LED grid design and diffusion for color mixing

[amaschas]

Over the last few months I've been putting together a project that involves a array of color LEDs in specific wavelengths of red, green and blue. Because the wavelengths required for my use case are so specific (440nm, 535nm, 660nm), and the blue and red particularly are outside the bounds of what you would readilyfind in a multi-color emitter I've had to use separate emitters per-color, which is causing some problems when it comes to mixing the three colors. I've attached a photo of my emitter board, which as you can see uses alternating rows of each color. The problems that arise from this are what you would image: the individual rows of color are discernible through most simple diffusers, and the problem is particularly bad on the edges where one color dominates. I've been thinking of few possible solutions to this:

1) Decrease the spacing between the emitters and alternate them in a repeating pattern (RGBRGBRGB) the way pixels in a monitor or the photo sites in a camera with a bayer array are arranged. This would potentially help with mixing, but would reduce the size of the array and make routing much more complex.

2) Explore more complex diffusers. Currently I'm actually using the multiple layers of material I pulled out of a broken tablet LCD, which appears to be a combination of light channeling sheets, multiple fresnel lenses and a diffuser. This is a big improvement over the simple plastic diffuser, but I'm not sure how I would source this material, and there are still issues on the edges of the panel where one color dominates.

3) Move the emitters to the edge of light, or add a mixing chamber, which would probably work well but would drastically reduce the brightness.

4) Get custom RGB emitters made. I know there are vendors that do this, but I don't know how much it costs and what the minimum order quantity is. Seems like it would be a big resource commitment.

Any suggestions or advice would be helpful.

[tooki]

FWIW, you can get the diffuser films and panels on AliExpress. But I do suspect you’d do better to arrange the LEDs in a pattern that distributes them more. (FYI, RGBRGBRGB, which is what I’d use, is not the Bayer pattern, which is specifically RGRGRGRG on one row and GBGBGBGB on the next, so that it’s an even checkerboard of 50% green pixels, with the red and blue filling up alternate rows.) Your LEDs aren’t densely packed, so I can’t imagine how it’d be problematic to mix the colors up. Think about how you might lay out an array of RGB LEDs, then just ignore 2 colors per LED.

[Benta]

Arranging the LEDs as a military parade gives bad mixing.
You'll have to put more effort into placing them. Yes, routing will be more difficult, but who cares today?
Another option to consider (I don't know your application) is to use two green LEDs per one red and blue. The human eye is especially sensitive to green spatial response; this might be irrelevant here, though.

[amaschas]

Seems like my best bet here might be alternating the colors and moving the emitters a bit closer together. I'm going to give this a shot on my next board design and see how it works!

[tooki]

Arranging the LEDs as a military parade gives bad mixing.
You'll have to put more effort into placing them. Yes, routing will be more difficult, but who cares today?
Another option to consider (I don't know your application) is to use two green LEDs per one red and blue. The human eye is especially sensitive to green spatial response; this might be irrelevant here, though.

This would be a reason to use fewer green LEDs, not more! (Assuming they each had equivalent energy output.)

We use more green subpixels on camera sensors because our eyes are more sensitive to green and thus perceive the green channel as having more spatial detail, so doubling its sensitive area reduces apparent noise overall.

[ConKbot]

I'm no optics expert, but I suspect a lens array that is a bunch of cylindrical lenses to blend them aggressively on one axis before the diffuser could help. (Linear Fresnell looks like a good starting keyword) and a one-off with a lens then diffuser can probably be done for cheaper than re-doing the array. A better distribution of LEDs would be better in terms of overall optimizing, but if this is a one-off or prototype, then the linear lens may be just the ticket.

[Benta]

This would be a reason to use fewer green LEDs, not more! (Assuming they each had equivalent energy output.)

Quite the opposite. Using more green LEDs would dilute the green points (at reduced power of course). We're talking about avoiding the greens being perceived as light points. This is about light diffusion, meaning avoiding spatial recognition.
An arrangement of RGB LEDS in a 2/3/1 pseuso-random pattern through a diffusor could present itself as a uniform area of light (depending on the quality of the diffusor)

[tooki]

This would be a reason to use fewer green LEDs, not more! (Assuming they each had equivalent energy output.)

Quite the opposite. Using more green LEDs would dilute the green points (at reduced power of course). We're talking about avoiding the greens being perceived as light points. This is about light diffusion, meaning avoiding spatial recognition.
An arrangement of RGB LEDS in a 2/3/1 pseuso-random pattern through a diffusor could present itself as a uniform area of light (depending on the quality of the diffusor)

Specifying “at reduced output” is a massive difference.

[Benta]

Specifying “at reduced output” is a massive difference.

I presume that people doing this are able to run the LEDs at appropriate light output, not at full blast. Stop nitpicking.

[tooki]

It’s not nitpicking, you left out a major piece of information. Your fail, not mine.

[Siwastaja]

Just reduce the distance between different color LEDs to bare minimum. I mean, generate the footprint courtyards according to your manufacturer / your own assembly skills and place as close as possible. I'd say, from solder mask opening to another solder mask opening 0.5mm. This is going to be so much easier and cheaper than negotiating for custom RGB emitters, and if you pick smallest possible SMD package, almost as good.

I don't understand how this affects routing. You do use at least 2-layer PCB, yes?

This change alone makes color mixing so much easier you probably get along with your salvaged LCD diffuser sheet or whatever.

But still, you have another problem: poor color mixing becomes poor intensity flatness. If this matters to you, you need to look at better diffusers anyway; or, to make your life easier, just moar LEDs. If you can have this machine-assembled, then it's a no-brainer. If you do it by hand, maybe even then just bite the bullet. I had a friend who designed a 32x32 LED matrix (1024 LEDs) and hand-soldered those SMD LEDs (no reflow; pad at a time). Learn home re-flow if you don't know how to do it already. You pick and place 500 LEDs with tweezers in maybe two hours.

Currently you have just 72 LEDs! Assuming you need a square, make that something like 12*12*3 = 432 and now we are getting somewhere.

Increasing distance to the diffuser obviously helps, too. Then you can have another gap, and a second diffuser.

[amaschas]

Just reduce the distance between different color LEDs to bare minimum. I mean, generate the footprint courtyards according to your manufacturer / your own assembly skills and place as close as possible. I'd say, from solder mask opening to another solder mask opening 0.5mm. This is going to be so much easier and cheaper than negotiating for custom RGB emitters, and if you pick smallest possible SMD package, almost as good.

I don't understand how this affects routing. You do use at least 2-layer PCB, yes?

This change alone makes color mixing so much easier you probably get along with your salvaged LCD diffuser sheet or whatever.

But still, you have another problem: poor color mixing becomes poor intensity flatness. If this matters to you, you need to look at better diffusers anyway; or, to make your life easier, just moar LEDs. If you can have this machine-assembled, then it's a no-brainer. If you do it by hand, maybe even then just bite the bullet. I had a friend who designed a 32x32 LED matrix (1024 LEDs) and hand-soldered those SMD LEDs (no reflow; pad at a time). Learn home re-flow if you don't know how to do it already. You pick and place 500 LEDs with tweezers in maybe two hours.

Currently you have just 72 LEDs! Assuming you need a square, make that something like 12*12*3 = 432 and now we are getting somewhere.

Increasing distance to the diffuser obviously helps, too. Then you can have another gap, and a second diffuser.

I think in my use case adding more LEDs (which I would love to do) becomes a power issue. My current setups is using rows of 8 LEDs, and each driver on my board drives 3 rows of 8. Each row draws 120mA, or 360mA per driver. The highest voltage row has a voltage drop of ~25V, so the board in total requires a 28V source, and draws between 1-2A depending on peripherals like my microcontroller. At this stage I can get away with a 50W power supply, but if I start adding LEDs the power requirements are going to increase pretty drastically. I could try to source lower-power LEDs, though the ones I've chosen have a pretty outstanding light output/power ratio which I'm not confident I can duplicate with alternate emitters, particularly at the very specific wavelengths I need.

I could also try to increase the number of parallel strings, but driving LEDs in parallel is already a tricky business. I'm using an aluminum PCB as a heat sink to try to avoid the issues resulting from parallel LEDs, but I don't want to push it.

[Siwastaja]

Of course if the light output is currently enough, adding more LEDs means you drive each at lower power, for the same total output. The only consequence to power consumption is a slight decrease due to better efficiency when running LEDs at lower current.

But yes, this either requires going for higher voltage (and less current), or driving many more parallel chains. And by that I don't mean paralleling LEDs directly, no, each chain needs its own driver - some kind of current limitation device (simplest being just a series resistor). It means some replicated electronics because you can't just copy-paste LEDs, you need to also copy-paste drivers (at very least resistors). But that's the price to pay from the greatly simplified optical design. Given that resistors cost < cent each, and full-fledged programmable constant current drivers less than 1$ per chain (I'd guess, haven't checked lately), I would not hesitate to go that way. That being said, I'm better in electronics than in optical design which is why I would tackle the problem from that perspective.

It truly helps if you have a complete controllable RGB chain layout (supply voltage and control signals in - light out) which you can just copy-paste in PCB EDA, so that you don't have to place 500 LEDs and then start thinking how to route them all to 20 control ICs somewhere else.