Fun with positive feedback

ation

,

ent

hich

Why the outboard oscillator? The main switch and the current sensor can be made to oscillate.

If your main switch is a MOSFET you can start it up nearly for free, then run its gate drive from your feedback winding.

That's pretty decent and widely used. Two transistors.

-- Cheers, James Arthur

They can, but it's usually boundary conduction mode (BCM). Away from the Bench, I don't want to take a stab at brute force driving it to points unknown. I'd much rather have it oscillate around a Q point.

That's true, and if the savings is worthwhile I might explore MOSFETs. BJTs are still cheaper though. Even if I'm not likely to create a circuit that winds its way into mass production (if it's not there already), I'd like to investigate the kind of circuits they use. And they don't use MJE13003's (or less!) in CFLs for nothing.

This is exactly what I wanted:

(which works amazingly well in simulation, by the way), except the 1k burns total supply voltage, which is no good at 160V, and even worse when hFE = 5. If I had a way to communicate inductor current down to ground level, it would be easier. Maybe I'll have to go with a transformer. Not a terrible idea, and a CT can have arbitrarily low voltage drop, even if it has to handle DC.

Tim

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

You could use a current-limited on-time with a constant off-time. Might possibly even be done with two transistors.

That topology can run continuous-mode over a wide range, especially at the higher loads where you want that (to maximize output for a given peak inductor current).

=A0BJTs

hat

to

's

5 (which works amazingly well in simulation, by the way), except the 1k bur= ns

=3D 5.

Yep.

Don't need a CT--it's easier sensing emitter current with a resistor. Cheaper too.

-- Cheers, James Arthur

Yeah, that's what I drew, maybe not to that level of simplicity. It's not hysteretic though, more of a "give me this much run time, and I pray it's the value I wanted" kind of thing.

Yeah. I did that with the other one, but it's a flyback supply:

the oscillator kicks on sharply by itself, dumping one full-sized cycle of charge. A few more bumps and there's enough voltage for it to start up (amazingly, it even starts into a 4 ohm load resistor!).

To make this general type of circuit self-starting, it would have to start kicking at a lower current, from a lower bias. That could go on in the usual self-excited manner for a few cycles. Once the bias supply comes up, more current is available, which drives the oscillator harder, and gets the driver/error amp sort of things operating.

But that only senses the current to turn off at -- fine for a flyback, but I want it to turn *on* at a specific current, too. That seems to be something fundamentally different. The low voltage circuit does it quite excellently, but it only does it with a high side shunt. The whole shunt--inductor--LED--diode circuit is at +V during the off cycle (the lowest point is the bottom end of the LEDs, which might be +V - Vf, lower but still arbitrarily high), so the only choice seems to be coupling current through a transformer, which somehow has to include DC, or a DC bias servo is required.

Tim

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

not

f t

up,

he

The circuit in your 1st post is a flyback, as have all been all the others. Isn't that the goal?

thing

ntly,

ent

o

For hysteretic you'll need a buck arrangement and a high-side current- sensor / mirror thingie for the inductor current. Or a half-bridge oscillator, plus inductor, and some way to throttle it all. Or your CT and split ac/dc feedback paths.

Kinda messy compared to a simple current-mode flyback, dontcha think?

-- Cheers, James Arthur

Noooo! :-(

The LEDs are in series with the inductor. The "flyback" is clamped by a diode, forcing discharge current to run through the load. Charging current runs through the load too, therefore it's a buck. Load current is continuous (or it's supposed to be). You could put in a filter capacitor and get a DC output (not very useful since it's bouncing up and down, and not isolated, but that's fine for driving LEDs).

This circuit at least:

is definitely a buck converter, with constant current control, and hysteretic action (turn on/turn off). The two diodes and transistor look just like a current source, since they are, and +FB makes it oscillate.

That's what I was thinking. Any elegant ways?

Yeah, but discontinuous current isn't what I want. Pulsed LEDs are less efficient, and I want to be able to point at it and say:

"I have exactly nnn amps through the load." "How do you know?" "Because the voltage across this current sense resistor is DC plus a little triangle wave, and the average voltage is nnn." "I'm convinced!"

It's a lot harder to say that when you've got some intermittent triangle or trapezoid. What should peak voltage be? Peak current? Average voltage? Average current? It also makes more RFI -- like I said somewhere else, this method has exactly one node that's moving a lot. A flyback has the switch side transistor and the load side diode, both of which can make interesting noises. That's acceptable for an isolated, voltage regulated DC power supply, but there's a simpler way for this.

Tim

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

Yeah--I was too hasty.

Why the obsession with enforcing a particular inductor ripple current?

The "flyback" is clamped by a

None come to mind. Those always wind up as kludges.

I say, ahh say that's why we have filter caps, son.

Oh, this is an LED driver? I thought t'was a general-purpose supply, part of your "I've got B.O." thread project. I have to be careful-- this is a potential conflict of interest for me.

I've tested pulsed LED efficiency. With a simple filter cap on a flyback (boost topology) the LED efficiency loss is reduced to insignificance even for low flyback duty cycles.

Running on rectified mains voltage you can size a flyback winding so the power supply's in 'flyback' most of the time, with the inductor acting as a quasi-constant current source that's driving the LEDs; filtering is then trivial. Even a small cap easily runs the LEDs with minimal ripple current during the short inductor recharging period.

A sense resistor in series with the LEDs + filter cap. fixes all that.

If you're dead-set on hysteretic, you could sense i(L) on the high- side as in your 12v CC_Buck.gif and use that signal as input to a low- voltage-powered base-driver circuit. That's efficient.

A continuous-mode current-controlled flyback supply is simple and easy to isolate. Add an "LED current" sensing resistor if you want to.

I suggest not worrying about the "zero LED ripple-current" thing-- you'll work too hard solving something that isn't a problem.

-- Cheers, James Arthur

Just 'cuz. I could do it with a flyback (as below), but that would be too easy. :)

Close: it's still transformer feedback and BJT technology, but I'm trying variations now. I'm pretty confident in flybacks, so I'm moving on to other styles. I don't think I'm going to be very successful at forward converters -- well, maybe a half bridge style ala CFLs, but that's for later, and much higher power levels, for which you might as well use a TL494. (Notably, Sony seems to have used self-excited forward converters in their Trinitron monitors -- with saturable reactor control! See this power transformer for evidence:)

I did that on my Internet's Largest Joule Thief. It seems to work okay, and it makes the waveforms look nicer.

I drew something new today, but it wasn't running correctly in the simulator, and it uses kind of a lot of parts (~30), one of them a high voltage PNP, which is probably on the expensive side (cents, but in principle, something to do without).

Tim

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