A Switcher! Boy I love switchers!

Obviously not happy with the blindcurrent arrangement I decided to tackle the LED problem differently. Inspired by Supertex, who makes really neat LED string converters that operate from line voltage, I decided to make one myself. Only design criterion: I wanted a string of not just 12 like Supertex's HV9921 but 37 LEDs, which would nicely fill out my lamp holder. Also this string would be much easier to build as I needed to solder only once for each LED, and not make a ridiculous affair with the serial/parallel matrix in Leeslamp Mark 1. But how to drive these LEDs? Making a string would require about 140 V, so that would not be an issue when running off 230 VAC. First I drew some really weird and exotic schematics which I am now so embarrassed about that I won't show them here. They were just way too complicated, comprising of video transistors, optocouplers and what all.

No, the way to go turned out to be the good old blocking oscillator. The transformer was quickly found in a dud laptop backlight converter. The secondary coil measured 300 mH, the primary total I have no idea as my LCR tester wouldn't give a good reading. The switching FET that I found in my junk box happened to be a Toshiba 2SK3067, an insulated 600 V device intended for off-line power supplies. The concept of driving the LEDs was first tested using Linear Technology's SwitcherCAD iii, a free development and simulation tool that I find very useful. I concocted up some new devices and started to simulate the circuit.

A very important design requirement is that the brittle LEDs stay happy. So whatever happens the current through the string and the reverse voltage across each LED must never exceed the maximum values. Even a slight excursion across the limit will result in immediate destruction. So first the current. When the FET switches off, the current through its drain coil wants to keep on going because of the stored magnetic energy in the ferrite core. The drain voltage rises until the diode string starts to conduct. At this point the current is transferred into the LEDs, and is exactly the same as the drain current at switchoff. This current is set with the source resistor and the gate zener diode. You see, when the supply voltage is applied the gate is biased through the transformer and will switch on the transistor. When the gate voltage is 6 V the source voltage is the gate threshhold voltage lower, in this case around 3 V. This sets the source (and drain) current to 3/180 is 0.0017 A or 17 mA. This is also the peak LED current.

During conduction the LEDs are reverse biased. This is not a desirable situation, so they need to be isolated from the 300 V or so that exists across the drain coil. This is achieved using one of those snubber diodes (from what power supply I had liberated the thing I have no idea anymore!). Also a small capacitor is added across the LED string which serves to balance out the much smaller diode capactitance. This arrangement perfectly effective. It turned out that touching this junction with my scope probe (10 meg, 15 pF) immediately destroyed my test LED that I had in series with some power zeners to simulate the string. This also proved the need for this measure. Phew! So now both criteria seem satisfied.

When the Tosh FET starts to switch off the gate is driven negative through the feedback winding, speeding up the process. The coil is now delivering its energy into the LEDs. This energy depletes after some time. This can be beautifully seen both in the simulation and in the real world. Now the gate gets biased positive again due to the collapsing drain voltage and the transistor switches on again, starting the process anew. All this in just 60 µs. Simple but effective! The LED current is also beautifully linear. The little spike is caused by the FET switching on again. And the 1.2k resistor? I just had it lying around, and wanted to have some extra protection as the bridge is rated for only 400 V. The circuit only draws a few milliamps from the mains power so it does next to nothing.

After I built my leeslamp I did some more tinkering around. I made another haystack kludge, this time using a IRFBC20. The reverse diode was a BYV26E that I liberated from the same junker digibox I had the IRF from. The transformer had been taken apart as I had removed the primary windings. I restored this using some 0.2 mm CuL wire. No idea how much wire I put in there, it seemed about right this way. As the threshhold voltage was a bit higher I reduced the source resistor to 100 ohms. Increasing the zener voltage turned out not to be such a good idea as the circuit started to become unstable, and the FET was not turned off properly. The resulting peak current now was 20 mA. The test setup squealed like crazy at 11 kHz, in the process proving that my hearing hadn't detoriorated that much from listening to loud house music. It turned out that the "OFF" voltage was only 100 V, and it just took too long for the current through the LED (and zeners) to bleed away. Also taking the two "E" cores apart to restore the primary (feedback) winding didn't help much. Upping the voltage to the intended 140 V sort of solved that. If need be I could also remove some drain windings. And I also found another failure mode: touching my probe to the source immediately destroyed my LED due to the resulting current spike, tiny as it was. Once I left it alone the replacement was happy once more!
















© Zappy TV 2005 (updated 24 april 2005)