Math Puzzle...

Hi Spehro,

Here's how I handled "hardness" with my version of an IDC element...

"Compliance" is the breakpoint. "Transition" is the equivalent of your "hardness". ...Jim Thompson

Fred, You just reinvented the Flyback topology. Must inform you, it has been done before. Cheers, Harry

The job of this converter is to charge a capacitor bank. It will actually be about 4000 uF, but that's not practical to simulate. The ultimate load is a laser, pulsed at about 25 volts, 120 to maybe 150 amps. After the shot, the converter has to recharge the caps before the next shot; it's the efficiency during the recharge that really matters... startup is infrequent.

The converter design is tricky because the entire driver has to fit on a 1.5 square inch board, and I want to use standarr magnetics for the first articles.

Right. This circuit just makes an occasional single blip to top off the capacitor. The only load, between laser shots, will be the feedback divider, which could be megohms.

You could use 0.5*C5*V^2 for the output energy at the end of the charge. Heck, just make that a constant.

But that's not what I want to see. I want a reasonably smooth graph of efficiency vs time, as the cap charges. I also want to see the averaged currents vs time, which are meaningful to my customer, hence the current filters.

I never did integrate. I lowpass filtered the currents, computed input and output power, divided, limited, and filtered a bit more for plotting. The LIMIT is inactive except for nanoseconds around startup where divide-by-zero sorts of things sometimes happen.

So does filtering. I need to graph the filtered input and output currents anyhow.

That's what I did. What you did is pick the "stupid" option.

He just wants to be right, Let him.

I worked with one of these gurus and couldn't convince him that RMS was a better measurement than an average reading analog meter.

Even did Calculus on the waveform image and still didn't convince him.

Cheers

If my graph doesn't show smoothed efficiency vs time, somebody might say why, instead of complaining.

Then why do we get different answers? I averaged the input current way out on the tail where you've reached equilibrium. At that point, capacitor charged, "load" is only feedback... efficiency is ~55%.

If you're defining efficiency as energy available at the output, versus how much energy you put in, there might be a different answer, but you've defined efficiency in a non-standard way. ...Jim Thompson

You might explain the math reasoning behind your equation rather than hand-waving. ...Jim Thompson

I've defined efficiency as Pout/Pin, as a function of time, assuming some reasonable filtering on the 9 volt supply. What's nonstandard about that?

Efficiency varies with time. You picked a time, after the cap is charged, to pounce on the 55% number. It doesn't matter then, because the power is tiny. The caps need to be topped off, and the only load is the feedback divider, which can be megohms. Efficiency matters when we're loading the customer's battery hard, which is during the cap recharge interval after a laser shot. The powerup phase is interesting but not very important, as it happens infrequently.

Rob is pushing the LT3420, which is a photoflash part. That's a possibility. I'd still need to graph filtered Iin and efficiency vs time.

Sure. Tanh(kx) is k times sharper than tanh(x).

Cheers

Phil Hobbs

It's Pout/Pin, in plain sight. What's to explain?

Ummm... your efficiency is greater than 100%. That doesn't cut it in the real world.

I think it's real. During startup, it pulls a lot of current and stores energy in the inductors. A little later, that energy gets dumped into the load. That part doesn't last long, so it's interesting but doesn't matter.

A black box that can store energy can have over 100% apparent efficiency for finite periods of time. Like a UPS for example.

I don't think it is. I expanded the interval and manually measured the efficiency during the >100% trace interval. It is not over 100%. I think your 3 integrators are messing up your function report.

Tanh is fairly nonlinear, even near the origin. It needs work to linearize it and make it sharper, which will require more runtime computing. LIMIT has zero error and is probably faster.

Sometimes I just use a resistor and a couple of ideal-diode clamps. Megawatts of power dissipation are no problem in Spice.

They're not integrators, they're lowpass filters. The third one, after the B source, can't cause an overshoot. Time lag differences in the in/out filters could cause apparent over 100% efficiency. I am shunting current through caps, around the sense resistors, to smoothe the current waveforms. But that is also a short-term transient thing, so doesn't matter. In real life, I'll have a huge output cap that will take hundreds of milliseconds to charge.

How did you compute efficiency during a finite interval?

Wow! Totally wrong! TANH is NOT non-linear... it has all derivatives.

LIMIT is the smack-dab wall.

Wow! Such ignorance!

Il sprach the Spice "expert" >:-} ...Jim Thompson