White LED lumen testing

[NewBie]

The meter is supposed to be within 10%, so I assume this is the upper bounds of my error. In practice, based on my relative measurements in Silverfox's light meter benchmark testing, my light meter is reading maybe 5% high for white LED type light compared to the average (although the average isn't necessarily dead accurate).

Unfortuately, the majority of the error is in the spectral response. These meters are "calibrated" for incandescent, sodium vapor, fluorescent, and such. Some have correction tables built in you select the type of light source, to adjust for the error, others have the correction factor in the manual that you have to apply manually. Read your manual, if it worth beans at all, they will mention those correction factors (the LM631 is only accurate for tungsten lamps as I recall).

An example is one of the higher end, unfortunately more expensive light meter (though it is not accurate for LEDs):
http://www.extech.com/instrument/products/400_450/407026manualregist.html

Unfortunately, nobody I know of has created a spectral correction factor for white LEDs, and that would be somewhat futile, since the white led spectrum varies from white LED to white LED, even from the same manufacturer.

One of the few instruments that can remove this *very* significant source of error is a spectroradiometer, and they are darned expensive.

 

[NewBie]

Results of the certified testing of a batch of lights, as compared to various folks light meters can be found here:
http://candlepowerforums.com/vb/showpost.php?p=1462364&postcount=461

Note the large error, especially for the Red Green and Blue.

Hopefully, some day, someone will have a chance to test "white" LEDs from several bins the same way, vs. light meter reported results.

 

[LumenHound]

Newbie, I would like to know how these leds would test out on your light meter using the identical testing procedure jtr1962 used.

Jeled 40K, 50K
BestHongKong 40K
Light of Victory 35K
Peak Snow 29
Nichia CS B1U white

I think it would be very interesting and informative to see what variations there are from one consumer level light meter to the next.

Anxiously awaiting your results...

 

[NewBie]

Newbie, I would like to know how these leds would test out on your light meter using the identical testing procedure jtr1962 used.

Jeled 40K, 50K
BestHongKong 40K
Light of Victory 35K
Peak Snow 29
Nichia CS B1U white

I think it would be very interesting and informative to see what variations there are from one consumer level light meter to the next.

Anxiously awaiting your results...

Variation testing has been done, along with Certified Lab Testing, see this thread:
http://candlepowerforums.com/vb/showthread.php?p=1461977

 

[LumenHound]

Excellent thread. Thanks for the link.
Looks like I'm going to need to get a fresh cup of coffee before I get comfortable and start to wade through it all.

 

[jtr1962]

I haven't been keeping up much on things around here since my father died on March 28 but this thread being bumped caught my attention. Three or so months ago Pinter analyzed my testing procedure and found that in theory my reported results would be higher than the actual results due to my using 5° increments instead of smaller ones. The difference between actual and reported would be higher for narrower beam LEDs than for wide-beam ones, but for most 15° to 20° LEDs it was on the order of 10%. By sheer coincidence my light meter reads about 10% low compared to the lab standard so the errors more or less cancel. Of course, the spectra of the white LEDs I test is different from the spectra of A1 white but nevertheless it gives some idea of the typical errors of my meter when measuring white LEDs. Anyway, while I certainly won't say my lumen measurements are dead accurate, I think it's reasonable to say they're within 10%. As for the color LED measurements, after seeing how much my meter is off, and also how a slight shift in center wavelength can make any "corrections" based on the standard meaningless, I've decided to refrain from testing any more colored LEDs. The results, other than for relative light intensity versus beam angle, are utterly worthless since I don't possess the equipment to accurately measure the spectral distribution.

I plan to eventually add to this thread but for now there don't seem to be any new white LEDs worth testing and I'm still grieving over my father. Perhaps when the 85 lm/W and 100 lm/W Nichias are in circulation later this year I'll continue. I'd also love to test one of Cree's new 131 lm/W LEDs as it represents a good 50% increase over my best efficiency measures so far.

 

[jtr1962]

Unfortunately, nobody I know of has created a spectral correction factor for white LEDs, and that would be somewhat futile, since the white led spectrum varies from white LED to white LED, even from the same manufacturer.

True, but in much the same way corrections for fluorescent sources are somewhat futile as well. While many light meters may have a correction for old-school halophosphor cool-whites there are so many types and color temps of fluorescents available now which differ markedly in their spectra from the "standard" fluorescent used in the corrections.

 

[milkyspit]

Jtr, you rule!!! This is some GREAT information. Certainly is a Godsend for me!
:bow: :bow: :bow:

BTW, check your PM Inbox... PM sent!

 

[jtr1962]

Jtr, you rule!!! This is some GREAT information. Certainly is a Godsend for me!

BTW, check your PM Inbox... PM sent!

PM read and answered.

Thanks for the praise!

I'm certainly glad many are finding this info useful. Now if I could only get my hands on some of those 131 lm/W Crees soon for some testing...

 

[Xe54]

Unfortuately, the majority of the error is in the spectral response. These meters are "calibrated" for incandescent, sodium vapor, fluorescent, and such. Some have correction tables built in you select the type of light source, to adjust for the error, others have the correction factor in the manual that you have to apply manually. Read your manual, if it worth beans at all, they will mention those correction factors (the LM631 is only accurate for tungsten lamps as I recall).

An example is one of the higher end, unfortunately more expensive light meter (though it is not accurate for LEDs):
http://www.extech.com/instrument/products/400_450/407026manualregist.html

Unfortunately, nobody I know of has created a spectral correction factor for white LEDs, and that would be somewhat futile, since the white led spectrum varies from white LED to white LED, even from the same manufacturer.

One of the few instruments that can remove this *very* significant source of error is a spectroradiometer, and they are darned expensive.

Theoretically speaking, IF a highly accurate photopic filter could be achieved to cause a transducer such as a photodiode to be accurately matched in response to the human eye (according to the current standard response curve) then that detector would exhibit accurate lux measurements for any color temperature of white light and even for monochromatic sources.

I believe I have found the most accurate photopic photodiodes available in the marketplace. One is the OSI Optoelectronics (used to be UDT Sensors) PIN10AP device in the datasheet below:

http://www.osioptoelectronics.com/products/35-36_DetecFiltComb_OSIOpto.pdf

Note however that the $275 device (with 6 month lead time) is specified as having 4% max error in area vs. CIE rather than error at any particular wavelength. Unfortunately the error at some specific wavelengths in the 470-520nm region is very large. This still precludes the use of the device for highly accurate photometric measurements of monochromatic sources near the blue, and perhaps could lead to significant errors even with white LEDs with a large blue spectral peak. I do expect however that a device such as this is considerably more accurate than a typical lux meter sensor.

When searching for this device, I found devices from two other manufacturers. A Hamamatsu device in BNC package was ridiculously expensive, more than $1000, though I think around $60 but unobtanium in small quantity for the TO-5 package.

http://sales.hamamatsu.com/assets/pdf/parts_S/S9219_series.pdf

The other device was from Advanced Photonix, PDV-C406, but which is no longer shown on their website (I have a datasheet and 4 samples if anyone's interested). Actually was from a company they recently acquired, though they told me in an email that the product would still be available. They also said the response was never calibrated.

http://www.advancedphotonix.com/index.asp

Hmm, now I need to consider whether to cancel my backordered PIN10/AP device...

 

[jtr1962]

I added four samples (MPJA 15,000mcd 10mm, unknown 26,000 mcd, LS Diodes THC3, SMJLED) sent to me by CPF member milkyspit to the list in the first post and updated the graphs accordingly. You might need to refresh your browser to see the updated graphs. The relevant spreadsheets were added to the .zip file linked to in the first post of this thread.

 

[jtr1962]

I added the Nichia NSPW310BS wide angle 3mm white LED to the list. You might need to refresh your browser to see the updated graphs. The relevant spreadsheets were added to the .zip file linked to in the first post of this thread.

 

[milkyspit]

Jtr, this is admittedly a VERY picky little request... but if it's not overly difficult, could you post your charts as separate images rather than a single big, long, continuous image? I love your research and like to save a copy of this thread locally for quick reference purposes... even turn it into a PDF for convenience... but those ultra-long images are giving me fits! No matter what I do, they get cutoff at page bottom. (As in a physical page... piece of paper.) I would need to cut them apart by hand in Photoshop then recreate a hacked version of the thread to fix the problem... if you can generate them as separate images to begin with, it would be a BIG help!

:bow: :bow: :bow:

 

[SemiMan]

GREAT TEST!


"It's interesting that for narrow viewing angle (i.e. 15° or 20°) LEDs the traditional methods of calculating lumens in the main beam grossly underestimates the output. One can see why by examining the radiation pattern of one of these LEDs. Some light falls outside the main beam but this decreases in intensity rather quickly."

When reading your post, I got the impression that you used different distances depending on the LED? If true, this would effect your measurements somewhat.

If possible, it is best to place the LED as far from the meter as possible given the resolution of the meter. This ensures the most homogenous output on the light meter. At those short distances the cosine correction may come into play assume the meter is corrected.

Interesting that your blue output is higher than normal. Most low cost light meters (i.e. less than $2,000) are very inaccurate at the wavelength of blue LEDs used for white LEDs often reading 50% low or more. This changes rapidly around that spot as well making things worse. This can effect the measurements by about 10-15% overall (low), but not the end of the world.

Overall, great experiment and nice set setup.

One way to improve things would be to put a length of 2" PVC pipe painted inside with a very flat black paint in front of your light meter. This will get rid of a lot of the side reflections from the LED hitting other things plus general ambient light.You may need to be careful with this given how close your LED is though.

 

[jtr1962]

Jtr, this is admittedly a VERY picky little request... but if it's not overly difficult, could you post your charts as separate images rather than a single big, long, continuous image? I love your research and like to save a copy of this thread locally for quick reference purposes... even turn it into a PDF for convenience... but those ultra-long images are giving me fits! No matter what I do, they get cutoff at page bottom. (As in a physical page... piece of paper.) I would need to cut them apart by hand in Photoshop then recreate a hacked version of the thread to fix the problem... if you can generate them as separate images to begin with, it would be a BIG help!

I'm actually thinking of doing that because that long image is getting unwieldy. I simply have to see how much web space separate images versus one large one would take up as Roadrunner only gives me 5 MB of web space. It would make my life easier in that I wouldn't have to edit that huge picture every time I test a new LED.

 

[jtr1962]

When reading your post, I got the impression that you used different distances depending on the LED? If true, this would effect your measurements somewhat.

When doing the beam profile the LED is 17.5" inches from the light meter in all cases. My test apparatus sees to that. This is the distance at which one 5° increment is the same size as the diameter of my light meter sensor. Another reason for using the shorter distance to measure the beam profile is that the background light represents a smaller percentage of my reading. As I mentioned it's at most 0.3 lux with the brighter LEDs. If I took the beam profile at 1 meter instead then once I got off center axis my readings might differ from the background reading by less than 0.1 lux, which is the resolution of my light meter. If I had a light meter with an extra digit then I might be able to do this but 4.5 digit light meters seem very rare and expensive. Of course, I do measure the actual intensity in mcd of the LED at 1 meter. One interesting thing to note is that since my background readings are limited in resolution to 0.1 lux increments then the final lumen values technically have a granularity of roughly (17.5/39.3)² x 0.1 x 4pi or 0.25 lumens. This is generally way less than 10% of the total lumen output at 20 mA of most LEDs these days. Doing the beam profile at a higher test current than this would reduce the contibution of the finite meter resolution to the final calculation.

Interesting that your blue output is higher than normal. Most low cost light meters (i.e. less than $2,000) are very inaccurate at the wavelength of blue LEDs used for white LEDs often reading 50% low or more. This changes rapidly around that spot as well making things worse. This can effect the measurements by about 10-15% overall (low), but not the end of the world.

Yes, for colored and especially blue LEDs my readings are way off which is why I decided to refrain from further testing of colored LEDs. While the blue component can undoubtedly affect the reading from white LEDs somewhat, I have yet to find any white LED which gives significantly more than its rated mcd (with the exception of the MPJA 10 mm). In fact, most of the eBay LEDs I've tested thus far read under their rated mcd. The Nichias seem to be very close to what they're supposed to be so I suppose that tells me my light meter isn't terribly off.

Another sanity check are the tests done for wide angle power LEDs like Luxeons. Q-bin Luxeons test right in the mid-range of the Q bin. A Seoul Semiconductor LED tested right at the 52 lumen typical value shown in the data sheet. Wide angle LEDs are less prone to the sorts of errors which might affect more narrow beam LEDs. It might be interesting to one day check how some of my tested LEDs do in a calibrated integrating sphere. My guess is they probably wouldn't be more that 10% to 15% off except possibly in the case of 10 mm LEDs with very narrow beams.

One way to improve things would be to put a length of 2" PVC pipe painted inside with a very flat black paint in front of your light meter. This will get rid of a lot of the side reflections from the LED hitting other things plus general ambient light.You may need to be careful with this given how close your LED is though.

I may one day build a new test apparatus with some of these improvements, including detents on the angle wheel to make the testing yet more consistent. I do place the apparatus on top of a narrow cardboard box now so as to minimize the stray reflections. I plan to paint the top of the box and the rest of my apparatus flat black. One thing to note is that the stray reflections mainly make the readings in the main beam higher although hopefully subtracting out the background reading reduces most of this error. Since all the readings in the beam profile are relative, this tends to give the readings outside the main beam smaller relative values than they would otherwise have. The end result is actually a lower total lumen value. It would be interesting to retest a few LEDs if I modify my apparatus to see if I actually end up with higher lumen values.

 

[milkyspit]

I'm actually thinking of doing that because that long image is getting unwieldy. I simply have to see how much web space separate images versus one large one would take up as Roadrunner only gives me 5 MB of web space. It would make my life easier in that I wouldn't have to edit that huge picture every time I test a new LED.

If it gets down to it, maybe I could give you some space on my hosting account, or at least point you to a hosting provider where $10 per month would buy you something along the lines of 150-300MB of space. Let me know if you need the help... glad to assist!

 

[jtr1962]

If it gets down to it, maybe I could give you some space on my hosting account, or at least point you to a hosting provider where $10 per month would buy you something along the lines of 150-300MB of space. Let me know if you need the help... glad to assist!

So far so good-the separate files for the white LEDs were only about 15K more than the one big file. The white LED efficiency charts are now separate pictures. Stay tuned for the colored LEDs...

Update: The colored LED efficiency charts are now separate files. They actually use less space than the one big file since I was able to convert most of them to 16 colors.

Second Update: LED comparison charts are now separate images.

Third Update: Power white LED charts now separate images. I think that takes care of everything.

 

[jtr1962]

I adding the following chart summarizing my results at 20 mA to the first post:

 

[jtr1962]

I added the Jeled PLCC-2 2000 mcd surface mount white LED to the list. You might need to refresh your browser to see the updated graphs. The relevant spreadsheets were added to the .zip file linked to in the first post of this thread.