silly circuit

Why wouldn't it?

Constant current diodes are depletion FETs.

Current mirrors are more predictable, and they start up at lower voltages.

The circuit has more parts than a constant current diode equivalent - you'd use monolithic pairs for Q1 and Q2, and Q3 and Q4 so it two pairs and three resistors - but might well be cheaper.

The comments below the article answer Steve Woodwards question about startup - the author says leakage. I’d be tempted to throw in 10megohm resistor to guarantee that - but I don’t think I would use that circuit anyway.

Author probably got his startup leakage from silly white protoboard?

An LM317 and a resistor would work.

Or the shock of connecting the power supply.

"with no requirement for heat sinks" is a real breakthrough.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Cool. Post a circuit. Try to keep the cost below $60.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Any printed circuit board you put it on could cost more than that, unless you bought a lot of them. Two monolithic pairs might be a dollar each, and the resistors are going to very cheap surface mount parts.

EDN already posted the circuit. Working out what you might drive with it would take longer, and why would I bother?

Where is there a monolithic transistor pair for a dollar?

I knew you'd bail.

And JLC sells us boards for around $1 even in small quantities. Panelized, a small current reg board could be assembled for way under $1. The real issue will be power dissipation.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

You'd clearly think twice now because there's a small risk you might make an erroneous claim. And then instead of being happy to be corrected by others who have the necessary information, you might come across as a being just a little vindictive towards those who pointed out your error.

Did you enjoy the capacitor film article JM posted?

TOSHIBA Transistor Silicon Npn Epitaxial Type (PCT Process) HN1C03FU

is stocked by element-14 in Australia for $A0.55 ($US0.38) each - minimum order five. It's not monolithic, but both transistors are in a single package, which is close enough for the application.

They've also got the Nexperia BC847BS NPN for about the same price, and the Nexperia BC857BS PNP part even cheaper.

The data sheet says upfront that it's two devices in one package. Not monolithic.

I've tested some of those dual-chip things. The thermal coupling between chips is ghastly.

Post your circuit.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

If you stand back from the microscopic current mirror circuit interpretation, the transistor equivalent of the DIAC using Q2,4,3, with a threshold of 2.5V or so, comes into focus. Then add in some current limiting/ regulating resistors here and there, and you end up with the floating current source. The BC557C has an absurdly small Early voltage and RCE, combined with the 'C' version beta in the range of 200-800, and start-up is a done deal. Voltage compliance shouldn't be a serious issue for driving LEDs, but it would be nice to see an analytic to that effect. AI hasn't penetrated deeply enough into our freebie CADs to give us that capabilty, so we're stuck with looking at graphs.

I'm not gong to bother because there's zero chance that anybody would built the circuit.

Where's the error?

I did. I said so in the thread. It reads a bit like a Mullard application note, in that the author doesn't go into the chemistry or the physics involved, but oit covers all the bases.

But better than the thermal coupling of the chips to the board they are mounted on.

So what. Two SOT-23's would have to be mounted very close to each other to do almost as well, and the application doesn't need particularly high performance.

To make you happy? Not exactly a compelling motivation. And the circuit is intended to drive a bunch of LEDS. Until you specify the nature and the number of the LEDs there's no point in getting into detailed design.

If you are saying that the theta between chips is nonzero, I wouldn't argue the claim. But the dual-chip things will make terrible current mirrors.

I have noticed that scientific types have a universal affection for current mirrors, and design terrible current mirrors.

See comment about current mirror fetishes. The output transistor typically dissipates a lot more power than the input transistor, so the thermals suck. Lots of papers in RSI have mirrors with separate transistors!

Your circuits designs do make me happy. I wish you would post more.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

The Wilson three and four transistor current mirrors do a lot better, and the cascode transistor on the output transistor also balances the power dissipation to a considerable extent. The extra transistors don't have to be tightly coupled to the basic pair.

I probably do look like a scientific type to you - I have published papers in the peer-reviewed scientific papers - and I do use current mirrors from time to time. I don't think that any of mine are "terrible". Can you point to one?

You are I have both derided some of the circuit designs published in RSI, and I've gone to the trouble of publishing derisive comments there. But Win Hill came out of the scientific community and there are people in who really do know about circuit design.

I wish I had more to post. The stuff I do for the local branch of the IEEE doesn't involve any circuit design. Given your performance with the emitter-coupled monostable, the kind of happiness you have in mind does seem to be schadenfreude.

No need for component matching or thermal coupling. The emitter resistor voltage drops can be selected to swamp the VBE uncertainty, making for zero chance of the current hogging hazard anywhere in the circuit. Matching to within a manufacturer part number is adequate for this application. The purpose of the mirror is to divvy up the power dissipation among the components, nothing special there.

That circuit may be very inefficient, depending on the LEDs and the input voItage range, and could fry those dinky little transistors. It should be a switcher. That could be done with 4 or 5 parts for about

It wouldn't be a constant-current block of course; constant-power.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

EDN doesn't specify the transistors at all. They don't have to be dinky. A practical design couldn't start until the nature and number of LEDs being driven had been specified. If the designer was remarkably incompetent they could put together something that might fry the transistors. Since John Larkin may select his staff on their willingness to flatter him it might happen. None of the people I've worked with have ever been that silly - they made less obvious mistakes.

Why? You could do either. What you want from LEDs is constant brightness, which equates to constant current. The voltage drop across a LED isn't linearly dependent on the current through them, so that does come pretty close to constant power.

??????

It sure does, in plain sight. Silly little TO-92s.

No, but they are.

Conservation of energy requires that the circuit dissipate I*(Vsupply-Vled)... unless you use a switcher.

A switcher would be constant power in and constant current out. And very efficient.

Give it a try.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

Oops. My mistake. Also EDNs. They didn't specify the LEDs to be driven, or with what current so specifying specific transistors was a mistake.

So what?

It might be. Switching losses are real, and inept design can make them quite large.

Specify the LEDs you want to drive, and the current you want to drive them at, and I might. Without that information the problem is undefined and perfectly insoluble. Since you wouldn't build the device, it would a waste of effort even if you produced a specification of what you wanted.