EEVblog
Vložit
- čas přidán 24. 08. 2024
- How may photons does a LED emit at very low currents?
What is the lowest current where a LED will actually emit photons? Dave does some measurements to answer this quantum physics level stuff!
Forum: www.eevblog.com...
EEVblog Main Web Site: www.eevblog.com
The 2nd EEVblog Channel: / eevblog2
Support the EEVblog through Patreon!
/ eevblog
EEVblog Amazon Store (Dave gets a cut):
astore.amazon.c...
T-Shirts: teespring.com/s...
💗 Likecoin - Coins for Likes: likecoin.pro/@...
Neat experiment!
+Applied Science Isnt it inspirational to make some quantum physics experiment....? :-)
@@JanicekTrnecka Or some DIY semiconductors? :D
Negative curvature over right half of curve looks like detector saturation. As detection avalanche events start overlapping, pulse counts are missed. Actual LED efficiency is still increasing in that region.
+Walt S the sensor can go to 10MHz+
+EEVblog i think that does not matter. with increasing counts the probabilty that two photons appear next to ech other (in a timing sense) does increase. One may calc the probality that out of n events distributed with equal probability over 1s are located with Δt
+vlad tepesch
Yes, time between photons should have an exponential distribution, if I'm not mistaken. But just because we know the maximum of the probability density function is at Δt = 0, doesn't tell us what the mean time between photons is, and whether that's dangerously close to the 50 ns dead time between pulses.
I would recheck with a different LED or attenuate the light output and check if the curve still tapers off at the same counts/s.
+Walt S Apparently you didn't read the freaking datasheet of the thing... The dead time of the internal APD is 50 ns... therefore the error of the measured count is less than 1% at counts < 20kHz (see the formula in the freaking datasheet).
+vlad tepesch Totally agree. Pulses per second should not be expected to just directly translate into photons per second. I also think that with enough data available it is possible to statistically compensate for that inherent flaw of that measurement technique and to estimate a true photon per second count (unless the photon count per second gets way too high and the sensor gets really saturated of course). That is probably why they didn't bother to make their pulses shorter - you would most likely have to do the math on your data anyway.
You almost have an S shaped curve there. The non-linear portion at the top may be saturation of your equipment, but at the low end it must be real. Interesting!
+Kenneth Scharf I was also thinking about the detector saturation at the top part of the curve, especially since he's already shown that above 500na, the reported count stopped increasing... It would be fun to see how the curve changes with the LED temperature...
+Kenneth Scharf I'v been working with a NIM in the past, and also developed an X-RAY photon counter using an extremely sensitive ASIC. was able to count even a single X-RAY photon with over 95% efficiency.
Roll off at the higher count is defiantly caused by the measuring equipment as it is clear that the photon flex is getting higher, all tho the datasheet stated 5MC/s... strange.
I agree, the dead time after each count would have a significant affect when close to saturation.
+Eyal Teerosh I'm pretty sure it's just the crappy counter in the oscilloscope.
TheHue's SciTech Not the counter in the oscilloscope, the detector itself has a dead time after each pulse in order to quench the avalanche in the diode. But if the detector has a 5 MC/s rating then this should not have ben an issue at 1.5 kC/s.
I work with that detector in my research!
Although you are a long way from saturation, the detector includes an automatic gain/quenching adjusting mechanism (undocumented but discovered by Makarov's group when they did a full reverse engineering of it: dx.doi.org/10.1364/OE.19.023590 presumably its inclusion creates better apparent performance.. but is a problem for some Quantum Key Distribution systems!)
This is unlikely to cause a problem with your LED demonstration, but has potential to reduce sensitivity as the light increases in orders of magnitude. Saturation and statistical distribution vs. deadtime are really only a concern when you get to 30Kcts+.
If you are concerned by the odd negative curvature on the right of your graph it may be worth repeating the experiment with an optical attenuator (camera ND filter would do) to confirm the behavior is the LED and not a detector effect!
Although the datasheet allows very high dark counts, all 6 of my units are at approx 70 dark counts per second, with minimal shielding effort... so the donor may have marked it as an unsuitable device to perform low photon rate experiments with, the SiAPDs in these units are binned by dark count and better units cost more.
Some things that come to my mind that might warp this figure:
- Missed photons due to dead time (50-60ns for counts below 5MC/s it says). Someone might want to calculate this.
- Shifting of the LEDs spectrum with higher current
- Shift in emission direction of the active LED region so that more photons are emitted into another direction than the sensor.
Page 6 of the datasheet has an actual formula to correctly defer the count rate from the measurements.
For measuring higher currents you might want to place some ND filter between the LED and the sensor. For the spectrum part maybe there exist some very narrow optical filters, or you can run things through a prism and a slot.
This would be a Video idea for Ben Krasnow from Applied Science.
+Dennis Lubert The dead time correction is actually quite simple to calculate. A nonparalyzable model would say the real counts n for measured counts m and dead-time t is n = m/(1-m*t).
So for a 60ns dead time (worst case), his top measurement of 15,000 counts would correct to 15,013 counts--less than 0.1% correction. Dead time isn't the issue here--it's most likely an issue with the counter in the scope.
@@DavidOlson14 I immediately thought it might be a dead time issue in the APD circuit also. Another possibility is that the issue is in the oscilloscope doing the counting.... might there be a dead time/sampling issues there? The scope display quite obviously has a max pulse density that it can display. Neutral density filters would be nice, but apds are mostly setup for fiber optics, so you would need to be careful that the NDF isn't changing your coupling geometry
Just FYI, the eye is actually extremely sensitive to light, but you need to be dark adapted. This is normally achieved by being in complete darkness for at least half an hour. Under those conditions, you can detect on the order of tens to hundreds of photons per second. According to the source below, you can see down to around 5 photons per 100 ms. That's 50 photons per second (though I think 50 photons/s would actually be easier to see than 5 photons/0.1 s). But the eye filters out about 90% of photons before they get to the retina, so that's 500 photons per second from the LED. So I'd expect that you could detect the LED at 50-100 nA rather than the 800 nA if you were dark adapted.Source: math.ucr.edu/home/baez/physics/Quantum/see_a_photon.html
What about the people who had their lens surgically removed?
I just watched EEVblog 713 and i shit you not, your son is smarter than some people in my advanced electronics course.
That's a nice experiment ;)
Well, to make this measurement reliable a few extra steps need to be taken.
For once, the wavelength of the light emitted from a diode is not completely independent from the voltage applied. This means the wavelength shifts (and with that the efficency of the detector changes) during the measurement. If you have a spectrometer you can check this and the applied voltage at the diode should be measured during the experiment aswell.
Further more: as the LED gets warm, its efficiency lowers. Meaning that as you drive more current through it, you get a lower efficency for the light emittance. Coupling the LED to a heat baths at different temperatures could be used to mitigate the problem (or driving it with pulsed currents of equal average power so as to have the diode run at potentially similar average power but different currents).
Then, when you count high rates of photons you need to account for the dead time of the detector. As you have an avalanche detector, after detecting a photon, the detector stays insensitive for further photons for a short period of time. This error is largest for high count rates. You can easily correct the count rates for the dead time when you know the dead time of the detector (or measure it).
At very low currents you also have the problem of noise with the current source. So the current-errorbars need to be measured aswell. I might also have unintentionally underestimated the errorbars for the photon rates slightly. They should be calculated according to gaussian error propagation.
Depending on the precise construction of the LED, the directional distribution of photon emittance also might change as the applied voltage is changed. This is easy to control for as one has to simply repeat the measurements with a differently oriented diode and see of the resulting count rates are proportional to the original ones.
These, and other unaccounted for effects may lead to the drop of in count rates for increased currents you observed.
But generally yes: at very low applied voltages even the metallic contacts to the semiconductor material start to behave non-linearly. From experience it is already not that easy to get simple contacts completely linear on e.g. p-GaAs if less than 10 mV are applied. And of course the n-p-junction is also not completely linear in behaviour. Most electronics are only approximately linear over the range over their typical range of operation.
But well done ;)
1 photon! ah ah aahhhh!
2 photons! ah ah aahhhh!
[etc]
I bought a DS1054Z, which still believe is a nice tool for the price. It has several limitations, though. Try saving the full memory data to the flash drive as a CSV file... 2 weeks later you may tell me how it went. :-)
a 1 minute tech tip ? Dave to make a 1 minute video ? this i want to see !
+Lynton Rodda He should do a new series called "One Minute Tech Tips" and have the videos average 10 minutes each. Only the regular viewers will get the joke. ;)
Please build a cnc tripod for this collector, then go into the darkest room you can find and try to make a high exposure image.
The danger high voltage warning was hilarious. The current is so low you wouldnt even feel much of a poke. I repurposed the HV psu in these to operate the MCP in a night vision tube, and the output also did double duty as overload feedback. ❤
This measurement setup discussion is a nice companion for your synchrotron tour and all this "why there are 1 GHz LeCroy scopes standing everywhere and they're treated like basic pocket multimeters" in nearly every experimental physics lab:)
This was very interesting. Now you gotta go out and buy some fancy/expensive LED and give that a go. I need to know how the data changes!
Maybe the reason the diode started up is a combination of factors:
a) it had a few seconds to warm up
b) your shielding of AL foil and electrical tape is not going to stop AM/FM so I'd hypothesize that some of that got through and got detected and rectified just like it does in a Germanium Diode used in an AM radio set.
That is absolutely amazing, looks like so much fun.
Interesting experiment. I suspect that the rolling off you see at the highest current is an artefact of the experiment rather then a real effect. We know from the data sheets that the output increases with increasing current up to some point. If you extrapolated the graph you have the output would have to level off and then start increasing again to join up with the data sheet graphs. That would indicate that there was one effect at very low currents and then another at quite low currents and a third at high currents. To me that feels like one effect to many but who knows, we'd have to do the experiment and find out.
+Wobblycogs Workshop Personally i think that if photons hit the sensor "at the same time" they get counted as one. Similar to overexposing a photosensor, taking the average of the photo with increasing exposure will also show a tapering-off at higher exposures. So its probably just a limitation of the set up, measuring out of the linear range of the sensor. (Similar effects for measuring Optical Density (OD) of bacterial cultures at higher culture densities and high fluorescence levels in fluorimetric experiments.)
@@ReapingMiner measuring OD actually has this problem at low OD, actually, not high. At high OD, you are limited by noise (ideally shot), not overlapping pulses... high OD means few pulses, so less overlapping
Some of the shortest 30 minutes I ever spent. Feggin' great, M'Lud!
I haven't watched the whole vid, but I am wondering - how sensitive is an LED to incoming photons?
i found that most linear relationships when scrutinized at the extremes display non linear relationships, they are reported as linear because that is the region in which the thing under scrutiny usually happens at, which coincides with the rising limb of the curve and is mostly linear. So the reporting of linear relationships is usually to simplify the condition irrespective of limiting conditions at the extremes.
That device is from right in my back yard, at the west end of the island of Montreal. :-)
I used to work in a pre-press shop equipped with a photomultiplier tube drum scanner made by Dainippon Screen; that was an amazing device!
you can reduce the dark counts by cooling the sensor down, the device already does some of that, but some dry ice would probably make it even better.
Great video! Only problem is most people can't afford to buy one of these modules, so it would be really great if you could show us how to build a quenching circuit with a SPAD diode from scratch. There are some "safe" low voltage SPAD's that might work (
That's fun stuff. There's got to be a way to use that detector in a double-slit experiment.
Love it..Dave/EEVblog at its finest.
Very interesting video.
fun fact: an atom in a high brightness LED emits a photon every 15 minutes.
Maybe 250 count is the 'maximum' one would expect, but for their experimentation, they needed only ones with a 200 or lower dark count.
Dave, Why don't you use Universal counter with higher time period, let's say 60 seconds? You can use the single capture after changing current through LED? Can you guarantee that your oscilloscope is not interleaving data from different measurements?
+Ianislav Trendafilov the scope is real time single shot capture
+EEVblog excellent video. I got the same 6th order polynomial curve when I plotted the relationship between a PLC analog output sent to a phase angle controller on a 230v 2.6Kw water heater for a batch reactor. It looks like a cosine wave.
+EEVblog
Hi Dave!
please test some other stuff like
wire
germanium diode
1n4148
or hot iron tip
...
you can go higher if you use a Filter between the led and detector. the amount a filter takes away you can measure in a range you already know and then extrapolate, add more filters for covering more range
I think the rolling of is a feature of the SPAD sensor. The diode requires a finite time to recover (quench) after it avalanches, if photons are striking it during this period it fails to recover. I imagine if you continued to increase the current to the LED the SPAD will never recover and you'll stop detecting anything.
Look into your "detector" dead time (in this case you have two detectors, the APD sensing the LED output and the RIGOL scope which you are using to detect the pulses. Both of these "sensors" will have an associated "dead-time" and pulse overlap associated with the time constant of the "detection" process. This probably is the reason for the response roll off at high rates.
as already commented by others, the roll-off at higher currents ist certainly caused by the dead-time of the detector and the scope!
The junction temperature of the diode increases when you ramp up current, which would decrease the photon output efficiency, that might be why it rolls off with higher currents.
I tested some clear ultra-bright LEDs ($2 for 30 on eBay). In a darkened room it is possible to see RED LEDs down to 300nA and GREEN LEDs down to about 80nA. Its amazing sensitiviity!
love collecting unknown data like that, super cool!
Oh yeah! I saw that photon counter in your mailbag before but I didn't notice it came from Quebec. I live there!
You could use an optical attenuator to see what happens at higher currents. You could use some sunglass and measure the transmittance and include the factor in your data readings.
Superb! I'd love to see the curve all the way up to 20mA.
Have you done a video on how LEDs work? Love this type of video, great job!
holy cow dave. that is an awwesome experiment. fascinating.
If you ran out of range on the detector, can't you move the LED further away (or put a darkening filter in the way) and re-run the test? Then use the overlapping measurements to come up with a multiplier value that will align the two sets?
Repeating this until the LED blows?
(I'm assuming the response to distance/filters stay linear)
I'm fairly sure he said it was the scope.
Fascinating video. Didn't even know photon counters existed. It's pretty cool to see the equipment you use(or could use) to capture the data
I would like to train an animal with good night vision to press a button/lever when the LED is on to see how low they can go. In theory they should get close to the photon counter (many animals like horses, deer, lions, cats etc, have a night vision thousands of times more sensitive than a human).
Amazing that there is something that can COUNT photons. whoa
+Dave B Every single image sensor, be it ccd or cmos, counts photons.. Some are more efficient at it than others.
Excellent video Dave, I have two of those exact SPCM units which were scrapped due to high dark count, and haven't yet found a use for them! We use them for photon correllation spectroscopy. Not a cheap bit of kit, as you say!
Interesting to see the non linearity at a very low threshold level. I would anticipate it has something to do with an exponential in the energy of band gap algorithm.
Really interesting.
I'd like to see other stuff shoved up it's "clacker". Like various led colors and perhaps a laser diode.
ooh, interesting.. the transition between "lasing" and just producing light
I have designed a board for a photon counter, the other guy did the firmware for it and it was a pig to calibrate! Trying to get a batch of them to give below a rated dark count and all perform the same was like trying to catch a fart in a breeze! no wonder they were expensive! It was a very interesting project to do though: quantum physics, electronics - what's not to like! Interesting video! I often wondered this!
Cool. It looked like the light was purple at very low power. Color actually changes with temperature.
If you have a good recent DSLR camera you could try 1 second exposure or longer.
That time Dave tried to shake the electrons out of the cable).
Dave. On the Rigol you could have used the hw freq. measurement that you have still turned on all the time. It was showing the pulse frequency ( count per sec) as you wanted. Missing stat on this parameter is annoying but the tool can measure the count correctly.
One reason you might be seeing the end turn down is the sampling rate on the scope is too slow and you are seeing the pulses pile on top of each other and being counted as a single pulse, just like you saw at the currents above 500 nA.
It would be interesting to see if you see the same thing with the counter or by taking the integral of the scope waveform.
Maybe Mike Harrison should send you one of his PMTs and a frontend to go with it.
They clearly say the thing has a deadtime. No need to invoke quantum mechanics, dark matter, superstrings, multiverses, Alcubierre drive or false vacuum metastability to explain a basic common sense every day phenomenon of the upper part of the curve bending. The difference between a top optoelectronics scientist and an amateur is that the top scientist actually turns on the basic common sense thinking skills he learned in the kindergarten.
Where can i get a 2016 special edition MSO 3000?
Dave, did you try to touch the LED wires to see if the noise your body receives is enough to turn the led on ?
Shouldn't you wrap it in lead bricks to block any background gamma radiation? I think the dark count would drop close to zero.
Dave, when you showed the graph of number of photons by current in nA, I believe the shape of the curve would be called slightly "sinusoidal". :) Love your videos mate! Cheers from a younger player in the US of 'Murica.
what do you mean by "limits of our current setup". If its the test setup causing this saturation it would be irritating to take this data into account, isnt it?
The latest Rigol firmware update (10.2015) added the pulse and edge count feature. So you can upgrade the oscilloscope for new free features. Also, in the measure menu there is an option to set the area used for measurement, i.e. whole screen or cursors. It's just the problem of RTFM
Interesting! I bet that the saturation at high current is caused by the photon counter itself. If theres some small leak resistance / offset current, it might show up as a non-linearity at low current values?
nice use of that free kit boss.
I saw your count taper off at high currents so it appears sensor is non linear.
I would like to see a group of 5 LEDs in at the input and then measure pulses of each LED individually. Then turn on 1 then 2 then 3 etc. This ensures linear input so pulses per second should also be linear. If not then you will have to calibrate sensor accordingly.
I'm a bit worried about how the end of the test setup is the upper inflection of the curve. The bottom would be more intuitive and maybe accurate. Could the top be the result of the end of the test setup, like sample time, or sample blocking up? It is a bit unlikely you just happened to get the inflections ranges but maybe you did.
You can use 83.3 ms/Div on the Rigol ;-) Then you get a full second on Horizontal.
Thanks to the cool cat who sent in the photon counter. Cheers! :)
I wonder about one thing. You said that Rigol will miss some pulses because it doesn't have enough memory depth. On 100ms it was 10M and 10 MSa/s. On the other hand Tektronix on the same time base was set to 1M and 1MSa/s. I thought that the most important thing in such measurements is to have as much memory as possible to be able to set highest sample rate (I know that scope does this automatically).
Hypothesis: The measurement is showing the dynamic range of the photodetector ? looks excactly like the distortion curve for transistor amplifiers.
Might the sudden delayed jump in the count at 7:45 be due to the amount of time it took to charge the capacitance of the LED at such a low current? So once it finally got up to a high enough voltage it actually started emitting a few photons.
Dave Tech tips, coming soon :P
Very interesting video. And very good information about counting pulses, I'm experimenting with pulse counting and this was just what I needed to know for now so thanks. :)
You ought to do a video on Understanding your oscilloscope's limitations.
I have picked up bits and pieces from your videos.
Is your powersupply really linear? How does it work? Especially the upper end seems pretty suspicious to me.. You seem to be far away from the saturation of the detector..
Otherwise this might be eplainable by differents layerheights in the diode, some may conduct earlier others later, the laters may be eclipsed, so no light is emitted which hits the detector..
Cool experiment
So 100ms sampling time doesn't have the "vibe" but 1 second does? You could also use 10 seconds, sure the longer the sampling time is, the closer your counts will be to the average. But the average will still be the same. Also, are you sure the Rigol's frequency counter measures the distance between pulses? In that case one would expect it to be all over the place, but it wasn't it actually displayed pretty much the same values as your final setup with he Keysight did (just no average of course) and it also seems to refresh once a second which may suggest it simply counts the pulses (or rising/falling edges) every second and displays whatever it counted...
In the decrease in gradiant at the higher currents maybe because it wasnt measured correctly. The same reason why it dropped off at even higher currents?
I'm a bit sceptical about that.
+Hellsslave666
I agree. I don't trust the values beyond 10 000 Counts/s.
Checking with the universal counter might have been a good idea, to rule out the scope miscounting pulses.
Love this kind of geeky lab-grade "arm-and-a-leg-expensive" type of gear
+catt87 and a kidney
+meh mehabd and sometimes even a testicle
FOXER those are acutally only worth ~30k where a kidney is around 300k
It might be interesting to see a log/log plot of the results, as opposed to lin/lin.
@ 5:20: near perfect Poisson statistics: mean = 212.8, standard deviation = 14.4 → variance = 14.4² = 207.4. Of course for Poisson statistics the mean is the same as the variance, and for dark current like this you expect Poisson statistics.
To follow on on that: something is fishy when you actually measure the signal. The sum of two Poisson distributions is another Poisson distribution, but the stats seem to be off. Is that an indication that the signal is not yet stabilised?
Remember that as it is a diffused red package you will lose half the emitted photons to the package. A clear one would have a much better performance. The water clear hyperbright red I have had give enough light to see at under 10uA, though if i was to use a good dark room I probably could be seeing the light at 1uA or less.
+SeanBZA Do you have any data on losing half the photons to the package? Where does the half figure come from?
+EEVblog Rough guess, the die is not going to convert all the electrons with enough energy into light, and not all will escape the surface. Of those that do the non clear package will scatter some, absorb some, so you will not get all going out the front. Thus the loss compared to the water clear package.
ps you need to clean the switches on that current source, you can see some large spikes in changing steps on it in the final light test, Somewhere there is a contact that is breaking before the next makes.
+SeanBZA Maybe I should ask you - How sensitive is an LED to incoming photons?
+Noth Ankyou I use pairs of regular red LEDs as transmitters and detectors in some industrial applications. They are pretty sensitive, the transmitting beam doesn't have to be super close or accurately placed.
***** Interesting indeed, thank you.
Dave why don't you make a dark chamber and use distance to attenuate the light from the LED to extend and augment the readings from the photon counter?
I love these types of videos.
Please test this again using the universal counter and totalize over 60s. I think you hit the limits of your test setup...
The device should be able to count up to 10 million counts per seconds, so something might be acting up (or might it be the sampling frequency of the oscilloscope too low to catch the pulses)?
It amazes me I can make a white LED glow just by touching it (60Hz pickup)!
And I don't need a photon counter to see it ;-)
+qwaqwa1960 There could easily be passing 10uA through your body into the LED (from the capacitive coupling between your body and mains wiring). If I'm not mistaken, the smallest current humans can sense is around 100uA (that is through healthy skin), so with 10uA you shouldn't be able to sense it at all. ;)
cool experiment! very cool when you go into the science of stuff! I own the rigol ds1054z and yeah, good for the smaller projects for sure.
An interesting relationship between the LED forward voltage Vf and the colour of the light emitted.
Many LEDs have a Vf less then that predicted by the Einstein equation, suggesting the bad gap jump is greater than the forward voltage. I would really like to know the true mechanise for this, as the LEDs colour is a relatively poor guide to the minimum Vf value.
what about temp of the led does it make it move efficient when running at nA
You should do a video on transimpedance amplifiers. PMTs are super cool and photonics is a great subject.
Thank you for sharing this experiment. The mayor source of error is, I think, the coupling between the led and the fiber connection in the APD. Dave, have you used the formula in page 6 to calculate the "actual countrate"? . In this case afterpulsing does not appear to play a big role.
What do you think of using your low voltage sources to apply a very small differential between two opamp inputs and to check if the output is like the open loop gain describes it in the datasheet?
"Stupid Dual Bloody Button Pain In the Arse Interface, who invented that crap!!?" Couldn't agree more... What a shitty system! Nice investigation though, never thought any photons would radiate at those low currents, very interesting!
Just a hunch, but is this not looking like a Bell curve? The chance of high level electrons jumping to the lower level is dependent on the amount of those electrons, which is dependent of the voltage over the LED. I know you measured the current, which should be linear if every electron emits a foton, but there is maybe some leakage of electrons?
Interesting. As for the curve shape, are you measuring the characteristics of the LED, the current source or the detector - I would think all of these. I doubt you have the sensitivity to reliably detect single photons and in any case, not all of the emitted photons will reach the detector. Recalling 1A is 1 Coulomb per second and the electron charge is 1.6E-19 Coulombs, even 10nA represents a significantly large number of electrons passing through the LED junction. I suspect the photon efficiency will be lost in the noise.
The graph looked to me like it has 3 stages, the start of the graph at low current the signal to noise is very low so its possible that you are seeing some detector noise effects. However, it is very smooth looking so probably something else. At higher current the relationship looks linear and at the end there seems to be some saturation of the detector or counter. Fitting a straight line to the linear section looks like it will intersect the current axis at some small positive current. That's maybe a leakage current? The leakage may change with applied voltage? How about measuring the volts/amps relationship? Does it look the same?
This was a really cool video, I really like these kinds of experiments. Are these things poised to overtake PMTs at any point? We have brand new hardware (a flow cytometer) that uses PMTs for its detectors, and they're still being put in a lot of things.
I wonder why you did not uncap the led or use one that the covering could be easily removed. Couldn't the red plastic effect how many photons you detect?
Well done Dave! Should have graphed relative to number of electrons in an amp (coulomb (~6.24e18 electrons) per sec). Then you'd see actual efficiency down at that level. :) hint, given your data, pretty darn inefficient. :p
Need some optical attenuators so we can go up to the full 5mA.
don't you think using transparent led might be better, since red led (actually coloured, semi opaque) actually applying some filter?
Dave did you temperature compensate, the output of an led change a lot, depending on the temperature,