White LED lumen testing
- Thread starter jtr1962
- Start date
Help Support Candle Power Flashlight Forum
TorchBoy
Flashlight Enthusiast
Wouldn't it be fairer to compare an MC-E with two XP-Gs in parallel?
jtr1962
Flashaholic
Based on my rough estimates, a blue version of these should produce about 500 mW to 525 mW @ 350 mA. In other words, 50% conversion efficiency give or take. Such outputs would definitely help your experiment.
Maybe one reason Cree is withholding release of neutral and warm versions until late this year is to try and reduce the efficiency penalty. For any given CRI, there is no inherent reason a 4000K emitter should be less efficient than a 6500K one. In fact, one document I read shows efficiency peaking in theory at lower CCTs ( link-see page 12). It's just a matter of using the proper phosphor. Right now I'm reasonably sure most LED manufacturers just glob more of the same YAG phosphor on to get neutral or warm tints. This is pretty apparent looking at the spectra. The yellow hump for warm/neutral tints covers the same wavelengths as cool, only it's larger relative to the blue spike. They're effectively using globs of phosphor to block out some of the blue! Small wonder there's an efficiency hit (and only a marginal improvement in CRI from perhaps low 70s to high 70s). Now Philips/Lumileds may be doing something different with their Rebels, as evidenced by the neutral tints having the same rating as the cool ones.
High CRI is another animal entirely. Regardless of tint, you can expect a 20% or larger penalty when you use a special phosphor to get CRI into the 90s. So far Cree hasn't jumped on the high-CRI bandwagon at all, but it would be interesting to see a high-CRI LED based on the XP-G. Even an ~20% efficiency hit still implies 100-110 lm/W @ 350 mA. And I may be in the minority but I would love to see cool high-CRI tints as well.
In the meantime it is possible to both warm the cool tints somewhat, and improve CRI substantially, by adding small amounts of red and cyan. This won't work well for focused lighting, but for more diffuse general lighting using emitters with no optics mixing colors evenly shouldn't present a problem.
Last edited:
jtr1962
Flashaholic
I'll set up the test jig one last time and test a few of the LEDs you mentioned (Cree XP-G, Rebel, MC-E, perhaps the SST-90 and a few others) at 60°C baseplate. I don't plan to make this testing a part of my regular procedure however unless I can automate it somehow. Right now it's just too time consuming to set up and manually regulate the temperature at each test point.
In terms of die area you're 100% correct here. I was merely comparing a single XP-G to the multidie emitter to illustrate that we don't have long to go before the single die emitters reach the output levels of the MC-E. Right now the best binned MC-Es should manage perhaps up to 825 lumens at the maximum rating of 700 mA per die. The XP-G manages around 350-360 lumens at its maximum rating of one amp. The S2 bin should close the gap even further at roughly 400 lumens. Down the road in the not too distant future I can see higher S bins hitting 500 lumens @ 1 amp. I can also see Cree bumping the XP-G rating to 1500 mA as efficiency continues to increase. This could get us to 650-700 lumens, depending upon droop. This is close enough that it could make multidie emitters pointless. Not saying that will happen, but I tend to think with cost reduction per lumen being a priority, the industry would probably like to get away from multi-die emitters. They've always been more expensive and harder to make than their single-die counterparts. Granted, they may have some advantages in terms of matching Vf to power supply, but they also have lot of drawbacks as well. Note that there is an exception to this. Multi-die emitters using different types of dies (RGBA, for example), do and will continue to make a lot of sense.
This would be a good way of seeing the droop at higher current without temperature effects. For example, I could PWM the LED at a 1% duty cycle with peak currents of both 350 mA and 1000 mA, and compare the relative outputs. I have no plans to do this however at this time. As with the 60°C baseplate testing, it just adds to an already time-consuming test procedure (as it is each LED takes me over two hours to do between setting it up, getting the raw test data, entering it in spreadsheets, and doing the write-up here). With my business picking up lately, my time to do these tests is very limited compared to even a few years ago.
HarryN
Flashlight Enthusiast
Hi JTR - would you consider switching your standard test for these power LEDs to 60 C instead of 25 C ? It seems that most of the power LED mfgs now are not far from their specs for 25 C, but this data still remains - well - interesting but not that valuable.
Your test has the potential to be more realistic compared to the mfg specs as a 60 C test would be what real flashlight and lighting uses are like. Frankly, the only downside of your exhaustive testing is the new false claims that people make about their OTF lumens claims - and then they reference your data as the basis for the claim.
Just MHO, take it FWIW.
Thanks for all of the work.
Harry
Your test has the potential to be more realistic compared to the mfg specs as a 60 C test would be what real flashlight and lighting uses are like. Frankly, the only downside of your exhaustive testing is the new false claims that people make about their OTF lumens claims - and then they reference your data as the basis for the claim.
Just MHO, take it FWIW.
Thanks for all of the work.
Harry
jtr1962
Flashaholic
Right now I'm not really measuring at 25°C baseplate. The LED is mounted on a 60 mm square heat sink which has some air flow. The heat sink does indeed get somewhat above ambient temperature as I ramp up the current (this is especially true with 4-die LEDs such as the P7 or MC-E), and the ambient temperature itself in the room is subject to variability throughout the year (although not much because the room is heated in winter and cooled in summer). What I'm doing is kind of halfway in between constant 25°C baseplate and constant 60°C baseplate. I'd roughly estimate my standard setup has a thermal impedance on the order of around 1.5°C/W (plus additional losses at the thermal interface between the emitter and heat sink). Overall this setup is a lot more realistic than data sheets which usually assume a constant 25°C junction temperature. In fact, my relative measurements between 350 mA and 1000 mA in my recent XP-G test were less than the data sheet precisely because I didn't keep junction temperature at a constant 25°C (indeed I couldn't do so unless I cooled the heat sink below ambient).
Besides that, I'm confused as to whether you would want me to standardize on 60°C baseplate temperature or 60°C junction temperature (manufacturers tend to use junction temperature). I probably couldn't even do the latter except by using the manufacturer's estimate of junction to thermal pad impedance, measuring power consumption, calculating the rise of junction temperature above heat sink based on the measured power consumption, and then adjusting heat sink temperature accordingly. As heat sink temperature changes, Vf and therefore power consumption changes, so that means some recalculation of junction temperature rise. I dread to think of the logistics of trying to do this at one data point, much less at 30+ data points. Controlling base plate to a constant temperature is a little easier, but still not without problems. It takes a while until you adjust the thermoelectric module current just right to keep your heat sink at constant temperature. Maybe I can do this automatically via a PID controller, but even those still take some time to stabilize at setpoint.
I'll grant that not controlling either baseplate or junction temperatures precisely introduces some variability, but it also represents what can be achieved in the real world with halfway decent cooling. At low currents the emitter will be fairly close to ambient temperatures just as it is in my tests. At higher currents it may indeed be warmer than in my tests but this is relatively easy to correct. Looking at most of the spec sheets the difference in output between 25°C and 60°C baseplate temperatures is about 5-6%.
Believe me, it might be nice to standardize on some sort of constant temperature for testing purposes, but after testing a few emitters this way the HUGE amount of extra work relative to the knowledge gained seems minimal. For example, let's look at these two curves for the neutral and cool K2s:
The differences here between 60°C baseplate and my standard testing amounts to 6-7%. However, I could have obtained virtually the same plot by just looking at the data sheets, and applying a correction for temperature. The only noteworthy item I discovered here not obtainable from applying the data sheets was that the K2 neutral white falls on its face at a lower drive current than in my standard testing. But note that this occurred above the rated current of 1.5 amps anyhow. It would be of little interest to any reputable manufacturer who would keep current ratings within specs.
Anyway, it's not that I'm against doing things the way you suggest in principal but I see a couple of problems. First off, I lose any direct comparison between previous tests and current ones (unless I perform tests both the new way and old way). Second, after doing a few LEDs this way it's just too time consuming due to the logistics mentioned earlier. It takes about an hour to set up everything, and then probably another few hours to gather the data. Fortunately for us white LEDs don't vary their output by a huge amount with temperature. You need to increase temperature by 45°-50°C to decrease output by 10%.
It's fairly easy to guestimate output at 60°C baseplate or any other temperature by applying the relevant correction from the manufacturer's data sheet. I strongly encourage anyone citing my data to correct for temperature rise in this manner (and also to correct for any optical losses). The thing is with so many different applications out there I can't possibly have data for all scenarios. Some lights may get much hotter than 60°C. On the other hand, I'm aware of larger flashlights, or general lighting applications, which actually mimic the conditions in my tests (i.e. heat sink doesn't get much higher than 10°C over ambient).
I hope none of this post came off as overly dismissive as it wasn't meant to. You offer valid suggestions to make my tests more meaningful. And in fact if the logistics weren't against it I would probably follow your suggestions. It's just that during the last two years I've find myself with a lot less time to devote to the experimentation I enjoy due to the need to earn money and/or take care of personal matters at home. I WILL try to do a few constant 60°C tests on some common emitters in the near future as I said I would just to see if they reveal anything new. But after that I doubt I'll have time to do much beyond my usual testing. Truth is after doing a lot of the things I have to do I'm just too fatigued to do the things I want to do.
EDIT: To add a bit more here the way LEDs are trending I'm seeing less and less point doing testing at elevated temperatures. For example, look at this chart for the XP-G:
The red line represents total input power to the LED and the white line represents waste heat. In the coming years the white line will be trending towards zero even though it will obviously never get there. In light of this, I tend to think our flashlights and general lighting will operate ever closer to room temperature. This is especially true of portable lighting. There are limits to the amount of energy any cell can store, and this is improving very slowly compared to LEDs. For any given runtime, this represents a hard limit on the amount of power delivered to the LED. As more of this power comes out as light due to improved LEDs, thermal pad temperatures will continue to approach ambient temperature.
Last edited:
spencer
Enlightened
I would like to see this as well. Maybe not really cool tints like WC but something in the 5000K range. Mmmmmm.
Thanks again, jtr1962, for your testing, and for responding to my question above.
I have ordered some warm white LEDs from eBay seller shop4leds (the same as TopBright?):
8mm (straw hat) 140 degree warm white 210,000mcd (300mA)
8mm (straw hat) 140 degree warm white 80,000mcd (100mA)
4.8mm 120 degree warm white 12,000mcd (20mA)
5mm (piranha high-flux package) 120 degree warm white 23,000mcd (20mA)
I will be happy to send some for testing if you want. The figures above sound impressive, but of course they are just figures. Your testing of LEDs has been quite an education!
I have ordered some warm white LEDs from eBay seller shop4leds (the same as TopBright?):
8mm (straw hat) 140 degree warm white 210,000mcd (300mA)
8mm (straw hat) 140 degree warm white 80,000mcd (100mA)
4.8mm 120 degree warm white 12,000mcd (20mA)
5mm (piranha high-flux package) 120 degree warm white 23,000mcd (20mA)
I will be happy to send some for testing if you want. The figures above sound impressive, but of course they are just figures. Your testing of LEDs has been quite an education!
According to this site those would be:
~870lm
~330lm
~38lm
~72lm
Would be nice to know The Real Specs. Sadly there are too many leds in the eBay that don't have correct specs. :thumbsdow
TorchBoy
Flashlight Enthusiast
It also says the calculation is inaccurate for wide angle LEDs (which you've realised).
jtr1962
Flashaholic
Those might be interesting to test although I can tell you right off those specs are exaggerated. The XP-G R5 had a narrower beam angle than those LEDs, and even so it only managed 160,700 mcd at 2500 mA. And the XP-G R5 is the most efficient production LED at this time. If the LEDs you listed really had those specs, they would be achieving in excess of 1000 lm/W which is physically impossible (maximum theoretical for phosphor whites is around 265 lm/W).
If you want to send them to me anyway PM me for my shipping address. Not sure when I'll get to them, however. Right now I'm in the process of re-landscaping the house. I might be at this until late November.
Yardwork over LED testing........pfffftttt!
Just kidding, thanks for the great work that you do.:twothumbs
jtr1962
Flashaholic
Truth is the yard looked terrible and it really needed a makeover. I was mowing weeds all summer.
Besides that, we may start growing vegetables again next year. I might try starting seedlings under LEDs, kind of combine two of my hobbies into one. 
jashhash
Enlightened
Thank you very much for your testing JTR. I really appreciate and respect your work a lot.
Justin Case
Flashlight Enthusiast
- Joined
- Mar 19, 2008
- Messages
- 3,797
I have some XP-G R4s mounted on two different MCPCBs: 1) a 10mm diam, 2mm thick board, and 2) an 8mm diam, 1mm thick board (actually closer to 0.85mm). I have a total of seven XP-Gs, two on the 10mm board and five on the 8mm board. For whatever reason, the XP-Gs mounted on the 8mm boards consistently show a higher Vf than the two on the 10mm boards (sorry about the non-uniform drive currents, but the measurements were not originally done for Vf evaluation purposes).
8mm XP-G #1, 3.3V@950mA
8mm XP-G #2, 3.5V@960mA
8mm XP-G #3, 3.3V@950mA
8mm XP-G #4, 3.6V@910mA
8mm XP-G #5, 3.4V@1047mA
10mm XP-G #6, 3.1V@1042mA
10mm XP-G #7, 3.3V@1196mA
XP-G #4 seems particularly poor wrt Vf.
jtr1962
Flashaholic
If the 8mm board has better thermal transfer then the Vf will on average tend to be higher (Vf increases as temperature decreases at the rate of roughly 2.2 mV/°C). Since the 8mm board is thinner, it probably does indeed provide lower thermal resistance. So actually the higher Vf here is a good thing.
But of course the effiency drops with junction temperature.
Would be interesting to see if there is an optimal region.
When i cooled my P7s with liquid nitrogen it turned out to make them much more inefficient...
Is there posibility to measure Flashlight lumens with setup like using for LED testing. I thinking that step must be at least 2.5°, because beam of Flashligt is more narrow than LED, and better will make measurements on both directions from maximum. I understand that results will be far from perfect, but for comparing lumens before Flashlight upgrading and after it will be OK. I also assuming, that beam must be simetrical in vertical and horizontal axis.
Is there possible get modified Excel spreadsheet for these measurements, for one fixed power level?
Thanks in advise, for any help!
Dainis
Is there possible get modified Excel spreadsheet for these measurements, for one fixed power level?
Thanks in advise, for any help!
Dainis
jtr1962
Flashaholic
It probably is possible although for the very narrow beam of most flashlights you would probably need 1° increments. You would also need some means to hold a variety of different flashlight bodies on your test jig. Modifying the spreadsheets for this is trivial-you just change put increments to 1 instead of 5 in the first column-the spreadsheet does the rest. I personally have zero interest in doing any lumens testing of flashlights, however. Don't have the time for starters. And there are far too many other variables such as regulated/unregulated, temperature, state of batteries. etc.Is there posibility to measure Flashlight lumens with setup like using for LED testing. I thinking that step must be at least 2.5°, because beam of Flashligt is more narrow than LED, and better will make measurements on both directions from maximum. I understand that results will be far from perfect, but for comparing lumens before Flashlight upgrading and after it will be OK. I also assuming, that beam must be simetrical in vertical and horizontal axis.
Is there possible get modified Excel spreadsheet for these measurements, for one fixed power level?
Thanks in advise, for any help!
My first post! Hurray! I just bought some of these xpg\\\'s on a bunch of different boards including the 8 mm and 10 mm ones mentioned above and all of the boards are aluminum. The 8 mm boards are only about 1 mm thick and the 10 mm boards are about 2 mm thick. does this small difference in thickness really matter for thermal resistance when measuring vf? the way I have measured vf, I set everything up ahead of time before I power up the LED. I also start from a cold state and put my xpg/boards on a larger piece of metal to help keep the aluminum board cool. Then I turn on the power, light up the LED and make my vf measurement. That takes about two seconds, which doesn\\\'t seem like it would be long enough to get anything hot enough to make a difference of several tenths of a volt. and if vf decreases at a rate of 2.2 mv/C, to get a vf difference of two or three tenths of a volt would mean a temperature difference of about 100 C. That seems very unlikely to me for a test of about two seconds in duration.
