Before anyone worries we are boiling our brs's, that temp channel is .1 C, so 220=22.0C.

Edited wrong units (you'd have to ask my subconcious how that passed my sanity check).

Back to Observing at 1827 UTC. Automated relocked, but we paused for 10 minutes Jim to swap an ITMY CPS card.

After the lockloss this morning, I went to ex to look at the enclosure and we didn't completely seal up one of the seams when we shut the box. I fixed that, temps should normalize over the next day. I've bumped the voltage on the heater up, but will probably return that to it's previous value this evening, so we don't overshoot.

Today I turned on the Jitter and Laser noise subtraction during commissioning (without having retrained it), and it performed at a similar level to this plot from last week.  The LSC subtraction was not as effective as this plot, but that's because the LSC in-loop feedforward was tuned later in the week last week, and so there just was less that needed to be subtracted.

Elenna, Sheila

We ran the noise budget code for this no squeezing time. 

• We updated the optical gain (Calibratations.py, in the closedLoopSenseing function) using the calibration report here which Jeff pointed us to.  (The most recent one with the exported tag is the current calibration.)  This is the same as the inverse of the gain in H1:CAL-CS_DARM_CFTD_ERR filter called O4 gain if you have the correct calibration report.  Elenna also found the ratio of DCPD mA to DARM err counts by running the template that measures their transfer function aligoNB/H1/couplings/check_DARM_gain.xml for the braodband pcal injection from alog 72047.  These two steps are what we needed to do to get the calibration right so that the semiclassical shot noise caluculation seems correct.  
• We are using CAL_DELTAL, since this noise budget was run before the calibration fix of last week that channel is more accurate.  We are also using the TDCFs kappa_c and the cavity pole to calculate the shot noise. The radiation pressure noise is calculated with the arm transmitted power monitors. 
• There is no DAC noise included in this budget, as we aren't confident in our usual DAC noise projection right now.  68869  With recent improvements in the low frequency noise it would be good for us to revisit the PUM DAC noise projections soon, by running in the higher noise state and running the noise monitor estimate for that higher noise state. 
• We have changed the frequency noise in this budget. We measure frequency noise by switching CARM onto one sensor, (REFL B), then using that sensor as a out of loop frequency noise sensor.  There was a mistake in the code where the psd was instead multiplied by sqrt(2).  We have now divided the PSD by 2, which should give us the correct frequency noise projection where we are frequency noise limited.  Interestingly, this correctly predicts the darm noise at 4-5kHz where it should be an underestimate by sqrt(2) in the ASD.
• Vicky is working on modeling the quantum noise with Kevin, we are just doing the no squeeze budget here and will return to the noise budget with squeezing with Vicky.  
• Also attached are the sub budgets for laser, asc and length noise, with the DARM noise plotted from a time with squeezing.
• Laser noise: The 120Hz peak is well predicted by the IMC PZT yaw injection.  The frequency noise has the factor described above applied, and there may be some double counting at low frequency although we aren't sure.  
• ASC: CSOFT P, and the hard loops are the dominant contribution.  CHARD Y in particular is large, this is because of Gabriele's recent loop change which helped reduce the RMS and improve some nonlinear noise in DARM. 71927 Elenna is looking at if we can roll that off to keep the low frequency gain without this additional noise at 40Hz. 
• LSC: MICH is the dominant contribution to LSC, it is similar to the level of the ASC total. 

This is all commited as 0f9ffe0e

Sheila, Vicky - we have re-run the noise budget for following times:

Noise budget with squeezing. Changes here: using GDS instead of CAL-DELTAL, closer thermalized FDS time to no-sqz, using updated IFO gwinc parameters related to quantum noise calculation.
(Edit: was a glitch in the old time; updated to an FDS time without glitches. All plots updated.)

PDT: 2023-08-10 08:45:00.000000 PDT
UTC: 2023-08-10 15:45:00.000000 UTC
GPS: 1375717518.000000

PDT: 2023-08-10 09:35:52.000000 PDT
UTC: 2023-08-10 16:35:52.000000 UTC
GPS: 1375720570.000000

Noise budget with no squeezing. Same time as above, now calculates using gwinc quantum noise calculation instead of semiclassical calculation used previously.

PDT: 2023-08-10 10:18:11.000000 PDT
UTC: 2023-08-10 17:18:11.000000 UTC
GPS: 1375723109.000000

Both sqz & no-sqz noise budgets now use the correlated quantum noise calculation from gwinc, instead of semiclassical calculations for SN & QRPN. The gwinc budget parameters related to quantum noise calculation are consistent with the recent sqz data set (8/2, alog 72565), with readout losses evenly split between IFO output losses that influence optical gain (20%) and SQZ injection losses (20%), parameters in plot title here. This is high on SQZ injection losses, and slightly conservative on IFO output losses. This updated FDS time is thermalized and closer to the No-SQZ time; the time used previously was several hours earlier near the start of lock, w/ ifo not yet thermalized.

Unlike before, both budgets now show GDS-CALIB STRAIN, which on 8/10 was more accurately calibrated (see Louis's alog on Aug 8, LHO:72075, comparing CAL-DELTAL and GDS vs. PCAL sweep, and his record from 72531). CAL-DELTAL was previously overestimating range due to calibration inaccuracies. We got GDS-CALIB_STRAIN data from nds servers, and at first weren't able to get input jitter data from nds, due to the sampling rate change of IMC-WFS channels from 2k to 16k, 71242. Jonathan H. helped us fix this issue, so we can now pull GDS data and input jitter data from nds.ligo-wa.caltech.edu:31200 -- thank you Jonathan!! With this, the input jitter sub-budget is kind of interesting, looks to be mostly IMC-WFS in YAW.

A quick thought on discrepancy between expected and measured DARM between below several hundred Hz-- I don't know if this could be related to the recent update to gwinc CTN parameters (high/low index loss angles), related to quantum noise, or mystery noise. The recent gwinc CTN update seemed to have dropped the calculated CTN level slightly (maybe 10-15% or so). In April 2023, Kevin helped update CTN parameters LHO:68499 to reconcile H1 budget with the official gwinc parameters, while Evan made a correlated noise measurement 68482 where noise in the bucket seems more consistent with the older CTN estimate from gwinc (or very slightly higher). Another idea is that it could be related to quantum noise, such as SRCL detuning or sqz angle which could've changed since the sqz dataset, as quantum noise can also affect the noise in this region.

All pushed as git commit 28cf2664.

Edit: All pushed again as git commit 33ffd60b.

Added noise budgets with squeezing for HVAC off time on August 17 from alogs 72308, 72297.

When comparing this HVAC off time on Aug 17 with the noise budget from above on Aug 10, it's interesting to note the broadband difference in input jitter (Aug 10 vs Aug 17, HVAC off). Between these times, worth noting that I think there were several additional improvements (like LSC FF or SUS-related) as well.

Edit: updated 8/10 input jitter budget to the less glitchy noise budget time.

Much of the gap between expected DARM (black traces) and measured DARM (red traces) in the noise budget looks compatible with elevating the CTN trace. Budget plots with 100 Hz CTN @ 1.45e-20 m/rtHz are attached below for the no-HVAC times. This is almost 30% higher than the new gwinc nominal CTN at 100 Hz (i.e., 1.128e-20 m/rtHz --> 1.45e-20 m/rtHz). Compared to the old gwinc estimate of 1.3e-20, this is ~11% higher. Quantum noise calculation unchanged here.

This CTN level is similar to the 30% of excess correlated noise that Evan H. observed in April 2023, see LHO:68482. His cross-correlation measurement sees ~30% excess correlated noise around 100 Hz after subtracting input jitter noise, where that "30%" is using the newer gwinc CTN estimate of 1.128e-20 m/rtHz @ 100 Hz. This elevated correlated noise, if attributed to CTN, corresponds to CTN @ 100 Hz of about 1.3*1.128 = 1.46e-20 m/rtHz. See this git merge request for the gwinc CTN update ; this update lowered the expected CTN at 100 Hz by ~15%, from 1.3e-20 (old) to 1.1e-20 m/rtHz (new), based on updated MIT measurements.

For reference, I have plotted these various CTN levels as dotted traces in the thermal sub-budget.

To elevate CTN levels by 30% in the budget code, I scaled both high+low index loss angles by a factor of 1.8, specifically Philhighn 3.89e-4 --> 7e-4 ; Phillown 2.3e-5 --> 4.14e-5. It seems like much higher than this level ~1.45e-20 might be difficult to reconcile with the full budget.

Noteworthy w.r.t. squeezing: from the laser noise sub-budget, laser frequency noise looks within 33% of squeezed shot noise with ~3.7dB of squeezing. By contrast, the L1 noise budget from Aug 2023 (LLO:66532) shows laser noise at the ~20% level of squeezed shot noise with 5.3 dB of squeezing -- i.e. a lower laser noise floor past shot noise.

The following plots can be found in /ligo/gitcommon/NoiseBudget/aligoNB/out/H1/lho_all_noisebudgets_081723_noHVAC_elevatedCTN, and not yet commited to the git repo.

Plots with higher CTN are attached here for the SQZ / no-SQZ proper noise budget times from 8/10, when injections were run.

Comparing the sqz vs. no-sqz budgets suggests there might be more to understand here, to tease apart the contributions from coating thermal noise (CTN) vs. quantum noise in the bucket. In particular, something disturbing that stands out, is that I imagined that if elevated CTN is the physical effect we're missing, it would reconcile both NBs with and without squeezing. However, there is still some discrepancy in the un-squeezed budget, which was not resolved by CTN, and seems to have a consistent shape. I'm wondering if this is related to the IFO configuration as it affects the quantum noise without squeezing. I think this could result from a non-zero but small SRCL detuning since it looks like elevated noise, with a clear shape, that increases below the DARM pole. Simply elevating CTN to match the no-sqz budget would put us in conflict with squeezed darm, so I don't think it makes sense to elevate CTN further. The budget currently has 0 SRCL detuning as it "seems small-ish", but this parameter is somewhat unconstrained in the quantum noise models.

In models, the readout angle is upper-bounded by Sheila's contrast defect measurement, though in principle it could probably be anything lower than that too, which could be worth exploring. It might be helpful to have an external measurement of the thermalized physical SRCL detuning, or in the models allowing the SRCL detunings to vary, to explore how it fits or is constrained by the fuller noise budget picture.

Plots with squeezing can be found in /ligo/gitcommon/NoiseBudget/aligoNB/out/H1/lho_all_noisebudgets. No squeezing plots are in /ligo/gitcommon/NoiseBudget/aligoNB/out/H1/lho_darm_nosqz_noisebudget.

I pushed to git commit 70ca191c without elevated CTN and the associated extra traces. The relevant parameters are left commented out at the bottom of the QuantumParams file, and relevant code to plot the extra traces is commented out in the lho_all_noisebudgets script.

Attached the data times as a .py, for easier analysis.

Kevin, Sheila, Daniel, Vicky

Here is the squeezing dB data from this time, which shows the quantum noise reduction with squeezing, relative to a model of quantum noise without squeezing. To sanity check IFO parameters, here is the no-sqz quantum noise model (dashed purple) is plotted alongside the full budget (solid red), plotting CTN (the -. and -- green lines), and the technical non-quantum noise estimate (grey), using the no-sqz times taken in this dataset. IFO & SQZ parameters used for modelling are shown in the title.

Note: this squeezing subtraction (estimated quantum noise w/sqz - quantum noise model w/o sqz) is model-dependent. We are still working to use a cross-correlation estimate of the technical noise, so we can calculate the quantum noise without squeezing w/o relying on a shot noise model (b/c the model depends on optical gain / ifo readout losses). In this analysis, the method this is reversed: we use the quantum noise model to get an estimate of technical noise, then subtract that technical noise estimate from squeezed darm, to see the quantum noise reduction with squeezing. Specifically, in PSD here we do the following:

     - tech noise = (meas darm no sqz) - (qn model no sqz)    --> this is better to get it from the dcpds correlated noise (but correlated noise includes QRPN, need to account for that accurately)
     - quantum noise w/sqz = (meas darm w/sqz - tech noise)
     - sqz dB = 10*log10( (quantum noise w/sqz) / (gwinc qn model no sqz) )

How I set the IFO modeling parameters:

1) Homodyne / readout angle = - 10.7 degrees. This is constrainted to < -10.7deg from sheila's recent contrast defect measurement 71913.
2) SRCL detuning =0.  It is likely small, though possibly non-zero; anyways, setting it to 0 for this analysis, while working with Louis to see the SRCL detuning from calibration measurements.
3) IFO readout efficiency - based on sqz wiki, expect ~13% broadband optical readout losses of the IFO signal, without squeezing. Based on no-sqz measured DARM, circulating arm powers in the 370-380kW range, homodyne angle < -10.7 deg, and negligible SRCL detuning -- here it is estimating 73% output efficiency (27% loss) to reconcile shot noise sensitivity with measured DARM at 1 kHz. With the no-sqz quantum noise model and 73% output efficiency, the estimated technical noise level is ~0.9e-20 m/rtHz at 1 kHz. This 1kHz technical noise estimate lands somewhere between Craig's correlated noise estimate 71333 (like 0.5-1e-20 m/rtHz @ 1kHz), Sheila's measurement of correlated dcpd noise 70891 (like 1.2e-20 m/rtHz @ 1kHz), recent noise budget projections, 72245, and a lil higher than Jenne's NonSENS noise budget 72578 (considering laser frequency noise ~0.3e-20m/rtHz at 1kHz). Given that subtracting the shot noise model suggests technical noise within a factor of 2-3 of other metrics, it lends at least some some confidence to the no-sqz quantum noise model w/excess output losses that impact optical gain.

Next question:  How well constrained is the breakdown between readout / injection losses? Not all the IFO readout losses that affect optical gain have to be SQZ readout losses (two beams could have different alignments/mode-matchings through the IFO / AS port). But to look for sqz losses, it'd be helpful to have a hint of where to look, if we can find whether losses are IFO side or else SQZ side. I don't see a clear sign in our measurements either way.
--> I considered 3 scenarios:
1) Equal IFO readout losses 20% + SQZ injection losses 20%,
2) Using same IFO readout losses (71%) for SQZ, then rest is SQZ injection losses (9%),
3) All SQZ injection losses (35%) and perfect readout efficiency (100%).
Noticeable features that could distinguish between the two scenarios (high readout losses vs. high injection losses) are mostly the low frequency QRPN SQZ rotation+gain from the arm cavities optomechanics. If more quantum noise is injected to the IFO (so injection losses * gen sqz dB are higher), then the arm cavities have more sqz to rotate, and so we see the squeezing efficiencies increase at low freqencies as a result. Kevin has nice plots that show this (squeezing efficiency within the arm cavity bandwidth decreases w/injection losses but not with readout losses; or like Kevin says, injection losses source QRPN and shot noise, while readout losses only source shot noise.

Some next steps to understand in the analysis/models:
1) How is the squeezing relatively flat given the known mode-mismatches of both IFO-OMC and SQZ-IFO/OMC? It seems that hot OM2 curvature leads to ~10% IFO-OMC mode mismatch, while the SQZ PSAMS are railed without yet optimizing SQZ-OMC mode-matching, so SQZ-OMC is clearly not perfectly mode-matched either. What is the scenario where we get flat squeezing given the known mode mismatch?
2) Use cross-correlation to calculate squeezing dBs. Is the squeezing definitely flat, or is that an artifact of how we compare to a quantum noise model?

Small notes about this analysis:
-- for 8/2 data: for calibration, I am using GDS-CALIB_STRAIN_CLEAN, with the text-file magnitude correction from Vlad's alog 71787.
-- for no-sqz noise budgeting, I am plotting both what gwinc has as CTN, and a trace that is 1.3e-20m/rtHz at 100 Hz. Reminder of this alog from Kevin K. and Evan H. regarding the CTN gwinc noise budget update 68499, which dropped the CTN estimate from ~1.3e-20 m/rtHz previously (green -. trace) to the ~1.13e-20 m/rtHz calculated in gwinc now (green, -- dashed line).