Some Qs on LEDs for biasing

LEDs enjoy a good reputation for making a bias element (i.e. for current sources or output stages bias) which is silent and has a low impedance, i.e. the voltage drop is rather independent of forward current. So let me ask a few questions:

* can someone recommend a red LED which is know to be reliable, i.e. the voltage drop amongst different specimen is very similar?
* What tolerance has the voltage drop typically?
* how do I simulate a LED? My simulator has thousands of diodes, but as it looks no LED. So far I used ideal voltage sources and looked up the right voltage in the datasheet. There has to be a more convenient way...

Thanks!
Samuel

This looks like a good overview of types:

http://members.misty.com/don/ledc.html

The "Orignal visible red" is the one I have used that seemed to have the best tempco match to bipolars like 2N3904's---GaAsP on a GaAs substrate. Some called them "standard red". I believe the first reference about using them to bias current sources was in a Precision Monolithics app note for a low noise amplifier, where it is merely asserted that they match the Vbe tempco. Sumo used them extensively in some power amps.

BTW the impedance is similar to regular diodes---it's the tempco match that is usually the appeal, coupled with a useful voltage over Vbe.

When I proposed using them in an automotive application for a low-cost temperature-compensated level shifter, for the trigger input of a power amp, I met strong resistance from some who insisted that LEDs were inherently unreliable. However, the references cited for this had to do with their application as solid-state lamps. I prevailed, only to get serious egg on my face when manufacturing f*cked up and hand-inserted the leads into the wrong set of holes on the PCB, which was in turn due to the auto-insertion program people not finishing the program in time for the build. The lead bend put stress on the internal bond wire and some of them broke in wave soldering. Worse, in flagrant violation of QA protocol, the units that passed final test were shipped to F*rd and a few failed at Body & Assenbly :evil: . Very bad news. I was out of town for my first vacation in many years and came back to this and another crisis---that will teach me.

The source of the parts was H*wlett-Packard, generally known for high quality including a stable manufacturing process. I do not recall what the statistics of the forward voltage were, but it was a pretty tight distribution. I conjectured that the consistency of light output would correlate with other parameters like forward voltage, but this was never proven.

More recently I called out some for another client and found that H*wlett-P., now Agil*nt, still made a chip that acted about the same, now in a ribbon-leaded part. I don't know if this is still available, and a search on the site shows things divided into through-hole and SMD, apparently omitting this type. I will try to find the P/N. As well, I have heard that someone was making a part specifically for bias applications.

There is a discussion that rambles on (way worse than this) in another forum about current sources that mentions LEDs and noise, and IIRC asserts that they get noisy at higher currents. I've used them at 100uA to a few mA and they seemed to be quite quiet, comparable to standard Si.

Some people have expressed concern that, being somewhat light-sensitive, that they would pick up hum from exposure to fluorescents and even incandescents. The degree of light sensitivity and the typical photocurrents involved are both miniscule IMO, but if this bothers you you can cover them up. Oddly the same people don't seem to object to the use of glass case normal diodes, which are generally quite a bit more sensitive.


EDIT: found the Agil*nt parts: HLMP-6000. They are apparently still manufactured: http://cp.literature.agilent.com/litweb/pdf/5989-1708EN.pdf

Thanks for your contribution!

BTW the impedance is similar to regular diodes.

Click to expand...

Doesn't "low (dynamic) impedance" mean that the voltage drop is rather independent of forward current? In this respect, LEDs seem to be better than regular diodes.

BTW, I found four LED models in SPICE - I must have overlooked them in the flood of other diodes.

HLMP-6000

Click to expand...

And I can even source them easily - a bit pricier than two 1N914Bs though!

Samuel

I'm not sure I trust the simulator becuase it shows the dynamic impedance almost twice what I though it should be for 1N914B's, but for circuitmak*r's model of a red LED the dynamic impedance at 2mA bias is 48.4 ohms---i.e., not all that low, although it is doing this with a 1.63V drop.

The only properly behaving "diode" I was able to simulate was a diode-connected transistor, a 2SC1815, which at the same bias current shows a Z of 13.0 ohms.

I'd be interested to know what others' simulators show for the regular diodes. I'm perplexed---this is fairly basic stuff. I thought I was maybe seeing bulk resistance effects but the Z is also about double theory at 100uA.

EDIT: Hmmm. I guess these diodes have always been higher Z than I thought---the Nat Semi databook gives dynamic Z curves and shows something pretty close to the simulator. The curves also show no sign of bulk resistance out to 100mA, although it's hard to see much from a log-log plot.

FWIW, the setup is a V to I converter driven from a sine wave oscillator with very low amplitude relative to a d.c. voltage source in series setting the quiescent current. The 914B at an average current of 2mA has an average forward voltage of 724.6 mV, which also sounds a bit high.

Z is calculated by dividing the peak-to-peak voltage by the peak-to-peak current.

I may go to the bench on this one and see what the real parts do.

EDIT 2: H and H point out that a correction factor between 1 and 2 is needed for the thermal voltage V sub T. This would account for the discrepancy, although they give no explanation for this assertion, at least when it is first mentioned.

Pease (in Troubleshooting Analog Circuits) has a series of curves in the back of the book for various diodes. As I recall the slopes are widely variable. EDIT 3: I found the book. Pease's comment: "Nobody ever tells you about these widely varied characteristics!" :green:

No convenient internet refs to explain this that I can find---looks like I will have to dust off Sze, Physics of Semiconductor Devices.

EDIT 4: Ahh, Peter Dunn to the rescue (Gateways Into Electronics, an excellent book if a tough read): the culprit mostly is recombination in the depletion region, and at higher currents the onset of injection.

I'd be happy to e-mail you the models of my simulator, if interested.

Samuel

Sure Samuel. But now I realize that the circuitm*ker ones are probably accurate---it was my naivete about real diode operation. To paraphrase the Firesign Theater, [almost] "everything I knew about diodes was wrong."

What is intriguing about the departure from the first-order Shockley theory are the implications for using diodes to bias transistors in things like class B output stages. It also sheds light on why using a diode-connected transistor produced a poorer result in a recent Ideas for Design piece where a diode was used to linearize a simple transistor amp stage. I thought I would improve performance and found that it didn't work as well to use the transistor.

Not sure I understud you last post right - shall I send them or not? If yes, please tell me which ones.

Could you in turn send me your LED-model? Mine don't model the tempco... :?

Samuel

Samuel, I don't think the models are different, although now that I think of it the 1N4148 doesn't work right, so maybe I should look at yours.

I will try to copy over the red LED model in a bit. I'm not sure if it has a temp parameter or not.

EDIT: Slogging through, here it is:

*Typ RED GaAs LED: Vf=1.7V Vr=4V If=40mA trr=3uS
.MODEL LED1 D (IS=93.2P RS=42M N=3.73 BV=4 IBV=10U
+ CJO=2.97P VJ=.75 M=.333 TT=4.32U)

The 4.32U TT (transit time?) seems to be a very stupid default applied by Circuitm*ker to any damn diode they don't have a spec. for---the most irritating being schottkies, which has led me to making a number of invented parts fro doing anything useful with switching power supplies and amplifiers. Here, for biasing, it is not likely important, but beware nonetheless.

the voltage drop is rather independent of forward current

Click to expand...

This is true only if you are in the linear portion of the transfer curve... if the voltage/current swings cause the LED to get close to the "knee" part of the curve, the output will be non-linear.... so, you have to make sure the variations never take it down to the lower levels.

What tolerance has the voltage drop typically?

Click to expand...

Even LEDs out of the same batch may have a 5% - 10% variation.

regards, Jack

Thanks, Brad. Here the 1N914/1N4148:

*1N914 MCE
*100V 80mA Si Switching Diode pkgIODE0.4 1,2
.MODEL 1N914 D(IS=7.075E-9 RS=0.78 N=1.95 TT=7.2E-9 CJO=4E-12 VJ=0.657
+ M=0.4 BV=100 IBV=0.0001 )

This is true only if you are in the linear portion of the transfer curve.

Click to expand...

The onset of the knee seems to be very dependent on type. Most start to get weak below 5 mA, but the one Brad recommended is very linear between 2 mA and 20 mA, and still above 1.5 V @ 0.2 mA.

Samuel

[quote author="Samuel Groner"]

{Jack wrote} This is true only if you are in the linear portion of the transfer curve.

Click to expand...

The onset of the knee seems to be very dependent on type. Most start to get weak below 5 mA, but the one Brad recommended is very linear between 2 mA and 20 mA, and still above 1.5 V @ 0.2 mA.

Samuel[/quote]
I guess when we are talking "linear" we also mean a relatively high slope, i.e., low dynamic Z.

There was some talk in the other forum thread about LEDs getting noisy at higher currents, I believe Curl stating above 6mA for some part. I'd be interested in knowing if anyone else has data on this, and what the nature of the noise is (white, 1/f, etc.). The bench is crowded at the moment in the middle of a design crisis, and I'm about to be off for a day at that.

Is 6 mA for the forward current or for the CCS-current?

Had a quick look at the model you posted; isn't the tempco positive instead of negative?

Good luck with your designs!

Samuel

BTW: which is the other forum?

>6 mA LED forward current.

I'll have to loiok at the model but LED forward voltage tempco is negative up to high currents I believe. Remember it is a fair match for transistor Vbe tempco hence one of the appeals.

Other forum is the diyAudio.com one.

I'm off to Oceanside and San Diego. There may be internet access in the hotel. Back sometime tomorrow in any case.

Time to resurrect this thread--finally found time to do some measurements on the HLMP-6000 parts Brad recommended: HLMP-6000_forward_voltage.pdf and HLMP-6000_impedance.pdf

The forward voltage distribution is very tight--all measured parts are well within 1% of the mean reading. Impedance (measured at 1 kHz but found to be consistent from 100 Hz to 100 kHz) is very low. Looks like this part is perfectly suited for BJT current source bias.

All parts were from the same batch, so I might need to verify the consistency with another Farnell order.

Samuel

Glad to hear the HLMP-6000 is still available and evidently consistent in Vf behavior. I was just setting up a while back to do a really careful noise measurement when I foolishly dismantled my noise amp to reconstitute it as a still lower noise fixture and then never finished it.

Could you conjure a capacitance measurement at roughly zero bias Samuel? Reason: the good Dennis Colin, who seems to be drunk on scrivening for aXp of late, has, besides yet two articles* (!), a letters exchange in the latest issue (August) with someone pointing out that his "measurement preamp" could have much higher input impedance if he used LEDs as protection diodes instead of 1N4148 parts.

I'm going to warn him that they have high capacitance IIRC, but it would be nice to see some data.

The old warning about light sensitivity comes up in the exchange as well, which in this case has a bit more validity since they are talking about potentially very low currents. Although I do wonder about what D.U.T.s the letter writer envisions this preamp measuring, where such a high input Z would be required.


EDIT: *btw the second article, "Optimum Stages for Minimal Distortion" is quite interesting, albeit controversial. I don't have time to critique it at the moment (although a big problem is the curve-of-growth of distortion harmonics with signal level), but I do love his remark to the minimalist crowd: "For those who believe that having the fewest possible components is always best, I suggest listening to an Edison horn Victrola!" :razz:

Could you conjure a capacitance measurement at roughly zero bias Samuel?

Click to expand...

At home I don't have a suitable meter yet for anything below 1 nF. But the datasheet says 100 pF typical which should be enough to counterfight the idea. But perhaps he could use two 1N4148 (or even FDH300) in series? [Edit: 100 pF is for standard red--orange goes down to 4 pF.]

The second article, "Optimum Stages for Minimal Distortion" is quite interesting, albeit controversial.

Click to expand...

I quickly looked at it (it's online). Needs a more detailed read though.

BTW, did the third Jung CCS article ever got to paper? It never made it online at least.

Samuel

Ah yes I just checked the Avago datasheet and saw the 100pF.

Note the gigantic window for forward voltage at 10mA. I'll bet those limits are set for about 10 sigma process capability :razz: The variations may also reflect a bulk resistance term that becomes important at the higher currents, but that doesn't much affect the light output for a constant current drive.

Interesting how the other color is so much lower C. It probably has lower light sensitivity too.


The distortion versus number of stages argument Colin makes seems to assume that a given stage distortion is level-independent, unless I'm missing something---as I say I haven't worked through his math yet. But typical nonlinearites follow a power law with level, with 2nd growing linearly with level, third rising as the square, etc. If one is building an overall multi-stage 60dB gain amplifier, obviously the early stage sees a much smaller signal than the later ones, presuming the real-world constraint of some limits on the later stages' output swings. The bulk of your distortion will arise in the final stage, surely.

But he's trying to show that a single classical opamp stage can be inferior to a multiple-stage arrangement when a lot of gain is required. Frankly I don't know who would argue against that point for overall high-gain chains anyway, except out of an audiophile minimalist Zen etc. attitude.

Another quibble is that almost everyone plugs in open-loop distortion and then says the distortion reduction when the loop is closed is simple arithmetic based on the thrown-away gain, but almost never take into account where you are at the frequency of measurement on the open-loop gain curve. Few opamps are still in the flat gain region at frequencies of interest for audio. When you are in the usually-predominant Pi/2 phase region it makes things better than the magnitude arithmetic, IIRC.

Rich May advocated, for good sound, open-loop gain constancy out to the highest fundamental of interest in the signal, even if the value of that open-loop gain at a given frequency was not as staggeringly high as a typical opamp.

I do like Colin's extracting e out of his argument though. When we can get Euler's equation then we're getting somewhere :green:

Oh well here we go wildly off topic again

PS: haven't seen Jung III

The distortion versus number of stages argument Colin makes seems to assume that a given stage distortion is level-independent, unless I'm missing something---as I say I haven't worked through his math yet. But typical nonlinearites follow a power law with level, with 2nd growing linearly with level, third rising as the square, etc. If one is building an overall multi-stage 60 dB gain amplifier, obviously the early stage sees a much smaller signal than the later ones, presuming the real-world constraint of some limits on the later stages' output swings. The bulk of your distortion will arise in the final stage, surely.

Click to expand...

That's what I thought as well. The presence of crossover distortion (which might be very level, frequency and load dependent) and common-mode effects likely make matter of truth much more complicated (though I surely welcome the contribution of a Swiss mathematician to the problem at hand). Just apart from the fact that if you'd like to keep noise equally (or at least similarly) low than with a single stage the first opamp will have to face a pretty low-impedance feedback network which again messes with the distortion characteristics.

When you are in the usually-predominant Pi/2 phase region it makes things better than the magnitude arithmetic, IIRC.

Click to expand...

Sure? I'd say that because it's not the loop gain at the fundamental but the loop gain at the harmonics that counts for reducing distortion it's actually worse.

Samuel

[quote author="Samuel Groner"]

When you are in the usually-predominant Pi/2 phase region it makes things better than the magnitude arithmetic, IIRC.

Click to expand...

Sure? I'd say that because it's not the loop gain at the fundamental but the loop gain at the harmonics that counts for reducing distortion it's actually worse.

Samuel[/quote]

Sorry my sloppy---what I should have said is that the gain accuracy at a given frequency is better than the magnitude number suggests. And thus the available distortion reduction is not as bad as the magnitude number at a given harmonic would imply. But yes, flat gain would be desirable, all other things equal.

Also, my statement about the curve of growth for distortion products of different orders is something typical for many circuits. But one can contrive nonlinear networks, however unlikely, to do all kinds of different things including constant ratios among harmonics independent of level. Look for example at the spectrum of a sinusoid run through a cubing operator, or a cube root operator. Note how much the former, or other monomial odd-exponent operators, resembles crossover distortion.

EDIT: But this is not typically what one will find inside a 797 etc.

EDIT: Ironically this pondering has led me, precisely, to a method that I need for another purpose. Now can I bill for it?? :green:

[quote author="bcarso"] But typical nonlinearites follow a power law with level, with 2nd growing linearly with level, third rising as the square, etc. [/quote]

Not exactly true. If we are talking about polynomial operators and if input
is sin(wt), x^3 will produce 3w AND w therms, x^4 will produce 4w AND 2w,
x^5 will produce 5w AND 3w AND w etc.
So if you look at second harmonic in THD spectrum, you will have
contributions from all even order therms in polynomial.

Now nice part. Few years ago I developed expression for all harmonic
produced by 7th order polynomial. While that paper is lost amid junk
in my "archive" I do remember that contribution of higher order is
converging (that is x^6 will give highest DC therm, than x^2 than x^4 and
x^6 will be lowest). OTOH all these components are proportional to
coefficient in front of x^n exponent. So they are usually small cus usually
these coefficients for high order exponents are small.

Now even nicer part. Note that all odd parts of polynomial will produce
component at fundamental. Those components ARE DISTORTION.
When you look just
sine wave, this effect resembles compression. With infinite time constants.
I have no idea what is audible effect of this "fake fundamental", cus
I havent figured out how to extract it. Fun
part is, THD wont show these, IM wont show these. BTW even order
exponents will have largest contribution to DC. This is DC when you
use constant amplitude input signal. With real world dynamic signals, even
therms will produce funny "fake transients".

cheerz
urosh