[GPS_Standard] Question about averaging mechanism

[Dave Platt]

> In Summing Mode (haven't looked at voting) I see that the current average is discarded and zeroed when a sampling session ends.  Could someone explain the reasoning behind this to me?  Essentially I'm trying to figure out why each session starts a new averaging count, rather than just applying a change to the DAC and leaving the averaging to run continuously.  I could understand this if there was a coarse correction, but not for a fine correction.

Seems to me that leaving the count intact could result in overcorrection, and an
oscillation in the DAC control voltage.

In control-system terms, I believe that the VE2ZAZ design is a P (proportional)
controller... the control correction is (roughly) proportional to the current
error (as measured over one averaging period).  It has roughly one
proportionality constant (the coarse/fine threshold).  P controllers never
quite catch up to the correct control value, but with proper
control-constant tuning they can be kept from oscillating.

Not resetting the summed error after each averaging period would turn
it into an I (integral) controller.  An integral-only controller can be
subject to "wind-up", overshoot, and oscillation... it will dither
back and forth on either side of the correct value.  That's likely
what would occur if you reset the error after a coarse correction
but not after a fine, I suspect.

A more complex control algorithm is PI, which makes its control
adjustment based on both current error (proportional) and
accumulated error (integral).  There are separate "weights" for the
two corrections, and there's usually an "anti wind-up" limit which
clips the absolute value of the integrated error.

Somebody could certainly rewrite the VE2ZAZ code to be a PI controller,
perhaps with a larger range of control value adjustments (rather than
just coarse and fine).  Tuning the P and I proportionality constants,
the wind-up limit, and averaging period for any given oscillator and
GPS setup would probably require months of experimenting
due to the long averaging intervals required.