Marc, Michael, Yuhang

Today we added the PR pointing loop to medm.

It is basically a copy & paste of BS pointing loop except that we had to change the pitch gain sign.

We could close the loop and things seem stable.

On medm there are now 2 buttons to close pointing with feedback on either BS or PR.

However, as pointed in entry 2493 PR is drifting more than BS so closing the pointing loop on PR instead of BS should be useful for long term stability.

I calculated the IR misalignment effect on CCFC error signal in I phase.

As you can see, even if the IR mode matching changes between 90% and 100%, the zero crossing point of CCFC error signal (detuning) changes by only ~ 0.06*54 = 3 Hz. So this effect cannot explain ~10Hz FDS detuning fluctuation.

The Bayesian method to evaluate parameter in a function is a well-known prevalenty used method for parameter estimation in many research of fields. (such as gravitational wave detector signal analysis) You can check from this link, they did a simple demonstration that Bayesian method gives a more accurate parameter estimation. In our experiment, we are trying to fit the FDS measurement result to FDS degradation model to get information of cavity detuning and homodyne angle. To achieve this goal, we were using a least square curve fit method but the error of many parameters were not considered. Therefore, to better evaluate the filter cavity detuning information, I would propose to use Bayesian method.

To do Bayesian estimation, there is a package called 'emcee' developed in python environment. The method to do this Bayesian estimation is called 'MCMC'. To use this package, we need to move the FDS degradation function from matlab to python. I did this based on the FDS code provided from Eleonora. To make sure the python function I wrote is correct, I did the comparison of python and matlab results as the  attached figure 1. I found the python code I wrote gives quite the same result with Eleonora code.

Then I made a preliminary run of the mcmc code I wrote. (It was taking ~30 hours, but can be reduced if I use parallel calculation) Based on the result of MCMC, I made a violin plot of the cavity detuning fit result estimation, as attached figure 2.

I would like to do this analysis for the future measurement of FDS.

Using Inventor 2015 I tried to design the holder for KAGRA QPDs that will be used for all our OpLev sensors.

Figures 1 and 2 show the 2 parts I'm forseeing to use. Basically (as in figure 1)  a vertical spacer to mount the QPD board using 4 screws, a long hole to use a L to fix to a translation stage and 2 holes on top to fix the 15pin D sub connector.

Figure 2 shows the 15 pins D sub holder.

The main point is that there will not be vertical or horizontal translation stage (similar to current situation). I'm also planning to put a laser line filter on the translation stage along the beam axis.

If this preliminary design is fine, I'll translate it into documents usable by manufacturer.

Aritomi, Marc, Michael, Yuhang

Yesterday during the work on CCFC, we found out that the FC can be unlocked if we touch some cable of the rack or some ground of mixers also connected to the rack.

The rack power supply ground is connected to tama ground. Using a powermeter we checked the voltage between the power supply ground and the ground cable and found that it was zero...

We might have to investigate directly on the rack.

Also we found out that even if the signal generator is turned off, if it is connected to some voltage driver it might introduce noise and cause some unlocks.

Although the locking accuracy is 1.8Hz with CCFC, the FDS detuning fluctuation is ~10Hz even with CCFC/Z correction/pointing/AA. The possible reasons are parameter estimation error and IR misalignment change and FC macroscopic length change.

sqz_dB = 10.5;                    % produced SQZ (dB)

L_rt = 120e-6;                    % FC losses

L_inj = 0.31;                     % Injection losses 

L_ro = 0.24;                      % Readout losses

A0 = 0.05;                        % Squeezed field/filter cavity mode mismatch

C0 = 0.05;                        % Squeezed field/local oscillator mode mismatch

ERR_L =   1e-12;                  % Lock accuracy (m)

ERR_csi = 30e-3;                  % Phase noise (rad)

Note: In this calculation, shot noise level is assumed to be -133.2 dBVrms/rtHz, but actually it may not be true. So the result in this log could be wrong.

All AA filters are 'DC_damp2' with an additional gain of 0.002 for yaw and 0.005 for pitch.

I plotted as an example in the figure 1 both OLTF and filter of AA input yaw. But similar results for all other dofs.

The right  column shows the product of the OMTF with the associated filter.

It can be seen that the gain never reach 1.... This might be due to poor coherence during the measurement (even if it is not possible to drastically increase the excitation while keeping the FC locked).

Another possibility to do this measurement could be to estimate Type-C suspension TF from simulation and use a single line excitation to estimate the WFS gain.

We have seen detuning change more than we expect, such as elog 2512 and elog 2537. In elog 2512, we expect detuning change less than ~15Hz, but we observed detuning change of ~50Hz. In elog 2537, according to the alignment change calculated in elog2540, we expect detuning change of ~3Hz, but we observed detuning change of ~10Hz.

To understand in more detail why this can happen, we take into account the errors of other degradation parameters (shortly DP) (squeezing level, RTL, mode matching, optical losses, phase noise, locking accuracy) and use Monte-carlo estimation to see how the fit result changes. In this entry, I used data of one curve from elog2512, but the same calculation can be applied to other data. The code used in this entry is based on Eleonora's code.

From least square curve fit, the expectation and standard deviation are estimated by giving fixed DPs. The result is angle 11.22+/-0.13,  detuning 105.09+/-0.54.

The Monte-carlo estimation gives expectation and std based on random chosen of DPs within normrnd(mean,std). The means of DP are chosen as the same with least square curve fit, while std is chosen as indicated in PRL paper. This means that, for one calculation, 8 DPs are chosen randomly while cavity detuning and homodyne angle are left free, then matlab uses least square method to give a fit of the FDS measurement and gives center values of detuning and homodyne angle. After calculating for 1000 times, we have 1000 center values for [cavity detuning, homodyne angle], which took 1.5 hours. This result is plotted as two histgrams as the attached figrue.This gives result of angle 11.22 +/- 0.35, detuning 105.04 +/- 1.18.

We can see basically the fitting error becomes about two times larger.

The CCFC error signal you analyzed is Q phase signal which is not used for CCFC lock. How about the detuning change with I phase?

[Aritomi, Yuhang, Michael]

Today we found that there was a 250Hz noise in CCFC signal and this noise came from low UGF OPO lock due to OPO demodulation phase change. After reloading the DDS1, this problem was solved.

After that, we found that FC lock was unstable even when Z correction, pointing, AA were engaged. In the end, FC could not be locked at all. We will investigate it next week.

Today I brought the KAGRA spare viewport that was sent together with the 2 dirty ETMY viewports.

It seems to be 10 cm diameter and 1 cm thick.

I tried to install it inside the holder for the shinkosha evaluation plate (similar thickness) BUT

- only one stabilizing screw can be used making me afraid that the viewport might move during a measurement

- from the lowest point of the mirror up to ~3.5 cm will be hidden by the holder.

I guess a possible quick fix could be to buy ~6cm stabilizing screw and use the TAMA size holder or drill new holes in the SHINKOSHA holder to move the screws and metallic rode to a good position.

In elog2187, we measured FDS with CCFC loop locked. Meanwhile, there were no GR automatic alignment (AA) and input mirror feed back (InputFB). And we have seen detuning changed by about 10Hz, which should not happen since CCFC works to remove the detuning change with precision of 1Hz (could be a bit worse in elog2187). I have thought about this in more detail, and I found misalignment can impact on CCFC lock.

I have a code Aritomi-san provided about CCFC error signal calculation. By using this code, as elog2300, I added mode matching influence for this error signal and found the shape change of this signal, which was confirmed with experiment. I made this simulation again with mode matching level of 100%, 95%(our current situation), and 90%. The simulation result is shown in the attached figure (figure 1 is the zoom in of figure 2). We can see that the CCFC error signal zero crossing point changes with different mode matching level. Let's remind that when CCFC is used to lock filter cavity, we lock filter cavity to the zero point of CCFC error signal. Since CCFC zero crossing point changes with mode matching level, a misalignment for squeezing field will translate into FDS detuning change.

Attached figures show zero crossing point changes from 1.954 to 1.895, which are normalized by detuning frequency 54Hz. Therefore, this corresponds to a frequency change of 3.2Hz. Without automatic alignment, and considering the experience of our mirror suspension system change, this misalignment induces detuning change and prevents us from the goal of 1Hz detuning stabilization. Even with GR automatic alignment, the alignment is actually fluctuating and results in ~95% misalignment, which should introduce already 3.2Hz detuning change.

To achieve really detuning change less than 1Hz, IR automatic alignment will be necessary!

On Thursday morning I measured the opto-mechanical transfer function of the pointing loop with FC locked with all loops.

For BS pitch the amplitude was 9000 and BS yaw 3000 (higher would cause unlock).

Results are presented in figure 1 and 2.

Marc, Michael, Yuhang

On wednesday morning we started the AA characterization.

First, we measured the SNR of one QPD segment (here QPD1_I1) but similar results are expected for other segments.

You can see in figure 1 that we have reasonable SNR up to about 100 Hz.

Then we measured the input opto-mechanical transfer function by injecting white noise in input pitch (amplitude 9000) and yaw (amplitude 6000) and checking the signal reconstructed by the AA.

Note that to make these measurements with diagui we had to close the mirror damp loops.

The results are presented in figure 2 and 3.

Then when we tried to inject noise to end we unlocked the FC.

Actually, even if the damp loops are not used (gain to 0), there were several filters and large offsets that caused issues.

So on Thursday morning I repeated these measurements for end mirror  with identical amplitude without particular troubles.

Next step is to combine these measurements with the filter that we are using to check the bandwidth, UGF, phase margin to see if we can improve the AA filter.

[Aritomi, Yuhang, Michael]

Today we found that the SMA cable for GRMC demod was not fixed well. After we fixed it, the 100Hz bump disappeared!

The attached figure shows FDS with CCFC. CCFC gain was 2000 and CC2 mass feedback gain was 2.7. Although the 100Hz bump disappeared, there is still a small bump at 150Hz and many 50Hz harmonics. We might have another unfixed cables.

The detuning is a bit high in this measurement so we will measure FDS with correct detuning again.

sqz_dB = 10.4;                    % produced SQZ (dB)

L_rt = 120e-6;                    % FC losses

L_inj = 0.31;                     % Injection losses  

L_ro = 0.24;                      % Readout losses

A0 = 0.05;                         % Squeezed field/filter cavity mode mismatch

C0 = 0.05;                         % Squeezed field/local oscillator mode mismatch

ERR_csi = 30e-3;                  % Phase noise (rad)

ERR_L =   1e-12;                  % Lock accuracy (m)

I have installed 2 beam dumps on spurious green beams.

Figure 1 is spurious beam due to a galvo

Figure 2 is spurious beam due to the PR window.

While I don't think it will improve performances, it will make the work more confortable as these beams were quite powerful.

If my understanding is correct, this measurement was made with the FC half detuned. So it means that the mode-mismatch computed here mixes the mode-mismatch between BAB/FC together with BAB/LO, which means that we should not use this value in FDS degradation budget.

Also note that the measurement was performed over 50s and not 200s as this 50s corresponds roughly to the FDS measurement duration.

We forgot to tune the OPO temperature during this measurement that could explain the difficulty to fit..

Marc, Michael, Yuhang

Yesterday morning we started the measurement of opto-mechanical transfer functions to check the AA filter design (results will follow in another entry)

During this activity we injected noise in input pitch and yaw using diagui.

In order to send these noises from diagui we had to close the damp loop of input.

When we tried to do the same for end mirror we first could not see any noise so we remove many filters (especially 50 and 100 Hz notches) from this loop and then had a FC unlock due to an extremely large offset that we forgot to remove.

I think it is the only modification between your last 2 measurements, but as this loop had a gain of 0, it might not affect the apparition of this peak...

I measured FDS with CCFC around optimal detuning (attached figure).

CCFC filter gain is 2000 with 30Hz LPF. CC2 mass feedback gain is 2.7.

The spectrum for squeezing quadrature (red curve) seems distorted by backscattering noise and the fitting result is not reasonable.

Now the problem is a bump around 100Hz. I tried the different LPF cut off frequency and gain of CCFC filter, but the 100Hz bump cannot be removed...