Stomping in Clown Shoes

Here is a design for an overhead light fixture (see the Creative Commons license notice below).

I’m not really sure I’m going to save any money with this.  I’m spending a lot for a very nice lamp, but economically, I’m not sure the energy and light bulb savings will be worth it.  But then again, that’s not the only reason to do anything.

This lamp will replace the aging standard fixture in the ceiling of my study in a house built in 1965.  This existing fixture is made for two Edison base light bulbs, and switched at the wall near the entry door.  Bo-rrring.

The idea is to stop the 120V wiring at the switch.  There I’ll put in a small DC power supply and a circuit inside the wall box with a special wall plate for the switches mounted on the PC board.  A three wire cable (ground, 24 volts DC, and a 0 – 5 volt square wave) will connect to the overhead fixture.  The lamp will be made of 6 pentagonal printed circuit boards connected together on the edges like the lower half of a dodecahedron (12 sided solid).  Each board will have 5 Cree XLamp MX-6 high-brightness LEDs, 6500 Kelvin (cool white color), for a total of 30 LEDs.

I selected the Cree XLamp MX-6′s for a couple of reasons… they have similar or better light output at 300mA as other LEDs that use a lot more amperage.   They are touted by Cree to have a nice long life with good reliability, retaining 70% of their luminosity after high-temperature age testing, which Cree seems to be taking very seriously even if there aren’t any international standards for it yet.  And… they are generally a dollar cheaper per LED.  That’s obviously not my biggest concern, but it’s a factor.  In reality, this board should support almost any LED with a minor change to the SMD pads.  I tried to limit costs where possible, but I didn’t skimp on the quality.

Here is the schematic for the dimmer board.  It features a triangle wave generator for the PWM signal, and uses a “Dallastat” digital potentiometer for 64-step up/down dimming.  There’s an adjustable hysteresis so that “all the way down” is really 0% and “all the way up” is really 100%.  A MAX6816 debouncer chip and a ‘D’ flip-flop are used to create an on/off signal, and these are  ‘and’ed with the PWM and buffered with an op-amp for the lamp on/off control output.

I used Paul Falstad’s circuit simulator to work out the component values for the triangle wave generator.  Here is the data to ‘import’ (in extremely small text so as not to bore everyone else) if you’re interested:

The dimmer printed circuit board is 1.25 x 2.5 inches, small enough to fit inside the single-gang electrical wall switch box.  The pushbutton on/off switch and the up/down dimmer rocker switch mount on the board, but the pushbutton switch is a whole centimeter shorter, and so it rides on a short riser board held up with standoffs.  The four square ground pads at the corners will use 2 cm standoffs to mount to a specially cut wall plate.  There’s also an interface for digital control of up/down/on/off through a home automation mechanism.  Here is a gerber-generated drawing of the dimmer board:

Anybody who’s spent a few days in Michigan in February knows I’ll be popping big static sparks off this switch, so there’s plenty of ESD diodes and varistors.

The uncredited protagonist of the lamp printed circuit board is a constant current driver called the CAT4101.  It’s very easy to use, takes the higher voltage needed to drive a string of LEDs, and features a PWM/Enable input.

This sucker is gonna generate some heat.  As many of the parts as I could get are operational up to 125 degrees centigrade… however all of the fun components start to derate over 70 degrees.  The back side of the board has been cleared for room for a TO-3 transistor style heat sink, with mounting holes.  I specifically selected a heat sink with prongs on it that I could bend, as all the heat sinks pointing into the center of a polyhedron might poke each other… and I admit that I didn’t want to actually do the solid modeling or math involved in worrying about that.  The CAT4101 LED driver will need to handle about 5 watts, so it has it’s own tin-plated surface mount ‘D2PAK’ heat sink, which I really hope will be enough.  The lamp boards can be daisy-chained together by two connectors on the back of each, up to the amperage of the power supply.  The PWM signal from the dimmer board is buffered again by a receiver op-amp, and this need only be on the first board of a chain, the others can shunt the op-amp with a solder jumper.  Another solder jumper permits the board to be used ‘always on’ without the dimmer.   There’s a big honking mounting hole in the center of the board, and again, there’s lots of ESD protection.

The 6500K white LEDs generate 100 lumens at 300 milliamps, the suggested and my projected load (Cree MX-6′s are available in warmer colors, but with less light output).  This should be 3000 lumens total, which is pretty much enough for surgery.   But… the digital potentiometer and the response of the LED driver should be near flat-line linear at 400 Hz, so I don’t have to run it full blast all the time.  It will be nice to be able to turn the lights UP when I really need them (like during teach-yourself brain surgery).

February 15th, 2010: I worked all weekend and a holiday, and I’m convinced that any further soliloquizins’ will just screw me up, so I pushed the buttons.  Everything is ordered.  The Gerber files went into Sierra ProtoExpress’s “No Touch” automatic verification system on the first try.  That’s scary.  I have no idea if this is even going to work, but it a be a good circuit board dere bubba.  I doubled up the dimmer board and ordered 2, so I’ll get 4 of them for about $100.  The lamp boards were 6 for $200.  The parts were $300.

February 18th, 2010: Most of the parts arrived, and they look pretty good.  Digi-Key did a terrific job again. The power supply is really pretty, and I can adjust it down to 22 volts.  The 5 Cree MX-6′s on each board are projected to drop 3.3V at 300mA, and the CAT4101 drivers can’t take more than 25V, so between 17 and 24 volts should be okay.  From the data sheets, the 24V ESD devices will work just fine with less than ‘typical’ voltage.  The heat sinks look like they’ll do the job, putting on the surface mount D2PAK heat sinks looks daunting, I believe I may invest in an electric griddle and try my hand at pancake-style reflow soldering.

February 19th, 2010: I have ordered a Reflow Soldering Controller kit from SparkFun Electronics.  This kit will help me use an infrared toaster oven as a reflow oven.  The LEDs and the D2PAK drivers and heat sinks will just never work out right without this, and I’ll be needing it in the future for the much finer components I’ll likely need to use for the wireless home automation.  I’ll be heading over to Target for the toaster oven.  I wonder how the wife will react to me baking semi-toxic fumes.  I guess I better clean the garage while I’m waiting for the Fedex and UPS deliveries.

February 28th, 2010: The dimmer works perfectly.  All the dimensions are good, all the parts fit.  The voltage regulator isn’t even getting warm.  I can adjust the hysteresis potentiometer so that one click up from the bottom starts the PWM square wave.  One click from the top, it’s almost completely on, and then at the top it’s fully on.  The switch risers and the standoffs measure up well, the LED in the pushbutton switch works.  If I hold the dimmer one way or the other, the Dallastat digital potentiometer clicks one notch, pauses, and then starts ticking along.   When power is lost and restored, it always comes up with the lamp off, and the Dallastat remembers it’s last setting.  I didn’t do any breadboarding to test any of this, so I’m pretty impressed with it.
Here is a scope picture of the PWM output, 1 volt / division vertical (center mark is zero), 1 millisecond / division horizontal.  It puts out a very clean 5V square wave at ~330 Hz.  This is 32 clicks (of 63 top to bottom) from the top of the Dallastat, so right around a 50% duty cycle.

Here is the built dimmer circuit attached to the power supply.  The pushbutton is lit (it has it’s own white LED inside), the black thing is the up/down rocker dimmer switch.  The four stand-offs at the corners will connect to the back of the wall switch plate.

The Reflow Toaster controller is not working out very well.  To write any kind of meaningful control program, I had to buy a compiler.  It also won’t work well with the goofy boot loader that’s on the chip, so I have to get a programmer.  This is not what I expected, and very disappointing.

So instead, I’m going to try hand-baking one of the boards.  The Cree lamps are what I’m worried about the most, but the tin heat sink for the driver transistor would be very difficult to solder by hand, and I’d likely destroy the transistor in the process.  Here’s my manual stopwatch profile:

1. cook at 175-200F for 5 minutes (pre-pre-heat)
2. bump to 275F, cook for 2 minutes (pre-heat)
3. bump to 410F.  As soon as it melts, count to 10, turn off heat and crack open the door.
4. Open the door a little wider after a minute.  Cool down shouldn’t happen faster than 2 – 3 minutes, so walk away and have a smoke.

I worry too much.  It all worked out great, and I’ve cooked up one pentagon.  And the result…

It works.  It would work better if I remembered that LEDs are polarized.  I got too excited and got two of them backwards on the first pentagon.  However, I soldered the bypasses on the two and calibrated the amperage to 300mA, so the working three are fully loaded.

It’s using a lot more wattage than I suspected.  The three working LEDs are using 7 watts according to my Kill-A-Watt meter, which measures at the plug.  It’s bright, but it’s also getting hot.  I’ll have to risk it and let it cook for a while and see if it cuts out.  The driver transistor has a thermal shut-off if it gets too warm, and I suspect it will.  I may only be able to use two or three LEDs per pentagon with this design.

On the second pentagon, I discovered a design flaw.  I’ve made the driver pins too close to the ground plane, and pins other than 5V shorted through the solder mask.  I fried the driver transistor.  When investigating this on the first pentagon, I found a solder ball behind the pins, which shorted the LED return to ground putting full amperage to the LEDs, and fried one of the LEDs.  Hard lesson learned.  The first now has two working LEDs (at least I didn’t fry the driver on that one) and the second is toast.  It’ll be difficult to rework them.  I repaired all the boards with a dremel.

I made a third pentagon, and that one is working great.  With 5 LEDs, it’s drawing only 6 watts at full amperage and full on PWM, which is much closer to what I was expecting.  It’s hot, both the back heat sink and the driver surface mount heatsink are hot, but not so hot that I can’t hold them and keep my fingers on them.  This suggests that they are working within acceptable limits, and I can turn down the amperage a little and be pretty comfortable that this circuit will last a long time.  I’ll get a temperature sensor soon and find out for sure exactly how hot they’re getting.  The amperage adjustment trimmer starts at 0.20A fully clockwise, and turned up to 0.25A is plenty bright and allows some comfortable lee way, Cree MX-6 nominal amperage is 0.30A, absolute maximum is 0.35A.

Here’s the third pentagon working

Here’s a shot of my desk

March 7th, 2010: I’ve been doing some thermal testing.  The programmer for the PIC chip came, and the toaster controller works.  SparkFun also sent the wrong PIC chips I ordered, big surprise… they got something else wrong.  I’m not sure I want to pursue it.  But the controller has a working thermocouple on it, and taped to the bottom of one LED, I ran the light at full blast under a porcelain food bowl.  I ran it long enough to stabilize the temperature, and it does get hot.  At full amperage, 300mA, it goes up to 78C.  At 250mA, it goes to 71C.  This is just over the limit, at 70C most of the active and critical parts start to derate rapidly.  With the bowl removed, it stabilizes at around 57C.  This is pretty good.  I can’t truly seal the thing inside a glass fixture, but given any amount of air flow, it should be just fine at nominal 300mA amperage.

Cree does “High Temperature Output Life” testing on their stock, they cook LEDs at 85C for 1008 hours and ensure that light output is not diminished by more than 15%.  Running at less than 60C, I’m guessing that I’ll have plenty of light from these for a long time.  My goal is about 250K hours, or 25 years.

I’ve cooked up the remaining boards.  I’ve also successfully repaired my hot air pencil, and removed the parts I messed up from the first two boards.  Replacements are on the way, I just hope I can successfully reflow them again.  A tip o’ the hat to my friend John who pointed me to Silver Circuits ( http://custompcb.com/ ) who I’ll be trying next time.

It’s continuing to use more wattage than I expected.  The third board with 5 LEDs at full dwell reports 12 watts on my Kill-A-Watt.  The first pentagon was doing 7 watts with three working LEDs, so this makes sense.  I’ve ordered the next size power supply.   Sometime I can make another lamp with only 3 or 4 pentagons and use the other one. I measured this one at 6 watts before? It’s possible that this one is goofy, I’ll have to finish the last 3 pentagons and see what they do.

Wednesday, March 10th, 2010: I am a knucklehead.  I didn’t need the new power supply (which is here and very pretty) it’s only using 7-8 watts per pentagon.  I had the Kill-A-Watt set to volt-amps (VA), instead of watts.  It’s pulling 12-13 VA, and 7-8 watts.  I don’t even know what a volt-amp is.  Well, I’ve got a big honking power supply for the next project and this sucker is working per specs I was expecting, woo hoo!  Replacement parts have arrived, and the Fluke 62 Mini infrared temperature sensor gun is here, it confirms the temperatures I was seeing with the thermocouple.  Nothing in the way of final assembly now, just need the time and energy.  The weather has really warmed up, so I may even get a chance to ‘splore the attic a bit and see what’s going to have to happen there for the installation.

Monday, March 15th, 2010: Driver transistors are dropping like flies.  I’m not completely sure what I’m doing to them, but when everything else looks like it’s working perfectly but there’s just no current, I’m replacing the transistor.  I’m pretty sure it has to do with me testing the 5V supplies without the filter capacitor in place.  I’m actually getting pretty good at using the hot air rework to remove the transistor and the surface-mount heat sink, I just wish they weren’t electrically so fragile.  I’ll be adding more protection in the final version of this circuit.

I’ve wired the reflow toaster with the controller, it works.  It will need some calibration, it heats up 1 degree Celcius per second, but doesn’t really cool off very fast.  I don’t have a serial port on any of my working portables, so I’m waiting on a USB-to-serial cable so I can capture and graph the temperatures.

More spare parts, more hurry up and wait.  This is what I get for going straight to prototype without component level testing.  But hey, the dimmer worked right the first time, so it could be worse.

Thursday, March 18th, 2010: More transistors are here, and USB to serial port is here.  I plugged it into the old laptop and it recognized and started the serial port driver instantly, I love Kubuntu Karmic Koala.

I’ve also purchased an Atmel EVK1101 AT32UC30256 evaluation / demonstration circuit board, and an Atmel “Dragon” programmer/debugger.  These are for experimentation with using the One-Net protocol for wireless home automation… but this is for another posting.

So once again… barring any more stupid errors on my part, I may be able to do the final assembly this weekend.  Big maybe, baby, I’ve got the next project all lined up and ready, but there’s one other problem.  I’m thinking I can resist it long enough to get some work done.

Sunday, March 21st, 2010: It works pretty well.  I abandoned the power connectors, they were just too much trouble.  Copper foil tape also didn’t work well connecting the pentagons, I used a piece of wire in between the boards and just leaded the hell out of them.  It’s messy, but it’s pretty solid.

The lamp, all 6 pentagons assembled

This is the lamp with the LEDs at full brightness, and using a camera flash.

Here’s the whole setup: power supply, dimmer, lamp and tape measure for scale

Here’s a shot in the box with no camera flash and the lamp turned down to one click above off… 1/64th, or ~ 1.5% of maximum brightness.

Saturday, April 3rd, 2010: I had to turn down the amperage.  At nominal 0.3 amps and full brightness (100% PWM dwell), it gets too hot.  A few places were creeping above 75 degrees C, and I really have to draw the line there.  Turned down to 0.25A (individually controllable per pentagon), I haven’t seen any scanned spot go over 65 degrees C.  Turning down the amperage will cost 15% of light output, but that’s why it was over engineered in the first place.

I’ve been thinking about another design using a snap-together linear railed model rather than this polyhedral design, but again, that will need it’s own post someday, and if I ever build it.

It’s time for me to get practical on this, I have to buy a hard box to put the power supply in, and mount the sucker in the ceiling.  The weather in improving rapidly, I think I may have to get on this for this Easter weekend.

What I’ve missed so far:

• explore option to bypass the 5V regulator on daisy-chained successor lamp boards,  100mA should be enough for 6 boards or more
• move the pushbutton switch ESD components to the riser (maybe not)
• add lots more ground plane vias around the LEDs
• reverse the dimmer rocker switch!  The pushbutton switch should be above the rocker when mounted, and then the rocker should have ‘up’ make the lights brighter.  The switch is currently (electrically) upside down I must have gotten this backwards twice, because it’s working correctly.

Copyright (c) 2010, Roger Cody Venable (MDVE.NET).  Some rights reserved.

This design is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License .

This entry was posted on Thursday, February 18th, 2010 at 19:13:24 and is filed under Linux Development, Mad Scientist - Boo. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.