NAOJ GW Elog Logbook 3.2

Aritomi, Yuhang
It was reported in elog2130 that there is discrepancy between measurement in the path for QPD1 and QPD2.
This is due to the implementation is not the same with design. Due to the space limitation, QPD2 is not placed in the correct position. Therefore, we should move QPD2 further three holes.

As reported in entry #2117, the large offset in WFS2_Q3 is due to a broken channel (n°14) in the Anti-Aliasing board (AA1). It can be solved by using some spare AA channels but so far we considered that we won't use Q signal for AA. So we decided to wait untill the BNCtoDsub issue (that is likely to cause this kind of problem) is solved.

[Aritomi, Matteo]
After we locked both of galvo loops, we checked WFS2 signals.
We found that WFS2_Q3 had large offset (~4000) compared with others(<10). Even when we put 50 Ohm to DGS input of WFS2_Q3, WFS2_Q3 is still very large. So this problem seems to come from DGS.
To avoid using WFS2 Q phase signals, we injected 12Hz pitch line in input mirror and tried to minimize the 12Hz peak in WFS2_Q1 (maximize WFS2_I1) by changing the LO demodulation phase from DDS. The measured result is as follows.
DDS demodulation phase (deg) | WFS2_I1 | WFS2_Q1 |
70 | 0.031 | 0.008 |
80 | 0.034 | 0.010 |
90 | 0.047 | 0.008 |
95 | 0.040 | 0.010 |
100 | 0.012 | |
110 | 0.019 |
The original DDS demodulation phase is 90 deg and it is already good demodulation phase. So as long as we use WFS2 I phase, this will not be a problem.

What I did
I re-connected Q-mass to the flange and tightened the bolts in order to improve the vacuum level.
This improved the vacuum level as before.
Then I tried to measure the residual gas by Q-mass.
Q-mass, hoever, showed a fatal error as shown in fig. 2 --- filament 2 defect.
Actually I could not measure the residual gas as shown in fig. 1 though I could do it yesterday...
It's like "yak shaving".
Next Step
I will open the chamber and install two temperature sensors with indium sheet.
In addition, I will adjust the optics and then close the chamber and do some measurements.

What I did
Today I checked the operation of Q-mass which was installed on the cryostat.
It seems that Q-mass can measure some data as attached.
However, the pressure was not good enough to measure reasonable data by Q-mass.
I tried to improve the vacuum level, but could not reach below 0.3 Pa with Q-mass.
Notes
At first, there was a leakage around the tube between the cryostat and Q-mass which prohibited vacuum evacuation, though Q-mass operation requires less than 10-2 Pa.
Actually, it could not reach below 20 Pa.
This was due to the poor workability around Q-mass.
The bolts were not tightened well.
I somehow tightened them and the vacuum level was improved but not enough.
Next Step
I will try one more time to measure the residual gass by Q-mass.
After that, since indium sheet arrived, I will install temperature sensors and adjust the input optics and then start cooling down the cavity.

[Aritomi, Matteo]
Recently we had a problem that QPD2 DC centering loop cannot be locked.
Today we found that when we move the green beam horizontally, QPD2 DC signal somehow moves vertically and vice versa. This means horizontal and vertical correction signals for QPD2 galvo should be swapped. After we swapped the horizontal and vertical correction signals, QPD2 DC centering loop locked stably even when the green is reflected from filter cavity. Maybe the cabling of QPD2 is wrong.
This issue is originated by the fact that the servo galvo and the DGS are following two different convention for the WFS segment order:
The galvo servo from initial TAMA is following the convention in pic 1 (take from this document), while the DGS is following the convention on pic 2 (KAGRA convention).
The cables from the WFS DC segments have been connected into the galvo servo followig the convention from KAGRA/DGS and this resulted in a swap between X and Y error and correction signals.
We decided to keep the current configuration to be coherent with the DGS beam position monitor. Note that for WFS1 the cable for the correction signal for X and Y were originally inverted by mistake.

[Aritomi, Yuhang]
We measured homodyne AR reflection and it is around 4 uW while LO power is 840 uW. It means we have 0.5% loss for each homodyne PD. In total, we have 1% additional loss from homodyne AR reflection.

I re-installed Q-mass in order for checking of operation.
Tomorrow, I will evacuate the chamber and confirm the operation of Q-mass and vacuum gauge.

Aritomi and Yuhang
According to elog 1659, the simulation result of beam size and gouy phase for AA system is shown in the attached figure 1. We could see that there is only 1000mm lens and QPD1 is located close to beam waist while QPD2 is located far from beam waist.
The measurement was done recently to check the real beam size. The measured points and beam diameter is shown in the attached figure 2. The measurement done for QPD1 (9 points) is indicated by 'cross' wihle the measurement done for QPD2 (3 points) is indicated by 'point' 21,22,23.
Since the beam size should be symmetric for two path after BS, the 'cross' and 'point' can be imagined to be located in a single beam. These measured ponits are fit with gaussian beam as shown in figure 3. The measurement done for QPD1 and 2 seems to have a wrong distance, we will check that.

[Aritomi, Yuhang]
Yesterday we tried many galvo servos for QPD2 DC centering, but no galvo servo can lock QPD2 DC centering loop. WFS2 DC signal either oscillates or goes away when we tries to lock.
We found that QPD2 DC signal moves a lot compared with QPD1 (attached movie). This may be due to larger beam size at QPD2 (farther from waist position). We measured spectrum of QPD DC signals without DC centering loop (attached screenshot). QPD2 DC signals are larger than QPD1 by a factor of ~2.
We also characterized beam size of green around QPD1 and QPD2 (Yuhang will report).

Today, I soldered another temperature sensor.
Then I tried to install the Q-mass, but it was not successful due to the weight of Q-mass...
Anyway, I will order indium sheet or Apiezon N Grease in order for thermal contact.

[Aritomi, Yuhang]
We measured sensing matrix of WFS1. We injected a line to pitch and yaw of input mirror. The frequency of the injected line was 12 Hz and the amplitude was 2000. Then we measured WFS1_PIT and WFS1_YAW signal.
First we injected a line to pitch of input mirror and measured 12 Hz peak of WFS1_I_PIT and WFS1_I_YAW with demodulation phase R = 90 deg. With R = 90 deg, WFS1_I and WFS1_Q signals were almost the same. The measured 12 Hz peak is as follows.
WFS1_I_PIT | 0.25 |
WFS1_I_YAW | 0.038 |
This means we have about 10% coupling from pitch to yaw of input mirror. This may be due to coupling of coil-magnet actuator driving.
Then we optimized the demodulation phase R in order to maximize WFS1_Q_PIT signal. The optimal demodulation phase for WFS1_Q_PIT was R = 20 deg. After that we injected the same line to yaw of input mirror and measured WFS1_I_YAW and WFS1_Q_YAW. Measured sensing matrix of WFS1 for pitch and yaw with demodulation phase R = 20 deg is as follows. Screenshots of the 12 Hz peak for PIT and YAW are also attached.
I phase | Q phase | |
WFS1_PIT | 0 | 0.48 |
WFS1_YAW | 0.07 | 0.47 |
The optimal demodulation phase R = 20 deg for pitch was not exactly optimal for yaw, but it seems we can use WFS1 Q phase signal for both pitch and yaw.
We'll replace galvo servo for QPD2 and measure sensing matrix of WFS2.

What I did
I replaced the vacuum gauge which was repaired by the company as attached picture.
Then I installed a temperature sensor.
First, soldered the connector to the controller.
Then I soldered a sensor to the cable inside the cryostat.
After that, I checked the temperature which was displayed on the controller.
The value was about 430 K and was strange.
I suspected that there was a contact failure at some point.
Eventually, it was due to the poor connection between the cable and the pin.
After solving this problem, the displayed value was reasonable as shown in the attached picture.
Next step
I will install another temperature sensor and Q-mass.
And then I will start cooling down.

Shimode-san, Miyakawa-san
After experiencing the failure of AA channels and PSD reported in entry #2096. We shipped the PSD and its power supply to Kamioka where Shimode-san investigated the circuits.
His report is reported below (the japanese version is in the attachments):
1. Features of 100V of NAOJ power supply
3pin Earth terminal: Earth line connected to cold and grounded into earth through the power distribution. (Pic 6. "100v_Earth.jpg")
3pin Cold terminal: Power output connected to Earth on the power distribution. (Pic 5. "100v_Cold.jpg")
3pin Hot terminal: Power output not connected to Earth on the power distribution. (Pic 7."100v_Hot.jpg")
* Note that the tester voltage is based on the Earth (rack) of the building. (pic 3. Earth_point.jpg)
2. Features of connection between PSD and power supply
(1) The PSD chassis is connected to the power supply chassis by a shield on the power connection cable. As a result, the PSD chassis and the power supply chassis have the same voltage.
(2) The NAOJ power supply's GND (black) is isolated from the power supply chassis. Therefore, the PSD chassis and the PSD board are at different voltage levels from the PSD's GND.
(3) The voltage difference between the NAOJ power supply's GND and the 3pin 100V power supply's ground terminal is about 21V! or 1V depending on the way to plug to AC100V power (forward and reverse). (pic.1, pic.2 "100vON_A.jpg" "100vON_B.jpg)
3. GND in each chassis.
(1) About PSD output
As shown in "PSD_connector.jpg", the outer of LEMO is GND because of the unbalanced output. If conversion to BNC without connecting to chassis GND, the voltage at "2. (3)" is applied to both positive and negative.
(2) BNC to Dsub chassis
BNC is from parallel input to Dsub parallel output. GND is not connected in the chassis. (See https://gwdoc.icrr.u-tokyo.ac.jp/DocDB/ 0047/D1604782/001/BncDsub_ADC.pdf )
(3) About AA/AI Chassis
The 5pin of Dsub INput in the AA/AI chassis is not connected. (The AA/AI chassis is designed to use the power supply's GND connection.)
(See https://gwdoc.icrr.u-tokyo.ac.jp/DocDB/ 0099/D1909971/001/ADC_AA_interface.pdf )
(See https://gwdoc.icrr.u-tokyo.ac.jp/DocDB/ 0099/D1909969/002/AA_AI_Filter.pdf )
* The new AA/AI Filter Board has a 5pin connection, but the 5pin is disconnected on the interface board.
4. From the above
The GND among PSD → BNC-Dsub → AA/AI Interface Board → AA/AI Board is not connected. As a result, a large voltage difference can happen between [GND of PSD] and [GND of AA/AI circuit].
It is presumed that the very high common-mode voltage originating from the power supply (from noise around) is applied to the AA/AI board and the first stage ICs are all damaged.
In addition, I guessed that the high voltage from the power supply, GND and Earth went to AA/AI input, but the same things can happen even if different voltages are applied to the power supply chassis, power supply cable or circuit configuration chassis due to other factors.
5. Solution.
Temporal solutions: (1) Pull the GND line from the GND (black wire) inside the power supply, and connect to the GND(0V) of DC power supply supplying the AA/AI along the signal line. (2) Connect the ground wire between the "PSD chassis + NAOJ power supply chassis" and the "AA/AI chassis".
Fundamental Solutions: (1) Use 15V power supply boards AEL made in all chassis. The power line should be along the signal line. (2) Connect the chassis and the rack to the GND (0V) of the power supply. (3) Connect the ground through wire between the PSD chassis and the AA/AI chassis. Connect them with a relatively thick wire. (4) Insulate the PSD from surrounding metal objects.

Matteo, Yuhang, Eleonora
We measured the QPD2 electronic noise with spectrum analyzer using 32 dB amplification.
Pic1: Amplified QPD2 noise compared with 32 dB amplifier noise and spectrum analyzer.
Pic2: Same plot of entry #2121 where we added QPD electronic noise. Note that the data were divided by a factor 40 to subtract the 32 dB amplication used for the measurement. We didn't plot data below 1Hz as we assumed they are dominated by the spectrum analyzer noise.
We verified that the QPD measurement doesn't change when we added a bandpass filter (https://www.minicircuits.com/WebStore/dashboard.html?model=SIF-70%2B) before the 32 dB amplifier.

Yamamoto-san, Aso-san, Yuhang, Eleonora
Thanks to Aso-san's help Guardian PC is now connected to the TAMA DGS network. We used a USB-> ethernet adaptor since the system was not recognizing the ethernet port.
Aso-san solved also a problem caused by the fact that space on the disk was completely occupied by log files generated by an error.
Then Yamamoto-san could remotely access the PC and set up the guardian. Below a memo of his work:
##################
Today, we have some progress about guardian deployment in TAMA. Thanks to the help from Yamat. The detailed information can be found in the #filter_cavity channel of gokagra in Slack. Here, I put a summary of my activity today.
1. Yamat helped us check simulink file and found no problem.
2. Yamat suggested to run caget K1:FDS-FC_GR_TRA to check potential issue on client workstation of desktop1 and k1grd0. Running it on desktop1, I got -27.7212 for K1:FDS-FC_GR_TRA. But running it on k1grd0, I got channel connect timed out for K1:FDS-FC_GR_TRA.
3. Yamat suggested to check environment values in guardian computer. To do that, I used 'env > tama_filter_cavity_k1grd0_env_out.txt' to save the environment variable of k1grd0 workstation. Then I used 'scp tama_filter_cavity_k1grd0_env_out.txt controls@192.168.11.110:/home/controls/Desktop' to copy environment variables from k1grd0 workstation to desktop1 workstation. To share with Yamat on Slack, I firstly uploaded the txt file to dropbox and then share link with him. (Note that I didn't just take screenshot because the environment variable information is large.)
4. After Yamat checking env output, he found that a variable called 'EPICS_CA_ADDR_LIST' seems to have a wrong IP address. He suggested to change this from '/home/controls/.bashrc or /kagra/apps/etc/client-user-env.sh.' and reboot k1grd0.
5. I found the channel 'EPICS_CA_ADDR_LIST' is in /kagra/apps/etc/epics-user-env.sh. So I did the modification and reboot k1grd0. But the problem is not solved.
6. I found after modifying from /kagra/apps/etc/epics-user-env.sh, the env output in k1grd0 is still as before and didn't change. Maybe this is why the problem is not solved. Now I am waiting for the answer from Yamat.

Yuhang, Eleonora
Pic.1 shows the QPD2 (seg 1) signal for FC (oplev OFF), aquired with DGS and spectrum analizer respectively for 1.5 mW sent on QPD
Pic.2 shows the QPD2 (seg 1) signal for FC (oplev OFF), aquired with DGS and spectrum analizer respectively for 2.5 mW sent on QPD
We checked that the QPD dark noise is below the spectrum analyzer noise.
# About green power to FC
In order to have more light on QPD we increased the light injected into FC by incresing modulation amplitude of AOM/ ( from 2.5dBm to 5.5dBm).
The green refleced light when FC is locked is about 1/4 of the input light. The reflected light is splitted by a BS (92:8) and 92% is send to QPDs and 8% is sent to FC lock PD. (note that an additional attenuation of a factor 3 is present before the PD)
The usual amount of light sent to FC is ~14.5 mW, the maximum we can send is ~25 mW. The limitation is due to 78 MHz EOM damage threshold.
Here the two cases corresponding to the shown measurement:
Injected into FC |
TOT reflected |
to each QPD | to lock PD(before attenuation) |
14.5 mW | 3.6 mW | 1.5 mW | 0.3 mW |
25.5 mW | 5.8 mW | 2.5 mW | 0.5 mW |

Yuhang, Eleonora
We installed 12 dB attenuator on the LO of the QPD2 demodulation board, as we did for QPD1 demodulation board 1 (see entry #2112)
We performed the same test and confirmed all the channels are working fine.
The LO was -6dBm, we notice that if we put additional attuantion the demodulated signal is slightly increased. For a LO of -21 dBm the demodulated signal reaches the maximum amplitude and it is a factor 1.2 higher that in -6dBm case.
Anyway we will split -6dBm LO between the 2 demodulation boxes. The actual LO for each board will be -12dBm which is fine.

Yuhang, Eleonora
We report three additional failures of Anti Aliasing (AA) channels.
Symptom: ADC channel reads large, almost constant offset (order of ~1000 counts). The offeset is still present if we disconnect cables towards AA board -> it is not coming from the BNC2Dsub converter. The offset is reduced when we switched off AA board and disappear when we disconnect AA from ADC.
This time the broken channels are:
- Ch.10 of AA0 used for INPUT YAW OPLEV (We realized it while charachterizing input oplev on 13/07)
- Ch.16 of AA0 used for END L SUM OPLEV (not currently used, we know it is broken since several months)
- Ch.14 of AA1 board, used for WFS2 Q3 (We realized it while charachterizing demodulators on 10/07)
For the moment I rearranged the model to used spare Anti Aliasing channels for INPUT YAW OPLEV (chan 9-12 of AA0 moved to chan 25-28 of AA1).
We shoud understand the origin of such problem and solve it. According to Miyakawa-san It might be due to a ground-connection issue in the design of KAGRA Bnc2Dsub converter JGW-D1604781 https://gwdoc.icrr.u-tokyo.ac.jp/cgi-bin/private/DocDB/ShowDocument?docid=47

Eleonora and Yuhang
To have an idea about how much amplification is required for demodulated signals from QPD, we compared the QPD demodulated signal with DGS and network analyzer.
The test is done by taking signal from QPD2 segment1 with green beam reflected form the locked filter cavity. The beam was centered with galvo on the QPD and power was ~1.5mW
In the first attached figure, the in-phase demodulated signal (measured in DGS/network analyzer) is shown together with DGS ADC noise. We see that the measurement in DGS and network analyzer agree with each other. However, in the measurement of DGS, the peaks are barely higher than the ADC noise.
In the second plot, there is a comparison of demodulated in-phase/in-quadrature signals.
The large peak at 88 Hz is due to galvo loop, we adjusted the gain of galvo servo and it is reduced.
As reported in entry #2117, the large offset in WFS2_Q3 is due to a broken channel (n°14) in the Anti-Aliasing board (AA1). It can be solved by using some spare AA channels but so far we considered that we won't use Q signal for AA. So we decided to wait untill the BNCtoDsub issue (that is likely to cause this kind of problem) is solved.