This circuit was posted to the files section of the Yahoo group Joulethief (you may need to join to see this). I´m going to try to attach the pic but I couldn´t do it earlier, You might also be able to see the picture here.
I built four versions of this circuit and made some measurements of each. The first version (shown in the picture) used a 2N3904 for the output transistor and a 180 uH, 2.5 ohm DC resistance RF choke for the coil. The second version I removed the RF choke and used a toroid with 10 turns of 24 AWG giving 171 uH and a DC resistance of less than 1/10 ohm. The third and fourth versions were the same as the above two, but with the 2N3904 removed and replaced with a 2N4401.
Ver. — Xstr —– Coil — Isup —- Iled —– Freq
1 —2N3904 – Chok – 30mA – 9.2 mA – 71kHz
2 —2N3904 – Tor — 20mA – 6.3 mA – 61kHz
3 —2N4401 – Chok – 50mA – 15.7 mA – 38kHz
4 —2N4401 – Tor — 40mA – 15 mA — 48kHz
I noticed that with the 2N4401, when I changed from the lossy 2.5ohm RF choke to the toroid, the supply current went down but the LED current remained nearly the same. But with the 2N3904, the LED current dropped a lot when the toroid was used. The only reason I can think of for this odd behavior is that the 2N3904 is mostly responsible for the poor results with the toroid – the loss doesn´t happen with the 2N4401.
With the 2N4401, the circuit has decent performance, putting out 15 milliamps to the LED. This could be increased a bit by reducing the 470 ohm resistor to 220 ohms, and would give a few more milliamps LED current. I would not consider using the 2N3904 due to its poor performance.
I have not yet tried changing the values of the capacitor to see if it will improve performance. Most similar circuits I´ve seen use a smaller value, around 220 pF.
Back to experimenting…