Spice is great!

It’s probably pretty linear, so you could solve the DE by hand. Then you’d have more info than a stack of sims.

But then your fencing would suffer. ;)

Cheers

Phil Hobbs

It's a mess of PWM modulators, mosfet half-bridges, cycle-by-cycle current limiters, 3-phase common-mode chokes, ferrite beads, about 50 bypass caps. I'm not smart enough to do that analytically.

The sim includes the FADEC power supply, which rectifies the 3-phase alternator output and shorts it when the DC voltage gets too high.

What a mess: I have the simlated alternator frequency, some low KHz. The PWM driver stage at 250 KHz. And the simulated FADEC which PWM shorts my fake alternator at about 30 KHz, when it feels like.

This represents the load that I'm driving, what I imagine a FADEC power supply might do. PM alternators don't mind being shorted.

It's under 10% of the total sim schematic.

My brain would explode. Spice avoids a lot of hard thinking.

Of course they interact!

The 200 watt floating dc/dc conveter is reasonably independent so that's a separate sim. Ditto my soft-start circuit that ramps up the raw 48v bus in to the dc/dc.

High school trig was some time ago, so I have a couple of sims to verify the basic 3-phase math. Spice could eventually destroy all our math skills; I use it for voltage dividers and RC timing circuits too.

Is there an anlytical way to express delay lines? I think I saw the delay case once, in a test for PE registration as a Controls Engineer.

H(f) = exp(-j 2 pi f tau)

Cheers

Phil Hobbs

Of course there is. Williams and Taylor (ISBN 0 -07-070434-1) tell you how to build linear phase low pass and all-pass filters, which is to say structures which offer a constant delay over a range of frequencies, and they give you the analytic forms of these filters.

Yes, but I think he means the Telegrapher's Equations:

.

Joe Gwinn

Maybe. I use the one I gave much more often. It’s dead useful for modeling temperature controllers—you model the plant as a delay followed by an integrator.

You then do a bump test, i.e. dump a half-scale current step into the TE cooler and measure the temperature response.

You can read the model parameters right off the scope trace, or use a micro and auto tune.

Cheers

Phil Hobbs

Which says, in the frequency domain, that the phase of the output is turned backwards in proportion to the delay and to the frequency.

Delays are a nuisance in closed-loop feedback systems.

Jeroen Belleman

What about startup slewing?

We use Spice to model fairly high-order tremperature controllers and the small-signal loop response is only part of the concern. A recent customer wanted an optical modulator to get stable to 0.05c in one hour from cold start, and that's substantially nonlinear.

The hard part of simming that is modeling diffusion and estimating heat losses to ambient from various surfaces of the oven box.

That's actually the beauty of the bump test autotune approach.

Any time you've got a slow plant that you want to speed up by adding gain, you have to worry about windup. That's not too hard though--you just stop updating the integrator during slew.

With TE coolers, the main issue is actually plain ol' linear overshoot. For a given cooler and heat sink, there's a drive current value where d(Tcold)/ dI crosses zero. At that point, the sign of the loop gain inverts, which leads to Bad Things.

If your controller is underdamped, it's easy for the overshoot to put you in that territory, so you design the autotuning to give you 70 degrees or so of phase margin, which will prevent that from happening. (Obviously you put in some belt'n'braces safeguards as well.)

To get good bandwidth with a TEC loop, I usually make the cold plate out of FR4 with lots of via stitching, with a bare ENIG pour up against the TEC. One or two 0603 thermistors solder to the pour, right beside the edge of the TEC. That's good for control bandwidths up to a couple of hertz with small coolers.

(TECs are quicker than you might think--the cooling happens right where the PbTe bars solder to the alumina, whereas the I2R heating is distributed fairly uniformly along their length.)

How hard that is depends almost entirely on its physical size. If it gets too big, you wind up having to use multiple zones, which gets complicated in a hurry.

Diffusion isn't that difficult in homogeneous materials. If you have a squint at Section 20.3 in my thermal control chapter,

,I go into a lot of that stuff in a semi-analytical way that has been super useful. Dunno about your particular case, of course, but most of mine have used a lot of chunks of homogeneous material. ;)

Cheers

Phil Hobbs

Yes. Is the PE test for Control Engineers that advanced?

Joe Gwinn

Here's the eo gadget oven, with the big top off.

The eo modulator is inside the oven, on a secondary platform on some not super thermally conductive spacers. The temp sensor, a two-thermistor wheatstone bridge, is on the bottom of the platform with a flex cable going down to the controller. The heaters are a lot of mosfets on the bottom of the big box.

The platform adds a second lag to the loop, which lowpass filters temp transients a bit but mostly prevents temp gradients from getting into the eo modulator.

That's a dual-stage Mach-Zender modulator, which depends on cancellation of a zillion light wavelengths against a zillion and some plus 0.5 wavelengths, so it's pretty temperature sensitive and the customer wants extreme extinction.

It was Spiced and came out about as expected.

The coax cables conduct heat too, which was measured and included in the model.

LT Spice has both ideal and lossy transmission lines. A lossy line is a decent model for a diffusive thing, like a thermal delay or a really terrible PCB trace.

I think conductors inside digital ICs are essentially all diffusive, R and C and not much L. That's terrible for rise time.

My super slow sim now is the PM alternator simulator board, which doesn't have any delay lines.

That took a half hour or so to run. Even zooming on a slice of time takes something like a minute. I guess I need to tell it to not save so much stuff.

I have used lossy delay lines to simulate thermal systems, where the lag between the heater and the temp sensor is diffusive. Thermal systems have the equivalent of R and C but no equivalent of L.

I tried .save of a few nodes and it didn't change the sim time much. I'm compute bound.

Zooming should improve if I don't save gigabytes of data.

Give SIMPLIS a try, you'll speed simulations up by an order of magnitude.

I'm really used to LT Spice, and it's great for most things. This one case runs really slow.

One selling point for LT Spice is all the part models that are available. How does SIMPLIS compare in that respect?

I really want 5000:1 speedup and sliders on parts so I can tweak and see results instantly, like a breadboard with a trimpot and a scope.

Our big e/o modulator oven was slow in sim *and* in real time! Realtime was worse. What a nuisance.

The way you go about it, no amount of computing power is ever going to be enough.

Jeroen Belleman

The way *I* go about it? How would you sim a polyphase switching supply, simulating an alternator, driving a FADEC shorting regulator?

My old Win7 computer and LT Spice is a bit leisurely. But good enough.

I want Spice to run on an Nvidia board.