Camilla, Sheila.

The drop was because the OPO PZT ran out of range, see t-cursor on attached plot. It's a known issue that the SQZ angle (and alignment) changes with different PZT1 voltage.

It seems like that the sqz angle was set for 60-70V and was bad once we relocked at 100V. This would have been improved by taking SQZ_MANAGER to SCAN_SQZANG_FDS so it can find the best sqz angle again. If we were running the SQZ_ANG_ADJUST servo, this may have improved itself as you can see the ADF reported SQZ angle change at the time.

With the higher (19 vs 11) NLG 83665, the range and 350Hz SQZ (yellow BLRMS) does seem to be improved, but the high frequency SQZ did seem to be less stable as we thermalized.

Code for analyzing this data

I had a brief look at some of this data to put bounds on losses and arm power in 83953:

The first attachment shows a plot of more of this data against models, focusing on the unexplained low frequency noise that we don't see with the filter cavity . The measured NLG matches the NLG infered from anti-squeezing and squeezing for the NLG 19 measurements, but for the NLG 35 measurements the infered NLG is 27.3, so that is what I've used here.  As Camilla wrote above, the NLG 35 measurements were made with a different SRC detuning than NLG19, so that is included in this model.  Squeezing angles are fit to the band from 2100 Hz to 2300 Hz. 

The first plot shows the measured data in solid lines, the quantum noise model in dashed lines, and the dotted lines show the non quantum noise from subtraction added to the quantum noise models.  There is a discrepancy where many of the measurements seem to have extra noise from 20-50 Hz, I've tried to make an easier to read version in the second plot, and finally removed some traces to try to make it easier to see.

In the above alog we thought perhaps that this could be explained as an excess noise that was larger with higher nonlinear gain but consistent with squeezing angle, the last attachment shows the residuals between the model and measurement for the measurements that had clear discrepancies, they all seem to be different, so this excess seems to depend both on squeezing angle and nonlinear gain. 

The script used to make these plots can be found at this repo

Appears to have been an ETMX glitch

Ibrahim, Oli, Camilla

We had two 30+ minute range drops last night, plot attached. Looks to be mainly under 60Hz from the OMC BLRMs, plot.

Doesn't appear to be sqz, EX ground motion 83233, or CO2 ISS channel 82728 related. Oli also checked this wasn't PI related 82944.

Confirmed not the PIs for either range drop (drop1, drop2). There was a small drop in range a few minutes before the second range drop around the same time as PI31 had a small ringup, but zooming in, the ringup happened multiple minutes before and so doesn't account for that quick drop either(ndscope3).

Here are some plots of Camilla's first dataset above, changing the SRC detuning while adjsuting the squeezing angle for high frequency squeezing, made with the same code used for 80318, which is available here

For the gwinc model, I've set the generated squeezing to 23 dB based on Camilla's measured NLG of 58.  Based on the loss estimates from 83457, I've set the Injection loss to 0.178 (17.8% loss) and the PD efficiency (readout efficiency) to 0.815, and the phase noise to 0.

The third attachment shows the model where I've manually adjusted the SRC detuning to roughly match the subtracted squeezing, and the second shows a linear fit of SRCL offset to these detunings.  This suggests that the SRCL offset should be at -306 counts to reduce the SRCL offset, and that we are currently running with a SRCL detuning of 0.013 radians. 

This morning we put SRCL offset to -306, FC de-tuning -28Hz. I then ran SCAN_SQZANG which changed the angle form 171 to 161 and compare the before and after DARM, attached, SQZ looks alot better at higher frequencies, however the range, attached, is similar or a little worse, maybe the 300Hz (yellow BLRMs) squeezing is slightly worse.

Updated DTT legend as had typo.

Here are some preliminary plots from Camilla's data set of different squeezing angles taken at an NLG of 58 with the SRCL offset at it's nominal -191 counts setting, which we believe is about 13 mrad SRC detuning. 

The first plot shows some assumptions that go into making this model, we start with an assumption about arm power, use the noise budget estimate of non quantum noise at 2kHz (which may be out of date now), and set the readout losses to fit the no squeezing data at 2.1-2.3kHz.  Then subtract this quantum noise model without squeezing  from the no squeezing data, and use that as an estimate of the non-quantum noise, which can be added to all of the quantum noise models for different squeezing angles to compare to the measurement. (second plot is a somewhat overwhelming plot of all this added for completeness).

I've set the phase noise to 0 based on 83457.  Using the level of sqz and anti-squeeze at 2.1-2.3 kHz, we infer that the NLG was 63 and the total efficency was 66.5%.  Camilla measured the NLG to be 58, for 120uW circulating power, but in 83370 she measured 61-63 for 120uW.   The third plot here shows the data that Camilla took with the LO loop unlocked, so that the squeezing angle is averaging and rotating freely.  Using this and knowledge of the NLG, we should be able to infer the total squeezing efficiency as a function of frequency. Doing the subtraction of non quantum noise increases the infered efficiency, (compare thick lines to thin), the two different values of NLG suggest rather different efficiencies. There is evidence that the efficiency frequency dependent, which could be caused by a number of effects. Below 200 Hz there is some excess noise in the mean sqz trace, as you can see here, which causes the efficiency infered to be above 1.

The next two plots show the model broken into more readable plots, with the only thing I've adjusted by hand being the SRC detuning.  There is a discrepancy between the model + noise for the anti-squeezing and anti-squeezing +/-10 degrees traces without the filter cavity, which seems like it could be some excess noise that is similar for the different traces.  This is similar to the discrepancy seen in the last plot in 82097, but it is larger in this higher NLG dataset.

ah HA! There is already a SENSALIGN matrix in the model for the M1 OSEMs - this is a great place to implement corrections calculated in the Euler basis. No model changes are needed, thanks Jeff!

If this is a gain error in 1 of the L osems, how big is it? - about 15%.

Move the top mass, let osem #1 measure a distance m1, and osem #2 measure m2.

Give osem #2 a gain error, so it's response is really (1+e) of the true distance.
Translate the top mass by d1 with no rotation, and the two signals will be m1= d1 and m2=d1*(1+e)
L is (m1 + m2)/2 = d1/2 + d1*(1+e)/2 = d1*(1+e/2)
The angle will be (m1 - m2)/s where s is the separation between the osems.

I think that s=0.16 meters for top mass of HLTS (from make_sus_hlts_projections.m in the SUS SVN)
Angle measured is (d1 - d1(1+e))/s = -d1 * e /s

The angle/length for a length drive is
-(d1 * e /s)/ ( d1*(1+e/2)) = 1/s * (-e/(1+e/2)) = -0.85 in this measurement
if e is small, then e is approx = 0.85 * s = 0.85 rad/m * 0.16 m = 0.14

so a 14% gain difference between the rt and lf osems will give you about a 0.85 rad/ meter cross coupling. (actually closer to 15% -
0.15/ (1 + 0.075) = 0.1395, but the approx is pretty good.
15% seem like a lot to me, but that's what I'm seeing.

I'm adding another plot from the set to show vertical-roll coupling. 

fig 1 - Here, you see that the vertical to roll cross-couping is large. This is consistent with a miscalibrated vertical sensor causing common-mode vertical motion to appear as roll. Spoiler-alert - Edgard just predicted this to be true, and he thinks that sensor T1 is off by about 15%. He also thinks the right sensor is 15% smaller than the left.

-update-

fig 2- I've also added the Vertical-Pitch plot. Here again we see significant response of the vertical motion in the Pitch DOF. We can compare this with what Edgard finds. This will be a smaller difference becasue the the pitch sensors (T2 and T3, I think) are very close together (9 cm total separation, see below).

Here are the spacings as documented i the SUS_SVN/HLTS/Common/MatlabTools/make_sushlts_projections.m

I was looking at the M1 ---> M1 transfer functions last week to see if I could do some OSEM gain calibration.

The details of the proposed sensor rejiggling is a bit involved, but the basic idea is that the part of the M1-to-M1 transfer function coming from the mechanical plant should be reciprocal (up to the impedances of the ISI). I tried to symmetrize the measured plant by changing the gains of the OSEMs, then later by including the possibility that the OSEMs might be seeing off-axis motion.

Three figures and three findings below:

0)  Finding 1: The reciprocity only allows us to find the relative calibrations of the OSEMs, so all of the results below are scaled to the units where the scale of the T1 OSEM is 1. If we want absolute calibrations, we will have to use an independent measurement, like the ISI-->M1 transfer functions. This will be important when we analyze the results below.

1) Figure 1:  shows the full 6x6 M1-->M1 transfer function matrix between all of the DOFs in the Euler basis of PR3. The rows represent the output DOF and the columns represent thr input DOF. The dashed lines represent the transpose of the transfer function in question for easier comparison. The transfer matrix is not reciprocal.

2) Finding 2: The diagonal correction (relative to T1) is given by:

I will post more analysis in the Euler basis later.

Here's a view of the Plant model for the HLTS - damping off, motion of M1. These are for reference as we look at which cross-coupling should exist. (spoiler - not many)

First plot is the TF from the ISI to the M1 osems.
L is coupled to P, T & R are coupled, but that's all the coupling we have in the HLTS model for ISI -> M1.

Second plot is the TF from the M1 drives to the M1 osems.
L & P are coupled, T & R are coupled, but that's all the coupling we have in the HLTS model for M1 -> M1.

These plots are Magnitude only, and I've fixed the axes.

For the OSEM to OSEM TFs, the level of the TFs in the blank panels is very small - likely numerical issues. The peaks are at the 1e-12 to 1e-14 level.

@Brian, Edgard -- I wonder if some of this ~10-20% mismatch in OSEM calibration is that we approximate the D0901284-v4 sat amp whitening stage with a compensating filter of z:p = (10:0.4) Hz?
(I got on this idea thru modeling the *improvement* to the whitening stage that is already in play at LLO and will be incoming into LHO this summer; E2400330)

If you math out the frequency response from the circuit diagram and component values, the response is defined by 
    %  Vo                         R180
    % ---- = (-1) * --------------------------------
    %  Vi           Z_{in}^{upper} || Z_{in}^{lower}
    %
    %               R181   (1 + s * (R180 + R182) * C_total)
    %      = (-1) * ---- * --------------------------------
    %               R182      (1 + s * (R180) * C_total)
So for the D0901284-v4 values of 
    R180 = 750;
    R182 = 20e3;
    C150 = 10e-6;
    C151 = 10e-6;

    R181 = 20e3;

that creates a frequency response of 
    f.zero = 1/(2*pi*(R180+R182)*C_total) = 0.3835 [Hz]; 
    f.pole = 1/(2*pi*R180*C_total) = 10.6103 [Hz];


I attach a plot that shows the ratio of the this "circuit component value ideal" response to approximate response, and the response ratio hits 7.5% by 10 Hz and ~11% by 100 Hz.

This is, of course for one OSEM channel's signal chain. 

I haven't modeled how this systematic error in compensation would stack up with linear combinations of slight variants of this response given component value precision/accuracy, but ...

... I also am quite confident that no one really wants to go through an measure and fit the zero and pole of every OSEM channel's sat amp frequency response, so maybe you're doing the right thing by "just" measuring it with this technique and compensating for it in the SENSALIGN matrix. Or at least measure one sat amp box's worth, and see how consistent the four channels are and whether they're closer to 0.4:10 Hz or 0.3835:10.6103 Hz.

Anyways -- I thought it might be useful to be aware of the many steps along the way that we've been lazy about the details in calibrating the OSEMs, and this would be one way to "fix it in hardware."