NAOJ GW Elog Logbook 3.2

Yuhang and Michael fitted this data with mcmc. However, the detuning fluctuation is larger than that with least square... In this fit, the fit has been started from 60Hz and the detuning fluctuation could be smaller with higher fit starting frequency.
Left: mcmc (detuning: 50-68 Hz)
Right: least square (detuning: 49-61 Hz)

In the last calibration calculation, I didn't consider the loop gain. Therefore, the calibration factor must have some error.
Nevertheless, we can use another way to do this calibration without considering the loop gain.
0. Lock filter cavity.
1. Change slightly the temperature of main laser.
2. Read how much main laser frequency is changed.
3. Check how much length correction is sent to end mirror.
I did these procedures. The frequency change is read from the attached two figures. The correction signal change is in the attached figure three.
And get calibration factor (frequency difference)/(correction signal) = (248.6-235.2) [MHz]/ (5200) [counts] = 2.56 [MHz] / 1000 [counts]
Since 1pm = 1Hz, we can calibrate the factor above as 2.56 [um]/[kcounts].

The mcmc fit result of four parameters from published FDS data
sqz (dB) | loss (%) | phi (deg) | det (Hz) | |
1 | 8.3 +0.5/-0.3 | 34.1 +1.0/-0.8 | 0.1 +0.4/-0.5 | 46.3 +1.8/-2.1 |
2 | 8.2 +0.5/-0.5 | 36.0 +1.5/-1.9 | 14.7 +0.9/-0.8 | 68.7 +2.7/-2.6 |
3 | 8.9 +0.1/-0.2 | 34.4 +0.5/-0.3 | 26.0 +0.4/-0.3 | 59.8 +0.9/-0.7 |
4 | 7.8 +0.4/-0.3 | 40.1 +0.7/-1.3 | 43.3 +1.8/-2.1 | 66.0 +2.4/-2.9 |
5 | 8.9 +0.1/-0.1 | 34.2 +1.9/-0.9 | 55.4 +1.7/-1.0 | 63.6 +1.9/-1.2 |
6 | 8.6 +0.2/-0.2 | 36.7 +2.9/-2.5 | 91.6 +1.8/-2.0 | 70.9 +1.6/-1.9 |

[Aritomi, Yuhang]
Today we found that OPO automatic lock doesn't work. The reason was that OPO somehow cannot be scanned automatically with servo. For the moment, we locked OPO manually. We checked the UGF of the manual OPO lock and it was 4kHz.
We also found that current of p pol laser was not optimal value and the mode hop appeared in the OPO p pol transmission. We brought the current to the optimal value and the mode hop disappeared.
OPO automatic lock recovered by itself.
OPO automatic lock doesn't work again...

Yesterday, I took a new mixer (not the old TAMA one) and monitor its IF channel with two identical frequency RF signals as RF/LO.
The result is attached. Comparing this monitoring with elog2616, we can see much smaller drift.

To investigate the origin of bumps at 50 and 100 Hz in FDS measurement, I removed a phase shifter for CCFC LO and directly connected the DDS output for CCFC LO to the mixer with brown+green LEMO cables.
I tuned the CCFC demodulation phase by changing the LEMO cable length between DDS and mixer for CCFC so that time difference between center and 0 crossing point of the CCFC error signal becomes 44.4 ms, which corresponds to 44.4 ms*1.2 kHz/s = 53.3 Hz detuning. By using brown+green LEMO cables, I could realize the time difference of 44.4 ms. The sign of CCFC error signal is opposite compared with the one with the phase shifter.
The attached figure shows CCFC FDS without the phase shifter. The 100Hz bump becomes better without the phase shifter, but the 50Hz bump is still present. Also the 50Hz harmonics become larger without the phase shifter.
The detuning drift is 36-52 Hz, but this will be better with mcmc fit. The detuning is a bit smaller than the optimal value, so I will change the LEMO cable length for CCFC LO.

To compare least square fit and mcmc fit in a fair way, it is necessary to make both of them have both four parameters free with the four parameters defined in elog2618.
The information of mcmc fit has been already summarized in elog2618. The fit of least square information is summarized in the attached four figures.
Figure 1 and 2 are FDS with detuning ~200Hz. Figure 3 and 4 are FDS with detuning ~70Hz.
The least square fit gives similar result with mcmc if detuning is around 200Hz. However, the least square fit gives not-expected and seems-unresonable result as figure 3 and 4. By just changing the fitting method from least square to mcmc, we extract information more precisely and more reasonably.

For detuning around 200Hz data, the fit result of generated squeezing level and optical losses are
sqz | loss | |
data1 | 11.1 | 38.3 |
data2 | 11.2 | 39.9 |
data3 | 11.1 | 39.9 |
data4 | 10.8 | 37.4 |
data5 | 10.6 | 37 |
data6 | 10.5 | 42 |
For detuning around 70Hz data, the fit result of generated squeezing level and optical losses are
sqz | loss | |
data1 | 10.7 | 40.6 |
data2 | 10.4 | 40.6 |
data3 | 10.4 | 40.4 |
data4 | 10.0 | 37.5 |
data5 | 9.6 | 36.8 |
data6 | 10.0 | 37.1 |

Michael and Yuhang
In this elog, we compare the published FDS fit result and the new mcmc method we are using.
least square detuning (Hz) | mcmc detuning (Hz) | |
data1 | 42.6 | 46 |
data2 | 69.2 | 69 |
data3 | 62.2 | 60 |
data4 | 60.4 | 66 |
data5 | 67.9 | 64 |
data6 | 71.4 | 71 |
The mcmc fit result of four parameters from published FDS data
sqz (dB) | loss (%) | phi (deg) | det (Hz) | |
1 | 8.3 +0.5/-0.3 | 34.1 +1.0/-0.8 | 0.1 +0.4/-0.5 | 46.3 +1.8/-2.1 |
2 | 8.2 +0.5/-0.5 | 36.0 +1.5/-1.9 | 14.7 +0.9/-0.8 | 68.7 +2.7/-2.6 |
3 | 8.9 +0.1/-0.2 | 34.4 +0.5/-0.3 | 26.0 +0.4/-0.3 | 59.8 +0.9/-0.7 |
4 | 7.8 +0.4/-0.3 | 40.1 +0.7/-1.3 | 43.3 +1.8/-2.1 | 66.0 +2.4/-2.9 |
5 | 8.9 +0.1/-0.1 | 34.2 +1.9/-0.9 | 55.4 +1.7/-1.0 | 63.6 +1.9/-1.2 |
6 | 8.6 +0.2/-0.2 | 36.7 +2.9/-2.5 | 91.6 +1.8/-2.0 | 70.9 +1.6/-1.9 |

Interesting result! By the way, how is the fitting result of generated squeezing and optical loss for each curve? Are they consistent with each other?

Michael and Yuhang
We took FDS with filter cavity GR control about two weeks ago. The measurement contains 12 effective data with 6 for detuning around 200Hz and 6 for detuning around 70Hz. The data below around 70Hz is contaminated by back scattered noise. To have some margin from back scattered noise, we start fit from 100Hz.
The mcmc code needs a good enough initial value and corresponding range. We start with a least square fit with detuning, homodyne angles free and other parameters fixed. The fit result was used as initial value for mcmc code. The least square fit results are attached as figure 1 and 2.
We used the result of least square for mcmc and set four parameters to be free, including homodyne angle, detuning, optical losses, generated squeezing level. The result is attached as figure 3 and 4. The FDS with 200Hz detuning has more information about the squeezing quadrature rotation. Therefore, the error of fitting result is more precise. But the FDS with 70Hz detuning has less information, which makes the fit result has larger error on detuning.
The mcmc result gives more stabilized detuning, which means data favors a more stable detuning. The least square mothod gives larger detuning change may just comes from the fact that we are fixing other parameters but leave only two free.
Interesting result! By the way, how is the fitting result of generated squeezing and optical loss for each curve? Are they consistent with each other?
For detuning around 200Hz data, the fit result of generated squeezing level and optical losses are
sqz | loss | |
data1 | 11.1 | 38.3 |
data2 | 11.2 | 39.9 |
data3 | 11.1 | 39.9 |
data4 | 10.8 | 37.4 |
data5 | 10.6 | 37 |
data6 | 10.5 | 42 |
For detuning around 70Hz data, the fit result of generated squeezing level and optical losses are
sqz | loss | |
data1 | 10.7 | 40.6 |
data2 | 10.4 | 40.6 |
data3 | 10.4 | 40.4 |
data4 | 10.0 | 37.5 |
data5 | 9.6 | 36.8 |
data6 | 10.0 | 37.1 |

First I checked IR injection alignment. There was yaw misalignment and the mode matching was 89%.
After the alignment of yaw, the mode matching became 92% as follows. The injected BAB was 447uW. The misalignment is more or less fine, but LG is a bit larger than before.
Mode | IR transmission |
TEM00 | 480 |
yaw | 102 |
pitch | 104 |
LG | 111 |
offset | 95 |
By the way, during the alignment work, I noticed that the injection BAB power drifted a lot between 435uW and 465uW within a few minutes.
Then I locked CCFC and measured FDS (attached figure). CCFC calibration amplitude was 124mVpp, which is somehow lower than before. CCFC gain was 1000 and CC2 mass feedback gain was 3. The CCFC was stable and it kept locking during FDS measurement other than the squeezing quadrature. The 50, 100Hz bumps and detuning drift still exist.
Finally, I checked the nonlinear gain as follows. The nonlinear gain was 4.5 which corresponds to the generated squeezing of 10.2dB.
green power (mW) | 0 | 20 |
p pol PLL (MHz) | 245 | 185 |
OPO temperature (kOhm) | 7.163 | 7.163 |
BAB maximum (mV) | 57.2 | 256 |
nonlinear gain | 1 | 4.5 |
I will replace the electronics for CCFC to investigate the 50 and 100 Hz bumps.

By using TAMA demodulator, I monitor its output with two identical RF frequency signals as inputs. The signal drifts from 83.1 to 81.9.

Marc, Michael, and Yuhang
When we lock filter cavity with GR, IR detuning has change related to alignment. When GR automatic alignment (AA) and pointing loop is closed, IR detuning change can be stabilized.The filter cavity pointing loop is working mainly to fix the injection beam on end mirror. AA works to align filter cavity to the incident beam.
To check how detuning change for different alignment condition, we can change the pointing. By pointing the incident beam to different positions on end mirror and keeping AA loop closed, we can get AOM frequency for each point on end mirror when BAB is on resonance. The change of position on end mirror gives us a dependence of detuning as end mirror beam hitting point. In this way, we call it a detuning map for end mirror.
The changed parameters are not only beam hitting position on end mirror, the other changed parameters are input/end mirror angles and cavity length. A typical change for input angle is 40urad, end angle is 10urad and cavity length is 0.2um. Since the optical axis of filter cavity is almost the same for GR and IR, the GR AA should work also for IR. In addition, GR length control should also work for IR. Therefore, the map we get should just depend on beam hitting position of end mirror. The corresponding map is attached.

Today, I went to filter cavity end room and centered the FC GR transmission camera in yaw. But I didn't move pitch.
The pointing offset is still a good reference, since I didn't move PSD.

The filter cavity IR detuning is monitored from 2021-07-08 11:40 to 2021-07-09 11:40. (The time used in DGS is JST minus 9 hours) The minute trend data is saved in standalone desktop/detuning/20210709.
The screenshot of this monitor is attached.
Although there are many peaks in the detuning data, only 4 of them come from the unlock of filter cavity. Others are due to the suspended mirror sudden position changes but pointing loop has limited bandwidth.
We see change of detuning even when fc length is controlled.
The FC length control change may also come from the main laser frequency change, which is due to we use laser frequency as a reference at low frequency. Especially, we don't have a reference cavity as used in gravitational wave detectors, such as KAGRA.

Abe-san, Aso-san, Marc, Michael
For reference, ETMY cleaning is summarized in KLOG entries : 17219, 17271, 17292, 17311, 17397, 17409
We brought the ETMY inside PCI room using the crane.
We added HEPA filters and put an ion gun at the end of the pressured air.
We remove optics from the small optical table on the side of the PCI setup and installed ETMY box on it.
Using a strong green light and a strong white flashlight we inspected and slightly cleaned the AR surface using ultra pure water and the ion gun.
On this table, we installed ETMY inside its holder using 4 jacks.
In order to avoid incident, we removed the entire imaging unit optical table to ease the ETMY installation on the translation stage.
Before doing so, we had installed pairs of forks on 3 of the 4 pillars to be sure to recover the same imaging unit position.
We installed ETMY on the translation stage (the additional weight due to the jig is negligeable because the translation stage can hold few hundreds kg).
We removed the HR surface first contact while using the ion gun.
To avoid scratching ETMY surface or magnets, we decided to let a metal ring at the edge of the mirror surface.
We reinstalled the IU and turning on the probe beam showed that this beam was still hitting well on the IU optics.
I don't have so much pictures but we'll add beautiful ones to this entry.
This morning we set up translation stage limit along the Z axis. We are now not letting ETMY get closer than 1 or 2 cm to both side.
There are now really strange troubles with Zaber that does not recognize the translation stage (or any com port) even if it is working fine in labview..
We will solve this issue before any translation stage motion.

Here is the calibration performed before the installation of ETMY on PCI setup :
AC_surfref = 0.45825;
DC_surfref = 3.987;
P_in = 30.3e-3;
abs_surfref = 0.22;
R_surf = 17.24 /W
The AC peak is located at Z=39.6 mm.

It was noticed that filter cavity z-correction was feeding back some high frequency components recently.
I modified the filters and now less high frequency components are feed to end mirror.
The new filter is called dc_damp2 (gain is adjusted so that we can use gain 1 in medm). Let's use this filter in the future.
Poles: 1e-4, 0.1(1), 20, 20, 300
Zeros: 0.04, 0.05(1), 3
The comparison of signal sent to end mirror is attached.

Marc and Yuhang
Recently, we found a new spot of filter cavity elog2573, which makes the IR locking accuracy much better (the spectrum below~3Hz reduce by up to a factor of 10). At the beginning, we thought we found a more stable optical axis. However, we did a test of GR length correction signal when using old and new spot, which shows pretty similar spectrum at frequency region below ~3Hz (attached figure 1). Since the GR length correction signal below ~3Hz tells us mirror motion information, this means the mirror motion is similar for the old and new spot.
Meanwhile, the correction measured in this time is different from elog2312. Especially, it seems more high frequency signal is sent to end mirror.