LED reference current source

The whole idea is that the two have about the same drift, i.e. ~ -2 mV/K, but different terminal voltages, e.g.

Vbe = 650 mV - (2.1 mV/K)(T-25C) and

Vled = 1650 mV - (2.3 mV/K)(T-25C).

You wire them up so that the voltages subtract, so you get

Vdiff = 1000 mV - (0.2 mV/K)(T-25C) .

JL's circuit and the more common LED-plus-emitter-follower voltage reference both work that way. If the drift rates are close, the temperature stability is improved a lot. Not to the level of a monolithic reference, of course, but you can reduce it from 1500 ppm-ish to a few hundred, which can be pretty useful since it's so quiet.

Cheers

Phil Hobbs

ce.

The paper that Lasse pointed us to

shows different LEDs having rather different temperature drifts - from

-1.5mV/C for a red LED to -5mV/C for a green LED, with a UV LED giving

-2.3mV/C in the middle. Regular red and orange LEDs presumably come closer to -2.1mV, otherwise the trick wouldn be a popular as it is. John Larkin's two-point measurements also show up his green LED as well away from -2.1mV and much worse than the other two LEDs that he tried.

As you push up the current through the LED the coefficient of voltage change against temperature is presumably going to get less negative, as the positive temperature coefficient on the resistive part of the voltage drop begins to become significant.

nction? I've

LED

hat

are

'scribble scribble'. Red LED reverse biased at ~20 V's counts photons at ~10k Hz under 'room light' illumination. LED capacitance is ~10pF with pulses of about 1V... 10pC of charge per pulse, 10k per second.. 100nA. Hey that's spot on!

Of course I'd expect forward biased leds to be much less sensitive.

George H.

So assuming it has a few milliamps' forward bias,

t -

I don't want a voltage reference, I want a low-capacitance current source.

There's a zero-TC point on even a 1N400x diode, but unfortunately it's at a rather high current (tens of amperes, IIRC).

My Iout tc is measuring around +90 PPM/degC, so the transistor Vbe tc is bigger than the LED tc, but not by much. Presumably I could get the current tc lower by running the LED at lower current, but I'm already below the specified tc of the emitter resistor. And the brightness is about right for a pilot light.

Small schottky diodes often have a zero tc at a useful current, like

I think some science-project fun could be had with a windowed eprom.

The first CMOS imagers were de-capped DRAM chips, where light affected refresh times. I think some guys at JPL did that.

That's not what's being aimed at here. The voltage drop at constant current across an ideal diode decreases by 2.1mV/C around room temperature, but LEDs aren't ideal diodes.

and the green LEDs that John is complaining about seem to be rather further from ideal than most.

-- Bill Sloman, Nijmegen

A guy I worked with some years ago, told a sad tale of the development of one of the first very-high-performance "clones" of the Sun Microsystems SPARC processor.

The development team got the first sample back from the fab (in a quartz-lidded ceramic package), carefully powered it up, and it began running diagnostics just fine. One of the engineers pulled out a camera and took a photograph of the system board... with flash. There was immediately a second flash from within the CPU package, as a whole bunch of the circuitry on the chip latched up and crowbarred the power supply to ground. Poof. That particular CPU never ran again.

One of the more expensive "victory" photographs ever taken, I believe.

You are being disingenuous.

But you want a more or less stable current source, and the compensating drifts of the LED and the transistor gives the stable voltage that controls the current output. The transistor makes for a simpler, cheaper and more compact output than anything fast enough/low capacitance enough for you that you could wrap around an LM4140.

If you needed something more accurate, you'd need a better reference voltage, but it seems that here you can get away with being a cheapskate.

-- Bill Sloman, Nijmegen

Maybe Jim Janesick?

Cheers

Phil Hobbs

Congratulations! You've just won this months SED Snarkiest Way Of Agreeing With Somebody Award. This prestigious honour is hotly contested--it's almost as sought after as the Stupidest Flame War Award.

Cheers

Phil Hobbs

The board is 2" x 2" and it's brickwalled with parts. This current source is small, cute, glows a nice warm orange color, and is plenty good enough for my needs here.

Is there some virtue to wasting more board area and more money, for no advantage? Actually, I've known some engineers who would usually select a more expensive part, because it apparently made them feel more important.

I measured a TC of 90 PPM/degC, which is better than the spec on the metal-film emitter resistor on the breadboard. And as Phil points out, this circuit will have much lower noise than something based on a bandgap reference and an opamp.

The LM4040s that we have in stock have TC specs of 150 PPM/degC.

Cute circuit. I like it.

Hey, it's only September 11. Give some other guys a chance.

(+)----+-----+ | | \-\ R1 ^ | / \----+ / \ | --+-- | | V +---| | \ R2 | | (current out) (GND)

with added bypass capacitors in appropriate places... It has a temperature compensated voltage reference, gain, and drives the transistor irrespective of Vbe drop. The LM4041-adj is a kinda backward connected TL431, it regulates the cathode-to-sense voltage instead of the anode-to-sense voltage.

My apologies for the wrong part number in the first posting...

Typical tempco (not worst-case) for the LM4140 is better than the measured performance of the LED circuit; if the LED were a perfect planar diode, and matched the diode equation, both the transistor Vbe and LED Vak would track at 1/T (at room temperature, circa 3400 ppm/C), and so would the transistor emitter current. Collector current will have a bit of correction for base current, of course.

My apologies for the wrong part number in the first posting...

Typical tempco (not worst-case) for the LM4041-adj is better than the measured performance of the LED circuit; if the LED were a perfect planar diode, and matched the diode equation, both the transistor Vbe and LED Vak would track at 1/T (at room temperature, circa 3400 ppm/C), and so would the transistor emitter current. Collector current will differ a bit because of base current, of course.

That would work, but I'd want to bypass the 4041, and add a base resistor, to better control the impedance that the base of the PNP sees. But it wouldn't glow in the dark!

ct

Absolutely none. One definition of an engineer is somebody who can do for $1 what any fool can do for $2. Cheapskate is a term of approbation in this context, though Phil Hobbs - as a physicist - doesn't seem to realise this.

The world is full of lunatics, but lunacy tends to take different forms in different people.

I tended to use 15ppm/C metal film resistors where it mattered - they aren't that expensive and they are easy to get hold of.

Perhaps not. You've probably got a lot more bandwidth than you'd end up with if you were engineering something around a bandgap reference. Phil's being a physicist rather than thinking about the circuit.

You invented it - you mightn't have been the first to invent it, but you still love it in the same way that we all love our brain-children.

-- Bill Sloman, Nijmegen