The SOFT loop engagement has in the past always been a source of trouble, mostly because the initial offsets can be really big. Together with the double-integrator design (in control filters and suspension top stage filters) this usually is a recipe for disaster.

This should now be a thing of the past:

- re-ordered the filter modules to keep the switchable integrator and lead filters ahead of the always-on low-pass (see alog 44124).
- Set error point limiters before the control modules, but after the input matrix (in the blend filters) to 0.02.
- During engagement only a 0.1Hz pole integrator in FM5 (along with the DC gain, inverted plant and 10Hz LP in FM1, FM9 and FM10) are on. Plot 1 (red traces) show the total control filter frequency and step response. After this filter, we still have a gain of 2.5 and the suspension top stage 1/f integrator.
- Thus, during engagement with limiters on, the max signal to the suspensions can only be
   0.02 (limiter) x 10000 (control filter) x 2.5 (gain) = 500 cts
  This is small enough that the suspensions don't run away. Instead they perform a constant velocity move to shrink the SOFT error signals.
- The Guardian then looks for
   a) convergence of the SOFT control signals below 100cts, and
   b) convergence of the SOFT error signals below 75% of the limiter (or 0.015)
  When these conditions are met, Guardian turns the limiters off and turns on the true integrator and lead filter (changing the transfer function and step response to the blue trace in plot 1).
- Finally, after waiting for the 10 sec filter ramp time of the lead filter, Guardian turns up the gain to the nominal values of 4 for DSOFT and 6 for CSOFT.

Plot 2 shows the turn-on sequence for all SOFT control and error signals