LED reference current source

OK, here's some numbers.

The orange looks good for my application. I don't understand what's going on with the green.

The supply regulation isn't as good as a bandgap or something, but is OK for what I'm doing.

There's zero evidence of light sensitivity. I can blast my mini-maglite, or the Mantis illuminator, into the LED and nothing happens.

Osram wasn't the only manufacturer that made liquid LEDs but it must vary with the particular part numbers. The ones I was using were primarily 4-lead SMT and 0603/0805 types. They would claim RoHS but some could only take 235C. Others were maybe 250 for five seconds but I found that parts specified for

An interesting aside: About a year ago I built a simple "LED curve tracer" that used an i2c DAC, an Arduino and some PC scripting to automatically generate pretty plots of the I-V characteristics of LEDs, and ran tests on just about every LED I could get my hands on.

Medium intensity diffuse yellow 3mm LEDs, like this one:

seemed to have a remarkably "stiff" forward voltage across a wide range of currents. On a log-linear plot of current v. voltage, with current on the x axis, the slope of the line was very slight over operating currents from microamps to about 35 mA.

Here's the plot of the yellow LED:

For comparison, here's a diffuse red:

and a high intensity green:

ction? I've

ED

at

re

Minimal. I'm just coupling your proposition that the LED can also be a power-on indicator with the earlier proposition that it's a photodiode.

There's obviously a potential problem - it could be negligible but without numbers, I can't say. Phil Hobbs thinks he's got numbers, and can be more explicit.

You seem to be willing to imagine a few numbers, which isn't quite as helpful.

"Numbers matter. Nanoamps over milliamps equals parts per million."

Real numbers matter. Imagined numbers are just cheap rhetoric.

-- Bill Sloman, Nijmegen

..

=A0 =A0 =A0...Jim Thompson

Strange data, and an implausible fitting function. You'd expect the LED voltage to be a logarithmic function of current at low currents, then start going up faster at high currents as the ohmic resistances became significant.

You "exponential" fit is crappy at low currents, okay at medium currents and crappy again at high currents. You need to fit a more physically realistic function to your data.

-- Bill Sloman, Nijmegen

..and for an _excellent_ reason that always works.

Whatever it is, it's also going on with the green LEDs measured in the paper that Lasse dug up

Their green LEDs come out at about -5mV/C, about twice what you get on a transistor Vbe.

Most cobbled-together voltage reference circuits need a bit of tuning

- usually by fiddling with the current through the "reference diode" - to give respectable performance.

Even the 1N823-29 reference diode series were only reference diodes at exactly 7.5mA, and to get the stability you were paying for with the

-- Bill Sloman, Nijmegen

junction? I've

Well, I measured it this afternoon. There is no discernable photoelectric effect, even with intense illumination. I didn't expect any, and Phil didn't expect any.

Can you elaborate on the noise aspect? After all, most bandgaps are filtered. Or are you referring to an op amp circuit using the bandgap to create the current reference?

It seems questionable to make such a blanket statement about the LED reference being 20dB quieter without describing the competing circuit.

The plots do show the yellow is the stiffest.

Curious idea, why don't i see any indication of a light tight box around it?

?-)

Because there isn't one? The LED is a voltage reference and a pilot light.

What's the schematic symbol for a light-tight box?

Sunlight on green LEDs is used for solar tracker applications, turns up on various DIY sites.

Grant.

LED:

red:

green:

Yes, I'm aware of the shortcomings of the fit. As you say, at high currents ohmic resistance causes a deviation from log/exponential behavior. At low current, there's deviation due to base current errors from the current source feeding the LED, which was just a jellybean op amp and a couple of transistors. There are also errors due to Early effect, and the fact that in this setup the current is not sensed directly. In this case I wanted the measurement circuit to be as straightforward as possible for Arduino users/hobbyists who might not be too well versed in electronics.

With a little work it should be straightforward to develop a much better current source for LED characteristic measurement, and the test firmware and script could accommodate it with very little modification.

Hey, I used to be able to crash a quartz-windowed 8749 microcontroller by taking a picture of the machine it was in with a flash camera! :)

Just doing a few searches found these:

Also, AN-211 from Analog Devices discussing the use of a pair of back to back 2N3904 BE junctions zenered to supply a 15V reference with very little noise (bypassed with 10uF tants.) See Figure 3 on it, Q14 through Q17.

And check out DIYAudio, from 2002:

Jon

Both the LED and the transistor have the same 0.3% per Celsius temperature sensitivity, and nothing cancels. The current through the resistor would only remain constant if the resistor had a compensatory thermal characteristic, and was heatsinked to the other devices.

There's also the issue of photons getting at the sensitive part of the LED junction. Photosensitive references: just say NO.

A way to make a temperature-dependent diode drop into a non temperature dependent reference is well known: it's the bandgap reference.

No. It was done as a public demonstration of early light emitting semiconductors and was entirely qualitative. I will try it with a spectrograph next time I am near some LN2 (and have spare time).

When I put a light chopper on a laser and detect a weak response signal with a lock-in amplifier, the sensitivity will be much more clear. Pay the extra nickel, use a REAL reference. LM4140adj is suitable, just add two resistors (the op amp is built in).