30W LED driver - EMI concerns?

[Martin v]

Hi everyone,

I have designed and built a boost converter for driving high power LEDs. Schematic attached. Part values are chosen for the supply voltage I'm working with as well as the LEDs I'm driving. In practice the input voltage is 12-24V nominal and it drives 1A LEDs that have a voltage drop around 30V (ie. 30W LEDs). I have released it under a Creative Commons license.

The circuit works a treat and I'm using it for lighting in much of our house.

Recently I heard of problems with EMI produced by sub-standard LED lighting being imported into the country.

EMI is something I hadn't considered when designing the circuit, and while I have taken much care with the PCB layout I haven't put anything in specifically to combat EMI.

I want this circuit to be as useful to as many people as possible and of course want it to perform wonderfully for all. I don't have easy access to sophisticated test equipment - even the price for obtaining the required EMI standard document cannot be justified (it's hundred of $$$).

I guess what I'm asking is if there are some sensible circuit mods I can make to ensure this thing will be nice and quiet?

Also is there some way to measure cheaply (ie. with the tools available to a hobbyist) how the circuit radiates so I can perform before and after tests that would be helpful and appreciated too.

 

[ronsimpson]

Martin,
I think C1, C7 should have a small capacitor in parallel. I did not check but the resonant frequency on the caps is low. If you are having 100mhz noise problems these caps will not be good at that frequency.

IC1-pin1 could use a 0.1uF or 0.01uF cap.

If the wires from main power OR the wires to the LEDs are long then you have an antenna.
>On the input side I think there is high frequency current in the wire. You might want to add a CLC filter. A current probe on the input wire will probably show that your PWM is pulling high frequency current from the power supply. Adding a inductor should open up you wire at high frequencies.
>On the LED side I think you have the same problem and probably much worse. I am not convinced that C7 will short out all the sharp edges.

Much of the noise issues will because of the PCB layout. Keep high current (fast high current) traces short. When the FET turns off it will shake the world. Depending on the mode of operation the FET may be loud at turn on.

Hope this helps.
Ron,

 

[dr pepper]

If its a one off dont worry about it too much, do some testing, put things you might think would be interfered with by it and see what happens.

Keep the wires short to the led from the circuit, maybe put the circuit in a conductive box, connect - to the box.
You could put a snubber across the switching transistor, and maybe use a coil that has a transfromer core or a torroid core to reduce stray inductance that can cause emi in more ways than one.

 

[Tony Stewart]

If you want to test it for EMI, in this design, use an AM or VHF radio between stations and compare the noise with an electric DC drill or a variable speed AC drill.

 

[cct314]

If one wanted to go "all out" to make sure that such a device was "clean", this is one of those situations that can be "made" or "broken" by a board layout - and the components chosen.

C1 and C7 should be of the low ESR types (electrolytic or monolithic ceramic) connected on the opposite side to a very low-impedance (e.g. "heavy" and short) single-point ground that connects also to R4/R5, the ground leads of IC1, and the bottom of R13/R14: Preferably, these should be connected almost together within a very small (a cm or so) area to minimize the probability of circulating currents and differential voltages - which would almost exclusively contain RFI! Ideally, one would use a committed ground plane for this if there is the luxury of such (e.g. double-sided board, or better yet, multi-layer.)

If RFI is going to be a problem with this circuit it will be, in order of most likely source to less likely:

- Differentially, between the grounds of CON1 and CON2.
- Out, via the DC in connection via CON2, superimposed on the V+ connection.
- If there is any significant lead length, out to the LED array, via CON1, superimposed on the V+ connection.

The first one of these is almost entirely up to board layout (e.g. good, heavy, single-point ground) while the last two can be controlled by component selection - particularly with the use of low-ESR capacitors of known-good brands from reputable distributers (e.g. Panasonic, Nichicon).

In the first of the above, if there are differential currents, a LOT of grunge can be radiated if there is significant lead length connected to the DC input side and to the LED output side as these will act as antenna elements - particularly if one (or both) can somehow couple into the mains wiring - say, back to a low voltage DC supply located some distance from the LED driver circuit.

Were I to construct such a device I might use higher capacitance values than 10 uF for C1 and C7 (actually, I'd use additional low-ESR devices in parallel - say, 100uF) to further-shunt the switching energy. I would also put a series choke at the input and output, say, 22-100uH with an additional 0.1uF to ground on the "outside world" side of each choke. A word of warning about that, though: If the ground to which one were connecting the 0.1 uF weren't *really* good (e.g. a really solid, single-point ground, bad board layout) then the addition of these capacitors could actually make things worse again by coupling RFI from the "bad" ground to the outside of the chokes again!

The way around this would be to use bifilar inductors (22uH or higher - 100uH+ is typical) in addition to the aforementioned series chokes (which would be on the "circuit" side of the bifilar chokes) for the DC going in and out with bypass capacitors (0.1uF or so) connecting to a metal case, with that metal case containing the circuit tied at just ONE place to that single-point ground. While this is one of the very best way to (almost) completely assure that RFI doesn't escape the circuit.

Best of luck!

 

[Martin v]

Thanks everyone for the (sometimes very detailed!) feedback. It is appreciated.

I am attaching the board layout if anyone is interested. The shape and size of board as well as location of connectors was dictated by other constraints. That said I don't think the layout is too bad as a first pass???



I have been trying some experiments with two different radios trying to use them as interference detectors.

I used a 12V battery, drive circuit and LED, all with short leads between for the tests. An 'ideal' configuration.

The first radio is a mains powered decent 3 band Sony AM/FM/MW. I tried a variable speed DC drill, variable speed AC drill and the LED drive circuit and didn't pick up anything noticable on any of the bands.

The second radio is a free give-away 2 x AA battery AM/FM unit that exhibits new levels of crappiness in almost every area. The selectivity on the AM band is particularly poor - I can tune across the whole band and go from one station to the next (sometimes two simultaneously) without any real quiet spots whatsoever. With the variable speed DC drill I nevertheless don't get any interference. With the variable speed AC drill I get interference on AM quite clearly! By moving the radio around I can locate where exactly in the drill the interference is originating! Trying it with the LED drive circuit I get quiet - as in I turn on the circuit and the radio station disappears and goes quiet! So not noisy interference like the AC drill, but it is having some influence on the radio, even if it is to make the radio lose station (whatever that means).

So the cheap radio does seem to be a detector of interference but I'm wondering if there is a slightly more up-market detector that could give me something a little more quantitative, without going into the hundreds/thousands for professional equipment.

...

More than one person raised the issue of length of input and output cables. While in most cases these can be kept short, particularly the input cable, the cable to the LED sometimes needs to be long.

I have just now tested another configuration where I'm running the driver from a mains->24V power supply connected by short leads, then long leads to the LED. The long LED leads enclose a large loop, which is asking for trouble, but is what aesthetics dictate. The driver schematic is the same as the one above, though board layout differs and incorporates an Arduino for PWMing the LED driver chip (brightness control). Here is a photo showing the LED with output leads. Photo of light

The cheap radio picks up quite a lot of interference on AM when held near the light. I guess something like a 250Hz square wave that kind of changes in tonal quality as I change the duty cycle. Which makes sense! So I'm very happy that I can identify this. It's less awesome that the output is noisy.

I think I will re-visit the circuit and add in the options for the various improvements:

1) bypassing caps with smaller caps
2) low pass filters in the form of L and C
3) common mode chokes
4) snubber for the FET
5) toroid for the coil (or a shielded inductor?)
6) bypassing the driver chip power supply

That way I can evaluate what effect each addition has.

Once again many thanks!

 

[ronsimpson]

I would use a smaller size wire to fill ground. Look at CON2 and CON1 how I filled on around the connector.
Also around R2 I filled in some ground.
I added a small cap from D1 to ground. The idea is to get current back to ground faster. (back to R4,5)
I wish you could connect the ground end of C1 to the ground end of R4,5 and C7. That should be a short connection.
For power supplies, using SMT, I do not use thermals. I talked with several production houses and they do not have a problem with that. I understand why thermals are used with flow solder, but with IR soldering or oven soldering the thermals are not needed. (some people will disagree)

 

[Martin v]

I added a small cap from D1 to ground. The idea is to get current back to ground faster. (back to R4,5)
I wish you could connect the ground end of C1 to the ground end of R4,5 and C7. That should be a short connection.
View attachment 90618

Thanks for the improvements.

I think I'll re-visit the layout and see if I can improve the ground between C1 and R4,5 and C7. Past experience tells me I'm always going to end up with one long run somewhere but maybe I can improve upon what's there.

As a matter of interest I measured the MOSFET drain with the idea of adding a snubber. Photo attached. The spike isn't too bad, going to just over 40V before returning to steady state 30V. The oscillation looks to be around 100MHz but dies out pretty quickly. It's right up near the top end of my 100MHz scope's capabilities so I'm wondering how accurate the reading actually is. I think I'll make a spot on the next PCB for a snubber in any case.

 

[Tony Stewart]

Trying it with the LED drive circuit I get quiet - as in I turn on the circuit and the radio station disappears and goes quiet! So not noisy interference like the AC drill, but it is having some influence on the radio, even if it is to make the radio lose station (whatever that means).


More than one person raised the issue of length of input and output cables. While in most cases these can be kept short, particularly the input cable, the cable to the LED sometimes needs to be long.

The cheap radio picks up quite a lot of interference on AM when held near the light. I guess something like a 250Hz square wave that kind of changes in tonal quality as I change the duty cycle. Which makes sense! So I'm very happy that I can identify this. It's less awesome that the output is noisy.

The AC drill is an example of a product that could fail FCC tests. The cable makes a great antenna and the brush arcs, a great impulse generator.
The cheap radio works well to verify this.

Putting this radio near the LED driver swamps the front end to turn down the gain of the AGC control thus quieting the radio.
Putting the radio near the light picks up the modulation with less dramatic effects on AGC but audible on speaker.
The cordless drill works as a good example of low EMI.

If you can find similar noisy products but with an FCC ID then you have a good benchmark for reference and a good cheap tester.
A large CM choke on the LED cable, should give good reduction without loading the MOSFETs.