High Power LED Controller

[MikeMl]

Here is a constant-current switch-mode regulator derived from Hero's reference. I show how it handles start-up, turn-off, and variation of input voltage from 12 to 15V. I set the output current to 1.7A. The current is set by the 0.33Ω shunt resistor and the Vbe of the 2N3904.

The inductor is 10mH, and must handle 2A without saturation. The dissipation in the FET is miniscule, in fact the only significant dissipation occurs in the shunt resistor (~1W) and the LEDs (~4.5W each).

Now all we need is someone to build it

I see one potential problem with a switcher vs the linear regulator I presented earlier. It will be a ***** to keep it out of any audio in your car. It switches at an audible rate, and varies frequency as the input voltage changes. I tried one like this (for incandescent lamp dimming) in an airplane once, and could not get it out of the headphone audio. I gave up and went to a linear transistorized dimmer instead. There is about 2000 square feet of aluminum to heatsink transistors to in an airplane.

 

[Hero999]

Simple, design it to switch at an ultrasonic rate and interference is less of a problem.

I would be tempted to use a much smaller inductor but the problem with that circuit is that R4 and R6 will slow the switching of the MOSFET down, making it unsuitable for high frequency operation.

 

[Hero999]

Here's my idea, LTSpice says it's 83% efficient.

The resistance of the inductor was 100mΩ.

Q1 and D1 convert the LM311 open collector comparator output to a proper push-pull output. I found adding R5 improves the turn on time of the M1 no end.

Using a low capacitance MOSFET helps to improve the efficiency and seems to as important than on resistance.

Using a high speed comparator with a push-pull output will probably reduce the switching losses as well as cut down on components.

EDIT:
The LED current is just under 1.5A and the switching frequency is 40kHz.

 

[MikeMl]

Even at 40kHz, I still see the a noise problem. Since the OP started out wanting 5 LEDs per headlight, the circuits posted drive them three at a time, which means like four strings of three each. If you just build four switchers, the beat frequencies between them would be as bad as running at audible frequencies. You would have to some how synchronize the four switchers so they run at the same frequency. At that, it would be preferable to interlace the PWM so as to reduce the ratio of peak-to-average current drawn by the four strings.

 

[Hero999]

You're right about the beat frequency issue.

An easy option would be to just stick them all in series and use a boost converter.

If should be fairly easy to add other synchronised switchers, providing they have the same number of LED with the same voltage drop. Just add more inductor, Schotky, LED and MOSFET units. The wavefrom can be easily tapped from M1's source and used to drive another MOSFET using a driver similar to Q1 and D1 (but using a PNP). I can post a schematic if you're interested.

 

[Mr RB]

Here is a constant-current switch-mode regulator derived from Hero's reference. I show how it handles start-up, turn-off, and variation of input voltage from 12 to 15V. I set the output current to 1.7A. The current is set by the 0.33Ω shunt resistor and the Vbe of the 2N3904.
...

It's not great. The load current has a lot of ripple and the drive to the switching FET is not as sharp as it could be.

The 3 transistor circuit I posted has less parts, has DC load current (for less noise and better LED life) and (i believe) faster switching of the main switch due to better pos feedback from the inductor itself. And it has separate current regulation and voltage regulation, so the open circuit voltage can be set a fraction above the LED voltage to ensure that LEDs are not killed when connected to a live supply.

 

[Hero999]

Don't get me wrong, the Black regulator is truly great, I just wouldn't recommend it in this application.

I can't see how it would be able to achieve fast switching speed from a MOSFET, not without adding another transistor, so I think it'd be better to stick to bipolar transistors for the Black regulator.

Another issue is the inductor; tell me exactly where you are going to find a 2A, 470µH inductor from and how much it'll cost? It'll be almost five times the size as the 100µH inductor and have a proportionally higher ESR which will hamper the efficiency.

Wow, I've just looked back and noticed that Mike's circuit uses a 10mH 2A inductor which will be huge, the size of a small mains transformer.

 

[Mr RB]

Well the black regulator can use a lower value inductor, as it oscillates around the inductor properties. The only problem with a lower L is the switching freq increases so switching losses increase.

2A and 3A pre-wound toroids can be bought from most of the hobby electronics suppliers now, they are about 1" to 1.25" across and available in the 100uH to 470uH range.

Mike's circuit is cool because it doesn't use an output cap and oscillates around load making it fairly failsafe and simple, but my circuit uses similar or less parts count and includes DC filtered output and regulates both voltage and current which can be set independantly.

As for driving the FET its always a problem, Mike's circuit uses a 1k pull down resistor to drive the gate, a weak point there. My circuit needs a logic level driven PFET and uses a pull-up resistor to turn the gate off, again a weak point. It would probably be limited to 2 LEDs as well, not 3.

The main thing I don't like about Mike's is the feedback is not inductor driven. I think feedback is best from the inductor as it helps tune the oscillation to the inductor properties AND provides a lot more oomph in the feedback to ensure fast switching. It woul dbe hard to add inductor feedback to his circuit as its polarity is inverted to the regulator stage.

 

[Hero999]

Here's a MOSFET version of the Black regulator.

I've added an extra transistor and diode to make the drive push-pull instead of open collector. I've also put C3 before R1 which helps.

Efficiency is just under 90%, the switching frequency is 30kHz and the current is just under 1.6A.

This circuit only works because I used a low threshold, low Ron MOSFET, the AO6047, which is being pushed to its limits as the output voltage is nearly 8.5V.

The ESR of the inductor is 25mΩ and the peak current is just under 3.2A.

The disadvantage of the Black regulator is that the peak inductor current is high but this is smoothed out by a filter capacitor so there's very little ripply in the load.

The black regulator also has the disadvantage of relying on a low threshold MOSFET to work properly, when the load voltage gets neart the supply voltage.

 

[Mr RB]

Very nice Hero.

I have a couple of questions;
1. Did you really need Q3 / D2? Was there a problem getting the fet to turn off quick enough? I would have tried just reducing R2 to a couple of hundred ohms or less, and seen if it turned off fast enough. The current used for R2 only ends up going through the load anyway.

2. The inductor current ripple is much larger than I would like or expect, I think you should change L1 to 100uH or 220uH, there is no benefit to a very high switching speed and it causes more switching losses and more I2R losses on the inductor peak current.
The inductor current is best in a range of +/- 0.3 Iout, so from 1 amp to 2 amp would be much better, 1.2 amp to 1.8 amp better still.

3. did you try putting another cap on the output? Putting the main cap there and reducing C3 should reduce current ripple quite a lot. The way you have it set up now slows the current feedback down a lot (hence larger current ripple).

PS. If you like I can put your circuit up on the web page after you fix the inductor current.

 

[Hero999]

Very nice Hero.

I have a couple of questions;
1. Did you really need Q3 / D2? Was there a problem getting the fet to turn off quick enough? I would have tried just reducing R2 to a couple of hundred ohms or less, and seen if it turned off fast enough. The current used for R2 only ends up going through the load anyway.

No, it won't switch off fast enough without Q3 and D2. The current through R2 may only end up going to the load but the increased switching losses ruin the efficiency, try it yourself.

2. The inductor current ripple is much larger than I would like or expect, I think you should change L1 to 100uH or 220uH, there is no benefit to a very high switching speed and it causes more switching losses and more I2R losses on the inductor peak current.
The inductor current is best in a range of +/- 0.3 Iout, so from 1 amp to 2 amp would be much better, 1.2 amp to 1.8 amp better still.

Increasing the inductor value doesn't make much difference to the peak current, with a 220µH inductor the peak current is 2.8A and switching frequency drops to just under 4kHz so it will emit a horrible noise.

Not only will a larger inductor be more expensive, but it will also have a higher ESR meaning more losses.

3. did you try putting another cap on the output? Putting the main cap there and reducing C3 should reduce current ripple quite a lot. The way you have it set up now slows the current feedback down a lot (hence larger current ripple).

I see why you had it set up the way it was before now.

C1 and C2 seem to change the frequency more though. C2 reduces the frequency and the frequency increases if it's removed.

PS. If you like I can put your circuit up on the web page after you fix the inductor current.

It will always be a compromise between inductor size, switching frequency and current ripple.

A larger inductor will give less ripple for a given current at a certain switching frequency.

I'll post a version with a lower ripple current if you like but it will use a larger inductor or switch at a higher frequency.

I'd be flattered if you put it on your site but I recommend actually building it. If you do post it, without testing it, I think you should say that it's only been tried in LTSpice.

Here's the improved design:
Frequency: 120kHz
Peak current: 2A
Output current: 1.6A
Efficiency: 91%

The problem with all these MOSFET designs is that a very low threshold MOSFET is required. The trouble is that lots of these MOSFETs have a maximum gate-source voltage of 10V which is fine as long as the output isn't short circuited. It's probably a good idea to add a zener between the gate and source to protect it.

If you find a MOSFET with a low Ron at a gate voltage of 3V with a maximum Vgs of >15V then let me know. The MOSFET used here (AO6407) has a maximum Vgs of 12V but I can't seem to find a distributor of this exotic MOSFET.

 

[MikeMl]

I have learned a lot from this thread. First, Its all about the inductor, which means use what you can buy readily, Second, its all about dissipation in the Fet. Third, to lower the dissipation in the FET, you have to charge/discharge the gate capacitance rapidly.

Here is a nothing special PFET (Vt doesn't matter), good current regulation, low-ripple, small inductor, moderately high frequency, low-parts count current regulator for High Power Leds that has ~0.3W dissipation.

I'm cheating, and using a low value resistor to charge the FET gate, but using an active turn off.

Try it out. I have included a zip file.

 

[Mr RB]

Mike, your circuit has the advantage of being able to use the full PSU voltage to control the FET gate, with easier FET turn on/off and can operate with less buck headroom. But all the regulator circuit current is to ground (ie wasted) and the regulator draws power to keep the FET off, so it will be inefficient at low duty cycles (like driving a single 3v LED from 24v) which is the time you need a buck design the most.

But, your circuit is definitely the best when there is only a small headroom, like say a solar 12v -> 3 LEDs where the solar battery might get as low as 11v driving 9v LEDs.

Hero's version of the black regulator shares it's benefits (regulator circuit is on when FET is on, all regulator current goes to output) but also shares it's main fault; the FET gate switching is all done within the buck headroom so with limited headroom like 12v in -> 7v out it's hard to get the FET to switch hard and fast.

Hero, I like your new circuit. You can see how moving that cap on the output speeds up the response of the current feedback. Now your current ripple is +/- 25% it looks good, although I would drop the freq from 120kHz to 20kHz with an increase in L1 to maybe 100uH which is a common enough value in low ohms buck inductors.

I think it would shine as a buck driver for a single 3v LED from 12v. You would have plenty of headroom for good switching and the topology would give better efficiency at low duty cycles like 12v to 3v. With the larger inductor and larger headroom it might even work ok without the extra transistor...

I'll try to make time over the next 2 days to put one together and write it up on the page.

 

[Mr RB]

Ok here's my entry in the LED wars, I drafted the schematic in HandNpen 1.0.

I cheated and turned the buck upside down, so it can use a NFET which will be cheaper and have better performance etc.

It is extremely simple, using only one transistor as the current regulator and one switching FET. The FET is auto on, then when current > limit the BC337 turns on and turns the FET off. The FET gate is driven with full PSU voltage and turned on via a resistor (like Mike's design). Positive feedback is via Cfb direct from the inductor so regulator switching should be sharp.

Ctime may be needed to reduce the freq to improve efficiency.

One clever feature; the current sense resistor only dissipates power when in the ON part of the duty cycle so R losses are greatly reduced there.

Since I have only ran the simulation in HeadSpice 2.0 there may be errors. Mainly the current regulation is based on Imax (not Iaverage) so the current regulation over 11-15v Vin will be imperfect. That can probably be tweaked a bit by tuning the capacitor pair Cfb and Ctime as increasing Ctime will tend to increase the OFF duty more than the ON duty. If that is not enough then Rcomp can be added to provide some Vin compensation to increase the OFF duty as Vin increases. Given the one application (to run 2 LEDs from 12v battery) I think it can be tuned to regulate well enough from 11v to 15v and efficiency and simplicity are both excellent.

 

[MikeMl]

Mike, your circuit has the advantage of being able to use the full PSU voltage to control the FET gate, with easier FET turn on/off and can operate with less buck headroom.

I was trying to address that specifically. The OP wanted to drive ~10 of the high-power leds, so putting three in series seemed like the way to go for an automotive application. Since the total voltage drop for all three is over 10V, and that means that the duty cycle is like 80% on, 20% off, I optimized for that.

But all the regulator circuit current is to ground (ie wasted) and the regulator draws power to keep the FET off, so it will be inefficient at low duty cycles (like driving a single 3v LED from 24v) which is the time you need a buck design the most.

But, your circuit is definitely the best when there is only a small headroom, like say a solar 12v -> 3 LEDs where the solar battery might get as low as 11v driving 9v LEDs.

As I said, the duty cycle is 80 to 90% on, so the current in the pull down resistor on the PFET gate only occurs when the PFET is turned off, so the wasted power is small. Besides, the application is powered with a car's alternator, so a few wasted mW is not too significant.

I also optimized my circuit for protection of the expensive LEDs. If you run the sim, you will see the nice soft-start behavior, and the good current regulation as the input voltage varies from the engine not running (12V) to alternator charging (15V) range. Putting the filter capacitor across the LEDs (instead of to ground) was a "light bulb" moment, too.

btw- my interest in this project is motivated by wanting to replace the landing/taxi lights in my airplane with high-power LEDs

 

[Hero999]

Mike,
How efficient does LTspice say your circuit is?

RB,
Why are you so obsessed with using large inductors?

Most switching PSUs try to use as smaller inductor as possible, going larger is a big step backwards in my opinion: rember lager inductors = more expensive, higher ESR and less efficient. If the MOSFET switches fast enough then switching losses become less of an issue than Ohmic losses in the inductor.

You're better off with a smaller inductor and keeping the extra transistor and diode in. The amount of money you save in cutting down on the number of parts you more than make up in having to buy a larger inductor - you're pinching pennies for pounds.

I agree with your decision to flip it round and use an N-channel MOSFET though.

 

[MikeMl]

Ok here's my entry in the LED wars, I drafted the schematic in HandNpen 1.0.

I cheated and turned the buck upside down, so it can use a NFET which will be cheaper and have better performance etc...

I modified my circuit per these suggestions. When simulated, it took a few tries to get it to work. The biggest problem with it is that the gate pull-up resistor, even when it was 680Ω could not source enough current to turn on the NFET fast enough. The power dissipation in the NFET during the first few tries was a Watt or two. I borrowed Hero's bootstrapped gate driver, and that fixed that. Now even with a crummy NFET, the dissipation is < 500mW.

Next issue, which always bothered me with most of RB's circuits, is that the base of Q1 was driven far enough below ground so as to exceed the Vbe reverse breakdown, so I added the diode clamp D3, and limited the current with R6. I played with R2 to equalize the output current both at 12V and 15V input.

Added: I changed the inductor to 33uH; that just raised the switching frequency without changing anything else. When the input is 15V, the average input power (from the battery) is 17W. The LED power is 16W, but somehow I dont believe that this circuit is 94% efficient?

 

[Hero999]

What did you set the inductor's ESR to?

Look at a typical inductor of the desired value and current rating in a catalogue and set the ESR to the listed value.

 

[MikeMl]

What did you set the inductor's ESR to?

Look at a typical inductor of the desired value and current rating in a catalogue and set the ESR to the listed value.

In the previous posting, the esr was 10mΩ, which is too low. I resimed using an actual 22uH 2.6A 59mΩ Inductor that can be bought from DigiKey, and not much changed. I still get 16W to the leds and burn about 1W in everything else. The shunt is ~ 1/2W, the Fet about 0.2W, the inductor ~0.2W.

 

[MikeMl]

Next question: I need a boost circuit to make 18V @ 3 A out of 12 to 15V. Anybody have a favorite place to start.