Here is a frequency noise coupling transfer function during tonight's lock.
I calibrated REFL9 into watts using a the factor 2900 V/W (from the diode to the output of the demodulator) and 12 dB, −21 dB, and 2 V/V for the analog and digital signal gains.
If the loop is shot-noise-limited (with 4.6 mW on the diode), this implies a noise of about 40 pW/Hz1/2. This would imply a noise in DARM that is more than a factor of 10 below DARM shot noise.
Daniel and I spent some time looking at the various CARM error and control signals we have on hand.
Here we have referred the fast (ao) control signal back to the error point in watts. The horizontal line is the shot noise for 5 mW.
The slow control signal (LSC-REFL_SERVO_SLOW) would also probably work as a proxy for the error point, once properly referred. The error readback (LSC-REFL_SERVO_ERR) is heavily contaminated by some kind of white noise. Ditto LSC-REFL_A_RF9_I_ERR.
