I started the reassembly of the broken FLDACCs. #4 folded pendulum block was replaced with #7 block (Picture #1), which was transported with the special box (Picture #2) from NAOJ (Mitka). The assembled pendulum looks fine (Picture #3). The FLDACCs for SRM moved to the BS area temporarily to prevent troubles due to the cleaning of the SR-OMC booth (Picture #4).

I installed the poles to cover the table to store the MSBs removed from the SRM chamber (Picture #5).

[Kenta Tanaka, Dan Chen]

Follow-up Investigation of PR3 Glitches (2026/02/26)

Following the previous report on intermittent large motion of PR3, we performed an additional investigation today during a period when the interferometer was unavailable due to an earthquake.

Summary

• We investigated the PR3 glitches, but the root cause has not been identified.
• We suspected possible origins related to the oplev sensors and/or the oplev laser source; however, we found no evidence supporting these hypotheses.

Time-Series Observation

From the time-series data, the glitches appear to be correlated between the TM P and Y directions.

In the figure below, the red circles indicate periods where glitches are present, while the blue circles indicate quieter periods.

Oplev QPD Signals

We examined the QPD segment signals:

The glitches are visible in all four QPD segments and appear consistent across them. No abnormal behavior is observed in any single segment. Since the glitches do not appear in the SUM signal, they are unlikely to originate from the oplev laser source.

Spectral Comparison (PAY_FLOAT)

We compared the spectrum in PAY_FLOAT with data taken during a previously quiet period(2024/11/28):

• No significant spectral differences were observed compared to the earlier quiet dataset.
• Although the previous dataset appears to have had the loop manually closed, the high-frequency region should still be comparable.
• Note: during the present measurement the system was in PAY_FLOAT, so it is unclear whether glitches were actively occurring at that time.

Current Status

Despite these investigations, no clear cause of the PR3 glitches has been identified. The spectral shape is largely unchanged compared to the earlier quiet period, and the QPD behavior does not indicate a sensor- or laser-origin issue.

Fig.1 shows the time series of the oplev signals in the optical path from IMMT2 to X-arm when PR3 was glitchy. Glitch size seems to be 1.9 urad in P, 0.97 urad in Y, respectively. These sizes are much larger than the usual RMS.

Even though the glitch was occuring, HANDOVER from ALS to IR seems to be fine (but not easy). However, the fluctuation of POP90 was terrible. Fig.2 show the timeseries including to POP90, PRC and MICH error/feedback. Between the cursors, there seems to be no glitch in PR3 oplev. in the out of this region, POP90 dropped to the less than half of the maxlmun.

Komori, Dan, Tanaka

## Abstract

We implemented the new filters for the High-Bandwidth ARM ASCs in the Yaw direction ({D,C}{HARD,SOFT}_Y) and engaged them. This time, the Pitch DoFs were controlled by not High-Bandwidth filters but the original ones. The bandwidth of {D,C}HARD_Y, CSOFT_Y control seems to be from DC to 0.6 Hz and around 1.7Hz. On the other hands, DSOFT_Y could not be enlarge only to 0.3 Hz due to 1.1 Hz oscillation.

Also, we tried to increase DHARD_Y bandwidth more, but DHARD_Y started oscillated at 10-11 Hz when we doubled the gain.

Moreover, the BPCs for ETMX and ETMY were oscillated at about 47 mHz. The crossover frequency seems to be changed since the SOFT mode gain at DC was smaller than before. So we reduced the BPC gains to 0.1.  

1.14 Hz oscillation seems to be still there (fig.4) and makes IFO down. We need more investigation.

## What we did

### BPC gain adjustment

First, we implemented the same filter as HARD_Y to SOFT_Y and engaged all ARM DoFs in Yaw direction. This time, we used the same actuator setting in klog36362. Also, the Pitch DoFs were controlled by not High-Bandwidth filters but the original ones.

By the way, DSOFT input matrix value was applied for the original value because the new input matrix value seems to be not reproducibility before BPCs were engaged according to klog36341. So we decided to change the matrix value after BPC were engaged.

We succeeded in closing the loops of ARM ASC Yaw. Then we tried to engage BPC. However, ETMX BPC started oscillated at 49 mHz as soon as engaging BPCs (Fig.1). Originally, the cross over freq. between ASC for SOFT mode and BPC was set to 30 mHz. So we suspected the 49 mHz oscillation was caused by the change of the cross over frequency due to reduce the DC gain of SOFT mode ASC. Therefore, we reduced the BPC gain (BPC-YAW_{ETMX, ETMY}_INF_GAIN) to 0.1. Thanks to this adjustment, the 49 mHz oscillation seems to be disappeared. 

After that, we changed input matrix value to new one.

### IY NBDAMP implement to damp 1.14 Hz mode

After that, we tried to measured the OLTFs but the 1.14 Hz oscillated gradually (fig.2). At last, IFO was down. Although the IFO was down and ASCs were turned off, only ITMY kept to oscillate at 1.14 Hz. we found that only ITMY has no damping control for 1.14 Hz mode. So I implemented the damping control for 1.14 Hz as NBDAMP_Y3. Fig. 3 shows the TFs of NBDAMP_Y3. It seems to be work well.

Then, we engaged the ASCs. Unfortunately, 1.14 Hz oscillation seems to be still there (fig.4) and makes IFO down. We need more investigation.

### DHARD Y 43.2 Hz oscillation

Sometimes, DHARD Y oscilated at 43.2 Hz (fig.5). we implemented the notch fliter at this frequency. The 43.2 Hz oscillation was disappeared.

### OLTF measurements

After that, we measured the OLTFs. Fig. 6,7,8, and 9 show the TFs of DHARD_Y, CHARD_Y, DSOFT_Y and CSOFT_Y, respectively. The bandwidth of {D,C}HARD_Y, CSOFT_Y control seems to be from DC to 0.6 Hz and around 1.7Hz. On the other hands, DSOFT_Y could not be enlarge only to 0.3 Hz due to 1.1 Hz oscillation.

Also, we tried to increase DHARD_Y bandwidth more, but DHARD_Y started oscillated at 10-11 Hz when we doubled the gain.

I prepaired a datalogger and portable power surpplyers toword the  power outage on March 4th, near the OMC booth.

Next Friday, I will connect the permanent PEM snesors (SEIS, ACC@OMC-leg, MIC@OMC-booth), swtching from the KAGRA ADC (IY0) to them.

I checked the magnetic field data today. A 1/f noise is found in both X & Y direction, larger than tha outside.

I susupect a noise from the DAQ and the portable powersurpply, and moved them as far as possible.
The 1/f noise amplitude was not changed. So it may be an environmental one.

Upon Ushiba's suggestion we looked at the raw inputs the the REFL QPDs.  Sure enough the whitening stage on QPD2 segment 3 and 4 in both I and Q channels are not engaged.  See the attached plot.
Red green and blue on the top two plots show expected behaviour with the raw channel (IN1_DQ) being boosted at high frequency with whitenning. In the lower 2 plots red, green and blue (quadrant 4) are all overlapped showing no analogue whitening was engaged.  The same is true for quadrant 3.
It is strange that segments 3 and 4 on both I and Q channels did not engage, we will check tomorow.

[Kenta Tanaka, Dan Chen]

Before starting ASC work, the IFO failed to achieve stable lock.
Since PR3 appeared to be flactuated, we requested LSC_LOCK down and examined the behavior of PR3.

We confirmed that large and continuous oscillations occasionally occurred. The motion was observed in both P and Y directions. In some cases, the oscillations lasted for several minutes and then naturally subsided, but after some time they reappeared, repeating this behavior intermittently.

To investigate environmental vibrations around PR3, we checked the WIT sensor and accelerometer data at MCF and compared them with data taken during the health check on 2024/06/01. The comparison result files are saved at (figures attached. Refs in the fig are the data in thew past):

/users/Commissioning/data/VIS/PR3/2026/0225/

The comparison shows that the ground vibration level today is higher than that on 2024/06/01. However, since the observed PR3 motion appears and disappears intermittently, the relationship between the increased ground vibration and the PR3 motion remains unclear at this time.

Although the frequency of the large motion has decreased, it has not completely disappeared as of 16:00. The cause of the intermittent increase and decrease in the motion has not yet been identified.

Detail of the work:

The CMS circuit diagram can be found in document JGW-D1503567-v5.

The work has the following two purposes:
     1. Increasing the gain of the FAST path by a factor of 10, which allows us to turn off the 0.1 gain at the SLOW path and increase the actuator range of the laser PZT by a factor of 10.
     2. Adjusting the offsets at the input gain stages reduces glitches that can sometimes cause lock loss when increasing the laser power during gain switching.
Note that the detailed modifications for purpose 1 are reported in klog35396.
For purpose 2, trimmers have been installed on U10, U13, U14, U60, U61 and U65 (see the modification log in JGW-S1809466).

The following strategies are adapted for adjusting offsets at the input gain stages.

Basic ideas for adjusting offsets:

The input signals (Vin) are converted from differential to single-ended by U2, U3, U4 and U5, and U10.
During this process, offsets (Vofs_in) are added to the signals.
The gain stage then increases (or decreases) these signals according to the corresponding stage gain (gx) and adds some offsets intrinsic to the gain stage (Vofs_gx), where gx is a gain of x at the input gain stage).
These signals are monitored at the MIXER DAQ channels, which include the offsets (Vofs_ mon) at the monitoring circuit.
Therefore, the final outputs that we observe can be written as the following equations:
     Vout = gx (V_in + Vofs_in) + Vofs_gx + Vofs_mon
Considering the condition when IMC is locked, Vin + Vofs_in ~ 0 thanks to the feedback loop, so what we would like to achieve is Vofs_gx ~ 0 for all gx and Vin + Vofs_in ~ 0 without signals.
Note that, since the trimmer is only installed for gx = 16 dB and 8 dB, not all gains are adjusted, but this is acceptable because Vofs_gx becomes small when gx is small.

What I did:

To achieve the above, I adjusted the gain state using the following procedure:
     1. connecting the signals from RFPD (IN1) and CARM CMS (IN2), closing all laser shutters in the process.
     2. Measuring the Vout when the input switch is opened and closed with a gain setting of 0 dB.
     3. Adjust the offsets of U10 so that Vout does not change when the input switches are opened or closed. This adjustment achieves the condition of Vin + Vofs_in ~ 0 without signals.
     4. Measuring Vout when changing gx from 0 dB to 8 (16) dB and adjusting the offset of U14 (U13) so that Vout does not change when the gains are changed. This adjustment achieves the condition Vofs_gx ~ 0 for gx = 8 (16) dB.

Offset adjustment can be done for IN1 (signals from RFPD), but failed for IN2 since the input offset without signals is very large and the adjustable range is insufficient.
So, I decided not to replace the CMS at this time.

[Tanaka, Komori]

Abstract:

At this stage, we have mitigated the DSOFT Y issue by waiting for the BPC to stabilize and adjusting the input matrix of the X trans QPD combination.

Details:

We had been experiencing a significant phase delay in the DSOFT Y control loop, along with poor reproducibility after each IFO lock acquisition, likely due to strong dependence on the beam spot position.
Today, after allowing the BPC to stabilize, we measured the openloop transfer function using the input matrix values previously tested in klog:36329 (red).

We found that the phase delay was recovered to the same level as in the earlier measurement (brown), indicating that this input matrix configuration provides good reproducibility once the BPC is stable.

Therefore, at this point, we have decided to switch the ASC from the low-bandwidth mode to the high-bandwidth mode only after confirming that the BPC has stabilized.

I tried to replace IMC CMS to increase the PZT actuator range and reducing the glitches when increasing laser power by adjusting offset voltage at the input gain stage.
However, I found that it is difficult to adjust the offset due to the input offset is larger than the adjustable range, so I gave up replacing the CMs this time.

Detail will be posted tomorrow.

Note:

During the work, I accepted the SDFs shown in fig1 and fig2 in the down.snap to remember the epics value changes during my work.

We checked the cause of dip at 0.7 Hz in the {IX, IY} MN+TM actuator frequency response. First, we measured the TFs in the same three settings in the orignal post. Unfortunately, we could not reproduce the results, that is, the large dip around 0.7 Hz could not be observed in any settings (Fig.1).

I'm not sure of the reason why.

Note: This time, we found that TM actuators was saturated during the measurement in this measurement setting. So we adjusted the excitation setting not to saturate TM actuators during the measurement. This point is difference from the previous measurement. Thanks to this, the coherence above 0.3 Hz seems to be improved (fig.2).

Komori, Tanaka

We implemented high-bandwidth ASCs for all ARM modes ({D,C}{HARD,SOFT}) in the pitch direction. Fig. 1-4 show the OLTFs of DHARD, CHARD, DSOFT, and CSOFT, repectively. The control UGFs were enlarged up to 1-2 Hz.

Today, we did not implement the guardian script to engage these high-bandwidth ASCs automatically yet. We restored the setting related ASCs to the original ones.

[Alex, Carl]

We analysed the beam jitter measurements from 250-500 Hz. We found that beam jitter coupling was very close to DARM between 400-500 Hz however we believe this is an over projection and that better measuremennts with a higher SNR are needed. When we compared this coupling to DARM at 10W it seems that the coupling is different at higher power since the beam jitter does not limit sensitivity in this region.

We conducted measurements of the beam jitter in yaw using IMC PZT1 and PZT2 as drive and using IMC-REFL_QPDA*_DC_YAW_OUT_DQ as the witness sensors. This was done using a white noise injection with a bandpass filter from 250  to 500 Hz. Notched filters were used to reduce resonances of the PZT mounts. Both PZT1 and PZT2 had a resonance at 447 Hz and PZT2 had an additional resonance at 469 Hz.

The coupling for PZT1 (fig 1, 2 and 3) seemed to match DARM from 440-500 Hz including explaining a peak in DARM at around 490 Hz (fig 1 middle plot). However when we compare to DARM when the detector was operating at 10W and use the sensor noise from that time, the beam jitter overprojects darm (fig 1 bottom plot). This implies that the coupling function has changed since the detector operated at 10W, whether this is related to the change in power is unknown.

The coupling for PZT2 (fig 3, 4) is close to DARM but only touches darm at a few points.

Pitch has not been measured yet, we will measure pitch as well as remeasure yaw now that the sensor noise of the QPDs has been improved.

We used a combination of PZT1 and PZT2 on the PSL table to try to replicate the coupling function of the shaker 4 injection on the PSL table. We drove at 245Hz. The geometric factor for the coupling to be dominated by Mirror M17 would be about 0.25 while the observed ratio to replicate the coupling factor was about 0.025. This might mean coupling at 245Hz is dominated by a mirror closer to PZT1. It more likely means the coupling function (used here) is not accurate at 245Hz. This injection should be done closer to the 215Hz resonance frequency to confirm.

We used the PSL layout here. L8 and L9 are ignored for the moment. To achieve a beam rotation about a position X where PZT1 is at position A and PZT2 is at position B a gain of Ag = 0.4 and Bg = -0.4*(X-A)/(B-X) is applied to the drive to each PZT. If M17 is the suspect mirror the expected gain to match the coupling function with the shaker and the coupling function with the PZT would be -0.4*190/770 = -0.1.

A signal was injected at 245Hz with the drive split between PZT1 and PZT2 with this ratio (brown trace in the attached figure). This is compared to the shaker 4 injection (red) as witnessed by the IMC REFL QPDA1 witness of jitter noise. It is many orders of magnitude off. By reducing the PZT2 drive to 0.01 the coupling function could be made to match the shaker coupling function in magnitude and sign. This would equate to a beam rotation around a point 35mm from PZT1. M16 is the closest mirror at about 50mm from PZT1. 245Hz is too far from the observed resonance at 200Hz to produce a good coupling function by taking the simple ratio in diaggui. But, I think this is a proof of principle that the coupling function could be used to identify which mirror is moving in shaker tests.

The purpose of this test was mostly to prove the principle of driving 4 degrees of freedom of jitter (pitch, yaw, beam position and beam angle) with the pzts on the psl table. We are still trying to work out how to diagonalise beam position and beam angle.

[Alex]

There was an error in the coupling function estimates, the correct plots are attached. There are now sufficient coupling factor points that a total upper limit can be found for the 150 Hz point which is now overprojected onto darm. (The crosses show upper limits and the dots show coupling factor)

[Alex]

There was an error with the calculations due to confusing amplitude and power. The corrected plots are attached.

The coupling factor should have been ~81