Lumen vs. current and power

I have added a nice Gossen Mavolux lightmeter (with calibrated V(Lambda) curve) to my equipment and have used it to characterize some of my luxeons with different bins.
I have tried to measure lumen output with the following methodology and assumptions (comments or critisism is welcome):
The fraen lp datasheet specifies 13.4cd/lm efficiency Fraen LP datasheet, therefore it is possible to estimate the lumen output of a Luxeon I and III colimated by a FraenLP by dividing the Lux reading at 1m by 13.4.
I have used such calculated lumen values to calibrate a simple white integrating box. The measurement with such a box is much easier and allows the tranfer to Luxeon V.
The TWOH was an emitter, while all other LEDs are stars. I have used a big copper plate (20cmX40cm, 2mm thick) for cooling.

Ok, here are the results:

Lumen vs. drive current:


Lumen vs. power (log/log)


The 4XTV1J curve is calculated from the TV1J data by doubling the voltage and current and multiplying the light output by 4. This should simulate the parallel/serial configuration of a LuxV.

Some conclusions:
- LuxI behaves the same way as LuxIII
- LuxV has its main advantages above 2W
- The percentage lumen increase of a better bin is independent of the drive level (parallel shift of the curve in the log plot)
- 4X LuxIII is better than LuxV above about 1W, probably due to the improved thermal resistance.

Very nice work, Peter. I like the innovative way you used the specs for the fraen optic to get from a lux measurement to lumens. Do you have any feel for the repeatability of this technique? With the same emitter, if you repeatedly setup and measure with the fraen, what is the scatter of the data? I would expect at least a small variation with placement of the optic and accuracy of picking the spot in the beam to measure.
EDIT Peter, I reread your post. I wasn't paying attention the first time. You made these measurements with your integrating white box? It was just the initial calibration that was done with the fraen optic. Right? I would suspect repeatibility of the relative box measurements to be pretty good. I would still be interested in the repeatability of the box measurement with the same emitter. Set up, measure, break down, set up again as if it were a different emitter, remeasure.

Cool Peter!! /ubbthreads/images/graemlins/thumbsup.gif

EDIT: I noticed that with some of the few TW0H's that I had that although they were of a lower Vf, their lux was not as impressive as some J Vf LED's. Your two samples here are also supportive of this. /ubbthreads/images/graemlins/icon3.gif It would seem that the efficacy may not be greater with the lower Vf.

AWESOME work!!!

Peter, would it be much trouble to greatly expand the first .3 or .4A part of the graph for greater readibility? Something I've noted in somewhat similar testing is occasionally an emitter will differ in relative performance at low and high currents. For example, an emitter may do very well at low current relative to it's peers but underperform at higher currents.

I have a bunch of data I can contribute for some Lux I high domes, and Lux V high domes too. Tested at 1M from 130mA to 1.5A. I can post my lumen estimation technique when I get home (it's quite a bit different).

/ubbthreads/images/graemlins/bowdown.gif great work ! /ubbthreads/images/graemlins/thumbsup.gif
bernhard

Okay, here's my estimation technique.

Rather than use an optic, I decided to test the bare emitter. I set up a light meter at 1M from the emitter, set dead center, so I would be measuring the brightest part of the emission pattern (at 0 degrees off center). I then took lux measurements at 130, 310, 620, 920, and 1540mA.

Now comes the tricky part. I used the lambertian beam profile in the luxeon datasheet to build a spherical surface profile of how much light falls on part of a half sphere illuminated by the emitter.

I then used some tricky geometry to construct a set of spherical caps (spherical surfaces), based on 20-degree wide cones (so 10 degree off-axis) angles. Translation: If you made an angle, then rotated it to trace a circle in the surface of a sphere, that area is a spherical cap. I calculated the area of a series of those, then subtracted the area of the inner cap from the outer cap in the series to produce the area of a set of spherical cap rings.

Then, I took the areas of each of those spherical rings, and multiplied those by the average relative brightness for the appropriate relative brightness measurement from the datasheet for the given outer and inner angles.

Then I summed up the totals of those rings.

What I wanted to get at was to take a single lux measurement at the center of the beam, where the lambertian profile says the brightess is at 100%. So that's my reference measurement. I can assume that the average brightness in a spherical cap in the center 20 degrees (10 degress off axis) is 96.75%. I also have the area of the cap. Since lux is lumens per square meters, if I have a measurement of 10 lux, and the cap is 0.1 square meters, and the average brightness is 96.75 percent, then in that center 20 degrees has 0.97 lumens falling on it. I can do the same for a series of rings on a sphere, by knowing the area (calculated geometrically), and the brightness at that ring (known from the lambertian distribution), I can calculate the lumens falling on each ring, then sum them up.

In the end, I get a "lumen coefficient" that I can multiply a dead-center lux reading at 1M from a bare lambertian emitter by, to get the estimated lumens of that emitter.

Here's that coefficient: [edit, updated]

3.29

So what this means is that if you have a luxeon lambertian emitter/star, and you set it up 1M from a light meter, dead center (so you're reading the maximum brightness from the emitter), and take a lux reading, you multiply that lux reading by 3.29 to get an estimation of the number of lumens emitted by the luxeon.

Here are my results:



[edit] I updated the coefficient, due to a calculation error in the sphere calculations.

Darin, nice work. As I am sure you know, the distribution pattern of a lambertian emitter differs a bit from the true mathematical lambertian distribution. The true lambertian goes to zero at 90 degrees but the emitter does not due to light emitted from it's edge. The datasheets give graphical data showing the upper and lower bounds for the possible distributions. I am curious: did you attempt to do a piecewise intergration using the upper and lower graphical bounds to get a sense of the degree of possible variation from the 2.63 factor that you computed using the true lambertian[I am assuming here] pattern?

Different question: what was the heatsinking and measurement protocol associated with the data you have presented?

Hi all,

As you can see I am a newbie in the forum, at least when it comes to contributing. I have been browsing the forums for some time now, but believe it is time to start contributing too.
I think that within the subject of this thread you will find interesting to consult the US patent #6686691 by Lumileds. You can do that at http://www.freepatentsonline.com/ for example. In that patent they describe a LED system capable of emiting 575 lumen / Watt at 535 nm!

The future is bright!

C6H6

Crap! Forum outage ate my post /ubbthreads/images/graemlins/mad.gif

Crap^2 - I found errors in my calculations for areas of spherical caps. I will update my OP to reflect that.

Doug,

I took the average of the upper and lower bounds, by reading the graphs in the datasheet - so what Lumileds tells us the distribution is. Here they are calculated individually:

Average: 3.29
Lower: 2.85
Upper: 3.74

Testing: I used an LM350 cc power supply. I also measured Vf with each, so I could measure power dissipation, then adjust to a 25C Tj "output" (to satisfy my own curiosity) I put the actual value in the graph, not the 25C adjusted. Each current was tested for 5 seconds or until the measurement settled. Ambient temp was 18C.

Both Lux V and 1 Lux I were in a Maglite, 1 Lux I in a 3AA brinkmann, and the remaining stars were bolted to a 4"x3"x2" finned Al heatsink (1lb weight).

Honestly, I think things tend towards the lower bounds, just because that number gives a more "beliveable" number.

Nice graph! There was a theory that "R" bins were ending up as "T" bin Luxeons once ramped up to their full rated drive current, it seems that it's plausible after looking at your graphs.

Closer to home, I don't feel so bad about getting a R2K - at first I was thinking a TxxK Lux3 might be better, but given your graphs, an R2K is functionally identical at 500mA, and that's great news, I'm not missing out on anything.

Thanks for the positive replies!


[ QUOTE ]
Doug S said:
Peter, would it be much trouble to greatly expand the first .3 or .4A part of the graph for greater readibility? Something I've noted in somewhat similar testing is occasionally an emitter will differ in relative performance at low and high currents. For example, an emitter may do very well at low current relative to it's peers but underperform at higher currents.

[/ QUOTE ]
Hello Doug, here is the zoom.


Unfortunately the smaller currents are not too accurate. I have used an agilent 6V/2.5A bench supply and since the 6V are insufficent for the LuxV I have boosted the voltage with a NiMH cell and recorded the voltage (bench supply+battery) with a Fluke DMM (=> the voltage readings are very accurate). The current reading of the bench supply is quite accurate (have checked this before) but shows only 10mA steps => there is an uncertainity of +/-5mA which could influence the low current readings. Therefore I'm not sure, if it is possible to judge the effect you have described. I really need a second DMM!
If I stay with the LuxI/III's I could rerun the test with a DMM current reading. What would be the relevant current range? From 100mA to 350mA?

Repeatability: I think it is quite good (feeling from the handling) but I will do your proposed test and report the results (probably in a few days).

Evan9162: Great! Your method is more an "ab initio" method without assumptions, and our results fit pretty well together. I will compare your factor 3.29 with my results.

Don: Yes, unfortunately the TWOH's seem to stay in normal efficiency ranges. They are still very attractive for LiIon guys in a downboy or SDC configuration.

Peter, Thanks. As to relevant range, that is up to you. I sometimes go as low as 10mA but I think any differences would be lost in the resolution of the plotting on the graph. Lacking a second meter, have you considered using a low value resistor in series and measuring it's voltage drop to get current?

Doug,

One question that the forum ate (that I forgot to put back in) - Do you think that lumileds measures luminous output by using an actual integrating sphere, or do you think they have a coefficient that they multiply a lux reading by to bin these guys? To me, it makes more sense (and is faster/easier) to measure lux on-axis, then it would be to ship each emitter into an integrating sphere for a measurement.

[ QUOTE ]
evan9162 said:
Doug,

One question that the forum ate (that I forgot to put back in) - Do you think that lumileds measures luminous output by using an actual integrating sphere, or do you think they have a coefficient that they multiply a lux reading by to bin these guys? To me, it makes more sense (and is faster/easier) to measure lux on-axis, then it would be to ship each emitter into an integrating sphere for a measurement.

[/ QUOTE ]
Darin, according to Lumileds Applications Brief AB08, Optical Testing, they actually use a 6" integrating sphere on the production line. This does seem surprising. I would have expected a method similar to yours. BTW, something never discussed here on the forum it the fact that Lumileds specifies the tolerance on their flux measurements on both the datasheets and in AB08 to be +/- 10%. A consequence of this is that it is theorectically possible that tested under datasheet spec conditions, an "S" ranked emitter could be as much as 22% brighter than a "T" ranked emitter and both could still be within spec.

Interesting. I will have to check out AB08. 10% margin of error in the binning process...wow....

Doug S,

Interestingly, Lumileds says that it tests 100% of its emitters, using "instantaneous" flux measurement, rather than temperature stabilized measurements.

Each unit is binned for luminous flux, forward voltage and dominant wavelength or CCT. All binning is done at the nominal current.

When it comes to binning, the wavelength bins are 2.5 to 20 nm wide, the Vf bins are .20 to .50 mv wide, and the flux bins have a 30% range.

Lumileds goes to great lengths to get designers to incorporate these variables into their designs.

[ QUOTE ]
Doug S said:
Do you have any feel for the repeatability of this technique? With the same emitter, if you repeatedly setup and measure with the fraen, what is the scatter of the data? I would expect at least a small variation with placement of the optic and accuracy of picking the spot in the beam to measure.
EDIT Peter, I reread your post. I wasn't paying attention the first time. You made these measurements with your integrating white box? It was just the initial calibration that was done with the fraen optic. Right? I would suspect repeatibility of the relative box measurements to be pretty good. I would still be interested in the repeatability of the box measurement with the same emitter. Set up, measure, break down, set up again as if it were a different emitter, remeasure.

[/ QUOTE ]

I have just started some VT drift experiments ( Luxeon Vf shift over time ) and therefore I have got also some figures for repeatability since I'm using the same LEDs:
LED old Lux new Lux %
R2K 7.98 8.23 3.1
QWOH 6.59 6.76 2.6

The values are for 350mA (which I use now for the drift experiments) => the repeatability isn't too bad. I would trust the values to +/- 5%

By the way: The R2K has drifted by 123mV over 24 hours and the QWOH measurent is just started. I'm interested in the influence of the VF@0h on the drift. I will report the result in Doug's thread.

[ QUOTE ]
McGizmo said:
Cool Peter!! /ubbthreads/images/graemlins/thumbsup.gif

EDIT: I noticed that with some of the few TW0H's that I had that although they were of a lower Vf, their lux was not as impressive as some J Vf LED's. Your two samples here are also supportive of this. /ubbthreads/images/graemlins/icon3.gif It would seem that the efficacy may not be greater with the lower Vf.

[/ QUOTE ]

Exactly.

Though there are others here that feel differently.

The instantaneous Lumen measurement by LumiLEDs doesn't really account for thermal resistance variations very well either....

I notice on your graph, it appears that the Greenish bin parts are doing more lm/w, see X1 and V1