I have a high-end Luxo desk lamp with the magnifying glass removed to shoot macro photographs. Matthias Wandel produced a YouTube video demonstrating how LED lights produce more accurate colors than fluorescents, due to the emitted light spectrum. I decided to replace the circular T9 fluorescent bulb in the Luxo lamp with a warm white LED circline lamp (obsoleted by a frosted LED circline lamp).
T9 fluorescent bulb and ballast replaced by LED
Upon powering it on, a large portion of the bulb did not light.
Defective bulb missing LED causes segment to not light
It was a missing LED (nice quality control there). The bulb consists of 6 independent groups of 30 LEDs each. Missing an LED in one of the groups prevents current from flowing in that group, while the other groups still operate fine in parallel. Fortunately, I had a similar LED in my parts box, and soldering in a replacement LED corrected the problem.
Missing LED replaced in LED T9 bulb
However, upon trying to use the LED lamp for photography, the camera display was unusable due to horizontal banding / stripes / flickering.
Horizontal banding on camera LCD
The root cause was the LED driver circuitry in the lamp. AC power goes up and down many times a second, which is why it is called alternating current (AC) rather than direct current (DC). LEDs need steady current to maintain steady brightness. Instead, the brightness of these LEDs was going up and down with the household AC. Thus, bands of bright and dark lines appear on the camera as it reads the camera sensor line by line as the LEDs are changing brightness.
There are off-the-shelf LED driver circuits / power supplies that provide steady output. Although they are relatively bulky and are as expensive as the bulb, they would definitely provide steady lighting.
Most commercial LED drivers seem to supply around 24 volts or less, which turns out is too low for this LED bulb. Recall that I said this bulb has 6 groups of 30 LEDs? If each white LED needs about 3 volts, then 30 in a row are going to need 90 volts, not 24 volts.
With power off, I inspected the LED bulb’s circuitry to determine if there was anything I could do to steady the power. This bulb uses a capacitive power supply with a 3 x 1.5 µF input capacitor to produce a peak of 203 mA based on the formula (120 VAC / (1/(2 * PI() * 60 Hz * 4.5 µF * (1/1000000))). Spread across 6 groups, each LED strand gets 34 mA peak. However, due to the sinusoidal AC wave, the actual current is lower most of the time.
If this were the entire circuit, the LEDs would flicker noticeably to the human eye. So, the manufacturer placed 50 µF of capacitance after the bridge rectifier to act as a small power storage unit. The capacitors store energy during the peaks and release energy during the drops. Thus, the light output from the LEDs is partially smoothed.
I noticed a number of spare holes throughout the LED bulb PCB. These holes connected to the LED power lines, almost as though the manufacturer anticipated the need to add additional capacitors. So, I figured, why not add extra smoothing capacitors myself?
Before soldering anything permanent to the LED bulb, I wanted to experiment with different capacitance to see and measure the effect. Anytime you hook an expensive piece of test equipment to household power, you want to be very cautious for your own safety and the safety of the equipment. Worse still, the transformer-less design in this bulb means that some paths lead directly to the household outlet. I used a continuity tester on the multimeter to determine the common ground between the oscilloscope and the bulb to make sure I connected the oscilloscope ground lead to the bulb ground. Had I accidentally connected the oscilloscope ground lead to the other AC wire (hot), a dead short would likely have killed my oscilloscope.
In the traces below, the yellow line is the voltage being supplied to the start of the LED chain. The blue line is the voltage being consumed by a single LED at the end of the chain. Because the yellow line is typically 80ish volts, and the LED is typically 3 volts, recognize they are independently scaled to fit on the same screen. That is, don’t compare the vertical size of the different colored lines to each other, simply compare the shapes.
With the bulb as delivered, the voltage to the LED chain goes from 78.08 volts to 87.36 volts. A difference of about 9 volts. During that time, the LED changes voltage by about 2 volts. Practically speaking, that means the LED is virtually off about a quarter of the time.
LED lamp 50uF
Where have you seen this waveform before? Oh, that’s right, on the camera display:
Waveform on camera display
Doubling the capacitance reduces the peak-to-peak voltage to 6.4 volts, but also decreases the peak voltage to 85.64 V. The LED is turned on a little bit longer.
LED lamp 97uF
Quadrupling the capacitance (now 207 µF) decreases the peak-to-peak to only 3.64 volts, from 80.48 to 84.12. The effect of the LED is significant. In fact, the scale is now down to 50 mV, so the LED is seeing a relatively more steady voltage the entire time. The average LED voltage is 80.48 V / 30 LEDs to 84.12 V / 30 LEDs = 2.68 V to 2.804 V.
LED lamp 207uF
At almost ten times the capacitance (now 458 µF), the output voltage remains within about 2 volts the entire time. (For some reason, I didn’t capture the LED voltage on this trace.)
LED lamp 458uF
The supplemental capacitors need to be rated for US household voltage and reasonably high temperature, as they are being installed in an AC LED lamp. The Rubycon 68 µF 160V 105°C TXW radial aluminum electrolytic capacitors (160TXW68MEFC8X30) are rated for 370 mA ripple current and will fit in the LED bulb shell. Given that six capacitors are being used, and they $1.51 each, I can understand why the manufacturer would choose not to include them.
Rubycon 160V 68uF capacitor color
The capacitors were soldered into the spare holes and laid flat on the circuit side of the lamp.
Capacitors added to LED bulb driver circuit
This generally fixed the flickering on the camera display. Occasionally I’ll notice something.
I have never worked with a capacitive power supply before, so I didn’t know if the increased bulk capacitance would strain the existing components. Also, I was a bit nervous that increasing LED on-time might lead to overheating.
The Flir infrared camera determined that the LED side of the lamp stays relatively cool.
Even heating on LED side
The component side of the LED lamp has a hot spot.
Hot spot on component side
That hot spot is not significantly higher than the surrounding board. It is likely the MB6S bridge rectifier.
MB6S bridge rectifier
The worst case heat dissipation for BR1 should be 1 volt * 0.203 amps = 0.203 watts. That’s much lower than the 1.4 W rating of the package.
Without a doubt, white LEDs provide much better color matching than fluorescent bulbs. I spend much less time in color correction image processing. Although I am frustrated by the defective LED bulb and the low cost driver circuit, it is still less expensive than something similar that I could make myself. I’ll call that a “win”.