Ed Nisley's Blog: Shop notes, electronics, firmware, machinery, 3D printing, laser cuttery, and curiosities. Contents: 100% human thinking, 0% AI slop.
Tag: Improvements
Making the world a better place, one piece at a time
MP1584 buck regulator – current feedback – red LED
I started with the same 1.65 Ω sense resistor and got the same 484 mA current, with the LED forward drop at a surprisingly high 3.3 V = 1.6 W. Ouch.
Adding a 1 Ω series resistor to get 2.65 Ω lowered the current to 300 mA with a forward drop of 2.45 V = 740 mW.
Running the numbers suggested a 2.3 Ω sense resistor made from a pair of parallel 4.7 Ω resistors, which produced 346 mA and an LED drop of 2.66 V = 920 mW. The resistor dissipates 280 mW.
The bench supply provides 6.3 V @ 200 mA = 1.26 W, so the overall efficiency is 94% and the LED burns 73% of the input.
The PCB is the generic MP1584 buck regulator, as seen before in its normal voltage feedback mode, rewired to get feedback based on the LED current, so that it adjusts the output voltage to maintain a constant LED current, regardless of LED forward drop variations.
Pin 4 normally sees the output voltage divided down to the 0.8 V error comparator reference voltage:
MP1584 – buck regulator – voltage feedback
Yes, the MP1584 is “not recommended for new designs”, which surely accounts for the myriad cheap regulators built around it. Somebody picked up a great deal on a vast pile of obsolete ICs and is passing the savings along to us; there are exactly zero hits for MP2338 buck regulators.
Putting the ballast resistor on the low side of the LED turns it into a current sensor:
MP1584 – buck regulator – LED current feedback
Pick R to drop 0.8 V at the desired LED current and It Just Works™.
The two 3.3 Ω resistors in the top photo produce a 1.65 Ω sense resistor to set the LED current at:
485 mA = 800 mV / 1.65 Ω
It actually works out to a bit higher than that, because I stuck a 100 Ω resistor in series with the feedback input. The PCB still has the 8.2 kΩ resistor from the original voltage divider, so the error amp sees only 99% of the sense voltage, but it’s close enough.
With 6.3 V and 0.28 A = 1.76 W from the bench supply over on the left, the regulator puts 490 mA through the LED. The LED drops 2.54 V = 1.24 W and the resistor drops 0.809 V (that 1% thing) = 0.4 W for a total of 3.35 V and 1.64 W. The regulator is 93% efficient, although the resistor burns a quarter of the energy.
One could use a Hall effect current sensor and an op amp circuit to deliver the proper feedback voltage without resistive loss, but I think burning half a watt is Good Enough for the purpose.
One could add parallel resistors with MOSFET switches to set the LED current. An unswitched resistor would set the lowest current, with switched parallel resistors lowering the resistance, raising the current, and brightening the LED.
The PCB leaves the Enable input floating with an internal pullup. Grounding the pin shuts off the LED as you’d expect, so I can blink the LED without any further hassle.
One could imagine simultaneously blinking and brightening the LED as needed.
Based on simpleminded testing, a 1 W amber LED drops about 2.5 V at 430 mA. A 1 Ω ballast resistor drops another half volt and burns a quarter of a watt, sufficient to cover some LED forward drop variation.
The trimpot is entirely too twitchy, so I replaced it with an SMD resistor:
Amber 1W LED – fixed voltage SMD
The trimpot read 26.5 kΩ after I extracted it, but I surely nudged it a smidgen in the process.
For the record (first column is SMD topmark, second is measured resistance):
3012 = 29.9 kΩ (!!) → 3.67 V into a 100 Ω resistor
2492 = 24.9 kΩ → 3.19 V : 2.63 V @ 550 mA = 1.45 W
2362 = 22.6 kΩ → 2.97 V : 2.52 V @ 450 mA = 1.13 W
223 = 22.0 kΩ → 2.91 V : 2.484 V @ 425 mA = 1.06 W
With 6.3 V @ 210 mA = 1.3 W from the bench regulator, the resistor now burns 180 mW at 425 mA and the LED burns 82% of the input power.
Letting it cook overnight settled out with the LED at 2.47 V and 440 mA = 1.09 W, with 6.3 V at 220 mA = 1.4 W from the bench supply. The LED dissipates 78% of the input power and the resistor burns 190 mW = 14%, so the regulator uses 120 mW = 8%.
I can come close to the final output voltage by plugging the new resistor value and the 8.2 kΩ resistor (on the PCB) into the MP1584 datasheet equations, but figuring the resistor to get a specific output voltage seems largely empirical.
Not counting the heatsink, you’re looking at less than three bucks of parts; living in the future is great.
Fitting the lens over the LED produces a shatteringly bright beam, at least in the Basement Laboratory:
Amber 1W LED – lens test
The lens has a conical cavity surrounding the LED lens to capture the light and redirect it to the beam forming reflector. It’s done with total internal reflection, there are no coatings, and it’s a wonder to behold: one-shot molded aspheric optics at work.
Not seating the lens firmly against the LED produces a dark spot in the middle of the beam. I soldered the leads directly to the LED and cut out the sides of the black lens holder, as soldering them to the convenient side pads would prevent the lens from seating properly.
The headlight output is good for 6-ish V and 3 W = 500-ish mA, so burning half the power in a simple dropping resistor or linear current regulator is a Bad Idea™. You can get constant current LED drivers, but apparently not with 6 V input and 1 W output, so stepping the voltage down makes more sense. You’d want at least a little ballast resistor in there to soak up small forward drop changes with temperature variations.
The regulator can handle up to 28 V input and the tiny trimpot must cover nearly that range of output voltages, so the 2.5 V output jams it near the minimum end of its rotation (which is, of course, backwards). This calls for a fixed resistor to eliminate the effects of vibration on a trimpot at 10% of its range.
The cut is just in front of the PCB and went slowly to avoid clobbering the SMD resistors very near the edge.
The cataract turned out to be crud adhered to the LED lens:
Side Marker E – LED cataract
Brutal surgery removed the LED and installed a replacement:
Side Marker E – replacement LED
The PCB had two 150 Ω SMD resistors for use with 12-ish V automotive batteries. While I had the hood up, I removed one and shorted across its pads to make the LED work with the 6 V switched headlight supply from the Bafang motor.
In round numbers, 6 V minus 2.2 V forward drop divided by 150 Ω is about 25 mA. The original LED ran at 35-ish mA, but it’s close enough.
Glue the lens back in place:
Side Marker E – clamping case
The bubbly stuff is solid epoxy from the original assembly, which is why removing the PCB is not an option.
The new LED is no more off-center than any of the others:
Side Marker E – new LED – front
It does, however, sit much closer to the lens, due to the ring of plastic I cut away to get inside. As a result, the beam is mostly a single centered lobe with only hints of the five side lobes; there isn’t much waste light from the side of the LED into those facets.
The Bafang battery charger uses an AC line cord “binocular” connector with what must be the weakest spring contacts ever made, which finally annoyed me enough to fix:
Bafang charger – AC line cord anchor
Also, the case now sports four thick fuzzy felt feet to keep it from sliding around quite so easily.
Another customer-does-the-last-ten-percent product …
The original BBS02 reaction spacer for Gee’s Terry Symmetry didn’t work quite the way I expected:
Bafang BBS02 – reaction block displacement
The motor evidently vibrates enough to propel the block forward, shearing the double-sticky foam tape which was never intended to resist force in that plane. I thought the block was located at the point where the motor casing was tangent to the frame tube, so as to equalize the forces in both directions, but … nope.
A revised design based on measurements informed by new knowledge:
Terry – Bafang motor spacer – improved – solid model
The upper curve is now symmetric and the whole block mounts more rearward under the bottom bracket lug, where some tedious work with a machinists square located the real tangent point:
Bafang BBS02 – reaction block improvement
The motor sure doesn’t look like it’s tangent, but a dry fit showed all the curves laid against the case and tubes.
The brazing fillet means the step fitting the downtube can’t sit snug against the edge of the lug, but most of the reaction force should go through the section into the lug, near the center of the block.
A crude marker will keep track of any motion:
Bafang BBS02 – reaction block marker
I think the symmetric curve against the motor has enough projection to keep the block from wandering off, even if I haven’t gotten the location exactly right.
The Smell of Molten Projects in the Morning 