I designed a simple linear voltage regulator for high voltage application around 500V which is controlled by a PI controller.

enter image description here

Schematic 1 shows the raw design using a bipolar transistor to control the mosfet. The PI controller is build using operational amplifiers and works well. The feedback is a simple voltage divider because the output is around 400V.

How would you guys implement a current limitation? My first idea was a simple comparator which adds a maximum value to the feedback deviation input of the PI controller but I'm not sure about the stability in this nonlinear case.

  • \$\begingroup\$ Are you really using an N-channel MOSFET at Q1? \$\endgroup\$
    – The Photon
    Commented Dec 16, 2018 at 17:27
  • \$\begingroup\$ With an n-channel pass element, if the system power up with no base current to Q2, or if Q2 fails, or if the controller fails with its output open, the output voltage will go up to near the input voltage. Which is probably not good for the load. \$\endgroup\$
    – The Photon
    Commented Dec 16, 2018 at 17:31
  • \$\begingroup\$ Thats a good advice but I really noticed that p channel mosfets with high voltage capability are hard to get. \$\endgroup\$
    – Gustavo
    Commented Dec 16, 2018 at 18:52
  • \$\begingroup\$ Maybe this? \$\endgroup\$
    – The Photon
    Commented Dec 16, 2018 at 19:18

1 Answer 1


The problem you have is that you've got different things that you want to have control of your system depending on your system state. I'm not sure if it fits the exact definition of a variable structure problem, but that may be the right key phrase.

The solution I like, which looks wacky on paper but has proven to work nicely in practice, is to take each of the relevant error signals and feed a proportional-derivative controller. Then (assuming that you've already effected the sign change in calculating the error or implementing the PD controllers) take the minimum of all those PD controller outputs, and apply it to an integrator. The overall effect when either error predominates is that you're controlling just on that error signal, with a PI controller.

The nice thing about using such a wacky arrangement is that if you're controlling in one mode but approaching the other rapidly is that the corresponding PD output will kick in early, helping to kill the overshoot you'd otherwise get, AND because the final element in the controller is an integrator you're pretty much guaranteed that the control signal will be nice and smooth.

Figuring out how to apportion gains between the PD controllers and the integrator, where to allow sign inversions, and how to apply a pair of precision rectifiers so that they form a minimizer (or maximizer) circuit, is, of course, left as an exercise to the reader (or for follow-on questions).


simulate this circuit – Schematic created using CircuitLab

  • \$\begingroup\$ I considered this approach. The minimizer could be a threshold which routes the proportional feedback to the integrator. When the current is going near the maximum current then a current control takes control over the system. A threshold keeps the system from oscillating when the current decreases and voltage control takes over again. \$\endgroup\$
    – Gustavo
    Commented Dec 18, 2018 at 8:42
  • \$\begingroup\$ In my experience you need both the proportional and (bandlimited) derivative if you want the transition to be smooth, or you need a pair of proportional signals that feed into a single PI controller. The problem with that second approach is that you don't have the freedom to adjust the ratio between the proportional and integral gain. \$\endgroup\$
    – TimWescott
    Commented Dec 18, 2018 at 17:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.