Today I convinced myself that I can reproduce, offline, the filtering and subtraction that the calibration pipeline does. 

This enables me to try different GDS FIR filters that Louis has created for me (also offline) by training NonSENS coefficients for that filter, putting the coefficients into the front end to calculate a noise estimate, and then use the noise estimate to offline filter and subtract, to see what the effect would be.  All of this last sentence is done out of Observing, so I tried several different combinations during our commissioning window today. 

I put a notebook up on git, to remind myself how to do it. In the plot at the bottom, the green and red traces are indistinguishable, which is good, and means that my by-hand filter+subtraction worked the same as the GDS calibration pipeline's filtering and subtraction.

Since I haven't yet gotten an offline front end model working (Erik and EJ showed me how, I just haven't had time yet), I'm still using the actual front end to do the calculation of the noise estimate for me, which is why these tests below had to be done while we were out of Observing.

The tests below are different sets of coefficients, trained on different FIR filter files that Louis created for me.  The start and stop times are written, so I can go back and offline filter and subtract, to see what would be the subtraction that we have if we had that FIR filter set in use in the calibration pipeline.

• Jitter coeffs in, trained on advance_highpass_lowpass.npz, starting at 1372799399 (~5 sec after NOISE_WHITENING_GAIN up to 1).  Using Jitter gain of +1.  Stopping at 1372800113 (~ 5 sec before NOISE_WHITENING_GAIN down to 0).
• Laser noise coeffs in, trained on advance_highpass_lowpass.npz, starting at 1372800319 (~5 sec after NOISE_WHITENING_GAIN up to 1). Using laser noise gain of +1. Still has Jitter noise estimate going too. Stopped 1372801717.
• Laser noise, same coeffs, but no jitter noise est. Start at 1372801749.  Stop 1372802262.
• Laser noise, new coeffs trained on advance.npz (only, no highpass, no lowpass). Note that simulines calibration ongoing throughout this time. Start 1372802476. Stop 1372802765.
• Laser noise, new coeffs trained on advance_highpass.npz. Start 1372803740. Stop 1372804127.
• LSC_modulated, coeffs trained on advance_highpass_lowpass.npz.  Also using BosLscSumOut_to_CalibStrainClean.txt. This means it wasn't using the .npz file.  Start 1372805375. Stop 1372805988.
• LSC_modulated, coeffs *actually* trained on advance_highpass_lowpass.npz.  Start 1372806084. Stop 1372807009.
• Reminder to self that there are other times interleaved where the NOISE_WHITENING_GAIN was 1, but those are, eg, times were the individual gains (like the Jitter gain) were set to -1, to see if there was an effect to flipping the sign.

After I was done testing things, I reverted all of the coefficients I had changed this afternoon.  It doesn't really matter, since no subtraction is engaged in Observing, but since I don't have good motivation to make them different, I put them back.