power supply regulation

A PI control is fairly simple to do and works very well for this sort of situation. You may want to do something like this:

The diodes restrict the range over which the "I" part of things can swing.

PI is good. You can also feed-forward a correction based on load current and/or unregulated input voltage.

John

Those diodes for anti-windup are a good idea, but you have to design the loop carefully. The wave-a-dead-chicken school of anti-windup design will often cause nonlinear oscillations during transients, and so make things worse rather than better. Back around 1981 I was building PLLs for satcom frequency references, and when I put in anti-windup diodes I had some pretty amusing looking settling transients until I figured out what was going on.

Cheers,

Phil Hobbs

When I used to design control loops for steamship throttle controls, I'd use a nonlinear function generator to minimize open-loop error, and a P+I correction with carefully restricted range, so the the thing wouldn't run away too far if the tach failed or something.

I once almost ripped a 900' LASH ship off the dock at Avondale Shipyards, turning a pot too far. I did "lights out" a tanker that was off-loading in Binecia.

John

If the DC regulation doesn't meet designed values, look for noise influences or sources of leakage.

DC regulation is highly predictable and repeatable, unit-to-unit. It is often tailored for controlled droop in parallel situations.

Stating 'a couple of volts' and not a percentage, masks the degree of regulation currently achieved.

RL

Brr.

If I have the space, I usually like to use another op amp for the I term and let it saturate to limit the swing, using resistor ratios to set the gains. That way you either have a well-behaved PI loop or a well-behaved P loop with an offset--none of the weirdness you can get with partially-turned-on diodes.

Cheers,

Phil Hobbs

I still like to do that when possible. Putting control loops on things that are already nearly linear tends to make their behaviour much more benign in unusual conditions or when there's a fault like your tach going south. Thermoelectric cooler control loops spring to mind.

Cheers,

Phil Hobbs

ue

As with many thing, if you do it wrong the results aren't very good. The advantage of the diodes over other simple methods of doing the antiwindup is that you don't cause large difference voltages to appear on the input of the op-amp.

You never want to do just an integrator and then let it clip at the rails unless the op-amp has nose to tail diodes on its input. Hitting the rail builds up a large difference voltage on that op-amp that can lead to all manner of strangeness and long recovery times.

You can do both if you want. The feed-forward gets it almost right.

You need much more gain.

Also, on the feed back path, you may want the shift the reference ratio as current is detected. This will force the output set point to elevate to get better regulation. Normally, a pot in the circuit is needed to calibrate that. Years ago, we used to put incandescent lamps in series as part of the bias network on the regulator circuit :) They have CLD's for that now..

I agree that you don't want the main loop filter amp to clip. In my favourite configuration, I have an second amplifier doing the integrating, and sum its output with the main loop amp, which provides the P contribution, so it doesn't matter much if the integrator's input error becomes large. For slowish things like motor controllers, power supplies, and the sorts of PLLs I was doing, that doesn't cause any bad behaviour.

It's also quite possible to make the integrator clip cleanly by using Widlar's cute saturating-BJT breakpoint approach (see National AN4). And I didn't mean that the diode trick can never be useful--you just have to design it carefully, and it takes more testing to be sure it's right.

Cheers,

Phil Hobbs