Date 12-28-2004 (with some minor modification for clarity)

This chart is the CIE color chart with the Constant Current dimming and PWM dimming, with the Luxeon bins shown:




Here is an exploded view of just the data, note the dimming % is listed, % for PWM is the PWM duty cycle, % for Constant Current is the % of the max current:



In the above chart, take note of the percent where the constant current and PWM dimming split, about 50% for the Constant Current. A person could dim with constant current to 50%, then utilize PWM, to minimize color shift, yet pick up the efficiency increase.



Here is the efficiency difference:





Yet that chart only shows half the story, here is the efficiency gain at each step, I was rather surprised here:




Basicially it all boils down to Current Dimming is alot more efficient than PWM dimming (so you get considerably more run time on down to 0.5% current dimming, but PWM dimming results in less color shift. A combination of the two approaches could sure be interesting indeed.

FYI, the current level this Luxeon was ran at is 380.1 mA = 100% Also, the Luxeon emitter slug had a hole drilled into the side and a type-K thermalcouple inserted, the slug temperature variation over the whole test was a maximum of 2.3 degrees Celcius (yes, a large aircooled heatsink was utilized for this). The same emitter is utilized for all sets of data.
So, the Luxeon emitter becomes significantly more efficient at lower current levels.

Where with straight PWM, at the same current, you are hitting the emitter with full current.

The info is right on the Luxeon datasheet, if you look carefully.

Next to measure the difference utilizing 1 Amp.

I am expecting even a larger efficiency difference, we will see.

The efficiency gain of current dimming of up to 136% was rather surprising, I was not expecting that large of a gain, more like 15%, but the datasheet doesn't go down to 11.4 mA...
http://www.lumileds.com/pdfs/DS45.PDF

It's kinda cool, how the PWM dimming turned out so flat, just like it should.

BTW, each individual measurement point was taken 15 times and the 15 samples were averaged, to increase accuracy. (in otherwords, 270 measurements were taken, thank goodness for computers).

For the technogeeks that wanna be in the know, the frequency that the PWM circuit was ran at was 133 Hz, which means the pulse width at the lowest level (0.05%) was 0.000003759 seconds (or 3.759e-6 seconds- 3.759 millionths of a second).
Due to the low light output, and the long integration time for 0.05% duty cycle dimming, the measurement takes 40 minutes for 15 samples.
Okay, I ran some numbers at 1140mA, such as you'd have with a Li-Ion, and using a PWM to dim the LED.

First off, at the higher current level, the tint variation increases substantially:





A close up at 1140mA:




And we have the increased light output efficiency of current dimming vs. PWM dimming:




It really adds up a lot at the lower dimming levels:




On a side note, I noticed that the curve fit for what is shown on the LumiLEDs datasheet (output vs. current), when measuring a blue emitter, ends up with nearly an exact match , but the curve is more for a white emitter.
Doug S said:
Nice work, Doug! I note that the emitter x-y chromacity coordinates fall outside the bin structure of the current Luxeon binning scheme. Do you know the bin of the emitter tested? BTW, On the Nichia datasheets for their high power white LEDs they provide some graphs related to those that you have done. They provide graphs of x-y chromacity as functions of temp and also of drive current.



I'm aware of the Nichia datasheet, haven't had a chance to look at it and compare. I don't know the bin, this BW01 is before hounding Future to give us the binning data on the emitters. It is a rather yellow emitter, fyi.

When you current dim, the wavelength of a blue LED shifts a lot, see below:

FYI, the data above is for the 380mA range.


Here is some more data for everyone.

This is what the blue LED underneath the phosphor is actually doing, as far as wavelength shift, note how it stays put during PWM dimming:






Here is what the spectral output is actually doing, notice how it gets less pure when current dimming:




To help visualize things, here is a visual spectral plot of the blue LED. Notice how the spectral plot for each is not symetrical on each side of the wavelength peak, and more is on the green side. Next, note how the peak has moved, with the different current levels. Finally, check out the width of each plot, and notice how it broadens out at low current levels. (this widening affects the spectral purity mentioned previously.



In contrast, with PWM dimming, the dominant and the peak wavelength do not change, and when you dim it the spectrum actually narrows, reverse of what current dimming does. And it doesn't shift in wavelength left or right. See below:



McGizmo, you could do a combo of both and take advantages of each...(CC down to a point, then PWM)

Anyhow, here is some more info in a different format:


This is the difference in the relative output when you PWM dim, shown in a spectral power plot:




Here is the difference of Current dimming in a spectral power plot:






Below is the difference of the 0.05% power input levels for current dimming vs. PWM dimming. Take careful note of how with the current dimmed plot, where the blue peak shifted to a longer wavelength towards the red, and how the output of the phosphor (the big hump running from cyan through green to red), increased it's relative output (went higher) in relation to the blue peak at 100%.

Not every emitter has the same wavelength blue LED die in it (think back to the luxeon lottery but in blue), and depending on where it's wavelength starts at, in relationship to the peak phophor conversion efficiency wavelength for the blue, emitters in the same batch and binning will do different things.

I tried to hold the slug temperature as constant as possible. But in a flashlight, it usually isn't possible to do provide such extreme cooling. Additional changes begin to occur once you start changing the temperature.


I hope this has been educational and enlightening for the general CPF crowd.



Doug S said:
Interesting data on the blue lux wavelength shift with current. If we take the 380mA as 100% and the 7.6mA as 2%, the shift from 2% to 100% is 5nm. Putting this into perspective, this is the width of a single color bin for the blue emitters. AB21 defines 6 bins for blue emitters, each 5nm wide. There is also a tempco associated with wavelength. The luxeon datasheets give this as 0.04nm/C for the blue emitters.



Yes, compare the shift for 100% and 2% for current dimming.

Next, compare the shift for 100% and 2% for PWM dimming.

Think about what is occuring during this time with heat in the die.

Remember, I held the slug itself within a 2.3C range over the 380mA test range. There is a 15 C/W thermal resistance for the 1W emitter to the slug. 3.6V * 380mA = 1.368 Watt
1.368 Watt * 15 C/W = 20.52 C rise in the emitter die. Add in the slug temperature rise of 2.3C (monitored with a K-type thermalcouple embedded in the slug) and we get to a 22.82 C rise.

22.82 C * 0.04 nm/C = 0.9128 nm shift. We are seeing alot more than this when Current dimming...and there is no peak or dominant wavelength shift when PWM dimming (in the blue).

In fact, the PWM blue dominant wavelength and peak wavelength stay put, and match the emitter running 100% duty cycle (on continously at 1140mA), even clear down to a pulse width of 3.67 millionths of a second.

; )


BTW, red emitters when PWMing do change their wavelength.


The whole reason I gathered this data is that was an un-answered assumption that a PWM'd LED would be more efficient, since it doesn't have power applied all the time, it would be cooler, and thus more efficient. This is why I thought I'd put some data up. Looking at it from the other side, you are pulsing it, but you are hitting it full tilt, whacking it with full power, during which time it would be less efficient. With the data, it shows that the efficiency of PWM is quite poor, and in one common scenario shown above, shows that current dimming is 230% more efficient. This equates into 2.3x the runtime. Additionally, you loose efficiency in the battery, due to I^2*R losses in the battery. Some day, I hope to put together two circuits that emit the same amount of light, one a dim current circuit, the other a dim PWM circuit, and hook them to a rechargeable cell, and do a runtime comparison.

Again, there should be a Holy Grail combination of both approaches, current dimming and PWM dimming, that gets a person limited color shift, without paying the high efficiency penalty of PWM dimming at low light output levels.

There was some discussion on nanotubes and stuff...continuing the discussion...
That and silicon nanocrystals to replace phosphor...

Looks like they are attacking it from multiple aspects.


Anyhow, I was testing some red LEDs today, and found out that their efficiency is barely any different from PWM dimming compared to Current dimming. Their color also shifts dang near the same with either method!

I do notice that it appears the die is bonded directly to the slug. Which is different than a blue/green/white luxeon, where the die is floating on a few solder balls, which connect it to an ESD protection diode, which is then bonded to the slug. I'm guessing the thermal time constants between the red and B/G/White parts are different.

Anyone venture to guess about the other mechanisms that would cause the difference?
Canuke said:
I'd wonder about capacitance or some other mechanism that "smooths" PWM enough to make it resemble CC dimming... but I speculate. I know nothing about parasitic capacitances in the different LED architectures.





No, when driven properly, the capacitance doesn't cause a smoothing of the PWM. I've actually made sub microsecond pulses that have a nice square shape to the optical output. (I've actually done regulated clean pulses now that are 4nS for their time period, none of the information shown here has changed when looking at these levels)

There is only 800pf of capacitance, that changes as the forward voltage changes. (fyi, depending on the brand of power LED, you will typically have ~800pf of capacitance in a white/cyan/blue/green, and 146pf in a amber/red.



I actually exchanged some E-mails with Mike Krames of LumiLEDs Advanced Laboratories.

It has nothing to do with the thermal time constants or anything.

"Efficiency increases with decreasing current for InGaN (blue/green/white) LEDs, to reach a maximum at very low currents (1-2 mA or so for standard small chip used in 5 mm lamps). So, decreasing the dc current level gives you better efficiency, than pulsing at a high current. However, blue shift also depends on the drive current level, so the peak wavelength will change under "current dimming" but not under PWM. This color shift is not due to dimming but is related to some details in the InGaN/GaN electronic bandstructure.

For AlGaInP (red), the blue-shift effects are not present so all you get is red-shifting due to Joule heating, at ~ 0.16 nm/K (which apparently you are avoiding, which is good). Also, for red the efficiency is relative flat with current (it actually *decreases* at very low currents), so you don't observe the effects as with InGaN/GaN. The differences between AlGaInP and InGaN/GaN all relate to bandstructure properties and active region designs, which get into many details."


I was referred to an article for further details:

On the Bandstructure in GaInN/GaN Heterostructures - Strain, Band Gap and Piezoelectric Effect

....The redshift approximately corresponds to one halfperiod of the Franz-Keldysh oscillation near the band gap. We therefore attribute this redshift to the Franz-Keldysh effect [23]. Spatially indirect transitions across the band gap appear at lower energy due to the tilting of the bands. At the same time the transition probability is reduced exponentially as the area under the triangular tunneling barrier grows without involving impurity or defect states (see schematic in Figure 4):



Within a halfperiod of the first Franz-Keldysh oscillation the transition probability has decayed to 4 % of its value at the DOS band gap [23]. In this framework initial and final states are states of delocalized carriers. The exponential "tail" is not induced by any disorder or impurities but rather by the tunneling process in the tilted bandstructure....


And a whole lot more here:

http://nsr.mij.mrs.org/3/31/text.html


Reasonable drive is key.

Notice the 253% increase in efficiency of the LED(in current dimming) at 3% PWM vs. 3% Constant Current, when the PWM is hitting the led with 1.140 Amps.

Thats less than a 50% efficiency level there.

It gets worse at higher PWM currents. Go back and compare it to the 350mA PWM chart.

If you go higher for current, with the PWM, the LED efficiency continues to drop when dimmed.

Yes, a well designed PWM, that takes in consideration of the LED, converter/PWM, and the battery.

Lately, Constant Current regulators in flashlights have made leaps and bounds, with efficiencies exceeding 90%.(over the years since I wrote this up, manufacturers have continued with cheap low efficiency converters-> higher profit margins, only a very few utilize high efficiency converters, fyi)

Another consideration is regulation. Except on high, PWM can also regulate the light by adjusting the duty cycle as the battery depletes, but that isn't done on the mainstream lights which use PWM, like the Photons, LionHeart, LionCub. As commonly implemented it is more like direct drive. Here is an example of the unregulated runtime output on one of the common PWM lights:




So, what to do with this information?
The whole light needs to be considered as a system for the best solution, not just the converter/PWM circuit alone. Heat generated, thermal paths, temperature, reflector, front lens, battery contacts, converter/PWM, LED, battery, high load pulses on the battery, etc.


The effect of CC dimming is even more dramatic if you go on down towards 3% dimming, where you get 240% more light, if the LED I used was driven at 1140mA (data from that example on page 1).

So, if we assume you had 50 candelas at 1140mA
-PWM would give you 1.5 candelas
-CC would give you 3.6 candelas

And this is for the same power consumed.


Another way to look at it, is to dim both down to 1.5 candelas, and consume 58% less power.


Something that is often a forgotten piece of the puzzle is the effect on battery efficiencies when you take a full blown current pulse out of the battery (PWM), vs. sipping ever so lightly on the cell. The battery will deliver quite a bit more energy if you just pull a light current off of it, instead of yanking a full 1140mA pulse out of it. Consider the I^2 * R power losses formula and the internal battery resistance for starters.


Now, a typical switcher starts getting less efficient down at those levels. Most chips these days have burst modes/pulse skipping, which will greatly increase the efficiencies at this level, and if you have a proper sized capacitor, you can still keep the ripple down at 0.8%. Another technique is to use two phase switchers, and just switch off one phase. If you think a little deeper, you can put heavy MOSFETs (higher gate charge, lower on resistance) on one phase, for the higher currents, and put light MOSFETs (low gate charge, higher on resistance), and keep the efficiency of the switcher up. You can also lower the frequency, to reduce losses due to gate charge and gate drivers at light loads.
Often you will save additional energy on the very light load side, if you replace the one of the MOSFETs (depends on buck or boost), with a schottky diode.


As you start going up in PWM frequencies, you start taking additional losses due to the 700-800pf of capacitance in the LED and other factors. Adding just a capacitor on the output of the PWM, without an inductor, will greatly increase the losses in the transistor.


In reply to a post by CM-

The paper on the PWM thing was done by Henry of HDS, where he used PFM to get around the PWM LED flashlight dimming patent.


Another technique is to dim with CC, on the upper end, where efficiency is important (high power draws), and then switch to PWM/PFM off the CC mode, where you convert down to lower the current, but then pulse that. (Note, there are a few chips where you can now actually do this, they have given the designers the hooks into the chip to accomplish these more advanced methods/techniques)


One note about dimming of LEDs, is that if you use something like an X1/WF bin, the green tint will become pretty obvious and re-inforced at the lower currents. Of course, this also depends on where in the X1/WF bin it is actually at- within the X1/WF bin.

Both PWM and CC have tint shift issues when dimming, and there is a chart at the beginning demonstrating the effect. PWM has a bit less shift, but notice how when CC dimming, it started shifting back towards the cyan direction at the very low levels. Something for you to think about...


Main site
Mirror:
Main site mirror