Sheila, Ryan S, Camilla.

First we used the SEED beam to see if we were getting 00 flashes on the FCES IR camera (CLF is too dim). We did a mode scan with GR_SUS_LOCKED and a 'z step' as below on the VCO, as attached. We think that with the higher order modes a lot of the beam is clipping the edge of the FC2 AR coating and making the sizes of the peaks in the scan not representative of real powers.

z step H1:SQZ-FC_VCO_TUNEOFS '+0.01,1200' -s 0.5

Better Centering beam on FC1

We then, with the FC locked on green, used the template naoki.aritomi/Desktop/FC2_dither and adjusted the P2L and Y2L gains. With our inital sliders P2L was 8.0 and Y2L was 1.0 to minimize the injected peaks. We then stepped ZM3 in negative Pitch and then compensated with FC1 FC2 to get green FC power back up. These moves did increase power on SQZT7 IR PD from 0.3 to ~0.5. Ending sliders attached with P2L was 0.5 and Y2L was 1.5. So not completely centered, but close. O4 values were 0.8 and -1.6. See template here. We do not need to completely center green now as if we need to pico this will need to be redone. 

TITLE: 09/25 Day Shift: 1430-2330 UTC (0730-1630 PST), all times posted in UTC
STATE of H1: Corrective Maintenance
OUTGOING OPERATOR: None
CURRENT ENVIRONMENT:
    SEI_ENV state: CALM
    Wind: 18mph Gusts, 12mph 3min avg
    Primary useism: 0.03 μm/s
    Secondary useism: 0.27 μm/s 
QUICK SUMMARY:

• The EX models are down again, verbal shows the first errors at 14:24 UTC
• Microseism is increasing towards the 90th percentile, the wind is also moderate this morning and it's predicted to get worse, peaking at 2pm from windy.com
• Temps and dust look good

Summary

Toward optimizing the JAC injection mode-match at 62 W, the 2 W LDWFS data (LIGO-P2600491) taken before the EOM swap, after it, and after the IO_MB_L2 move were re-analyzed with a full Finesse model of the injection and REFL paths. Mode mismatch power at 2 W: 1.8 % → 5.8 % → 4.5 % for the three states, consistent with the REFL PD lock/unlock ratio (1.9 / 5.7 / 4.9 %), which didn't enters the fit. The current lens position did not improve as much as predicted earlier because that prediction used the symplified model, which misses the near-resonance of the RF sideband's first-order modes in this cavity. The Finesse model includes it and now reproduces all three states. The 62 W analysis and the lens solution derived from it come next.

Result at 2 W

The table gives the fitted injected beam as a deviation from the cavity eigenmode, both taken at the output of IO_MB_L3. The eigenmode there has a waist of 545 µm (YAW) / 551 µm (PIT) located 4.93 m downstream of L3. dw is how much larger the injected waist is than that; dz is the position of the injected waist relative to the cavity waist, positive toward the cavity (downstream), so the negative values below mean the injected beam focuses before the cavity waist. Values come from the joint fit over the three data sets; the errors are statistical.

• Mismatch (Finesse) is the carrier power the model puts outside the cavity TEM₀₀ at the fitted solution, which includes all of the higher order mode.

Measurement

Three data sets were taken at 2 W input, one per state of the injection path: before the EOM swap, after the EOM swap, and after the IO_MB_L2 move. Each set is a series of 30 s records (24 records, 12 for the last one).

Two angular actuators are dithered, each in both planes, at separate frequencies: JM1 at 9.7 Hz (PIT) and 11.3 Hz (YAW), the injection PZT mirror at 13.1 Hz (PIT) and 14.9 Hz (YAW). A length dither is applied to the JAC PZT at 809 Hz. In addition, during each record one pico motor sweeps the beam spot on one WFS head along one axis with a slow triangle wave (fundamental about 1.5 Hz), cycling through the four head/axis combinations over the series.

The recorded channels are the WFS quadrant I/Q outputs of both heads (H1:JAC-WFS_{A,B}_{I,Q}{1-4}_ERR_DQ). Offline, the amplitude of each angular line is read in the WFS signal (RF) and in the WFS signal demodulated once more at 809 Hz (DD: double demodulation), each normalized by the head's DD sum: 2 actuators × 2 planes × 2 heads × (RF, DD) = 16 line observables per data set.

The pico sweep adds, for each head/axis combination, the ratio RF/DD of the head's response to the spot motion. Being a ratio it needs no pico calibration, and it is a direct mode-mismatch observable: it measures the c₂ content (together with the detuning x₀) as seen at that head (4 observables). Each data set thus contributes 20 observables.

Analysis

• Step 1, coarse fit with the paper's analytic formulas. Fitting parameters are WFS Gouy phases, c₂ per plane, line amplitudes, and actuator gouy phases. It gives |c₂₀| / |c₀₂| = 0.093/0.055, 0.109/0.041, 0.073/0.044 for each datasets. The mismatch is 1.2 / 1.4 / 0.7 % which are quite small comparing to the predictions from the REFL DC PD. The current lens position was computed from this calculation, which is not as good as I expected.
• Step 2, Finesse fit seeded from step 1. The model consists of the three IO_MB lenses, the PZT and JM1 mirrors, the four-mirror of the JAC, the REFL lens (JACR_L1), the REFL BS, the two pico mirrors and the two WFS heads.
• Fit parameters: (1) injected beam per data set (dw and dz in each plane), (2) the angular excitation amplitudes for JM1 and PZT, (3) the dither depth, (4) x₀ (static detuning of the cavity), (5) the WFS Gouy phases, represented by a curvature on each pico mirror in each plane (four numbers), which also carries the REFL-path astigmatism.
• Joint fit: (5) is shared by the three data sets, (1) and (4) are per data set (pre-post EOM swap and after lens adjustment); 60 observables, 19 parameters, 42 dof. The single-set fits and the joint fit:

All parameters, errors and the per-record fits are in the detailed analysis note (to be posted).

Why the analytic model was off

• The paper's formulas treat the RF sidebands as fully off-resonant in every spatial mode. In the JAC that is not true for the first-order modes: the round-trip Gouy phase is 100.2° (bench measurement), so the HG₁₀/HG₀₁ resonances sit 100.2/360 × FSR = 41.3 MHz from the carrier, only 1.7 MHz (about 1.4 cavity linewidths) from the 43 MHz modulation frequency. The first-order content of the lower sideband is therefore partially resonant, and its contribution to the angular-line signals is missing from the analytic model.
• With the mismatch small (before the EOM swap, |c₂| about 0.1) the error is tolerable; after the swap (|c₂| about 0.2) the analytic fit underestimates the mismatch by a factor 3–4 and its Gouy phases drift (heads by up to 8°, the PZT reference by 10–20°). This is why the earlier lens plan, derived from that model, predicted more than the move delivered.
• The Finesse model carries the sideband modes explicitly (modes up to order 10) and needs no extra parameter to reproduce the three states, the REFL PD ratio and the scans.

Lens-move reconstruction (check)

Only IO_MB_L2 (and IO_MB_L1 by 3 mm) was moved between the 09-15 and 09-16 sets. Back-propagating the fitted 09-15 beam through the injection telescope and asking which L2 displacement reproduces the fitted 09-16 beam gives 18 ± 3 mm toward L1. The survey gives the L1–L2 spacing as 60 mm before and 26.5 mm after, i.e. an L2 move of 33.5 mm, so the error is ~1.5cm. This error is bigger comparing to the quoted error, which is statistical only (from the fit χ²); lens focal-length tolerances, the optic positions (even between PSL and HAM1) and any thermal lensing are not included. 

Next

• Repeat the analysis on the measured 62 W data (and the 10 / 25 W steps) with the same model; the thermal lensing between 2 and 62 W is the quantity to be fitted there.
• Derive the lens solution for 62 W from that fit. Already at 2 W the reach of the present lenses is limited: L3 can move only about 50 mm, and with L3 at that limit and L1+L2 moved together by 90 mm the mismatch would go from 4.5 % to about ~3 %. Anything better than that needs a lens change.

Figures

Figure 1: Beam radius along the injection path from the IO_MB_L3 output to the cavity, YAW (left) and PIT (right). Black: the cavity eigenmode (waist 545 / 551 µm, 4.93 m from L3). Colored: the injected beams from the joint fit for the three states. In every state the injected beam focuses before the cavity waist with a larger waist, and the two planes differ.

Figure 2: The 16 line observables per data set — RF (top) and DD (bottom) amplitudes of the four angular lines on WFS A (blue) and WFS B (orange), normalized by the DD sum — for the three states. Bars: data with 1 σ errors; black marks: the joint-fit model (one shared WFS layout, per-state injected beam and x₀).

At 19:39:56 PDT all the models on h1seiex stopped running after h1iopseiex detected an ADC timeout on the first ADC.

PCI bus scans can see all of the IO Cards.

Uptime of this crash was only 3hr 26mins suggesting a restart at 16:16 this afternoon. Trending the IOP CPU usage also shows an anomaly between 14:15 and 14:29

We will diagnose the issue and restart first thing in the morning.

2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: h1iopseiex: ERROR - An ADC timeout error has been detected on ADC 0, waiting for an exit signal.
2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: ligo_roce_mcast: ERROR - ib_poll_cq() - returned 128 results suggesting the buffer is not large enough.
2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: h1isietmx: ERROR - Waiting for ADC cycle 1242, live read is 62682, mm: 0, ioMemCtr: 1242
2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: h1hpietmx: ERROR - Waiting for ADC cycle 1242, live read is 62682, mm: 3, ioMemCtr: 1242
2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: h1isietmx: ERROR - An ADC timeout error has been detected, waiting for an exit signal.
2026-09-24T19:39:58-07:00 h1seiex.cds.ligo-wa.caltech.edu kernel: h1hpietmx: ERROR - An ADC timeout error has been detected, waiting for an exit signal.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu rts_awgtpman_exec[11050]: H1:HPI-ETMXIOP cycle timeout
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu rts_awgtpman_exec[11207]: H1:ISI-ETMX_CDMON_ST2_V2_I_IIOP cycle timeout
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu rts_awgtpman_exec[10903]: CONIOP cycle timeout
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1hpietmx.service: Main process exited, code=exited, status=1/FAILURE
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1hpietmx.service: Failed with result 'exit-code'.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1hpietmx.service: Consumed 11.418s CPU time, 42.4M memory peak.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1iopseiex.service: Main process exited, code=exited, status=1/FAILURE
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1iopseiex.service: Failed with result 'exit-code'.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1iopseiex.service: Consumed 10.982s CPU time, 76.6M memory peak.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1isietmx.service: Main process exited, code=exited, status=1/FAILURE
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1isietmx.service: Failed with result 'exit-code'.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1isietmx.service: Consumed 11.149s CPU time, 44.2M memory peak.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu cps_xmit[10899]: Closing client
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu cps_xmit[10899]: message repeated 2 times: [ Closing client]
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu cps_xmit[10899]: Closing out OpenMX and exiting
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-transport@cps_xmit.service: Main process exited, code=exited, status=1/FAILURE
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-transport@cps_xmit.service: Failed with result 'exit-code'.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-transport@cps_xmit.service: Consumed 4min 46.006s CPU time, 28.1M memory peak.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1hpietmx.service: Scheduled restart job, restart counter is at 1.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1iopseiex.service: Scheduled restart job, restart counter is at 1.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: rts-awgtpman@h1isietmx.service: Scheduled restart job, restart counter is at 1.
2026-09-24T19:39:59-07:00 h1seiex.cds.ligo-wa.caltech.edu systemd[1]: Started rts-awgtpman@h1hpietmx.service - Advanced LIGO RTS awgtpman: h1hpietmx.
 

Jennie W, Marc P, Fil C,

Summary: We still haven't found the issue with the segment 4 on QPD B and why it's noise is ~40 times higher than the other segments.

Marc and I checked (with a voltmeter) that the cable that takes the QPDB signal from chamber to the TIA variant 2 chassis is not connected to chamber ground and that it is connected to chassis ground at the other end that plugs into the TIA chassis.

The chassis input socket shield is also grounded to the rack ground through the front panel of the chassis.

To check that the high noise in segment 4 is not due to any problems with the shield, we plugged in a breakout board (which disconnects the cable shield from the socket shield) and checked the dark noise spectras again using the method and offsets detailed in alog #92010.

The high noise on segment 4 was unchanged by this.

This is making it look more likely that the problem is in chamber.

Tomorrow Marc and will use the breakout board to swap segment 3 and 4 at the TIA chassis input just to check this.

If the problem is only shown on the channel connected to segment 4 on the QPD then the problem is in chamber.

If this test shows that the problem persists on the same TIA channel then the next thing we might try is opening up the chassis to check if whitening is engaged for each quadrant (see D10001974 for wiring diagram of aLIGO ISC QPD Transimpedance Amplifier Primary Filter PCB). If the problem is only shown on the channel connected to segment 4 on the QPD then the problem is in chamber.

As a cross check I then turned off the measurement offsets in the QPD filter banks and re-opened the SPI shutter. Here is a graph of the spectra with light on the QPDs and you can see all four quadrants respond similarly for each QPD.

(Jordan V., Travis S., Gerardo M.)

Since controller for the large ion pump (IP12) at X-End had railed, see attached plot, and following a suggestion from others, yesterday we added a second controller to the large ion pump.  Now, both of the controllers are railed.  Let's see how long it takes them to recuperate.

In 91998 I showed a timeline of the satamp whitening swaps. I forgot to think about the fact that on the satamp chassis, there is a physical switch that must also be switched to Whitening ON (for self-explanatory reasons). There are also switches to turn the sum current feedback ON for each QOSEM channel. I've made a(nother) little timeline table.

Whitening status disappointments
The entire time that we had the 'Original' satamp whitening configuration, we had the whitening switch OFF. We then swapped the analog whitening to the first modification on July 13 (91003), but we still were not able to turn the whitening ON until probably July 15th at the earliest* (I didn't actually turn it on until July 22nd, but that doesn't matter any since we had already made the first set of mods). This means that we can't make a comparison of the BBSS M1 DAMP IN1 noise for the original whitening filter to compare to the other two later filter configurations. We can't start actually comparing data until after July 22. After this is when we had the whitening ON, Current Feedback ON, and were in our second configuration until early Spetember when we modified the satamp further.

* We had been advised that we keep the whitening OFF until we were in a much quieter state. That could've been late in the day on July 15th, when we were in partial vacuum and had the ISI in a nominal quiet state

We have started looking at the synthetic supersensors we created in seiproc for SPI commissioning and there are some things going on there I don't understand. For the 2 attached plots I look at 3 different versions of the table motion or loop error point, some calibrated GS13 CAL channels, the synthetic supersensors and the ISO loop inputs. For each subplot, red is the supersensor, blue is CAL CART GS13s, and green is the ISO in, first image is HAM2, second is HAM3, they weren't taken at the same time, so the spectra don't look exactly the same. I (probably naively) would expected all of these too look the same, with a caveat about sensor correction for the supersensor channels, the cps used there is before the sum with sensor correction, so the GS13 CAL and supersensor not matching  between .1 and a few hz  for X,Y and Z is because of that.  What I  (and I think Brian and Jeff) don't understand is why the ISO IN channels don't match. The input to the iso loops should just be the sum of the blended cps +senscor and gs13s. The blend filters I put in seiproc to construct the synthetic supersensor are the same ones we run nominally on the ISI. For the rotational degrees of freedom these should be the easiest to understand, but RX & RY for HAM2 and HAM3 are very different. Not sure what is going on. I had found one case of wrong sensors connected to filters in seiproc for HAM2 Z, but that has been fixed.

I have checked the signal routing out of the ISI master blocks to ipc senders to ipc receivers in seiproc and down to filter modules that generated the supersensors and haven't found any more errors, but it's a lot to look through. I have also spot checked that the right filters are installed in a number of places where supersensors don't look right, and have not found any errors so far there, either. I will try to keep looking. One of the next things I  will do is read all the filters in matlab and plot them there, to make sure that all the filters are right.

Louis and I spent the day working on getting MICH ASC running in DRMI.

First, we phased the AS RF36 WFS to maximize a BS length line in Q. Attached are screenshots. Louis and I found that for each 36 WFS, one segment was 180 deg out of phase of the others and needed a sign flip.

Next, we set out to find a good error signal for MICH P and Y. This was very easy for P, once we had the signals phased, we could see clearly that AS B RF36 crossed zero at the buildup max and moved appropriately with BS movements. We engaged the loop and saw the buildups improve and stabilize. We chose an input matrix value of 0.1 for AS_B_RF36 P. We engaged with an overall loop gain of 0.5, which is what Keita and I had set for PRMI.

Unfortunately, there is not a clear winner for MICH Y error signal, so we might need to combine a few signals to create a good error signal. Louis is going to work on a full sensing matrix.

However, we lost lock so we used this opportunity to go to PRMI and measure the MICH loop gains. We got a MICH P measurement, and the loop is very slow, sub 0.1 Hz. The design is for a 2 Hz loop, which would require a gain of 15. We bumped up the gain, but we're not sure if we want the loop to be this fast in PRMI. I put the gain back down to 0.5 for now, we can decide later if we want this loop to be faster in PRMI. The screenshot attached shows MICH P with gain of 15 in PRMI.

MICH ASC is still False in DRMI because we only have one of two working loops.

Other notes:

• the alignment has to be very good in DRMI to avoid SRCL mode hopping. hard to make this happen
• edited ISC DRMI guardian to disengage Not88 filters for sensing matrix in FM4 of all [LSC]2 banks, since they had been left on accidentally

Robert, Miranda, Alex

We mounted a temporary accelerometer to GV20 to continue with the noise investigations. To access the data from the channel, use H1:PEM-EX-ADC_4_01_OUT. For CDS purposes, it is plugged into channel 10 of the patch panel and into channel 9 of the power conditioner, and channel 2 of the ADC. Will attach photo of accelerometer tomorrow.

TITLE: 09/24 Day Shift: 1430-2330 UTC (0730-1630 PST), all times posted in UTC
STATE of H1: Corrective Maintenance
INCOMING OPERATOR: Ryan S
SHIFT SUMMARY: DRMI commissioning continued, more GV20 investigations, and SQZer alignment work. Some SEI model issues at the end of the shift.
LOG:

I started the morning with an modified manual alignment following along the lines of what was done the other day alog91911.

• PRX, adjusting PRM to maximize AS_A
• MICH_DARK adjusting the BS to minimize AS_A and the camera spot
• SR2_ALIGN looked well converged
• SRY adjusting SRM, I touched SR2 by less than a microradian to further maximize AS_A

GVs 1 & 2 were opened today, I did a quick trend of the ITMs oplevs and top mass pointing (Y1 cursors are before the GV closed 1473628152 to be specific, Y2 are the current values at the time I looked), its pretty close.

• 17:03 UTC ETMX ISI ST1&2 WD trip from Scaffolding work
• 17:41 UTC GV1 soft close, 17:46 UTC GV is open
• 17:53 UTC GV2 soft close, 17:58 UTC GV2 is open
• 21:28 UTC the SEI_EX model froze, EJ restarted it but DK and DAQ errors returned
• 21:38 UTC ETMX HEPI and ISI ST1&2 WD trip, if we tried to untrip HEPI it would retrip
• 23:13 UTC EJ restarted the model again

J. Kissel, O. Patane

The QOSEM and analog CDS team have been adjusting the analog whitening filter applied to the X, Y, and SUM signals of the QOSEM satamps; Oli has a nice summary in LHO:91998.

I wanted to compare the raw ADC voltage for the QOSEM signals under the most recent whitening configurations; 
    - the current whitening filter parameters ( {z:p} = {0.0144 : 2.89} [Hz], the so-called "1st modification") 
    - the previous filter parameters ( {z:p} = {0.263 : 2.89} [Hz], the so-called "2nd modification")

Here I attach the ASDs of the raw ADC voltage (comes in as ADC counts, and I've converted it in DTT to ADC volts using a gain of 40/2^16 [V/ct]) compared against the expected typical 16-bit ADC noise from T0900450 for the X DOF and the SUM.

Observations/Conclusions
    (1) Even with much less whitening, the X signals are still above the ADC noise floor where the sensor is actually measuring suspension motion. Good!
    (2) In the {z:p} = {0.0144 : 2.89} [Hz] filter configuration, the SUM signals used to sit a factor of 2-3x above the ADC noise floor at all frequencies. Unclear if that's because the SUM was ADC noise or real LED intensity noise (if that's what limits the SUM, or if that's the "signal" of the SUM). There are some differences in shape between 0.1 to 10 Hz from channel to channel, so my guess is that it's LED intensity noise (or, again, whatever's limiting the SUM, or whatever the actual "signal" is in the SUM).
    (3) In the {z:p} = {0.263 : 2.89} [Hz] filter configuration, the SUM signals are all identical in shape, falling as 1/f^(1/2), and -- somehow -- surpassing the ADC noise floor above 30 [Hz]. I don't understand this.
    (4) The F1 QOSEM SUM voltage is saturating the ADC constantly. As such, there's no amplitude spectral density, because the number 2^15 doesn't have noise. That stinks.

What next?

    (A) Without having as much knowledge about the QOSEM system as Tom, I'm trying to understand how to calibrate normalized spot position, s_X or s_Y, in [ct/ct = V/V = radians] into physical beam spot and thus flag displacement, x or y, in [meters] or [microns]. In LLO:81079 and LHO:90154 aLOGs, 4000 [ct/m] (or (1/4000) [m/ct] = 0.00025 [m/ct] = 250 [um/ct]) is tossed around "citing Tom Roocke" -- and these calibration gains are applied to the normalized sum, so the units should be [m / (ct/ct)] or [m/rad] or something like that. 

I want to understand that number so that we can reverse engineer adding the ADC noise floor to any QOSEM noise budgets. If I understand how to calibrate a QPD (like derived in, e.g. LHO:91210), then
 
       x = sqrt(pi/8) * w * s_{x}
            and
       y = sqrt(pi/8) * w * s_{y}

where  
        - sqrt(pi/8) is the "centered beam" approximation to 1/erf(sqrt(2)) from integrating the Gaussian beam, and
        - w is the beam spot radius at the lens surface

So we need to know the spot radius at the lens surface.
        
Then, we can propagate the noises from individual segments (i.e. ADC noise) into a QPD's normalized spot displacement units, if we know the whitened SUM's ASD, S, or the SUM's DC voltage, S_{DC} and the whitening filter frequency response, W, as I expect the conversion from ADC voltage noise is something like 
        
        n_ADC,X = n_ADC,Y = n_ADC * sqrt(pi/8) * w * sqrt(2) / (W * S_{DC})

where 
        - n_ADC is a single segment's ADC voltage, 
        - sqrt(pi/8) is the "centered beam" approximation to 1/erf(sqrt(2)) from integrating the Gaussian beam, 
        - w is the beam spot radius at the lens surface
        - sqrt(2) is the "centered beam" approximation to sqrt(2 + 4*s^2), with s as either the X or Y normalized spot position signal
        - W is the whitening filter (in [V/V]), and
        - S_{DC} is the DC ADC voltage

    We know the whitening filter response, and (except for F1 :-( ) we know S_{DC} in ADC [counts], so easy to convert to ADC [V]. We again, just need to know the beam spot radius.

    Hopefully Tom has this number in his design documentation.

    (B) We're still trying to understand why the QOSEM noise floor is so much worse than expected. Tom's ruled out a lot of noise sources, but I'd yet to see ADC noise on any of his noise budgets. So, if we can understand where the ADC noise floor lies with respect to the signal -- especially as we reduce the amount of whitening, then we can either rule out ADC noise as the issue or put it on the suspicion list. These plots show that the X signal is NOT limited by ADC noise, but there's definitely something confusing about the SUM signal. So this should be understood.
    

Jennie W, Jeff K, Keita K,

Summary: Scattering nosie from IFO couples into QPD A, QPDB segment four has larger dark noise b y factor of ~40 compared to other segments.

After we broke the ground loops on the TIA chassis inputs last week I redid the QPD dark noise measurements.

First I measured the average value of each quadrant input on both QPDs using diaggui to take a triggered time measurement withe the shutter open.

Then I closed the shutter and set the QPD_{A,B}_SEG{1,2,3,4}_OFFSET values to the average value for all four QPD A segments (15050 counts) and all four QPD B segments (16750 counts).

Then I measured the dark noise using the template in /ligo/home/jennifer.wright/git/2026/SPI/20260922_SPI_dark_noise_ASD.xml

The first measurement looked like it had some glitches on QPD A so Jeff got me to check the time series. At his suggestion we checked for coherence with ASC_AS_C (anti-symmetric port of H1) and it seems to be correlated. The AS port was flashing through fringes hence the fringing seen on the time series.

Zooming in this becomes more obvious.

I unlocked and mis-aligned the IMC and unlocked the JAC to stop scattered light through HAM 2 and 3 while I retook the measurement.

The QPDA dark noise now looks better but QPD B segment four is still significantly noisier than the other quadrants.

Looking at the time series this noise is obvious also.

Keita and I then checked the raw signal inputs before the anti-whitening filter on the QPD segments and these still show this noisier segment four. NB: this measurement was done with the IFO back to locking so scatter is again visible in the QPD A spectrum. The noise is flat, which suggests that the whitening stages in the TIA are not engaged so its not a whitening gain problem.

We then did a test with the SPI unshuttered and the dark offsets back to nominal. QPD B segment 4 now looks not that different to the others, which suggests that the four segments do not have wildly different electronic gains.

Try swapping chassis for TIA or disconnectiong input and measuring output ASD for TIA.

Next steps: 

- Measure TIA output spectra from each quadrant on QPD B with input unplugged.

(Travis S., Jordan V., Richard M., Gerardo M.)
After the installation of CHeTA viewports, we connected an ISP-1000 scroll pump and a SS-500 cart to each of the main turbo pumps at the X beam manifold and Y beam manifold.  After getting power sorted out to the ISP-1000s, thanks Richard, we started the pumpdown for both of the manifolds.
Something that we did noted was the new gauge for PT-180 shows almost 1000 Torr for pressure at atmosphere, this is not correct, then the other gauge for PT-170 shows 750 Torr for pressure at atmosphere, also not right.
Pumpdown has been stopped at the end of the day, and will be restarted tomorrow morning by Jordan.

Attached is a plot of both manifolds initial pumpdown.

After closing up BSC9 yesterday we started pumpdown of the EX volume this morning using the two mobile ISP1000 scroll pumps.

There was minimal overpressure in the system, so I could not get a good dewpoint measurement of the blow down air.

It took ~6 hours to rough the system down from atmosphere to ~500 mtorr, where we start the turbopump.

The cooling water was adjusted on the turbopump stand so that the pump and controller pressures were <= 40C. Once the turbo was at full speed and the inlet pressure ~5e-5 Torr, I switched the backing pump to the dedicated ISP250 on the turbo stand.

Setpoints were adjusted to 5E-2 Torr on Channel 1 (Foreline) and 5E-5 Torr on Channel 2 (Turbo inlet).

We will need to leak check the TMDS port gate valve since it was removed to verify the gate o-ring was in place.

Also, the EX cleanroom was powered off at ~3:10 pm local time.