I tried a few permutations of turning on the retrained cleaning that I described in alogs 70043 and 70023.  In the end, I'm just leaving on the updated jitter subtraction.  The LSC subtraction worked well, but then we updated some parameters (see Elenna's alog 70052), and it now needs to be retrained. The laser noise subtraction is quite confusing, and so is not on.

The first attachment shows all 3 (LSC, jitter, laser noise) updated subtractions on.  You can see that the low freqs (below ~500 Hz) shows improvement, as does very high freqs (above 5 kHz).  But, the middle-high freqs (500 Hz to 5 kHz) has injected noise.  The green trace is the contribution from the Laser noise subtraction, and you can see that it is the dominant term above 500 Hz.  This implies that we've got something strange going on that we have a sign flip around 5 kHz somewhere. It's possible that this is due to not long enough thermalization, but I suspect that we're just missing some term or sign or filter phase somewhere.

The second attachment shows just the LSC subtraction, and it does very well.  This plot was taken with all old settings (old SRCL offset, old SRCL FF), so is just illustrative of what we should be able to achieve.

The third attachment shows just the Jitter subtraction, and it also does very well, except that I've had to put in a minus sign in the filter bank gain.  A difference between what I had been using for Jitter subtraction and what is now currencly running is that what had been in use was trained on a measured exported-from-DTT transfer function between the OAF-NOISE_EST channel and GDS-CALIB_STRAIN_CLEAN, and today I'm using the filters that should be in use in the GDS calibration pipeline. Those two things have a difference of a minus sign, which is why today I have a minus sign in the filter bank gain.

Since the SRCL FF was changed after these plots were taken, the LSC cleaning now needs to be retrained.  I'll get some data from overnight tonight, and retrain the LSC tomorrow.

The fourth attachment is the range for our 3 versions of CALIB_STRAIN from the summary page.  Around 2300 UTC, there is a marked difference between pink and blue - that's with both LSC and Jitter subtraction on, after Shiela tuned up the squeezer.  Then the last time the range is high in this plot, around 2330 UTC, only the Jitter subtraction is on, but Elenna had turned on the updated SRCL FF filter, which is why our overall pink and orange range is higher than at 2300 UTC. Hopefully once I've retrained the LSC cleaning, we'll again get blue up a little higher.

I've accepted all the associated SDFs for these in both the safe and observe snap files.