This is a delayed alog, with some preliminary results from two scans of the squeezing angle that we did with different CLF powers.  The main message that we see more squeezing with less CLF power and more nonlinear gain, with around 3dB of squeezing at 200 Hz. 

The first scan was on Nov 6th, with 120 uW of CLF power (in Reflection off the OPO).  We ave been running O3b with 55uW of CLF power, since November 9th (53107), the nonlinear gain (amplifed seed over unamplified) was 4.9 on Nov 6th. 

The second scan was on Nov 13th, and for the test we set the CLF power down to 5uW.  The nonlinear gain was 4.5.  Daniel and I adjusted the gains of the OPO, CLF, and LO loops once before starting the measurement:

• CLF servo IN1 gain increased by 8dB to 4dB for a 1.1kHz ugf, with 5 uW CLF power reflected off OPO (according to reflected diode). 
• OPO ugf 2kHz with -14dB gain (we didn't make any adjustments)
• LO loop 15dB in1 gain gives us a 13kHz ugf for squeezing with 110 deg, for anti squeezing (CLF sign flipped) we have a 8kHz ugf for 18dB of in2 gain.  The setting we had been using was -4dB.  We didn't change the gain of the LO loop as we rotated the squeezing angle, because there is plently of margin to accomidate the change in optical gain from squeezing to anti squeezing and there shouldn't be much impact on phase noise. 

The first attachment shows the squeezing level in dB, with the correlated noise subtracted.  Here I've used the DCPD spectrum, taken the cross correlation at a time of no squeezing and subtracted that cross correlation from all the other measurements.  In principle if we were limited by radiation pressure noise you would expect that to show up here the same way that a squeezing angle rotation does.  However, I think that in reality we would need to take longer measurements and more averages to see the effect of QRPN.  This plot is comparable to the ones in 51981 and 51546.

In those earlier measurements of scanning the squeezing angle I found median squeezing levels in a selection of hand picked frequency bands, and did fits to them.  This can be done for this data as well, and I will try to get to that next week.  As a faster way to get an understanding of what is happening, we can look at the max squeezing, anti-squeezing, mean squeezing and squeezing angle rotation for every frequency, shown in the second attachment.   The angles plotted there are CLF demod angles at which we get the maximum anti-squeezing and squeezing at each frequency, I've subtracted 180 from the anti-squeezing angles just to make it easier to see what is going on.  We moved the phase by 15 degrees at a time, which is why they are discrete, the angle could be found more precisely by fitting.

If we use the standard equations for squeezing and anti-squeezing from aoki  (or other sources), the mean squeezing level is given by:  (R_s+R_a)/2 = 1+2*x*eta(1/(1-x)^2-1/(1+x)^2) if R_s and R_a are variances normalized to unsqueezed shot noise.  If we use this to try to infer the efficiency based on the mean squeezing in the second attachment, the higher measured anti-squeezing in the high CLF case would indicate lower losses, in that case, which doesn't seem believable.  In both cases the non-squeezing reference was recorded before the scan of the squeezing angle, and not repeated after, however in both cases these measurements were made many hours into a lock stretch when we would expect the shot noise level to be pretty stable.