Hirose, Tanaka

DRMI LSC oscillated at 70-80 Hz whenever POP90 and AS34 power decreased.

In pararell of Hirose-san's work in klog37320, I looked into the trend of build up sideband powers and the oplev signals. Fig.1 shows the time series of the build up sideband power (K1:LSC-POP_PDA2_R90_I_ERR_DQ, K1:LSC-AS_PDA2_R90_I_ERR_DQ), and mirror oplevs' signals (K1:VIS-{IMMT2,PRM,BS,SRM}_TM_WIT_{L,P,Y}_DQ). The drop rate of build up power seems to be about 0.1 Hz. It seems to be close to the Type B IM YAW resonant frequency. According to the time series, ITMY, SRM and BS seems to oscillate at around 0.1 Hz

I measured the coherences between POP90/AS34 power fluctuations and mirror alignment fluctuations. Thanks to the stable lock of DRMI, (almost 3 hours), we could measure it with long time. Fig. 2 shows the result. There seems to be large coherences around 0.1 Hz between AS34 and ITMY L, Y, SRM Y, PRM P, Y, IMMT2 P. Basically, SRM, PRM, IMMT2 are controlled by ADS. So it maybe fine. On the other hands, ITMY seems to has larger coherence than ITMX. So we need suppress the ITMY Yaw motion.

There seems to be some concerns. 

1. coherence around 1-2 Hz between PR3 alignments and AS34. I'm not sure of the reason why PR3 alignment has coherence with only AS34. One of the possiblities is the SRCL alignment is not good rather than the PRCL alignment.
2. the coherence around 0.2-0.3 between SR3 and AS34.
3. coherence peak around 0.45 Hz between AS34 and PR mirrors' motions and BS L (Type Bp resonant?)
4. broad coherence bump from 7 Hz to 70 Hz between PRM L and AS34 (PRCL control?)

Anyway, we need to modify the suspensions' local controls

[Aritomi, Tanaka, Hirose]

This work is a continuation of klog37318. We try to perform WFS for DRMI. So we start the preparation for WFS.

Our goal is to improve the alignment by switching the alignment control from ADS to WFS while the interferometer is in the DRMI1f lock state (with ADS IMMT2, PRM, SRM).

As a preparation for the WFS commissioning, we introduced dark offsets for the DC, I, and Q signals of the WFS QPDs. The applied offset values are summarized in Fig. 1-4.

After introducing the offsets, we enabled the DC centering loops(Fig. 5).  We created the ENGAGE_WFSDC_DRMI" state pathing DRMI ADS on the ASC_LOCK guardian. However, the error signal of AS_QPDA2 Yaw showed no response to the PZT feedback, suggesting that the signal path may not be properly connected. In addition, REFL_QPDA3 appears to receive little or no light.

We also performed a rough excitation of the BS to obtain preliminary responses.(Fig. 6)

For the next session, we will first check the cabling of AS_QPDA2 Yaw and REFL_QPDA3, and then repeat the measurements after confirming that the signals are properly connected.

## Summary of 2026.07.31 and 2026.08.04 works

(2026.07.31)

We redesigned ADSs for DRMI. First, we decided the dither frequency in order to solve the degeneracy betwen other DoFs. We prepared the frequency set per 2 Hz, which is above 1 Hz, typical suspension's resonant frequency, 4.125, 6.125, ...., 30.125, 32.125 Hz. This is because we want to see the sideband from alignment fluctuation until 0.5 Hz around one dither frequncy. So it is necessary to set the frequncy interval between each dither frequency which is used as PIT and YAW of one mirror.

Then, we measured the spectra of POP90 and AS34 with no exciations to look for the frequency region, which the only noise floor is in. There are some peak in the region from 14 Hz to 20 Hz in POP90. From this results, we decided the frequencies for ADSs as below. This time, we decided to use POP90 for PRM, BS, IMMT2 dithering and to use AS34 for SRM dithering.

• BS PIT: 22.125 Hz, YAW: 24.125 Hz
• PRM PIT: 4.125 Hz, YAW: 6.125 Hz
• SRM PIT: 26.125 Hz, YAW: 28.125 Hz
• IMMT2 PIT: 30.125 Hz, YAW: 32.125 Hz

We implemented the bandpasse filter at each corresponded frequency to K1:ADS-{PIT,YAW}_{IMMT2, PRM, SRM, BS}_DEMOD_SIG filter bank. and performed the phasing for each signal from alignment dither. This time, we decided the dither amplitude so that the sidelobe around the dither frquency could be seen largely enough.

After that, we implemented the 1.5 Hz cutoff lowpass filter, comb filter for corresponding frequency and its harmonics, and 2 Hz notch for the beat note between PIT and YAW in K1:ADS-{PIT,YAW}_{IMMT2, PRM, SRM, BS}_DEMOD_SIG_I, which is a filterbank for a demodulated signal.

Then, we tried to engage ADSs. PRM seems to work because POP90 increased and there seems to be harmonics of diter frequency, 8.25 Hz and 12.25 Hz in POP90. IMMT2 and SRM seems to work but their harmonics could not be seen in POP90 and AS34, respectively.

On the other hands, BS ADS seems not to work. ADS loop itself can be close but the POP90 and AS34 power decreased on engaging ADS.

During the work, we modified the ENGAGE_ADS_FOR_DRMI in order to engage current ADSs except for BS (I hope Hirose-san will report soon).

(2026.08.05 work)

In daytime, DRMI lock was too unstable to proceed the ADS. We suspected the seismic motion around 0.1 Hz disturbed the lock because the 0.1 Hz seismic motion got 1- order larger than the one in last Friday (klog37313) and BS and SRM IM YAW motion at 0.14 Hz, which is IM YAW resonant frequency, seems to become 1-order larger due to the large micro seismic. We implemented the resG filter for 1.4 Hz of FM5 in K1:VIS-BS_TM_OLDAMP_YAW, which was already implemented. It seems to reduce the Yaw motion at 0.14 Hz. However, DRMI lock was still unstable.

We observed two kinds of lock loss pattern in daytime. One is the oscillation at 20-30 Hz. Sometimes, DRMI LSC began oscillated at some frequency between 20-30 Hz. We measured the OLTFs. But the phase margins around their frequncies of DRMI LSC loop seem to be large enough even if the only overall gain was changed by some reasons. So something might change not only gain but also the phase of loop. Moreover, this oscilltion kicked the ADS, especially IMMT2 and SRM because their dither frequencies are close to oscillation frequencies and this kick changed the alignment them largely and often made the relock difficult.

Another pattern is the sudden lock loss. Sometimes, POP90 and AS34 dropped to 0 within less than 1 second. At that moment, any alignment seem to be not changed

In night time, DRMI lock seemed to become stable for some reason. Especially, DRMI was stable without ADS rather than with ADS. I attempted some trials to improve the stabilty, inclunding to review of ADSs.

• Decoupling the PRM motion from MICH error signal.
  • As reported in klog, I tried to decouple the PRM length motion from the MICH error signal with the same procedure of PRM-to-SRCL. I remeasured the MICH eror signal response when PRM length displacement was excited, including to the reprocebility of the check of PRM-to-SRCL decoupling factor.
  • I applied the obtained decouplig factor for PRCL-to-MICH but it seems not to work.
  • I also confirmed the decoupling factor of PRM-to-SRCL seemed to be almost the same. It seems to be good.
• Review of DRMI ADS
  • Since DRMI was stable without ADS rather than ADS, I reviewed the ADS
    • PRM
      • If PRM actuators is not optimized in terms of eular DoFs or if the beam spot on PRM is off from the center, the alignment excitation couples to length motion. On the other hands, if cavity length is tuned, this length motion should not be seen at the dither frequency. But if not so, the length fluction also contaminates in the alignment signal. Current DRMI lock is performed with some offsets in their error signals. But these offsets are not optimized. Therefore, I started to check whether the cavity length tunes or not by exciting the length.
      • I excited the PRM L displacement from K1:VIS-PRM_TM_ISCINF_L_EXC at the PRM PIT dither frequency, 4.125 Hz. Although PRCL control UGF is ~30 Hz, I could see the length excitation in POP90 and PRCL error signal. In this state, I changed the PRCL1 offset from -4.7 by 0.5 step. When the offset value was 0, the peak of PRM Length excitaiton disappeared from POP90 though it was in the PRCL error signal.
      • In this state, I switched the excitation from length to alignments and performed the phasing for PRM ADS again. Also, I found that the CLKGAIN = 30 cnts seems to be large enough to see the sidelobe. The phase seems to be almost the same. By the way, I tried to implement the resG filter for dither frequency in PRCL but the phasing is not changed.
    •  IMMT2
      • Also, as for IMMT2,  I performed the phasing. Before this, I found that there was the small peak at 30 Hz in POP90 without excitation. So I changed the PIT dither frequency from 30.125 Hz to 34.125 Hz. As for YAW, the domod phase was not changed so much (~several degrees)
    • I implemented ADSs of PRM and IMMT2 after re-phasing Then, POP90 are slightly improved (fig.1). The harmonics peaks of PRM dither frequencies (PIT -> 8.25 Hz, Yaw -> 12.25 Hz) also got slightly bigger (fig.2). 
    • SRM
      •  As Aritomi-san and Yokozawa-san already reported in klog37315, SRM phasing value seems to be not reproducibiltiy, the phase easily changed more than 90 degrees. 
      • I implemented the resG filter for SRM dither frequencies in the SRCL loop but the situation seems to be not changed.
      • Also, I tried to tune SRCL as the same procedure of PRCL but the AS34 fluctuation seems to be too large to adjust the offset.

During ADS work, we found that there was no anti-aliasing filter in AS34 and there was also no lowpass filter in normalization filter. we implemented them. (klog37314) 

### Next

• investigate SRM and BS ADS good configuration
• (give up ADS and move to WFS for DRMI?)

We measued the current TM coils' status with the health check script. Fig.1,2,3,4 show the results. They seem to be healthy compared with 2024.Dec. results by Takahashi-san in klog31919.

## (2026.08.05 early moring work)

I tried to implemented the subtraction of the PRM motion from the MICH error signal accoriding to the previous work in the original post.

The PRM coupling in MICH from of fig. 2 in original post is 0.45 (180 deg). However, the coherence seems not to be good, less than 0.7 unlike the PRCL-to-SRCL case. Anyway, I implement the value -0.45*10*(-1, phase =180 deg) as LSC_INPUT_MTRX. Also, the offset value in MICH1 was adjusted, -7.5 +4.7/10*4.5 = -9.6. In this situation, we can lock DRMI but the MICH error signal spectra and coherence MICH-to-PRCL seems to be not changed. So it seems not to work.

[Kenta, YamaT]

This is yesterday's work.

Add 16kto2k filter for AS34 at PD filterbank
When we checked all RF PDs to trace the signal path for the future calibration work, we found POP90 has "16kto2k" filter at FM9 of PD filterbank but AS34 doesn't. These two PDs are used for power normalization in LSC and the reference signal of ADS. Sending AS34 without AA filter to 2kHz ADS model makes aliasing noise on the error signal of ADS, so we added same "16kto2k" filter at FM9 of AS34 PD filterbank. According to the comparison plot of AS34 signal with and without AA on k1ascads, AA filter makes ~2 times improvement of noise floor around 30Hz (near dither frequency). Orange curve in Fig.1 represents original AS34_I on k1lsc. Green and blue curves represents AS34_I on k1ascads with and without AA filter, respectively. Note that, blue curve (without AA) is simulated one with original signal and offfiline signal processing (simulated filterbank, phase rotation, power normalize and decimation) because there is no proper DQ channel on k1ascads for checking aliasing noise and I forgot to take a spectrum data befor applying AA.

Add LP1 before power normalization of error signals
We also noticed MICH and SRCL error signal were divided by raw (non-lowpassed) AS34_I as the power normalization. (PRCL error signal is divided by low-passed POP90_I.) Since this implementation seemed to simply introduce noise derived from high-frequency components, we also added "LP1" filter (same as POW_NORM_POP90_I_FILT) at FM1 of POW_NORM_AS34_I_FILT.

Aritomi, Hirose, Tanaka

we decoupled the PRCL motion from SRCL 1f error signal to expand the lock duration. The coupling seems to be decreased by this decoupling. Also, we succeeded in transitting to 3f signals with decoupling factors. At least, the duration of 3f lock seems to be increased to 1 hour.

## what we did

The 1f lock duration seems to be too short to implement ADS. So we tried to improve the duration.

First, we found that PRCL-to-SRCL coupling in 1f signals seems to be large. Fig. 1 shows the spectra of DRMI 1f error signals (top) and coherences between them (bottom) in the DRMI 1F LOCKED state with ADSs except for BS. Coherence between PRCL and SRCL in the 30-70 Hz region seems to be large. We attempted to decouple PRCL and SRCL so that SRCL error signal (POP17I) could not see the PRM motion.

We excited PRM and SRM at 150.125 Hz, respectively and measured the ratio between the PRM motion and the error signal response at 150.125 Hz. Fig.2 and Fig.3 shows the results of PRM and SRM, respectively. The signals in K1:LSC-POP_PDA1_{RF45_I, RF17_I, RF17_Q}_NORM channels are calibrated to nano-meter unit. From the fig.3 result, when SRM was excited, the PRCL error signal(POP_RF45_I) did not see the SRM motion. This results seem to be reasonable because RF45 does not enter SRC thanks to the schnupp asymmetry. Also, the MICH error signal, POP_RF17_Q, did not see the SRM motion thanks to the phasing in klog36990. Therefore, it seems not to be necessary to subract SRM motion from PRCL or MICH error signals.

On the other hand, when PRM was excited as fig.2, the SRCL errof signal, POP_RF17_I see the PRM motion rather than PRCL error signals, This results are also expected because RF17 resonates in the coupled cavity consist of PRC-SRC due to the schnupp asymmetry. So we tried to decouple PRCL-to-SRCL by subtracting the PRM motion.  From the top pane of the Fig.2 right panel the ratio of (calibrated POP17I)/(PRCL error) is +2.58. So we applied the value, -2.58*10(=POP_RF17Q calibration factor) = -25.8 in LSC_INPUTMTRX to subtract the PRM motion from SRCL error signal, and adjusted the K1:LSC-SRCL1_OFFSET value (SRCL offset:-3.2, PRCL offset: -4.7, then, -3.2 + (-4.7/10*(-25.8))) = 8.9 so that SRCL control keeps the resonance point of SRCL. 

We tried to implemented this value into the matrix during the lock but the lock was down. So we input the 1/10 of the value at first, and increased the value to the nominal one, step by step. Fig. 4 shows the spectra and coherences of error signals when the matrix value was changed. Fig. 5 shows SRCL error signal and coherence PRCL and SRCL. I wrote down each measurement setting of each color line in the legend of top panel of fig.5. As you can see, the spectra and coherence around 20-90 Hz decreased as the MTRX value became close to nominal value= -25.8. Also, we found that BS dither can be seen in the PRCL-SRCL coherence at their frequency.

Moreover, there seems to be gain peaking around 140 Hz. So we measured the OLTF of SRCL. Fig.6 shows the result. The OLTF gain when MTRX value was -15 seems to increased. So we decreased the overall gain to -6dB. Then, the gain peaking around 140 Hz seems to be disappeard. Furthermore, in low frequency region, the OLTF gain shape seems to be improved.

After that, we tried to transit to 3f signals with decouplinf factor of PRCL-to-SRCL. We succeeded in transitting to 3f signals. Fig.7 shows the spectra of error signals and coherences before (without REF in label, measured in 2026/07/30 in klog37297) and after (with REF in label, measured in 2026/08/03) decoupling. As for 3f signal seems not to be changed so much.

By the way, Current 3f lock duration is at most 1 hour (fig.8)

[Shaojin, Washimi]

We located the portable seismometer setup at the outside of the X-end and started measurements.

K-Log#37272

Two PCAS server instances were being launched for each device, so I removed the redundant server instance.
The affected devices are Picomotor, Stepper (VIS and CRY), and Traverser.
I verified that Picomotor and Stepper (VIS) are working properly. The code for Stepper (CRY) and Traverser has also been updated. 
Their operation will be verified the next time they are used.

[Before]

[After]

I offloaded the BF GAS with the FR.

Ushiba, Hirose, Tanaka

We tried to implement ADSs for DRMI DoFs (BS, PRM, SRM, IMMT2). PRM, SRM, IMMT2 controls seem to work but BS control seems not to improve the alignment in terms of POP90 optimization.

I will report the detail later.  

[Tanaka, Ushiba, Hirose]

As we work on redesigning the DRMI ADS, we updated the new DRMI ADS information in ASC_LOCK Guardian.
The changed states are ENGAGE_ADS_FOR_DRMI and DOWN.

In the ENGAGE_ADS_FOR_DRMI state, the following parameters are modified: oscillation frequency, demodulation bandpass filter, demodulation phase, INFGAIN, and I-phase filter.
In the DOWN state, the changes made in the ENGAGE_ADS_FOR_DRMI state are reverted.

I'll write more details later.

More pictures on 7/28: link

After installing the pylons, we had planned to place a waterproof cover over them. However, we forgot to do this during the work.
We will install the waterproof cover at the next available opportunity.

I found that REFL DC power decreased and AS DC power increased after the transition (fig.1). It indicates that some length DoFs seems to be detuned or ADS has some offset in 3f locks. We need more investigation 

Ushiba, Hirose, Tanaka

We succeeded in transitting from 1f signals to 3f signals for the DRMI lock just after initial alignment. Maybe we can lock DRMI with 3f sensors with good alignment. Current lock duration is less than 10 mins. Robust ASCs for all of DRMI DoFs are necessary. 

## What we did

(Conclusively, following work seems not to work for some reasons...)

• We remeasured the sensing matrix for DRMI 3f signals with the good configration as reported in klog37295. Fig.1,2,3 show the results of PRCL, SRCL, and MICH, respectively. There seem to be good coherences at the excitation frequency, 150.125 Hz
• From these results, we calculated the sensing matrix with the following procedure.
  • We assumed that PRM motion contributes only PRCL, and also, the SRM motion contributes only SRCL. On the other hands, the BS motion changes the distances of BS-IX and of PRM-BS, that is, BS changes not only MICH but also a half of PRCL and a half of SRCL. Moreover, the direction of PRCL change by BS is inverse from other two DoFs. 
  • Therefore, each length changes delta L_{PRCL, SRCL, MICH} can be written by using each mirror displasement delta x_{PRM, SRM, BS}
    $\begin{pmatrix} \delta L_{\mathrm{PRCL}} \\ \delta L_{\mathrm{SRCL}} \\ \delta L_{\mathrm{MICH}} \end{pmatrix} = \begin{pmatrix} 1 & 0 & -0.5 \\ 0 & 1 & 0.5 \\ 0 & 0 & 1 \end{pmatrix} \begin{pmatrix} \delta x_{\mathrm{PRM}} \\ \delta x_{\mathrm{SRM}} \\ \delta x_{\mathrm{BS}} \end{pmatrix}$
  • we can derive the relation between the cavity length change and sensor response by using measured response from a mirror displacement to a sensor H
    $\begin{pmatrix} S_{\mathrm{51I}} \\ S_{\mathrm{51Q}} \\ S_{\mathrm{135I}} \\ \end{pmatrix} = \hat{H} \begin{pmatrix} \delta x_{\mathrm{PRM}} \\ \delta x_{\mathrm{SRM}} \\ \delta x_{\mathrm{BS}} \\ \end{pmatrix} = \hat{H} \begin{pmatrix} 1 & 0 & -0.5 \\ 0 & 0 & 0.5 \\ 0 & 0 & 1 \\ \end{pmatrix}^{-1} \begin{pmatrix} \delta L_{\mathrm{PRCL}} \\ \delta L_{\mathrm{SRCL}} \\ \delta L_{\mathrm{MICH}} \\ \end{pmatrix}$ 
  • Then, we can obtain the sensing matrix S,
    $S = \left[ \hat{H} \begin{pmatrix} 1 & 0 & -0.5 \\ 0 & 1 & 0.5 \\ 0 & 0 & 1 \\ \end{pmatrix}^{-1} \right]^{-1} = \begin{pmatrix} 1 & 0 & -0.5 \\ 0 & 1 & 0.5 \\ 0 & 0 & 1 \\ \end{pmatrix} \hat{H}^{-1}$
• After that, we used the K1:LSC-{CARM,XARM,YARM}_IN1 channels as virtual 3f sensors for {PRCL, SRCL, MICH} by using the obtained sensing matrix, respectively. And we normalized the each virtual sensor by the response of the corresponded DoF. Then, we measured the responses between mirror displacements and virtual sensors. Fig.4, 5, and 6 show the results. Their virual sensor could see the motions of corresponded mirros. Therefore, they seems to be decoupled.
• Then, we tried with the decoupled 3f sensors by the guardian, one by one. Before this, we modified the LOCKING_DRMI_3F state so that the guardian input the values of sensing matrix to LSC_INPUT_MATRIX. However, PRCL began to oscillated.

(Following work seems to work.) 

• Previously, as for PRCL and SRCL, we succeeded in transitting 1f sensors to 3f sensors just by using REFL51I for SRCL and REFL135I for PRCL (klog37228). So we tried to check the reproducebility of that results. At first, we adjusted the gains and signs of REFL135I and REFL51I by using the ratios of RF135I/PRCL 1f error and of REFL51I/SRCL 1f error. And we decided the offset of 3f signals by using the DC values when DRMI was locked with 1f signals. Then, we tried the transition, one by one. PRCL was succeeded in the transition. SRCL was also suceeded in the transition but the lock seems not to be unstable.
• After the lock was failed, DRMI lock seems to become difficult even with 1f sensors.  Actually, POP90 seems to be not already maximum because BS alignments are not controlled from this moring. So we assumed the alignment already became worse. We decided to perform initial alignment.
• After the initial alignment, we prepared the state for the transition (DRMI_TRANSITION_TO_3F) in the VERTEX guardian. Then, PRCL and SRCL transition to each 3f sensor with their offsets were succeeded by the guardian. The lock seems to be stable. Then, we tried to transitting MICH control from 1f to 3f sensor (REFL51Q) with the offset. It failed but we observed the error signal became far from the setpoint. So we flipped the sign and retried it. Then, we succeeded in transitting MICH control from 1f to 3f. 
• Fig.7, 8, and 9 show the measured OLTFs of PRCL, SRCL, MICH with 3f. The UGFs seems to be the same as the ones with 1fs.
• Current lock duration of DRMI 3f seems to be less than 10 mins. We need robust ASCs for all of DRMI DoFs to expand the duration.

I subtracted the dark offset from REFL_RF51 and REFL_RF135 signals.
New offset vales can be seen in fig1.

[kTanaka, Ushiba]

Conclusion:

Measurement with time span of 64 s (corresponding to a bandwidth of 0.015625 Hz) with the ADS gains increased by a factor of 3 from their nominal values, except for PRM yaw, appears to provide good results for the sensing matrix measurement.

Detail:

We reviewed the sensing matrix measurement results today (klog37289) and found that the coherence was very low for some degrees of freedom (DoFs).

To investigate the cause, we measured the RF PD spectra while exciting the BS with an amplitude of 30000 counts from ISC_INF.
We found large side lobes located approximately 0.05-0.1 Hz away from the excitation frequency.
Since the sensing matrix measurement had a frequency resolution of 0.125 Hz, these side lobes were not sufficiently separated from the excitation lines.

To mitigate the side lobes, we implemented a second-order boost filter in the MICH filter bank, extending the bandwidth from 1 Hz to 0.1 Hz.
However, the situation did not improve significantly.
Therefore, the side lobes do not appear to originate from residual MICH motion.

We then investigated whether the side lobes were caused by angular motion by examining the OpLev signals and the ASC-MICH_{P,Y} signals.
Since the ASC-MICH_{P,Y} signals exhibit peaks around 0.09 Hz, it is likely that the side lobes originate from angular motion.
Furthermore, the coherence between the ASC signals and the OpLev signals is high for SRM, PRM, and IMMT2, all of which are controlled by ADS.
Therefore, we suspected that the ADS control was producing the peak around 0.09 Hz, which in turn generated side lobes in the LSC signals when the BS was excited at 150.125 Hz.

To address this issue, we increased the ADS gains for IMMT2 pitch/yaw, PRM pitch, and SRM pitch/yaw by a factor of 3 in order to increase the ADS UGFs.
Since the PRM yaw loop became oscillatory when its gain was increased by a factor of 3, the PRM yaw ADS gain was left unchanged.

Figure 1 shows the RF PD spectra measured while DRMI was locked and the BS was excited at 150.125 Hz.
The coherence between the MICH and RF PD signals is high, indicating that the measurement quality is improved compared to the previous measurement.
When comparing the new results with the previous measurement, several values changed by significant factors, suggesting that the previous measurement may not be accurate sufficiently.
Therefore, it would be preferable to repeat the sensing matrix measurement.