Edge response, dynamics, and really ugly time constants

Edge response from what? In case we haven't been following the development, you need to fill us in with at least a block diagram.

Well, I pretty well described the external characteristics right there. If you must have more particulars, the control input adjusts an SG3524 oscillator's frequency, which is passed through the drive circuits to the series inductor and tank coil.

Block diagram? There's like, six blocks, half of which are either power supply or deal with buffering and delivering the already-generated signal to the output circuit.

Tim

-- Deep Fryer: a very philosophical monk. Website:

Hi Tim,

I'm not sure if I understand whats going on, but here goes:

Firstly, is the jpg the response of your tank (V or I) to a step voltage fed into the analogue amplifier?

secondly, is that open-loop?

I think what you really need to look at is the envelope. IOW assume sinusoidal steady-state (multiple by exp(-jwt) if you will), so ignore the actual output frequency. Then the behaviour of the envelope can be used to model your converter, prior to closing the loop.

If you look at the envelope around the big-to-small transition, it looks like a 300us period (3.3kHz).

the first peak is 2.1div, the 2nd is 1.5div, so the ratio of 1st to 2nd peak is 1.4. I could do some calculus and work out damping directly, but its easier to look at a picture of a 2nd-order response in a control systems book. it looks like d = 0.25 or so.

so the characteristic equations is s^2 + 2*d*wo*2 + wo^2,

d = 0.25 w = 2*pi*3.3kHz

you can easily check this, by doing a bit of maths and drawing a picture, then twiddling d until it looks about right :)

so thats the "plant" you need to stabilise with your controller, which can be designed the usual way - either analytically, or experimentally using either the zeigler-nicholls method, or the "fiddle until happy" method.

a PI controller ought to do the trick. you probably should do an analytical design first, its not hard and will give you values of Kp, Ki that are probably not far from where you want to be.

Cheers Terry

IIRC you have a controller that is attempting to regulate a phase setpoint between tank coil voltage and current. So why are you showing us a tank amplitude waveform? You should be displaying a voltage transient response that corresponds to the tank V-I phase, and it looks like this test should be performed at several values of tank loading. Your block diagram would show a phase setpoint input voltage into a differencing element subtracting the V-I phase from the detectors, etc...

I'm really confused. Looking at the waveform envelope it looks like about 0.3 milliseconds between peaks of the fast ringing, 2 milliseconds (hard to tell, it's so damped) for the slow ringing.. That makes the ringing frequencies about 3 KHz and 500Hz, plus or minus a lot.

If you're trying to stabilize the average level, you're looking in too much detail. You need to look at the analog control voltage, not look at the PWM switching transients.

Or I could be all wrong-- you just havent described the system or the problem in enough detail.

But to answer your question, I've had the best luck at stabilizing these kinds of transients with an ad-hoc 2-minute session with a R/C substitution box.... Twiddle the knobs for least ringing and read off the settings. Then wire in that 4.7K and 330pf series capacitor, backnote the schematic, and nobody will know you didnt do a full theoretical Laplace and Hamiltinian expansion. Works for me!

I suppose that works if you follow it up with 12 hours of operating parameter sensitivity measurements...

That's one of them, yes. I'm also interested in voltage response (with a peak detector and some time constant).

Yes, I should. I didn't bother checking the phase demod at the time because for one, the integrator is a 22k resistor and a 0.047uF capacitor -- too slow to observe the faster variations seen.

Absolutely. That will make things ever the more "interesting".

Ya, but this is open loop. Testing response before closing it and all.

Tim

-- Deep Fryer: a very philosophical monk. Website:

Yeah, you got me there.. I was looking at it later, and I thought, huh, that slow transient is a *lot* longer than the other... like, six times longer per wave...

Not the switching transients- all the class D stuff is out of question, pretty well set in concrete and hopefully about as reliable.

I didn't capture the input, which would tell gain and bias with respect to the tank's parameters, but just to take a random look like this, I'm not real interested in it. You just need to know that it's going up or down, squarewave-style.

I'll take some more detailed analytic steps later.

Heh heh. I've got the phase lock stabilized, by chucking a 0.1uF cap across the op-amp's feedback resistor. Voila, integrate the error signal over about 0.1 second. (Only problem is when the tank signal clipper drops out, due to insufficient tank voltage capable of switching the comparator, which causes a discontinuous phase voltage above a certain range. Naturally, this causes hysteresis and it literally sings as a result..)

Tim

-- Deep Fryer: a very philosophical monk. Website: