joseph2k said:
I confess to some surprise at not understanding the difference between
control signal and error signal.
The control signal is usually the output from the error
amplifier that drives some control device (pass transistor,
in this case)
PID stands for proportional, integral,
and differential components for the the control process. Using an error
signal for the PID input is a fools errand.
This depends on whether the setpoint is changing or fixed,
and whether or not you want fastest response to both
setpoint changes and load disturbances, or just fast
response to load changes and slow, over damped response to
setpoint changes (like turning the voltage setting knob on
the supply). There are several useful combinations.
Properly used, proportional
sets the final response, integral provides smoothing,
Integral provides eventual perfection, since it keeps
changing the output (ever more slowly) till the average
error is zero.
and differential
provides response speed. Direct error signals are not useful.
Another way to look at PID response is as a soft notch
filter. Any feedback process develops a conjugate pair of
poles that represent the onset of oscillation (a peak in the
feedback response. If you can center the response notch of
the PID combination (falling gain with rising frequency from
the integrator term to some minimum gain from the
proportional term, with rising gain above that minimum from
the derivative term) this response peak, you extend amount
of closed loop gain tolerated and the frequency range of
stability. Some versions of PID terms have deeper possible
notches than others. The more deeply notched versions work
better with inherently resonant systems.
Lots of useful lead lag phase compensation networks used to
tweak feedback systems can be described as a P. I. and/or D.
terms.
This is the underlying concepts for "Kalman filtering" which take into
account device response characteristics to determine optimal control
outputs for the highest performance and reliable stability.
I think that Kalman filter control also takes the noise of
the measured variable into account, combining some part of
the control output run through a system model that predicts
the result of control changes with some fraction of the
measurement of the system, to improve the control you can
achieve based strictly on a noisy measurement. Ideally, the
proportions of predicted response and measured response
adjust in response to the noise. Please correct me if I am
misunderstanding it.