I tried to make optimal sensing matrix for each DoF.
The work is still on the way...

Detail procedure will be reported later

I checked the seismometer data for the blastings up to today

Ushiba, Yokozawa, Tanaka

We tried to make diagonalized WFS from previous results in klog31041. First, we tried to make sensors for HARD modes. Yokozawa-san and Ushiba-san chosed the combination between QPDs and calculated the mixing ratio (DHARD PIT:klog31065, CHARD PIT:klog31066  For Yaw, Ushiba-san will report later). We input the value to the {C,D}HARD_{P, Y}_B element in INMTRX_{P, Y} (P:Fig.1, Y:FIg.2), and measured the diagonalized WFS responce about some alignment DoFs with 10 cnts amplitude excitation.

Fig. 3 shows the DHARD_YAW result. The coupling ratio between DHARD and others in the DHARD_YAW sensor seems to be less than 1/5 with the DHARD_YAW sensor. It seems to be fine.

Fig. 4 shows the DHARD_PIT result. In the view or the spectra, the hight of the DHARD PIT peak seems to be the same level as DSOFT PIT. On the other hand, the coherence between the DSOFT excitation and the signal in the DHARD_PIT sensor seems to be less than 0.6. The coupling ratio between DHARD and others in the DHARD_PIT sensor seems to be 1/2 at most with the DHARD_YAW sensor. So the results is not good.

Fig. 5 and 6 show the CHARD PIT, YAW results. in the view of the spectra, the CHARD sensor could not see CHARD mode but PR3... We tried to find the optimal demodulation phase but we didn't make it. We need more investigation.

[YamaT, Aso, Tanaka]

Yamamoto-san pointed out that the current order of switching the filters in the MN_OLDAMP bank during the handover from local control to global control may change the alignment. Fig. 1 shows the filter configuration in MN_OLDAMP before the handover. First, the AC filter in FM2 is engaged. Then, the ASC_LOCK guardian opens the input switch of the bank. Finally, the null filter in FM8 is turned on. However, in this order, since the filter DC gain shape become flat at DC after engaging AC filter (fig.2), the DC output changes when the input value become 0 when the input switch is open. Fig.3 shows the test result. We copied the AC, DC, and null filter from MN_OLDAMP to the TEST filter bank. We used the offset instead of the input value. We switched the filter as the same order as the guardian. At the timing on the time kersol in Fig.2, we turned off the offset. As you can see, the DC output value decreased after turning off the input value. This indicates the current switching order changed the alignment during the handover. Therefore we need to turn on the null filter before turing off the input swtich.

[Kenta, Yuzu]
We performed the lockloss investigation of 2024/09/11 16:45:56 JST.

memo

• The message  of ASC_LOCK was 'PRMI seems to be unlocked'. So, the last cause of the lockloss was the failure of 'LSC-POP_PDA2_RF90_I_NORM_MON'] > 0.1.
• At -11 s before the lockloss, the DC control was taken over from OLDAMP control to ASC, in INCREASE_SOFTHARD_GAIN state of ASC_LOCK guardian. (Fig)
• After that, K1:VIS-ITMX_TM_WIT_Y_DQ seems to start the oscillation and glowing. We can observe the similar trend on K1:ASC-DSOFT_Y_IN1_DQ and K1:LSC-POP_PDA2_RF90_I_ERR_DQ . (right bottom of Fig)
  • Based on the chat with Kenta-san, this is happening the weak DC control of the ASC during the handover of the ASC and OLDAMP.
  • The detailed procedure of the takeover will be posted by Kenta-san SOON.
• We could see the same trend before the following timing of the lockloss.
  • 2024/09/11 16:02:02 JST (klog) (Fig)
  • 2024/09/11 16:09:05 JST (Fig)
  • 2024/09/11 16:38:00 JST (Fig)
  • 2024-09-11 17:00:36 JST (Fig)
  • 2024-09-11 17:09:45 JST (Fig)

[Ushiba, Aso, Hirose, Tanaka]

(2024.09.11 work)

## Abstract

We measured the sensing matrix for remained DoF. The linearity seems not to be so bad. Although we implement temporal ASC to improve the contrast, it seems not to change the results.

## What we did

• We continued this measurement for the remained DoFs (PR2, PR3, IMMT2, IMMT1 P and Y). We checked the linearity of the measurement by changing the excitaiton amplitude. The measurement datas were stored in "/users/Commissioning/data/ASC/2024/sensingmatrix/0910/". (And today(2024/09/12), I found we forgot to measure DHARD/DSOFT YAW with 300cnts. So I meaured them today.)
• Fig.1 - 20 show the result, The values below the "data = '/users/Commissioning/data/ASC/2024/sensingmatrix/0910/{Alignment DoF}_P (or Y)_13(or 6)-125_{100, 10}-cnt.xml'" line are the WFS responce when we excited the corresponding alignment DoF in Pithc (or Yaw) direction with 100 (or 10 in the case of PR3) cnts as the excitation amplitude. The values below the "data = '/users/Commissioning/data/ASC/2024/sensingmatrix/0910/{Alignment DoF}_P (or Y)_13(or 6)-125_{300, 30}-cnt.xml'" line are the WFS responce when we excited the corresponding alignment DoF in Pitch (or Yaw) direction with 300 (or 30 in the case of PR3) cnts as the excitation amplitude. In the PR3 case, the excitation by 100 cnt seemed to be too large because both trans. power shaked largely by the oscillation. So we reduced the excitation amplitude when we excited PR3 and PR2. (Honestly speaking, we can excite with 100 cnts when we excite PR2). The radial distance and the phase were obtained by the following equations. (In this time, we have not consider each coherence is large enough yet.)
  • $\text{Radial distance} = \sqrt{\left(\frac{\text{WFS I}}{\text{oplev}}\right)^2 +　\left(\frac{\text{WFS Q}}{\text{oplev}}\right)^2}$
  • $\text{Phase} = arctan(\frac{\text{WFS Q}/\text{oplev}}{\text{WFS I}/\text{oplev}})$
• For the almost DoFs, each value of WFS response seems to be somehow consistent even though each excitaion amplitude was changed. So the linearity is not so bad. (For DSOFT P, the linearity seems to be somehow suspectable.)
• After that, we tried to engage previous ASC of HARD, SOFT modes to check whether the results, especially ASQPD results were changed or not with the somehow good contrast. First, since PRFPMI seems to lose the lock within 10 mins when the previous ASC was engaged, we reduced the gains of HARD loops from -1 to -0.5 and engaged the previous ASC. However, the DHARD YAW seemed to start the oscillation at about 0.4 Hz. We tried to reduce the gain further, but the situation seems not to be improved. We restored the gain to -1. the situation was not changed.
• We measured the TF from actuators (Type-As) to sensors (AS WFS) of DHARD YAW and compared the TF (fig.21) with the model (fig.22). They seems to be consistent. Also the OLTF model used by the TF model seems to be stable anywherer below 1 Hz (fig.23). So the DHARD YAW itself seems to be stable and not to be the cause of the 0.4 Hz oscillation.
• Then, we tried to engage only HARD loops. We succeeded in closing the loops. we measured the TF from actuators (Type-As) to sensors (TMSX QPDs) of XSOFT in this state. Unfortunately, all of Type-As started to oscillate at 1.1 Hz maybe because the excitation for this measurement seemed to kick the 1.1 Hz mode (according to the TypeA eignemode model, RM Yaw chain-0th?). Then we reduced the HARD loop gain from -1 to -0.3. Then, the oscillation seemed to disappear.
• We found each trans. power decreased gradually. According to ETMX TCam image, the beam spot seemed to shift from the TM center. we considered each beam axis drifted with each SOFT mode. Then, we tried to engage the BPC for each ETM using each SOFT mode in the DC region. The XSOFT BPC seems to work well and the trans. power improved somehow. However, in the YSOFT BPC case, the error signal did not become 0 and did not move from some value. We tried to improve by increasing the gain or flipping the sign, but the situation was not changed. (Maybe current HARD loops cancelled the actuation of the SOFT mode.) Anyway, the trans. powers are not maximum but stable. And the AS contrast seems to be improved. We measured the WFS responce about CHARD P in this state.
• Fig.24 shows the result, The values below the "data = '/users/Commissioning/data/ASC/2024/sensingmatrix/0910/CHARD_P_13-125_100-cnt_withASC.xml'" line are the WFS responce with ASC. The values below the "data = '/users/Commissioning/data/ASC/2024/sensingmatrix/0910/CHARD_P_13-125_100-cnt.xml'" line are the WFS responce without ASC. Unfortunately, we could not see the significant difference of the linearity between with and without ASC. the bad contrast seems not to be the cause that CHARD P motion can see in AS WFS.

Aso-san,

I tried it, and found that the resonant frequency of this mode got a bit lower, (112.9 Hz → 109.6 Hz) maybe because the total mass of the parts loadad on the bolts got heavier.

From 12:30 to 13:30 (JST) of yesterday, the laser was shutdown, thus OMC QPDs did not get any light.

I checked the time series of the QPD outputs during this period.

Obviously, the outputs move around in the order of 0.2 counts.
Note that these signals are normalized by the sums of the QPDs. Usually, they get more than 10000 counts in sum, but it was order of 10 when no light falls on the QPDs.
Therefore, 0.2counts of noise becomes order of 1e-4 or less counts in operation. Probably this is not a big issue for initial alignment.

Zooming into a glitch, they are pretty fast change of the offset. We can see some transient response from the circuit (probably the whitening filter).
The cause of these glitches are not understood at this moment.

Takano-kun,

Can you check if connecting the two top plates helps increase the resonant frequency?
(As shown in the attachment)

It seems the two plates are moving in the same direction but with different magnitudes.

[Kimura and M. Takahashi]
 We have started vacuum pumping of the GVetmx pressure gauge (CC-10).
 k-log 29326
We estimate that it will take about two weeks to reach a pressure that can be connected to X-arm.
After reaching the target pressure, we will open the gate valve at the boundary to connect to X-arm. 
 

I did the simulation with two thicker supports of the top plates instead of four thin bolts to confirm whether the resonant frequency of the modes related to the frame structure gets higher or not.

The effects of the thicker blocks are apparent. On the other hand, in some frame resonance modes the motion of the blade springs get larger (e.g. No7 and No8).

I found a lot of mistakes in the CAD model, though the results are quite similar. The simulation results with the proper CAD model are listed below.

Comparison with the actual value (ref: Junko Kasuya's master thesis):

Note

In these simulations, I was not able to find any resonance modes that can explain a resonance mode at 74 Hz observed in the laser displacement sensor measurement (klog 30206). Perhaps that mode is not related to OMC suspension, but the structure of the laser displacement sensor.

I found the several bugs and fixed them.
Then, I plotted the graph of yaw DoFs with the same measurment files that Hirose-san used in klog31056.

Figure 1-3 show the REFL, POP, and AS QPD results.
All plots have the data having the coherence greater than 0.5.
Results seem to be consistent, so the analysis seems fine.
 

[Kimura, M. Takahashi and Ohmae]

 We replaced the supply piping (Pair=~0.79 MPa) from the air compressor near IXC to the air compressor near BS.
However, we could not complete the pipe replacement because the prepared piping was about 5 m shorter.
 The insufficient piping material is scheduled to be delivered after September 13.
Therefore, compressed air to the PSL room on September 13 will be supplied from an air compressor near the BS.
The replacement of the supply piping is scheduled to be completed after September 17.

Continued from klog31047.

Today, as a continuation from klog31042, the sensing matrix of the remaining degrees of freedom was measured. (It will be posted to klog later)

Using the results, I plotted the Sensingmatrix. I did not plot the results where the coherence of I and Q signals for the Oplev signal was less than 0.5.
And I will upload the results of the cross-check with Ushiba-san to klog later.
(The YAW direction as linear plot: FIG1-FIG4.The PIT direction as linear plot: FIG5-FIG8. The YAW direction as log plot: FIG9-FIG12. The PIT direction as linear plot: FIG13-FIG16. )

The measurement file I referenced is as shown in FIG17.

I saved these results and PDF version in {/users/Commissioning/data/ASC/2024/sensingmatrix/0910/save_plot/}.

We performed the lockloss investigation of 2024-09-11 16:02:02 JST (still ongoing).

memo

• The message of LOCKLOSS of ASC_LOCK was 'PRMI seems to be unlocked'.
• I checked the error and feedback signals of ASC.
  • ndscope -w '(-15, +0.2)' -t '2024-09-11 07:02:02 UTC' /users/Commissioning/templates/ndscope/ASC/DHARD_P.yml
  • There were no signals with the saturation.
• At ~10s before the lockloss, the FM8 of DHARD_P, CHARD_P, DSOFT_P was turned on.
  • After that, K1:LSC-POP_PDA2_RF90_I_NORM_MON started the oscillation and the amplitude decreased. The oscillation frequency is ~0.71 Hz as seen Fig.
• From my eye, there are no strange point related to Xarm (Fig)
• At ~1s before the lockloss, K1:LSC-AS_PDA1_DC_OUT increased three times and K1:LSC-REFL_PDA1_DC_OUT_DQ decreased, as seen Fig. At almost same time, the laser power in trans started the decrease.

With Hido, Mitsuhashi, Ohnish,

We performed Pca-Y power sensor calibration.
This is the first measurement for Pcal-Y after the Noto earthquake.

Results:
Alpha values are voltage ratio between Integrating spheres and WSK.
alpha_RxPD1pWSK = 0.80687 +- 0.00043
alpha_RxPD2pWSK = 0.80625 +- 0.00087
alpha_TxPD1pWSK = 4.50220+- 0.00070
alpha_TxPD2pWSK = 1.6808+- 0.0017
e values are optical efficiency values.
e_1 = 0.99042 +- 0.00025
e_2 = 0.9815 +- 0.0011

We will compare the result with O4a later.

Yokozawa, Aso

We turned on the noise subtraction path in the OMC QPD driver.

We also tried to isolate the chasis from the rack by adding insulation between the bracket and the rack as shown in the attachment no.3.
We confirmed that the chasis is isolated from the rack in this state with a tester.
However, when we connected cables, the chasis was electrically connected to the rack. Especially, the cables between the QPD driver and the AA chasis connect the two chasis through the shell of the D-Subs.

After resinstallation of the driver into the rack, the output of the driver moved around a lot (a few tens of counts). This is a well known behavior of this circuit, probably due to the temperature drift.
We waited for ~30min, and the strange behavior of the output signals went away.

The attachment no.4 is the QPD noise spectra after the above wait.

Here is the link to the LIGO DCC for the circuit design of the QPD transimpedance amplifier.
https://dcc.ligo.org/LIGO-D1001974

KAGRA uses basically the same design as above.

The noise subtraction path can be enabled/disabled by moving the jumper pin (P1).
It was disabled before.

We can adjust the subtraction gain by changing R13 of the noise cancellation amplifier.

Since the stationary noise level is better with the noise subtraction turned off (even with the optimized values of R13), we left it turned off for the moment.

We have two issues with this circuit.
1. The stationary noise is worse with the noise subtraction enabled.
2. We see glitches. They are sporadic and elusive.

The glitch issue is more serious than the small stationary noise increase.

We have only tested long operation with the noise subtraction disabled so far. So we will enable the noise subtraction and leave it for a while to see if the glitch problem is eased by the noise subtraction or not.

In the similar manner of klog 29931 or klog 30740, I simulated the resonance mode of OMC suspension frame. To simplify the simulation model I removed the breadboad and applied the equivalent load (5 kgf = 49 N) to each blade spring. The coordinates in the simulation are defined as follows:

• X: along the longer axis of OMC breadboard in horizontal plane
• Y: along the shorter axis of OMC breadboard in horizontal plane
• Z: vertical direction

The simulated resonance frequencies and the corresponding modes are listed below:

Comparison with the actual value (ref: Junko Kasuya's master thesis):

For the mode of each Blade Z fundamental, the simulated resonant frequency is much higer than the designed (and actual) one. In the simulation the deformation of the bodies is assumed to be small and linear. However, actual deformation is quite large and not appropriate to be treated linerly. That may be the reason why the simulation gives such a larger value. For the other blade Z modes the simulation results are consistent with the design (but the design value seems also simulated by FEM...).

It should be noted that for the resonance mode No.4 not only the stopper plate of the blade springs but also the blade springs themselves move a lot. The resonant frequency of this mode is close to the resonant frequency of the connected stack in the vertical direction (klog 31004). It might be better to remove these plates to remove this unwanted mode.

I analyzed the images taken and evaluated the relationship between the two images using Affine transformation. The images match very well when we rotate the ASI image by -1.5 degrees and perform the parallel shift of +85 pixels horizontally and +65 pixels vertically. We confirmed that the accuracy of the test bench construction is at an acceptable level that does not affect the analysis.

Details

• Fig1: the image taken by the ASI camera.
• Fig2: the image taken by the gige camera.
  • Due to the file size and format limitations, the uploaded images are smaller than the original ones. I also converted from TIFF format to PNG format.
  • The setup of the test bench can be seen on the slide.
• The image size and the image sensor sizes are different due to the different cameras. To compare the images fairly, I added the black blank to the gige image to be the same image sensor size as the ASI camera. After that, I resize the ASI image (similar to up-sampling). As a result, two images have the same size (4768, 7002).
• Fig3: the overlay of two images with same size, before the transformation. 
• I performed an Affine transformation and looked for the best rotation angle and the parallel shift.
  • We rotate the ASI image by -1.5 degrees. Fig4 shows the overlay of two images after the rotation.
  • We perform the parallel shift of +85 pixels horizontally and +65 pixels vertically.  Fig5 shows the overlay of two images after the parallel shift. The images match very well.
• By counting the number of pixels of the 10mm target around the center of Fig2, which Tomaru-san provided, the conversion factor between the pixel and the actual length could be evaluated as 10/261.15=0.038mm/pixel.
  • So, the 85-pixel shift is almost 3.3 mm, and the 65-pixel shift is 2.5 mm. This difference is slight compared with the mirror size and image size.
• When we install this new camera in the TCam system and take images with two kinds of cameras, we need this kind of correction every time. At that time, it's good to do these steps automatically. We can probably use several characteristic appearances in the images.