L_IN etc.

OTOH if you look at the calibration model actuation path, A1 and A2 are the product of three filter modules (DRIVEALIGN, COILOUTF and another one called ANALOG that represents the suspension response in analog world), output matrix that mixes  L1 and L2 ANALOG output, and some digital gains. All of these are shown in the 2nd attachment (all gains are positive in this screenshot which represent the status now, but when Craig "flipped the L1 gain" H1:CAL-CS_DARM_ANALOG_ETMX_L1_GAIN was set to -1 rather than 1).

As shown in foton and filter screen shot, ETMX L2 drivealign L2L is -40.7deg at 16Hz, H1:CAL-CS_DARM_FE_ETMX_L2_COILOUTF is just a pass through, and ANALOG_ETMX_L2 was 0 deg, so the total is -40.7deg.

L1 drivealign is a pass through, coiloutf is a pass through, and ANALOG_ETMX_L1 is about 179.1 deg at 16Hz, so the total is 179.1deg.

phase(TF1)=179.1deg

phase(TF2)=-40.7deg

phase(TF1/TF2)=219.8 deg (model).

Comparing the model and measurement, it seems to agree well, which means that the sign of L1 path relative to L2 is correct now.

(But it was not the case when H1:CAL-CS_DARM_ANALOG_ETMX_L1_GAIN was set to -1 because phase(TF1) in the model was -0.9deg due to additional 180 degrees, so the model didn't make sense.)

Here is a comparison of the ASD, TF, and uncalibrated time series between GDS CALIB STRAIN and CAL DELTAL EXTERNAL.

The calibration applied to GDS CALIB STRAIN was just a multiplication by L = 3994.5 m.  The calibration applied to CAL DELTAL EXTERNAL is attached as .txt: this is what is in the DARM FOM currently.  These calibrations are applied to both the ASD and TF plot.  
There appears to be around -170 degrees between GDS CALIB STRAIN / CAL DELTAL EXTERNAL DQ.  

Oh sorry I attached slightly wrong file for sign-flip measurement. This is what I wanted to show.

Test of DTT calibration for DELTAL (suspicious).

Since there was still some possibility that the DTT calibration was wrong, I asked JoeB to do the same measurement using LLO template but somehow he couldn't reliably pull the H1 data when he tried. I did it on my own following JoeB's advice.

I removed the DTT calibration from LHO DARM FOM template, and put 6 pairs of [z,p]=[30,0.3] as dewhitening. This should be good enough. For Pcal I didn't change anything from my previous measurement (i.e. remove DTT calibration which was two poles at 1Hz, and just used the gain of 1 as the calibration).

The result shows that the sign was actually correct before we put the overall sign flip in (first attachment), and got incorrect after the overall sign flip (second).

The DTT templates were saved as

/ligo/home/keita.kawabe/O3CAL/EX-L1_L2_L3_pcal_DELTAL_sign/PCAL_DARM_sign_dewhiteOnly_20190313235002.xml

/ligo/home/keita.kawabe/O3CAL/EX-L1_L2_L3_pcal_DELTAL_sign/PCAL_DARM_sign_dewhiteOnly_20190319103130.xml

I have no idea how DTT template calibration is officially generated, so this could still be an intended behavior. Investigation continues.

Nutsinee, Sheila

We had another look at the transimpedance and loss from these single sideband measurements.

To summarize the measurement:

In single bounce off ITMY we measured 13mA of carrier  (Ic) through the OMC after running both AS and OMC alignment loops and offloading them.  With the OPO temperature detuned so that we have only a single sideband of CLF, we adjusted ZM1/2 to align the squeezer to the OMC, turned up the OMC whitening to 24dB, and measured 1.21uA of CLF power.  We then locked the OMC on the carrier, injected the single sideband at 3MHz from the squeezer, and measured -18.6dBm of 3MHz on the demod . 

Finding Transimpedance:

• Using an OMC Finesse of 390 and FSR of 261MHz, the transmission of the OMC for the 00 mode at 3.125 MHz is 1/(1+(2*Fin*sin(pi*3e6/FSR)/pi)^2) = 1.1% of the transmission on resonance (the factor of 10 above was an error).  So the peak photo current we expect at 3 MHz is sqrt(Ic*Iclf*0.011) = 13.4uA. 
• After the OMC transimpedance amp, there is a split off chassis that amplifies the 3MHz signal from the DCPDs, D1700376.  The test report that we have found is E1700363, which reports a gain of 19.7dB for 3 MHz, so the -18.6dBm measured at the demod is 3.8mV peak of 3MHz signal out of the transimpedance amp.  
• The transimpedance at 3MHz is then 287 Ohms.  
•  Koji's measurement is roughly consistent with this. 

Estimating loss:

• With the OPO temperature detuned so we have no nonlinear gain and only a single 3 MHz sideband, Nutsinee measured 0.757uW of 3MHz on SQZT6.  With the interferometer locked, Nutsinee measured at the demod 117.6mV pp with the scope in high impedance mode, which means the peak voltage out of the transimpedance amp was 3 uV once we take into account the 19.7dB gain of the 3MHz pick off chassis.  Using the transimpedance from the first measurement, we have a peak photo current at 3 MHz of 10.6uA.
• Using the DC photo current of 20mA, and the photodiode responsivity of 0.858A/W, this implies that there was 7.1 nW of 3MHz single sideband reaching the OMC DCPDs during the measurement.  
• Using the 1.1% transmission of the OMC for the 00 mode at 3MHz, we get an efficiency of 76.5% for the propagation of the 3MHz sideband from the beam diverter to the 00 mode arriving at the OMC.  

Loss budget:

This 76.5% efficiency sounds similar to the efficiency calculated above, but that is because two errors are roughly canceling (overestimating the transmission of the OMC at 3MHz and leaving out the gain of the 3MHz pick off chassis).  

Looking into the SR3 servo some more, the oplev and pitch osems were not agreeing last night when the heater was on. The OSEM says that pitch starts coming back down after some time, whereas the OPLEV says it keeps sagging. Could it be the OSEMS getting hot from the heater?

Anyway, the cage servo has been swapped to use the SUS-SR3_M3_OPLEV_PIT_OUTPUT instead as we feel that's a more reliable witness. The setpoint is 2.2 and the gain 1e-3.

Jenne, Sheila, Jeff B, Ed

People were having a little trouble with the SRC alignment step of initial alignment, so Jenne and I took a look at the cage servo.  It looks like it was not fast enough to keep up with the thermal changes caused by the disk heater last night.  We increased the gain by a factor of 50, based on the alignment change (seen by the oplev) during the temperature changes last night.  This is able to move the optical lever 1.5 urad in ~15 minutes, this should be enough to keep up with the thermal changes.  

That is an unbelievably cool tool.  Just beautiful.

Couple of additional notes:

• I've added a script that will synchronize the measured couplings from the legacy location in simple-noise-budget into the new aligoNB location:

Be sure to git commit and push the updated couplings after synching.

• Users should never need to override the update() method for our usage here.  The included plotter handles fetching the NDS data running the base update method to make the data available to the calc() method in the self.nds_data attribute.
• Please report issues to the issue tracker:
  • https://git.ligo.org/NoiseBudget/aligoNB/issues
• The current online plotter puts a bit of a heavy load on the NDS server, so we probably don't want to run too many of them simultaneously.  I'm working on a more efficient version that will be available soon.

The --online plotting option has been disabled until further notice.

The prototype online plotting method was implemented in a crude and inefficient way, and seemed to be triggering instability in the NDS1 server.  I've disabled it until I can implement a more efficient method.

Just added a new feature to BruCo:

if the --range=xxx option is added, then BruCo multiplies the PSD of the target signal by the coefficient xxx (0.00025=1/4000 to convert CAL-DELTAL_EXTERNAL to strain) an compute the BNS range, for the target signal and the estimated coherence subtraction. So one can get a ballpark idea of how much that noise contributes to the range.

For example, removing the coherence with PRCL would bring the range from 100.23 to 102.95 Mpc (the absolute calibration might be a bit off since I'm using a old calibration curve for CAL-DELTAL_EXTERNAL)

This is really fantastic!

Additional clue: there are small but noticeable changes in comb strength in a couple magnetometer channels at EX, in the same time frame. I'm not seeing anything similar in EY or CS mags. (The 1-Hz comb is present in many magnetometers; it's the change points that seems to be unique to EX.)

Hey all, 

This looks similar to what we have seen in the past from the hartmann wavefront sensor cameras. It seems like we solved the issue by changing the camera power supplies.  We can double check this by changing the HWS sync frequency at low noise and provide you with a gps time. 

Sheila, Anamaria

The first thing that comes to mind from last Tuesday is a change to the ESD driver grounding at EX (47308), which did reduce some coherence between EX magnetometers and DARM (47323)  At first it looked like this work last tuesday removed a line at 73 Hz from DARM, although looking at the summary page you can see that this line has been coming and going and was at 72.5 Hz yesterday. summary page

It's probably not the HWS because those were always showing up a few tenths of a percent below their sync frequency, so if this comb is within half a mHz it's too accurate.

Ok, I'm posting more detail on what I'm seeing in the magnetometers, in case it helps to pin this down. I've checked for any obvious/clear change points, and also compared whether the shape of the comb is similar to what's observed in DARM (same pattern of variation in peak heights). In summary, SEIRACK and SUSRACK magnetometers at EX are particularly notable.

Also, the comb does appear to be precisely on 1 Hz, unlike the HWS comb.

Detailed magnetometer report:

[Corner station]
H1:PEM-CS_MAG_EBAY_LSCRACK_*_DQ: no apparent change; comb shape dissimilar from what is observed in DARM
H1:PEM-CS_MAG_EBAY_SUSRACK_*_DQ: no apparent change; comb shape dissimilar
H1:PEM-CS_MAG_LVEA_VERTEX_*_DQ: comb not really present

[End X]
H1:PEM-EX_MAG_EBAY_SEIRACK_X_DQ: changed (decrease); comb shape similar
H1:PEM-EX_MAG_EBAY_SEIRACK_Y_DQ: no apparent change; comb shape similar
H1:PEM-EX_MAG_EBAY_SEIRACK_Z_DQ: changed (increase); comb shape similar (?)

H1:PEM-EX_MAG_EBAY_SUSRACK_X_DQ: changed (increase); comb shape similar
H1:PEM-EX_MAG_EBAY_SUSRACK_Y_DQ: no apparent change; comb shape similar
H1:PEM-EX_MAG_EBAY_SUSRACK_Z_DQ: changed (decrease); comb shape similar (?)

H1:PEM-EX_MAG_VEA_FLOOR_*_DQ: comb not really present, but neither is anything else. These channels look way too clean; there might be an issue.

[End Y]
H1:PEM-EY_MAG_EBAY_SEIRACK_*_DQ: no apparent change; comb shape similar
H1:PEM-EY_MAG_EBAY_SUSRACK_*_DQ: no apparent change; comb shape similar
H1:PEM-EY_MAG_VEA_FLOOR_*_DQ: no apparent change; comb shape similar (?)

Robert, Anamaria

Even with just 0.01 Hz resolution, there is large coherence between various magnetometers in the EBAYs and DARM. Indeed this was not the case before last Tuesday.

Separately, the 72.5 Hz that appeared in the last two days is coherent with the corner EBAY magnetometer in the ISC rack. We think it's a real peak because if the sensor were picking up DARM the large calibration lines at 35ish Hz would show better coherence. Though it does not appear to be there this morning, so not completely clear what's going on.

Danny and I had a 72.33 Hz line driving some frequency noise over the weekend during some TCS tests. We also had lines at 3.25kHz, 4.5kHz and a bunch around 23-25Hz.

Shifted ETMX HWS sync frequency from 1 Hz to 5 Hz starting at 1236485181 and shifting it back to 1 Hz at 1236494796.

What about the GPS clocks that got turned on last Tuesday?

I had trouble finding a lot of clean DARM data during the period when the HWS frequency was changed, but what I found suggested the 1-Hz was still strong. Figure 1 is from when the HWS frequency was normal. Figure is from when it was changed.

The 1 Hz comb is still present in data from the third week of ER14, see first attached plot. There is also a very bad 0.8 Hz comb visible between 420 and 500 Hz, see second plot.

The 0.8 Hz comb Pep noted is a complicated structure, but it has some symmetrical features which are centered on the 516.78 Hz line. Keith tells me this is an EX violin mode (alog 47650, alog 47179).  A plot showing data from March 25 is attached.

It seems that the work reported in https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=47766 has made the 1 Hz comb disappear.

The first attached plot shows the spectra before this change, and the second one the spectra after this change.

On the other side, the lines and 0.8 Hz comb around the ~500 Hz violin modes are still there.

I took a PCAL to DARM measurement, and a DARM OLG, they are in /ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O3/H1/Measurements/FullIFOSensingTFs/2019-03-07*.xml
The PCAL to DARM was pretty bad, so I updated the Npct_ER14 front-end calibration filters according to the numbers above.  I also switched off Npct_O3 and switched on Npct_ER14 for ETMX L3, and adjusted L3's FM calib actuator gain from 1.073 to 1.0.  Will retake PCAL to DARM if I get a chance tonight.

I ran another PCAL to DARM broadband injection while adjusting the gains.  I believe there is some phase mismatch which makes a perfect front end calibration not possible at the moment: no matter how I adjust the gains, there is always a hump around 100 Hz, right around the crossover between the L3 and error signal authority.
I did the best I could, we are good to 10% everywhere now.  Range went from ~87 to ~95 Mpc.

Comparing the third plot above (called actuator authority) to the attachments to 47164, it seems like there must be a sign error in one of the stages which is creating the notch just below 2 Hz in the total.  It would be easier to debug these sign flips if we also could see the phase.  

@Sheila -- You're correct -- the model file used to generate the plot you mention (via darm_loop_critiue) had a reference to the wrong H1SUSETMX filter file (it wasn't as simple as a sign flip, but a previous filter file [before the PUM design was fixed] called with the same filter *banks* meant a report of a cross-over instability). 

I've corrected this since (apologies for not posting until now), but the new version is attached below, including the phase.

Keita, Craig, Sheila

Thanks Jeff for the updated plot.  We are using the model used (and compared to cross over measurements) in 47164 to try to reproduce your plot. 

We've tried to reproduce the plot you have above, but the only way we can do this is by removing the cascading of filters.  To say the same thing a different way, when you say LOCK IN to displacement, I think that you mean L3 lock in to displacement, which for L1 would mean L3 LOCK L * L2 LOCK L *L1 LOCK L * L1 drivealing *L1 electronics * L1 mechanical stuff.  The only way we can recreate your plot is to leave out L1 LOCK and L2 Lock from the L1 actuator.  However, if we leave these out the toal line in your plot is misleading/wrong.

In the first attachment (not cascading filters) is our reproduction of Jeff's plot above, where we have removed the L3 lock and L2 lock filters from L1.  The second one includes the cascaded filters, which is more correct if you want to add these up to make a total.  

I've processed the measurements taken from 1) Drivealign bank and 2) Test bank (measurement data in aligocalibration/trunk/Runs/O3/H1/Measurements/FullIFOActuationTFs; March 26)

The resulting plots are attached. The fitting through the test bank is better. This is a quick comment. The details need to be further investigated and discussed.

The residual comparison plot is made following the steps below:

1) Create two separate reference models for drivealign and test scenarios, based on the Mar 16 model, by writing back the MCMC PUM MAP values.

2) Compute the error residuals for two scenarios separately.

3) Plot Response_drivealign_with_error/Response_drivealign_ref, and Response_test_with_error/Response_test_ref.