DARM low frequency (< 4 Hz) is highly coherent with ETMX M0 and R0 L damping signals. This might just be recoil from the LSC drive, but it might be worth trying to reduce the L damping gain and see if DARM RMS improves

Bicoherence is also showing that the noise between 15 and 30 Hz is modulated according to the main peaks visible in DARM at low frequency.

We might be circling back to the point where we need to reconsider/remeasure our DAC noise. Linking two different (and disagreeing) projections from the last time we thought about this, it has the correct slope. However, Craig's projection and the noisemon measurement did not agree, something we never resolved.

Projection from Craig: https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=68489

Measurement from noisemons: https://alog.ligo-wa.caltech.edu/aLOG/uploads/68382_20230403203223_lho_pum_dac_noisebudget.pdf

I updated the noisemon projections for PUM DAC noise, and fixed an error in their calibration for the noise budget. They now agree reasonably well with the estimates Craig made by switching coil driver states. From this we can conclude that PUM DAC noise is not close to being a limiting noise in DARM at present.

To Chris' point above -- we note that the PUMs are using 20-bit DACs, and we are NOT using and "DAC Dither" (see aLOGs motivating why we do *not* use them in LHO:68428, and LHO:65807, namely that [in the little testing that we've done] we've seen no improvement, so we decided they weren't worth the extra complexity and maintenance.)

If at some point there’s a need to test DAC dithers again, please look at either (1) noisemon coherence with the DAC request signal, or (2) noisemon spectra with a bandstop in the DAC request to reveal the DAC noise floor.  Without one of those measures, the noisemons are usually not informative, because the DAC noise is buried under the DAC request.

Attached is a revised PUM DAC noisemon projection, with one more calibration fix that increases the noise estimate below 20 Hz (although it remains below DARM).

Late entry.

On Wednesday I had the opportunity to investigate the failure of the annulus ion pump, IFO was out of lock.  The controller for the annulus ion pump did not show signs of power, the front display lights were off, see photo.  Replaced the controller, and the system started working again.  However, the signal for the current was a bit unusual, on a plot of the past 10 years the signal never reached such a low level, see attached plots.

And today the AIP railed high.  Pump to be changed next Tuesday.

DARM_IN1(attachment1) saw some sort of glitch(left data tip) ~170ms before DARM registered the lockloss starting(right data tip).

ETMX L2 and L3 also saw this glitch(attachment2). Seems like the glitch hit DARM_IN1, then ETMX L3, then ETMX L2?

I noticed something similar with the 10/15 08:53UTC LL(attachment3), but in that case it looked like the movement was first seen in ETMX L3, then DARM_IN1 and ETMX L2. In that instance the glitch also happened a full 0.5s before the lockloss.

03:25 Back Observing

Very interesting find Oli! I've attached some more plots to hopefully help us narrow down where this is coming from. The L2 ETMX fastimon and noisemon channels see this same glitch, but the L2 OSEM sensor inputs on all of the quads don't. This makes me think that it could possible be coming from the OMC DCPD signals, the error signals forLSC-DARM. Tough to say based on data rates, but the second attachement also maybe shows the DCPD signal moving first, but take that with a grain of salt.

On the bad alignment of the green to the OPO shown in Camilla's photos of the scope:

We normally apply an offset to PZT1, which we were not doing this morning.  This offset impacts the cavity alignment, as you can see in Vicky's screenshots from April 2022.  62856

Here are some past alogs showing an OPO cavity scan with the green transmission.  

Dec 2022: 66527  (we should do a repeat of a scan like this now, with lowered green power and a slow scan, to see if things have really degraded). 

April 2022: 62691

Sept 22 we swapped the CLF collimator.

With this new crystal position, we have more squeezing in DARM, at first briefly measuring ~4.8 dB SQZ at 2kHz! It has since settled to ~4.2dB in Observe. As Camilla said, this is with the 2nd spot from the left edge (5th from the right). The optimal OPO temperature will be changing as it settles into this new spot, so there we may need to tune co-resonance temperature over the next few days to re-optimize.

Just before injecting SQZ to the interferometer, with OPO trans = 80uW, NLG was measured as 13.15. We took some sqz/asqz/meansqz loss measurements, and will do subtraction using the following times. Measurements using dtt cursors:

• SQZ, ~4.8 dB
  • 1381612908 - 1381612998
• No SQZ
  • 1381613053 - 1381614325
• Anti-sqz, 14.0 dB
  • 1381616445 - 1381616576
• Mean-sqz, 11.0 dB
  • 1381616609 - 1381616740
• After 22:33:40 UTC (gpstime 1381617238), we went back to Observe with SQZ at this spot.

Loss analysis and subtraction to follow.

NLG variation over an early ~6min of operating at this new spot is interesting. I think we see the crystal's green losses increasing rapidly. In the screenshot, the OPO was locked in green (~80uW green transmitted power, but no intensity stabilization so green power varied).

• We see the crystal's non-linear gain decrease by ~11.7% over 6 minutes for fixed input green power (NLG = (max_amplified/unamplified) ; the top red trend shows decay of the maximum amplified seed power. The maximum amplified seed power can be found by optimizing the OPO crystal temperature, but this decay of NLG was not recoverable by optimizing crystal temperature (bottom trend = attempts to optimize opo crystal temp).
• NLG decay does look consistent with the decay of transmitted green power over ~6 minutes. The decay of green transmission for fixed input green power could result from fast crystal losses/absorption/etc in green. This lowered green pump power, results in a lower non-linear gain for red/sqz. That is, I think this NLG decay can be explained by fast green losses in the opo crystal, and doesn't necessarily require fast variations in red (i.e. these fast green losses don't have to also be fast red sqz losses).

edited to include: NLG measured ~7 hours later in the lockloss. After ~4 hours with stabilized green trans = 80uW, the opo co-resonance changed from 31.638 degC to 31.617 degC, and we recovered the same NLG after tuning the temperature, so the interaction strength (probably also the red losses) did not degrade much in this time.

edited to include darm sqz 15 min after relocking: SQZ reached 4.8dB again after relocking; the darker bottom SQZ trace is taken ~15 minutes into this next lock. This is operating at the new spot for ~8 hours.

For these squeezed darms at the new crystal spot, here's adding in subtracted sqz data for these times of anti-squeezing, squeezing, and mean-squeezing (i.e. LO loop unlocked, sqz phase is unlocked from the IFO beam and spinning through all sqz angles).

I checked H1:OMC-DCPD_SUM_OUT_DQ and don't see the comb in any of the three listed intervals (neither state A nor B). Tested with a couple of SFT lengths (900s and 1800s) in each case.

Since it seems that the coupling is NOT a direct electronics coupling from Beckhoff -> OM2 -> DCPD, we fully connected the Beckhoff cable to the OM2 heater driver chassis and locked the OMC to the shoulder with an X single bounce beam (~20mA DCPD_SUM, not 40mA like in the usual nominal low noise state). That way, if the Beckhoff is somehow coupling to OMC PZT that might cause visible combs in the DCPD.

We didn't see the comb in this configuration. See the 1st attachment, red is the shoulder lock and green is when 1.66Hz comb was visible with the full IFO (the same time reported by Ansel in alog 73000), showing just two largest peaks of 1.66Hz harmonics visible in the green trace. (It seems that the 277.41Hz and 279.07 Hz peak are 167th and 168th harmonics of 1.66Hz.) Anyway, because of the higher noise floor, even if the combs are there we couldn't have seen these peaks. We've had a different comb spacing since then (alog 73028) but anyway I don't see anything at around 280Hz. FYI I used 2048 FFTs for both, red is a single FFT and the green is an average of 6. This is w/o any normalization (like RIN).

In the top panel of 2nd attachment, red is the RIN of OMC-DCPD_SUM_OUT_DQ of the shoulder lock, blue and dark green are RIN of 2nd loop in- and out-of-loop sensor array. Magenta, cyan and blue green are the same set of signals when H1 was in observing last night. Bottom panel shows coherence between DCPD_SUM during the shoulder lock and ISS sensors as well as IMC_F, which just means that there's no coherence except for high kHz.

If you look at Georgia's length noise spectrum from 2019 (alog 47286), you'll see that it's not totally dissimilar to our 2nd plot top panel even though Georgia's measurement used dither lock data. Daniel points out that a low-Q peak at around 1000Hz is a mechanical resonance of OMC structure causing the real length noise.

Configurations: H1:IMC-PWR_IN~25.2W. ISS 2nd loop is on. Single bounce X beam. DCPD_SUM peaked at about 38mW when the length offset was scanned, and the lock point was set to the middle (i.e. 19mA). DC pointing loops using AS WFS DC (DC3 and DC4) were on. OMC QPD loops were not ON (it was enabled at first but was disabled by the guardian at some point before we started the measurement). We were in this state from Oct/17/2023 18:12:00 - 19:17:20 UTC.

BTW Beckhoff cable is still fully connected to the OM2 heater driver chassis. This is the first observation data with such configuration after Fil worked on the grounding of Beckhoff chassis (alog 73233).

Detchar, please find the comb in the obs mode data starting Oct/17/2023 22:33:40 UTC.

The comb indeed re-appeared after 22:33 UTC on 10/17. I've attached one of the Fscan daily spectrograms (1st figure); you can see it appear in the upper right corner, around 280 Hz as usual at the start of the lock stretch.

Two other notes:

• The comb is now back to its original spacing of 1.6611 Hz.
• There are new strong lines visible at +/- 1.235 Hz from the comb teeth. If this structure has appeared before, I'm not aware of it. I've attached an image showing the lines (2nd figure, average of DMT-ANALYSIS_READY data between 22:00 10/17 and 02:00 10/18) and a comparison with older data from Sept 20th (3rd figure).

Just to see if anything changes, I used the switchable breakout board at the back of the OM2 heater driver chassis to break the thermistor connections but kept the heater driver input coming from the Beckhoff. The only two pins that are conducting are pins 6 and 19.

That happened at around Oct/18/2023 20:18:00 to 20:19-something UTC when others were doing the commissioning measurements.

Detchar, please look at the data once the commissioning activities are over for today.

Because there was an elevated noise floor in the data from Oct/17/2023 18:12:00 mentioned in Keita's previous comment, there was some doubt as to whether the comb would have been visible even if it were present. To check this, we did a direct comparison with a slightly later time when the comb was definitely present & visible. The first figure shows an hour of OMC-DCPD_SUM_OUT_DQ data starting at UTC 00:00 on 10/18 (comparison time with visible comb). Blue and yellow points indicate the comb and its +/-1.235 Hz sidebands. The second figure shows the time period of interest starting 18:12 on 10/17, with identical averaging/plotting parameters (1800s SFTs with 50% overlap, no normalization applied so that amplitudes can be compared) and identical frequencies marked. If it were present with equivalent strength, it looks like the comb ought to have been visible in the time period of interest despite the elevated noise floor. So this supports the conclusion that the comb *not* present in the 10/17 18:12 data.

Following up, here's about 4 hours of DELTAL_EXTERNAL after Oct 18 22:00. So this is after Keita left only the heater driver input connected to the Beckhoff on Oct/18/2023 20:18:00. The comb is gone in this configuration.

Screenshot summarizing homodyne measurements today. With measured carrier NLG=11 (for generated squeezing ~14.7-14.8 dB), we observe

• squeezing           = - 6 dB
• anti-squeezing    = +13.5 dB
• mean-squeezing = +10.5 dB

Comparing sqz/anti-sqz to generated sqz: ~7% unexplained homodyne losses. This is consistent with our last estimate of excess HD losses (8/29/2023, LHO:72802, ~7% mystery loss). Since then, we swapped the HD detector and improved readout losses (visibility). We now measure more homodyne squeezing at 6 dB, consistent with expected loss reductions. That is compared to 8/29 (LHO:72802), we have less total loss, less budgeted hd loss && more squeezing, but the same unexplained hd losses as before.

Comparing mean-squeezing to generated sqz: could be consistent with sqz/asqz losses. I think there is a mis-estimate of the generated squeezing level from non-linear gain. If we ignore our NLG11 measurement, and instead use the generated squeezing level to match observed 13.5 dB of anti-squeezing, then we allow losses to determine the measured 6 dB squeezing level, we would have an NLG=10 (not 11) for a generated squeezing level of 14.5 dB. This would suggest 7% unexplained losses, same as the sqz/asqz measurements.

For ~7% mystery losses, this is compared to total HD losses of 21%, of which we budget 15% losses. From the sqz wiki, the budgeted losses are:

• opo escape 98.5% 
• in-chamber ham7 95.3%
• beam diverter 99%
• SQZT7 on-table optics losses 2% (72604)
• HD PD QE 97.7%  (73241)
• visibility 95.6%  (from 97.8% fringe visibility on PD B) 

If we include phase+dark noise that degrades squeezing but is not loss, then 21% total loss can explain the 6dB measured squeezing, see e.g. from the gsheet calculator (edited to include ranges for NLG=10 and NLG=11):

DTT homodyne template saved at $userapps/sqz/h1/Templates/dtt/HD_SQZ/HD_SQZ_101023.xml .

Edited to include some history of homodyne measurements:

• 6 dB SQZ -  10/10/23 LHO:73365 = 21% total loss, 15% known loss, 7% mystery (this time)
• 5.2dB SQZ - 8/31/23, LHO:72802 = 27% total loss, 21% known loss, 7% mystery (bad PD-B QE, lower visibility?)
• 5.5dB SQZ -   2/3/23, LHO:67219 = 25% total loss, 15% known losses, < 10% mystery (tech noise -10dB), NLG was varied.

It could still be interesting to vary NLG to see if we can obseve any more squeezing, or if an additional technical noise floor (aside from dark noise) is needed to explain the NLG sweeps.

We revised the sqz loss wiki table again today, and are including it to explain what we think our current understanding of losses is. 

It seems likely that the 7% extra losses we see on homodyne measurements are in HAM7, so we've nominally added that to the loss budget. 

In addition to this, there would be an additional 8% loss on the sqz beam if we didn't correct it's linear polarization with a half wave plate.  72604  At the time of the chamber close out, (65110) we measured throughput from HAM7 to HAM5 that would implies that two passes through the OFI were giving us 97.6% transmission, so this is not compatible with the polarization being wrong by this much.  We haven't included this as a loss in the loss budget because it seems incompatible with our measurement in chamber. 

The wiki currently lists the OMC transmission as 92%, and the PD QE as 98%.  The PD QE may be worse than this (see 61568), but measurements of the product of QE and OMC transmission for 00 mode seem to indicate that is in the range 90-92%, so this is close. 

With the infered losses of from the measured sqz/anti-sqz in the IFO, the plausible range of losses is 30-35%, we are using 32%.  With only known losses (including the values for OMC trans and PD QE), we have 14% unexplained loss. If we include the 7% apparent HAM7 losses, we have 9% unexplained losses in the IFO.  This does seem similar to the 8% polarization problem, but it would also include SQZ-OMC mode matching.  

Possible future scenarios:  We may be able to reduce the 7% HAM7 losses, and we may be able to swap the OMC to reduce those losses from 92% to 97%.  

These numbers come from the aoki equations that Vicky added to the google sheet here: gsheet

Don G. and Sheila have very likely resolved the homodyne polarization issue as being due to the SQZT7 periscope. So, the mis-polarization is likely not an issue for squeezing in the interferometer.

The sqz beam leaves HAM7 via reflection off the sqz beam diverter. From the latest CAD layout from Don, the outgoing reflected beam (blue) is ~75.58 degrees from global +X. The periscope re-directs the beam to travel along SQZT7, approximately along +Y. The CAD layout thus suggests that the SQZT7 periscope re-directs the beam in yaw (counter-clockwise) by an estimated 90 - 75.58 = 14.4 degrees. 

From recent homodyne measurements LHO:72604, of the sqz light leaving HAM7 and arriving on SQZT7, ~8% of the power was in the wrong polarization, this calculates to a ~16.5 degrees polarization misrotation. Compared to this 16.5 degree misrotatation we were searching for, the 14.4 degrees polarization rotation induced by the periscope image rotation can plausibly explain the misrotation.