I've started looking again at DHARD, after Sheila's noise budget that shows that DHARD is contributing more than a passive coherence test will indicate (which also means that we cannot get the full impact of DHARD out of DARM with the offline cleaning pipelines). 

Sheila points out that one of the first things that we should do (and should have done several weeks ago when DHARD was causing us so much trouble) is take out the bounce and roll notches in the L2 stages of our quads.  We now have the bounce roll dampers, and we've been able to take the equivalent notches out of the DARM loop, so we shouldn't need to actively notch those modes.  This will win us quite a bit of phase below 10 Hz.  See the first attachment for a plot of our control filters with and without the bounce roll notches.

I measured DHARD this morning (shown in second attachment), with the notable changes from last time of new spot positions, and extra boost in DARM loop.  It seems like we're much better off, and don't seem to have the non minimum phase problem.  See the third attachment for an extraction (undoing the control filters) of the DHARD_P plant, as measured in late January and today.  I show different plant models though, since today's measurement much better matches the 30W model, even though we are definitely using active radiation pressure compensation to make our plant look like 10W.  More thinking to be done on this.

In the 4th attachment I show my measurement, manipulated to remove the bounce and roll notches, removing the gentle cutoff, and adding in the hard cutoff.  I'm happier using these manipulations of my measurement, to ensure that I'm looking at what the situation would *really* be, since we don't match any of the models precisely. I reduce the gain by a factor of 2 in the final proposed OLG shape with the hard cutoff.

The last attachment shows the error and control signals before and after we implemented these changes.  The residual motion RMS increases by about a factor of 2, but we can afford that and still be moving less than about 1 nrad RMS.  You can see that the control signal is much reduced at higher freqs.  We did see that a roll mode of one of the quads started ringing up, so our new plan is to instead using the roll-only notch filters.

Sheila has put these changes into the guardian.  We could definitely see an improvement in the DARM spectrum, although we've had some small earthquakes giving extra ground motion, so the range impact isn't quite quantified yet.  We didn't get as much range improvement as we hoped for, but that seems consistent with Sheila's noise budget from the other day that DHARD Yaw is more impactful.  So, that's next up!