Driving series LEDs methodology

[JimmyM]

I have an LED project I'm working on. I have a big RGB LED 3 dies in series per color. I'll be powering it from (probably) a 14.4V nominal NiMH pack. The LED is rated at 350mA per color. Since I want to vary the intensity of each color I have 2 methods I could employ. PWM and current control.
I've tested a 31kHz switched mode supply being controlled directly by my microcontroller. One switch/inductor/diode/cap for each color. It can measure the current on each LED leg and adjust the current to keep it at what ever level I have programmed. So 200mA stays constant as battery voltage falls. This is pretty efficient.
The alternative would be to use a single switched mode supply to drop battery voltage down to the maximum required by LED, probably 11V(Blue). And use a resistor on the other 2 so that each one will be limited to 350mA Max. To control individual brightness I can use PWM to switch the LED. PWM frequency can be high enough to avoid strobing effects.
But which is better for color/brightness control? At low currents, LEDs act a little wonky.
At low brightness, would PWM (pulsing 350mA to each color) at varying duty cycles be better? Or would varying the DC current be better?

 

[Mr Happy]

It's my understanding that a combination of current control and PWM is preferred by designers. Current control as a base to set the desired maximum current through the LED, and then PWM to switch this on and off with varying duty cycle to control the brightness.

 

[JimmyM]

It's my understanding that a combination of current control and PWM is preferred by designers. Current control as a base to set the desired maximum current through the LED, and then PWM to switch this on and off with varying duty cycle to control the brightness.

For white LEDs I can understand this. The color of a white LED can change based on drive level. You wouldn't want your dim LED to be warm while the bright setting to be blue/green.
Does this hold true for Red, Green, or Blue LEDs too?

 

[Mr Happy]

For white LEDs I can understand this. The color of a white LED can change based on drive level. You wouldn't want your dim LED to be warm while the bright setting to be blue/green.
Does this hold true for Red, Green, or Blue LEDs too?

I guess you are looking for a simpler design? But unless it is too expensive to achieve, I would be inclined to the tried and tested PWM approach.

The variation in color with white LEDs is due to the mix of different color phosphors being illuminated at different intensities by different drive currents. With red, blue, green LEDs there may only be a single color phosphor and so the variation in hue with drive current may not occur, and so direct current regulation may be a suitable option. However, I believe you get better linearity with PWM control, and also higher brightness for the same power delivered (since the emitter works most efficiently with currents in the design operating range).

 

[CKOD]

For white LEDs I can understand this. The color of a white LED can change based on drive level. You wouldn't want your dim LED to be warm while the bright setting to be blue/green.
Does this hold true for Red, Green, or Blue LEDs too?

One benefit of it is that it maximises your resolution of your PWM. If youre feeding each color with an 8 bit hardware PWM from a uC, you have 256 different pulse widths available. As soon as you start reducing your maximum pulse width(lets say red is overpowering so you need to back off the red at maximum) you no longer have 256 different steps.
If you set your maximum with current control and control the color with PWM, then you will always have 24 bit color, regardless of what you set your maximum at. And you can use digital pots if needed for tweaking your current, but still retain digital control

 

[georges80]

Red/Green/Blue LEDs do not have phosphors (as one posted stated).

Anyhow, in terms of current/pwm control. For accuracy you are better off running PWM. You can use current regulation to get to your target maximum drive current and then use PWM to do the dimming from there.

Regulating to low currents can be difficult (small voltages across small sense resistors), so PWM is a better choice and gives you larger dynamic range (typically).

Most of the new Linear Tech LED driver/switcher IC's provide two schemes a) adjustable current control b) PWM control. For large dynamic range and accuracy PWM is the way to go.

In addition, some high powered LEDs have a minimum input current requirement, below which they will draw current but not light up. The Luminus -50 and -90 parts have a minimum 1A spec. Not likely a problem with your RGB LEDs, but certainly something to be aware of.

8 bit PWM is often not sufficient if you want to really dim down some of the colours, so I'd recommend you look at one of the AVR's with 10bit (minimum) PWM hardware, especially due to the log nature of human vision, you need more PWM resolution at the dim end of the scale.

cheers,
george.

 

[JimmyM]

All good feedback. Thanks. I'll look into the LED driver ICs to see what I can get away with. If I limit current to 350mA per color, I can use PWM for dimming. I was just thinking about using current regulation since I could implement it directly from the uC. I'll see what's around.

 

[JimmyM]

I found a nice 3 channel buck-boost current limiting LED driver that has PWM inputs. It's adding complexity, but it will allow a lot of smooth dimming.
Should be nice.

 

[DIWdiver]

Red, Green, and Blue LEDs don't change color with current, and they are typically less and less efficient as you increase the power level, due to increasing resistive losses. This is another good reason to consider analog dimming (by adjusting the DC current), but it will only be significant if you operate well below rated power and battery life is very important.

Something to watch for when planning PWM on an LED driver IC, some ICs don't turn the LED on and off fast enough to work well at very low duty cycles. Maybe you don't care if you don't plan to go to very low light levels, but 1/256 duty cycle at 200 Hz is 20 uS pulse width. Also, you mentioned that the driver has PWM inputs, and most drivers with PWM inputs are at least decent in this respect.

I turned down the current limit on my power supply to 1 mA, and the SST90 was still putting out light. Moonlight mode would want to be 10mA or so, which is why the driver I'm working on will have 1000:1 dimming capability, probably about 30:1 analog and 35:1 PWM. The chip I selected isn't meant for PWM, so I'm not sure I'll be able to get to even 35:1. Hopefully I'll know before too much longer. But there's so much to do:hairpull:

 

[JimmyM]

Red, Green, and Blue LEDs don't change color with current, and they are typically less and less efficient as you increase the power level, due to increasing resistive losses. This is another good reason to consider analog dimming (by adjusting the DC current), but it will only be significant if you operate well below rated power and battery life is very important.

Something to watch for when planning PWM on an LED driver IC, some ICs don't turn the LED on and off fast enough to work well at very low duty cycles. Maybe you don't care if you don't plan to go to very low light levels, but 1/256 duty cycle at 200 Hz is 20 uS pulse width. Also, you mentioned that the driver has PWM inputs, and most drivers with PWM inputs are at least decent in this respect.

I turned down the current limit on my power supply to 1 mA, and the SST90 was still putting out light. Moonlight mode would want to be 10mA or so, which is why the driver I'm working on will have 1000:1 dimming capability, probably about 30:1 analog and 35:1 PWM. The chip I selected isn't meant for PWM, so I'm not sure I'll be able to get to even 35:1. Hopefully I'll know before too much longer. But there's so much to do:hairpull:

All good info, thanks, DIWdiver. The chip I'm looking at now is a SMPS capable of current control in buck, boost, or buck-boost configurations on 3 channels. Each can do PWM. The data sheet refers to 100Hz, but I figured I could use different PWM frequencies and see how it looked on the scope.

 

[georges80]

I turned down the current limit on my power supply to 1 mA, and the SST90 was still putting out light. Moonlight mode would want to be 10mA or so, which is why the driver I'm working on will have 1000:1 dimming capability, probably about 30:1 analog and 35:1 PWM. The chip I selected isn't meant for PWM, so I'm not sure I'll be able to get to even 35:1. Hopefully I'll know before too much longer. But there's so much to do:hairpull:

Just in case I misunderstand you - the Luminus -50 and -90 is NOT guaranteed to light below 1A - hence the minimum 1A spec. The -50 will light to maybe 0.5A, and the -90 to maybe 0.75A, though Luminus chose 1A as their minimum.

I have spoken to Luminus about this and they recommend 1A as the minimum driver current and to then transition to PWM for further dimming.

This is what I implemented in my 6.6A h6Flex driver.

So, just because one or a small sample set of -50 or -90 devices will light at 10mA (constant current), it is NOT indicative of the entire 'breed'. The -50 and -90 parts are a single monolithic LED emitting surface (versus something like a mult-die P7) and the minimum operating current to ensure it 'lights' is a device physics issue. Anyhow, just a heads up that with some of these HIGH power LEDs we can't just assume they'll operate at any low continuous current level.

cheers,
george.

 

[JimmyM]

I think I'll try the current control method first. It's the most straight forward and fewest components. Once the LED arrives I'll do some testing with the buck SMPS method.
Regardless of the battery supply I do want to use switched mode voltage/current control to keep efficiency as high as I can. If I move to using just the 8D cells that the lantern takes off the shelf, I will either have to use a buck-boost converter for the LED feed voltage, or move to using 3 Cree colored LEDs and a buck converter. We'll see what results from testing.
I really appreciate the feedback I've gotten. You guys have been a big help providing possible alternatives.

 

[DIWdiver]

Just in case I misunderstand you - the Luminus -50 and -90 is NOT guaranteed to light below 1A - hence the minimum 1A spec. The -50 will light to maybe 0.5A, and the -90 to maybe 0.75A, though Luminus chose 1A as their minimum.

I have spoken to Luminus about this and they recommend 1A as the minimum driver current and to then transition to PWM for further dimming.

Hmm....

The data sheet only mentions in a footnote, which I found after going looking for it: "Special design considerations must be observed for operations under 1A. Please contact Luminus for further information."

I was planning on running down around 300mA before going to PWM, thinking that the color shifts would be acceptable and there wouldn't be other concerns. I've seen LEDs put out zero light at nonzero current, but it was always down in the region where you had to turn out the lights and look close in order to see if they were on. I never imagined that at 10% of the max current, bright enough you can't look at it, there would be a risk they wouldn't light up at all!

I'll admit that my small sample size (1) doesn't mean much statistically, but it's hard to believe that if one lights up at 1mA, its brethren might need 1000 times that. The physics don't usually suggest 1000:1 uncertainties.

Could there be something else going on besides a concern that the LED might not light up? You spoke to Luminus, did they really say "it might not light up at all", or something more like "it won't perform well"?

By the way, at low enough light levels that you can actually look directly at the LED, mine shows some distinct blue spots, as though the phosphor coating on the die has some holes in it. These don't change with current, at least at levels well below the minimum recommended current.

Don

 

[JimmyM]

I was using a white LED during some testing. It's rated at 110mA. I wish I had been measuring current at the time. But, as I raised voltage, there would be this feeble bluish/white glow, then by raising voltage just a little more, it would "wake up" and brightness would increase significantly. From there, it just got brighter and brighter.
Note to self (and others maybe). When testing a buck converter, you need some load on the output at all times, otherwise the output cap will charge to the full input voltage and dump into the next LED you connect. Instantly burning it out. Of course I was using 100uF instead of 10uF.

 

[georges80]

Hmm....

The data sheet only mentions in a footnote, which I found after going looking for it: "Special design considerations must be observed for operations under 1A. Please contact Luminus for further information."

I was planning on running down around 300mA before going to PWM, thinking that the color shifts would be acceptable and there wouldn't be other concerns. I've seen LEDs put out zero light at nonzero current, but it was always down in the region where you had to turn out the lights and look close in order to see if they were on. I never imagined that at 10% of the max current, bright enough you can't look at it, there would be a risk they wouldn't light up at all!

I'll admit that my small sample size (1) doesn't mean much statistically, but it's hard to believe that if one lights up at 1mA, its brethren might need 1000 times that. The physics don't usually suggest 1000:1 uncertainties.

Could there be something else going on besides a concern that the LED might not light up? You spoke to Luminus, did they really say "it might not light up at all", or something more like "it won't perform well"?

By the way, at low enough light levels that you can actually look directly at the LED, mine shows some distinct blue spots, as though the phosphor coating on the die has some holes in it. These don't change with current, at least at levels well below the minimum recommended current.

Don

YES - I spoke to Luminus and YES, they do not guarantee the LED will light at lower currents. THAT'S why they have the 1A minimum spec and YES, they recommend PWM with 1A minimum pulses if you want to dim below 1A average current.

Remember, their spec is to cover full production quantities at all temperatures and process variations. I had a customer that specifically had problems at <0.5A constant current (the LED would NOT light, a -50 I think), so yes, your sample size of 1 means nothing...

So, if you plan to make a driver to sell in any volume I HIGHLY recommend you do not go below 1A constant current and transition to PWM at that point. That's what I implemented in the firmware of the h6flex - its switcher IC can handle PWM very nicely with very fast response time. If in doubt, the 'elephone (amazing device) can be used to talk to Luminus directly...

cheers,
george.

 

[georges80]

Note to self (and others maybe). When testing a buck converter, you need some load on the output at all times, otherwise the output cap will charge to the full input voltage and dump into the next LED you connect. Instantly burning it out. Of course I was using 100uF instead of 10uF.

Yep, and more so for a boost converter I have notes on my website warning users of my drivers of that very issue. Even a 10uF cap fully charged (especially low ESR ceramics) can provide quite a "punch" to a LED. I've not destroyed, but certainly damaged a few Luxeons in the early days....

Funny, the typical 'damage' to the Luxeons meant they wouldn't light at low current (say <100mA) and even at higher currents would not put out full lumens - it's as if part of the die was damaged.

cheers,
george.

 

[CKOD]

Yep, and more so for a boost converter I have notes on my website warning users of my drivers of that very issue. Even a 10uF cap fully charged (especially low ESR ceramics) can provide quite a "punch" to a LED. I've not destroyed, but certainly damaged a few Luxeons in the early days....

Funny, the typical 'damage' to the Luxeons meant they wouldn't light at low current (say <100mA) and even at higher currents would not put out full lumens - it's as if part of the die was damaged.

cheers,
george.

Yeah, its funny how stuff like that can go. I have a power supply at work that I use for testing out stuff occasionally. definitely NEVER connect it with it on, and set your V slightly above what you need, even though youre running it in constant current mode. I dont know how much capacitance is on the output, but all the way up, it would probably blow out some smaller power LEDs. And even if you have it conneted while its on, there is a brief flash before it goes to CC mode. No laser diodes on this PSU, at least not without a shut while the supply is turned off and on.

 

[JimmyM]

re: Boost converter. NO KIDDING!

I've found a few good tech docs on designing SEPIC converters. I like that they're not inverters. That makes adding a current sense shunt/amplifier a lot easier.
I'll mess around with a few bit I have here. TI T5001 Buck converter is one that comes to mind. There are also a lot of nice buck-boost/SEPIC controllers out there too. I'm not looking to make super small converters like George does. I would like to get the hang of designing and building buck/SEPIC converters for other projects in general.
Right now the only reason I wouldn't just use the 3 buck-converter approach is if I don't get acceptable fading performance. Or I HAVE to use a lower voltage source.

 

[DIWdiver]

YES - I spoke to Luminus and YES, they do not guarantee the LED will light at lower currents. THAT'S why they have the 1A minimum spec and YES, they recommend PWM with 1A minimum pulses if you want to dim below 1A average current.

Remember, their spec is to cover full production quantities at all temperatures and process variations. I had a customer that specifically had problems at <0.5A constant current (the LED would NOT light, a -50 I think), so yes, your sample size of 1 means nothing...

So, if you plan to make a driver to sell in any volume I HIGHLY recommend you do not go below 1A constant current and transition to PWM at that point. That's what I implemented in the firmware of the h6flex - its switcher IC can handle PWM very nicely with very fast response time. If in doubt, the 'elephone (amazing device) can be used to talk to Luminus directly...

cheers,
george.

Damn. I was just about to pull the trigger on the prototype boards. Now I'll have to go re-evaluate first. I wish they had put that in big bold letters under the heading "REALLY, NO KIDDING, WE'RE SERIOUS ABOUT THIS, AND HERE'S WHY". At the very least they should put it in the specs and not in an unreferenced footnote.

Maybe "moonlight" mode will be more like "full moon on a clear night in the tropics" mode.

Thanks for the info, George. Sounds like maybe I learned more easily than you did.

Don

 

[DIWdiver]

re: Boost converter. NO KIDDING!

I've found a few good tech docs on designing SEPIC converters. I like that they're not inverters. That makes adding a current sense shunt/amplifier a lot easier.
I'll mess around with a few bit I have here. TI T5001 Buck converter is one that comes to mind. There are also a lot of nice buck-boost/SEPIC controllers out there too. I'm not looking to make super small converters like George does. I would like to get the hang of designing and building buck/SEPIC converters for other projects in general.
Right now the only reason I wouldn't just use the 3 buck-converter approach is if I don't get acceptable fading performance. Or I HAVE to use a lower voltage source.

SEPICs are cool if the output voltage may cross over the input voltage. Otherwise, they are just more expensive and less efficient. Of course, as I say this I have a design proposal on my desk that includes one running substantial power levels.

In general, and especially if you are talking about a battery operated device, I would recommend you shoot for a straight buck or boost design before resorting to SEPIC if necessary, because of the efficiency issues.