Let's see how stumped JL et al. get this time

Yikes! What are the startup transients like?

John

Ah, young Grasshopper, it's not a question of stumping John, but whether he's interested enough to more than glance at it.

1n914 rectifryer? That's naughty.

How do you turn the FET off?

-- Cheers, James Arthur

[..]

(I assume you're going to say "Via Unmarked's recovery". Okay, but why? That introduces a bunch of dependencies, unnecessarily.)

((Hint: reference designators were invented to make discussing these things a lot more fun.))

James

The NPN in the fet source does it when the transformer primary current gets high enough; the dynamics are interesting.

These things would be easier to discuss if Tim used reference designators.

This is (potentially) a non-saturating blocking oscillator. Tube and transistor blocking oscillators could be non-saturating, too, if the grid/base current exhausted the charging cap before the transformer saturated.

Cute circuit. But maybe too tricky for its own good.

How about this?

ftp://jjlarkin.lmi.net/FlybackThing.JPG

(needs a +17 bypass!)

John

What, do I *look* like someone with a DSO? ;-)

(120 ohm load = 12V 100mA output)

I'll go see what I can do with a long exposure. Overshoot shouldn't be too bad, it's first-order after all (ferrite bead aside).

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Idunno... he got my old assed function generator circuit wrong on first glance. One would hope to be more careful in the future. ;-)

Yeah, I don't have anything better. It's rated for 200mA continuous, so I'm "okay". I'll order some MBR160 or BAT86 some day to replace them. And maybe a BAT46 for the gate drive diode.

Good question, isn't it?

It's either "Unknown" reverse recovery (I tried a 1N914, which didn't work -- too *much* Qr? -- "Unknown" is a short, unmarked glass body diode) or controlled by the 2N3904's storage time (~200ns should be plenty to yank the gate the rest of the way, once positive feedback takes effect).

It was on the breadboard that the 1N914 didn't work. I should try it again and see, now that it's on PCB. (Yes that's right, I successfully breadboarded a 430kHz power supply.)

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Ohh fine...

Go hit refresh :)

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Potentially?

One toroid I tried turned out to be some high remenance thing. Instead of a neat triangular current waveform I got this parabola thing.

Still transferred power though. Hurray for leakage inductance!?

It does like to latch into a "full throttle" state on occasion. I'm not entirely confident in its supply or load tolerance. It doesn't seem to be too critical on parts values, but parasitics, ah...

Ewww, needs its own +5V supply?

Also don't like seeing the opto as part of the timing loop. Come to think of it, if you want to run at HF like mine (~400kHz), you won't even need a capacitor per se, just use the phototransistor's miller capacitance. Rfb may still need to be pretty big.

Could do the same with a regular gate driver (that runs from 12V, maybe not as much as 17 though, but that's fine) and nix the 5V supply... schmitt thresholds are pretty flakey though. TTL level is the only thing that's guaranteed on those things that I know of. I'd make my own with a pair of transistors, but 2N3904s stop switching over a few hundred kHz.

Speaking of bipolar transistors, I originally started building this with a

2N4401, but soon realized I needed more speed and more current capacity. A 2N3866 worked quite well, but the IRF510 works even better (Rds(on) is less than the current sense resistor).

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

It's only a newsgroup. It doesn't matter.

And besides, it's easier to design circuits than to analyze them.

There's no transformer feedback in the turn-off direction because UD1 (the unknown diode) blocks the potentially useful negative flyback from turning off the gate. So solid turnoff depends on delay/storage in (reference designators! Magic!) Q3. I guess. If UD1 has a lot of capacitance, that will help some.

Heck, I breadboard 1 GHz pin drivers! And I did a 5 GHz sampler once. Light doesn't move all that fast.

John

You have resistors and zeners. Why can't I?

I think I've shown you this one before. That National gate driver is nice.

ftp://jjlarkin.lmi.net/Inverter.jpg

This could be jiggered around to work in your config, with the opto feedback tuning the duty cycle.

Fets are great for things like this. A 2N7002 might be enough; they cost 3 cents. Why would anybody sell anything for 3 cents?

John

But that just makes the FET go linear, a crappy constant current xxMHz oscillator. The FET won't actually turn off until it saturates and the drain starts rising--then the flyback pulse takes a bucket of charge off the FET's gate through , leaving the gate stranded at some DC voltage. Q3's cutoff at that point, so it's no help.

Depending on D1, the FET might not even turn off. That's inefficient.

Either a) kicking the current-limiting BJT in the base on flyback or b) more direct coupling to the feedback winding turn the FET quickly and securely off.

But Sahib, merely ask and they appear! (if you re-check Tim's drawing).

Yeah, a class-A switcher.

Looks decent, offhand.

--James

We shoot from the hip here, playing, but I assure you, John's aces.

o I'm

nd

)

nk

gain

Your circuit works now (relies) on charge stored in D1. Yeah, you could do that, but why depend on that property if you don't have to? Those are the things that get screwed up later, when production substitutes a new diode.

Put the d.c. biasing stuff and a bypass cap on the other side of the feedback winding, then drive the FET with the remaining lead. That'll rip the FET off the moment it starts to saturate. That's what you want.

-- Cheers, James Arthur

Not sure. Storage in Q3 will help, as will capacitance in D1. What hurts is d-g capacitance in the fet. The beautiful flyback edge is wasted as regards turning off the fet. (Unless D1 is a zener?)

Note that all of the fet's turn-on gate charge has to come from C3.

Abricadabra, y'all.

Still, all discrete designs are cute. Keeps us from just copying Linear Tech ref designs or running the dreadful National Webench mess.

John

Because mine are doing *work*, not totying to some other part's silly requirements...

...Just ignore that gate zener...

Hmm, come to think of it, if Q3 sucks enough charge on turn-off, the zener won't be needed to draw bias from C3. Vgs(max) isn't in any danger of breakdown in this circuit.

That works better because you have a negative supply -- that 0.8 to 2.3V threshold doesn't matter as much when you have +5 to -5V to work with. When you have 0 to something, it gets a lot squishier at the bottom end.

LM5112 looks to be a one-input-inverted schmitt-NAND followed by a house favorite driver (in this case, BJT boosted FET, which Fairchild is also fond of). TC4467 series has the same logic function, too. It's misleading to draw it as a comparator when it's not (yeah, fine, comparators don't have "A" and "B" inputs, but still).

The fundamental thing I like about this kind of circuit is it's current limited (or supposed to be). You don't get that with RC timed circuits unless you add a lot of logic (like a current mode controller e.g. UC3842), which doesn't work in discrete since the logic (2N3904 etc.) is so slow.

Of course, despite the openness of the loop, because it's always running, you're less liable to seriously f*ck up a transistor if it decides it doesn't want to run that day...

2N700x is a bit small (couple ohms Rds(on)), but anything beefier would do, too (use two in parallel!). Heck, D2 will have more charge than the damn gate (if it doesn't already). IRF510 is perfectly well suited, up to more output than I'm using here. A modern replacement would be any of the myriad SMT FETs with the same specs.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

So wait, the winding gets flipped, so it's no longer positive feedback???

The fundamental purpose of D1 is to keep D2 from recharging C3 when T1 reverses. If it does that, then the whole damn thing runs full throttle, and that's no good. Move D1 so it's antiseries with D2, though, that might work...

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Very Ugly Electrons! ;-)

--
Greed is the root of all eBay.

l

No, don't change the phasing. Conventional design just moves the bias and bypass stuff to one side of the feedback winding, and drives the gate with the remaining "hot" lead. Positive feedback remains, of course.

e,

ight

You might want to reconsider all that. Right now you're regulating by starving the switch's drive voltage supply. You're better off regulating by using feedback to vary Q3's current-limit setpoint. That's how it's done.

Think about it.

-- Cheers, James Arthur

That would definitely do it!

-James