Here's all the raw sqz data. Everything looks as it should be apart from the frequency dependent anti-squeezing, will need to look more closely at the model to understand 

FDS money plot and some very rough subtracted data. It looks like we have 4dB of SQZ at high frequencies and 4.5dB at lower frequencies. 

Using the same classical noise estimate as the above comment, I tried to compare all the traces to a model. I didn't try using any elaborate fitting routine, just wanted to sanity check some of the basic parameters such as SQZ angles. The model doesn't include any mode-mismatch either. Just a SRCL detuning of 0.4 degrees to make the FIS curve match as much as possible. The arm power is assumed to be 370 kW. The FC parameters are the same ones inferred from the ADF (including detuning). the injected squeezing is assumed to be 16.9 dB (inferred from NLG). 

I've truncated the plots below 60Hz because the subtraction doesn't seem to make much sense there. The rest of the data seems to largely align with the gwinc quantum noise model. We could use more careful fitting techniques to put better constraints on IFO and mismatch parameters. 

From the RPN increase at low frequencies, it looks very likely that the arm power is less than the assumed 370kW

Same exercise as above comment but assuming 320kW in arms (looks a little better)

Data relative to no squeezing (w subtraction). Model parameters same as above, FC detuning adjusted to -32Hz. No mode-mismatch assumed in model, possibly could account for the remaining discrepancy in the 100Hz region. SRC detuning is 0.5 degrees

I did a rough analysis on Hanford frequency-dependent squeezing data using the similar approach in Livingston (llo63533), in order to compare and cross-check our models. The goal is to find a set of paramters that characterize the ifo and squeezing given many uncertainties. 

TL;DR: Have trouble fitting FIS at 80-400 Hz and FIAS below 80 Hz, some feature that LLO data doesn't have. I haven't explored everything yet listed in the TO-DO list. For less than 300 kW arm power, the IFO/OMC mode-mismatch varies from 15% to 30% with unknown squeezing loss absorbed into it.

Preliminary processing

I pulled raw time series of the measured DARM length from H1:OAF-CLEAN_DELTAL_EXTERNAL_DQ at GPS times listed in Dhruva's alog. The PSD is computed with a log-spaced frequencies bins downsampled to 100 points for faster fitting. I applied a simple transfer function with 6 poles at 0.3 Hz and 6 zeros at 30 Hz to calibrate the channel to meters. I didn't apply any extra fixings using Detchar kappa variables for now. 

I tried to use Pcal line at 410.3 Hz to find the DARM offset using the method in llo63533. However, the calculated DARM offset is in micrometers instead of picometers, so I'm not sure if the Pcal response PD is calibrated in meters or not. I did see that there's also a nearby line at 410.2 Hz on OMC DCPD so I made sure the spectral resolution is fine enough. I assumed the DARM offset to be 10 pm fow now. 

I also assumed input carrier power to be 57 W from Sheila's alog (lho67610). The arm power ranges from 270-384 kW, corresponding to arm cavity RTL of 63-99 ppm. There is also a significant amount of contrast defect light, but I don't include it for now because I still don't know what's the cause of it. The differential loss in arm cavities can't give you such high contrast defect light. 

The known/assumed parameters are summarized in the table below.

Fitting

Unlike what I did in llo63533 where I made a mistake by fitting IFO/OMC mode-matching as a free parameter but forgot that it also changes the subtracted classical noise, I fit the difference between frequency-independent SQZ (FIS) and no SQZ directly so that the classical noise is subtracted automatically in this process. The free parameters are

• Readout loss (OFI + all the pickoff in OM1, OM3, OMC QPD)
• IFO/OMC mode-mismatch
• SRCL detuning
• Arm cavity round-trip loss

These affect the quantum noise without squeezing. And the squeezing parameters

• Unknown squeezing loss, degenerate with injection loss
• OPO/FC mode-mismatch angle (with 3% assumed mismatch)
• FC/OMC mode-mismatch

These 7 parameters are used to fit three spectra simultaneously, two of which are FISqueezing - unsqueezed and FIAntiSqueezing - unsqueezed DARM noise. I've cut the frequency range to 30 Hz - 2 kHz to remove nonstationary noise at low f and excessive laser frequency noise at high f. I've also assumed the classical noise in 800-1500 Hz to be roughly 20% of the total unsqueezed DARM noise in ASD, so the third spectrum is the unsqueezed quantum noise at this freq range. The PSd unit is converted to db(abs(PSD))/2. 

There is a lot of degeneracy among these 7 parameters because it turns out that the FIAS spectrum isn't taken at SQZ angle 90 deg away from the squeezing (a lot of extra FIAS quantum noise below 80 Hz), so there is another implicit parameter to fit. I chose to constrain one of the 7 and fit the other 6 and compare residuals. I did this for readout loss, SRCL detuning, arm RTL, and FC/OMC mode-mismatch, since they can be independently measured. The fitted result is shown in here.

Some of the assumptions are problematic because they push the fitted parameters to the bound (they are underlined). The fit tends to minimize the unknown loss parameter and the FC/IFO mismatch for some reason. It struggles to fit the extra noise at low frequency part of FIAS, which is not present in LLO data. The FIS fits have trouble at 80-400 Hz and the worst is the -0.54 SRCL detuning one. The last parameter set has the least residual of 172 dB. I've plot this one in DARM. 

Since the measured FIAS is nearly 30 deg out of the real FIAS angle, I'm not sure if FIS angle is good, which is much more sensitive than FIAS angle. If I don't assume perfect FIS angle, then the fits are better. Anyway, it's another DOF that makes everything even more loosely constraint.

To-Do

This would be my parallel work on LLO squeezing analysis. I need to add and distinguish contrast defect, SRC mode-mismatch, and IFO reflectivity in the model. I assumed no phase noise, at least negligible. I also need to treat the OMC mismatch phase better but it changes the effective squeezing angle to maximize squeezing like contrast defect. I also need to include the other 2 intermediate angles measured for FIS. The classical noise is assumed to be 20% of unsqueezed DARM at high freq here, but we can use cross-correlation to set it constant.