However, I never ran anything larger than an RGB diode with a Joule Thief (after adding a cap and a diode). How do you provision for the higher currents, do you use a power transistor and thicker coil wire? The one I built was without doing any accurate measurements. Just a simple coil wound from "some" wire (horrible multi-core wire from something that attracts a magnet with copper coating in fact), some resistor, and some NPN. It still worked well though for a small LED or two.
That works to simply get an LED to work from a low voltage battery, but if working on a better version in the future, some measurements would be in order.
First of all, this is your prototype, the blocking oscillator:
As often happens in the world, the oft-repeated instructions (Cbb = 0, Rbias + Rb = 1k) are wrong. Not so wrong that they don't work at all, but very far from optimal. Typical efficiency is 30%, it's practically a class-A (linear) oscillator. Abysmal!
Best performance (sharpest switching) is had when Rb = 0. Rbias supplies base bias, which is about hFE times less than the average input current. Cbb * Rbias is the base bias time constant: it should be near the inductor time constant, so that base bias runs out just as it's done with one switching cycle. If it's too large, the circuit runs for several cycles, in a burst -- this is called squegging. If it's too small, the pulse duration is too short, and you won't get much power output (or a high operating frequency, and more switching losses).
This circuit is ancient: it has powered TVs since the invention of television! One advantage of the vacuum tube: the voltage ratings are massive, so they didn't have to worry about voltage, even at very high duty cycles (typical TV sweep signals are 80-90% duty cycle, so the peak voltages are quite high indeed).
BJTs don't have this advantage. Consider when the transistor turns off: the collector voltage spikes up, and so the base voltage spikes down. If base voltage goes below about -5V, it begins to avalanche (i.e., the B-E junction is a zener diode too). Which pushes charge back into Cbb, which means the transistor is getting more bias than just Rbias supplies. The result? Too much output voltage, and the transistor latches on, burning itself like a runaway diesel engine!
So this sets the maximum transformer ratio. If you need a 4V output from a 1.5V supply, that's a maximum turn-off collector "spike" of 2.5V. A 1:1 transformer is okay here. When the transistor turns on, it only needs a fraction of a volt (it turns on at about 0.6V, but that's already in Cbb when it begins to flip -- the extra drive from the transformer only kicks it on harder). So you might use an even lower ratio, because the base doesn't need much drive to force it on. When the transistor turns on, 1.5V drops across the transformer; you only need maybe 0.5 to 0.75V (or a 2:1 ratio).
Then, Rbias sets power level. Rbias needs to supply base current, that was drawn out of Cbb when the transistor was on. And, note that the transistor sucks extra current out, because it drives current from Cbb into its own base, because of the transformer action. For this reason, the hFE we need to know, is not the linear-range value on the datasheet, but a lower value.
For example, my 1W light has two ranges: Rbias = 1k and 100 ohm. (They're really not that different, visually, and probably 4.7k or 10k would be a better "low" setting. But the power is indeed about 1/10th.) The transistor is PBSS303NX, a high hFE, low-Vce(sat) part that works very nicely indeed here. I forget what the inductor is; even 1uH is a lot in this application, because the voltage is so small, and the frequency can be pretty high (this runs at 500kHz, on the 100 ohm setting).
As for the 10W light, since the power level and supply voltage are higher, I created a semi-discrete SMPS controller for that, discussed
here. Its efficiency is still not very impressive, actually, probably because of switching and diode loss.
Tim