Using power LED at low current levels

[Tobias Bossert]

Almost all applications for power LED focus on getting as much light as possible out of a given LED model. Only for some special applications (like helmet lamps for caving) the light is used at maximum power for some seconds only, but at more than 99% of time in a dimmed state. For such applications the efficacy in the dimmed state is predominant.

There are two aspects to be optimized:
1. Use a LED optimized for low output levels
2. Use a driver with high efficiency at low output levels

The first aspect is covered in the following three posts in this thread (posts #2, #3 and #4) and aims to help you choosing the best LED for low output levels.

The second aspect for drivers may be covered in a future thread.
Color rendering, die size and special distribution are not in the scope of this thread.

 

[Tobias Bossert]

Performance measurement at low current levels

This post describes the measurement method and publishes the rough data.

Cree Product Characterization Tool (PCT) delivers performance values starting at currents of 100 mA for models with maximum current of 1.000 and 1.500 mA (XR-E, XP-E, XB-D and XP-G, XT-E) and at 150 mA for models with maximum current of 3.000 mA (XM-L). The new generation models XP-E2, XP-G2 and XM-L2 are still not covered by this tool.

For a project of small helmet lamp I’m interested in the behavior of power LED in the range below 100 or 150 mA too.

I have no means to measure absolute luminous flux output of a LED therefore I’m constrained to take a relative measurement method.

I used a cheap lux meter with a spherical opal diffuser, a second spherical opal diffuser (Khatod) for the LED under test and a black tube (old 24x35 film case) fitting closely over both diffuser spheres. With this arrangement I hope to reduce the influence of uneven spatial distribution and to achieve color mixing in case the LED under test shows different spatial pattern for different spectral components (colored rings). With this arrangement I’m sure to get relative illuminance values which are proportional to the total luminous flux of the LED under test. But the proportional factor remains unknown so far.

This relative measurement method isn’t suitable to compare the total luminous flux of different LED models precisely, because no information upon the spectral sensitivity of the lux meter is available. Thus the reading may depend on the color bin of the LED. The measurements on a single LED sample may be reliable as long as the spectral distribution remains unchanged only. I hope this is valid for all the LED under test in the range 10 mA to 700 mA approximately.
All measurements were done at a temperature of the starboard of 25°C. The LED were driven by a dc power supply. Forward voltage and drive current were measured using a circuitry to eliminate the influence of voltage drop at the current meter.

I recorded the reading of the lux meter for currents between 10 mA and 200 mA in steps of 10 mA and additionally at 250, 300, 350, 500 and 700 mA (for some LED I measured at 5 and 15 mA additionally).
PCT delivers absolute total luminous flux values in the range above 100 or 150 mA only. My measurements deliver relative results in the range from 10 to 700 mA and at some reference points above that level. In the overlapping range the relative measurements can be scaled by fitting them to the absolute values from PCT. The scaled results are ‘calibrated’ and thus somewhat like ‘semi-absolute’ values outside the range of PCT.
PCT doesn’t cover the new generation models XP-G2 and XM-L2. Thus I’m obliged to scale my relative measurements at one reference point only, at 350 or 700 mA, whichever is specified in the datasheet.
For almost all LED under test covered by PCT my results show a comparable curvature as PCT.

There is one exception: The curvature for XR-E R2 differs severely from that one derived from PCT. Since my curvature compare nicely to that one found in 2009 by jtr1962, I’m convinced that there is something wrong with the data of XR-E in PCT! (Another bug…) As a consequence I calibrated the results for XR-E at the reference point 350 mA only.

The calibration method forces the luminous gain (lm/A) to fit to PCT immanently. This doesn’t mean that the LED under test really behaves like it is stated in PCT.
The calibration method doesn’t force the efficacy (lm/W) to match with PCT, because the measured voltages may differ from those specified in PCT.

All measurements were performed with one sample only. Consequently I don’t know anything about series tolerances.

All rough measurement data and all derived values can be found in an Excel file

The optimized choice of a LED model depends upon the kind of driver.

The following post addresses the optimization with respect to luminous gain (lm/A) – this is of interest when using a linear driver.

The following post thereafter addresses the optimization with respect to efficacy (lm/W) – this is of interest when using a buck driver.

 

[Tobias Bossert]

Derived results for luminous gain (lm/A)

This post interprets the results with respect to luminous gain.

The quantity ‘luminous gain’ describes how much flux you can get out of a given current for a single die LED. This is of interest when a linear driver is used, since this kind of driver doesn’t profit from reduction of forward voltage.

(This diagram is part of the Excel file of post #1)

As you can see, the curvature (curve shape, not positioning of the curve) fit quite well in the overlapping range, so I’m convinced that my measurements do not suffer linearity problems. But at levels below the range of PCT this isn’t proven and additionally series tolerances may come in.

There is one exception: The curvature for XR-E R2 differs severely from those derived from PCT. Since my results compare nicely to those fond in 2009 by jtr1962, I’m convinced that there is something wrong with the date of XR-E in PCT!

The validity of my measurements seams to be strong enough to claim that there are three quite different types of LED models:

XM-L, XM-L2, XP-G and XP-G2 show affine curvature. They are optimized for high current levels at a small expense of luminous gain at lower current levels. The maxima (+0 / -5%) cover the range 20 to 190 mA and 60 to 500mA respectively; XP-G2 R5 achieves 507 lm/A at 50 mA and XM-L2 U2 achieves 513 lm/A at 160mA.

XT-E is optimized towards medium current levels. The maximum (+0 / -5%) covers the range 20 to 120 mA. At 50 mA it achieves the highest luminous gain of 525 lm/A amongst all tested models. The decrease of luminous gain at higher currents is stronger than for XM-L and XP-G.

XR-E is optimized to very low current levels but nevertheless it doesn’t achieve the maximum values of XT-E (385 lm/A at 40 mA). The maximum (+0 / -5%) covers the range 10 to 100mA. The behavior at higher levels is worst.

Conclusion
For current levels above about 110 mA XM-L2 U2 achieves the best luminous gain of all LED under test. Below 110 mA XT-E R5 achieves the best luminous gain of all LED under test.

XP-G2 R5 and XT-E R5 achieve comparable light output. XP-G2 R5 is a little bit better at current levels above 200 mA (but still worse than XM-L2 at those levels) and XT-E R5 is a little bit better at current levels below 200 mA.

If you use XM-L2 U2 for a dimmable light, use PWM with pulse current of 160 mA; up to pulse current of 500 mA loss is still lower than 5%. Above 500 mA you should use analog dimming.

If you use XT-E R5 for a dimmable light, use PWM with pulse current of 50 mA; up to pulse current of 120 mA loss is still lower than 5%. Above 120 mA you should use analog dimming.

If you use XP-G2 R5 for a dimmable light, use PWM with pulse current of 60 mA; up to pulse current of 190 mA loss is still lower than 5%. Above 190 mA you should use analog dimming.

With respect to luminous gain there is no reason to use XR-E any more.

 

[Tobias Bossert]

Derived results for efficacy (lm/W)

This post interprets the results with respect to efficacy.

The quantity ‘efficacy’ describes how much flux you can get out of a given power. This is of interest when a switching driver is used, since this kind of driver profit from reduction of forward voltage.

(This diagram is part of the Excel file of post #1)

As you can see, the curve shapes (curvature, not positioning of the curve) fit quite well in the overlapping range, so I’m convinced that my measurements do not suffer linearity problems. But at levels below the range of PCT this isn’t proven and additionally series tolerances may come in.

There is one exception: My results for XR-E R2 (bold grey curve) differ severely from the results derived from PCT (grey rectangles). Since my results compare nicely to those fond in 2009 by jtr1962, I’m convinced that there is something wrong with the data of XR-E in PCT!

The validity of my measurements seams to be strong enough to claim that there are three quite different types of LED models:

XM-L, XM-L2, XP-G and XP-G2 show affine curvature. They are optimized for high power levels at the expense of efficacy at lower levels. The maxima (+0 / -5%) cover the range 130 to 1000 mW and 40 to 310 mW respectively. XM-L2 U2 achieves 186 lm/W at 360 mW and XP-G2 R5 achieves 186 lm/W at 110 mW

XT-E is optimized towards medium power levels. The maximum (+0 / -5%) covers the range 30 to 225 mW. At 110 mW it achieves the highest efficacy of 192 lm/W among the LED under test. The decrease of efficacy at higher levels is slightly stronger than for XM-L and XP-G.

XR-E is optimized to very low power levels but nevertheless it doesn’t achieve the maximum efficacy value of XT-E (142 lm/W at 50 mW). The maximum (+0 / -5%) covers the range 20 to 200 mW. The behavior at higher levels is worst.

Conclusion
For power levels above about 250 mW XM-L2 U2 achieves the best efficacy of all LED under test. Below 250 mW XT-E R5 achieves the best efficacy of all LED under test.

XP-G2 R5 and XT-E R5 achieve comparable light output. XP-G2 R5 is a little bit better at power levels above 300 mW (but still worse than XM-L2 at those levels) and XT-E R5 is a little bit better at power levels below 300 mW.

If you use XM-L2 U2 for a dimmable light, use PWM with pulse power of 360 mW; up to pulse power of 850 mW loss is still lower than 5%. Above 850 mW you should use analog dimming.

If you use XT-E R5 for a dimmable light, use PWM with pulse power of 110 mW; up to pulse power of 220 mW loss is still lower than 5%. Above 220 mW you should use analog dimming.

If you use XP-G2 R5 for a dimmable light, use PWM with pulse power of 110 mW; up to pulse power of 300 mW loss is still lower than 5%. Above 300 mW you should use analog dimming.

With respect to efficacy there is no reason to use XR-E any more.

 

[SemiMan]

Re: Derived results for efficacy (lm/W)

Great work, thank you.

Semiman

 

[bshanahan14rulz]

Re: Derived results for efficacy (lm/W)

This is interesting! We long posited that high power LEDs would be very very efficient at very low drive currents, turns out there's more to the story, eh?

Can't wait to read this more in-depth later!

Thank you very much for sharing with us the end results of your hard work :-)

 

[uk_caver]

Re: Derived results for efficacy (lm/W)

That's really interesting work, thanks - gives me a bit to think about regarding optimum driver design.

 

[PocketBeam]

Re: Derived results for efficacy (lm/W)

This is very interesting. One question though, you suggest using pulsed power, pwm. But it doesn't look like you actually tested with a pulsed supply. Did you, and if so what was the frequency and off/on ratio, and the wave form? If not, then different pwm variations would be interesting.

So far it is very educational.

 

[uk_caver]

Re: Derived results for efficacy (lm/W)

Given the good heatsinking, I think that PWM results could be pretty accurately estimated, by using the results for the PWM 'on' current, especially at low 'on' currents - a max 100mA PWM is likely to have luminous gain/efficacy results close enough to those measured at 100mA constant, whatever the average PWM current is.

 

[SemiMan]

Re: Derived results for efficacy (lm/W)

This is interesting! We long posited that high power LEDs would be very very efficient at very low drive currents, turns out there's more to the story, eh?

Can't wait to read this more in-depth later!

Thank you very much for sharing with us the end results of your hard work :-)

Who would that "WE" be?

There are numerous threads over the last 5+ years that talk about the efficiency curve of high power LEDs and that the efficiency improves to a point as the current decreases, but at some point that efficiency starts to decrease.

There are quite a few threads as well that discuss the best way to drive an LED for maximum efficiency is to use current control above the efficiency peak, but then PWM the led at the maximum efficiency current below this peak.

Semiman

 

[SemiMan]

Re: Derived results for efficacy (lm/W)

This is very interesting. One question though, you suggest using pulsed power, pwm. But it doesn't look like you actually tested with a pulsed supply. Did you, and if so what was the frequency and off/on ratio, and the wave form? If not, then different pwm variations would be interesting.

So far it is very educational.

This would be PWM with constant current drive. The results for efficiency would be the same as at the efficiency peak, with a small bump in efficiency due to lower power making the LED cooler.

Semiman

 

[bshanahan14rulz]

Re: Derived results for efficacy (lm/W)

Who would that "WE" be?

Heh, should have been more of a "I've always assumed." Guess we should have our psychologist check that out ;-)

Yeah, I do vaguely remember similar findings in the past. These graphs really helped me visualize it, esp. across the product range. I can see how as the device grows in size (as current density goes down?), the efficiency hump gets pushed farther and farther out (max efficiency reached at higher currents, lower currents becoming less efficient). At least, that's how it looks in BShan's Birdbrain.