This is some data that I've been sitting on for quite a while, taken on Jan 30th (54822)

I've used a time when squeezing was not injected to get a cross correlation, and estimated the squeezing level by subtracting this correlated noise, as was done in 53443 and 51981.  One potential problem with subtracting the noise in this way is that the radiation pressure noise is correlated noise, but it will also change when the squeezing level and angle change.  We can't use this type of techinque to infer the squeezing level at lower frequencies, but it's fine around 100Hz and above, as seen by comparing the correlated noise in the first attachment to the radiation pressure curves. I am using the pyDARM sensing function model to convert noises in mA to strain, but the strain curves here are GDS strain; above about 4kHz the shot noise curves look wrong because I am not applying GDS corrections to pyDARM.

The second attachment shows the anti-squezing and squeezing level, estimated with correlated noise subtracted.  You can see that there is more squeezing and more anti-squeezing at frequencies below 300Hz than above 1kHz, indicating that there is something like a frequency dependent loss happening.  This is consistent with what we've been seeing for a while and could be explained by the modeling Lee did in 50610. 

There is also a frequency dependent squeezing angle shift, shown in the 3rd attachment.  Here I've plotted the median squeezing level in different frequency bands, as the CLF demod angle is varied.   The lower frequency points have higher levesl fo both squeezing and anti squeezing, which is consistent with a frequency dependent loss, but there is also an offsett between the lowest noise CLF angle for the high and low frequencies.