2013-01-14 Questions About The Joule Thief, Particularly The Coil

I received an email from Paul, with some good questions about the Joule Thief, particularly about the coil. He’s given me permission to use them in this blog. Here is his intro and first question:

Imagine 1st an average JT circuit just as frequently made. 1 transistor, 1 white LED, 1 resistor, 1 battery at say a steady 1.3v, 1 toroid. Assuming common components used where not specified. Standard circuit. E.g.

http://farm3.static.flickr.com/2327/1828270262_b76c867954.jpg

I have not got the test gear to answer these questions definitively myself. I wonder what happens if you make some single parameter changes seeking an ‘optimum’ …

1) Keeping the length of the bifilar winding wire the same (to keep the same resistance you see), but increase or decrease the inductance by changing number of turns actually used on the toroid. The frequency rises with fewer turns, but what about the input current and efficiency? If you wished a high output to give bright LEDs which way would you go? Where is optimum?

I go into detail about this in my earlier blog. Remember that you are not limited to using a toroid. You can use an air core too, or other shaped cores. I would start with the most turns I could easily wind onto the core, but I would not use very thin wire. Then after getting it running, I would unwind a few turns at a time and measure the LED current as I discuss below. If there are a few inches of wire left unwound, then it will be easy to wind that wire back onto the core if I feel I unwound too many turns.

2nd Q.

2) Is there any advantage to a different ratio of turns on the toroid rather than 1:1?

Using the 1:1 ratio makes it simple for the experimenter and also makes it easy to wind both windings at the same time – bifilar wound. But the windings can and should be wound with different ratios depending on other factors. For instance, if you wind many more turns on the feedback winding, the JT will have more sensitivity at low voltages and will run with a battery that has less than a half volt. Also, if you have a 1:1 ratio, and there are two or more LEDs in series across the output, then the voltage across them will be 6.6 or more volts. The transistor’s maximum emitter to base reverse voltage is 5 or 6 volts, and the voltage will exceed this and possibly cause damage to the transistor. So the way to reduce the voltage is to reduce the number of turns of the feedback winding, or increase the number of turns of the primary winding.

Third Q.

3) For a ‘standard / common’ ferrite type, does the physical size of the toroid matter, if you are winding to a defined inductance? I.e. is small or big better? If aiming for high output bright LEDs. Some of my salvaged CFL toroids are just ~5mm across, will they saturate? Or be OK?

Again much of this question can be answered in the blog I gave for Q.1 – here it is again. There are other resources on the ‘Net that deal with electromagnetic theory and practice, and I have tried to learn from them but I haven’t picked up that much to be able to say I know enough about electromagnetics. What I do know is by my experiments with the toroid cores that I’ve used in JTs.

One fact that many toroid core users don’t realize is that the cores are made to tolerances that are wider than capacitors and resistors, typically + or – 20 percent. So winding the coils with the same number of turns may have a quite wide spread of inductances. Some factors that may influence the core are the quantity of ingredients used and the temperature and time that the cores are fired to make the ferrite sintered.

I’ve built satisfactory JTs using 6 mm cores, with 30 AWG (0.25mm) wire. The number of turns and the peak current determine the magnetic flux, and I have seen JTs with small cores draw well over 100 mA average or 200 to 300 mA peak from the supply. They do a good job of lighting the LED brightly, but the typical JT has an efficiency of only about 50 percent, so a lot of that supply current is wasted.

I have used two or more of the same small core for a coil, by stacking them and winding the wires around both cores. But it may be difficult to find two CFLs with the same identical core.

4th Q.

4) I think I struggle to measure very small resistances accurately with my meter. Using wire salvaged from CFL transformers, is the resistance of just 1 meter or so likely to be limiting in any way? My hunch was that because the scale is so small in length and current, there is not an important difference in resistance, BUT finer wire (than the enamelled wire I bought) certainly facilitates making more turns on a small toroid.

First, assuming you know the wire size, you can get the resistance from a wire table. For instance, in the table, 30 AWG wire says it’s 104 ohms per thousand feet, or 0.104 ohms per foot.

But if you don’t know the size, you can measure the resistance another way. If you have a 1 amp power supply that has current adjustment, you can put 1 amp through the wire and measure the voltage drop across a certain length, such as 1 meter. A 1 volt drop is 1 ohm, a half volt drop is 0.5 ohm, etc.

But in the blog I gave for Q.1, I talked about how the coil does not have to have a high inductance or a lot of turns. But the wire should have low resistance which gives low loss. If your wire is very fine, then wind a coil like I do. I use four strands of 30 AWG (0.25mm) wire quadrifilar wound. I connect three of the four together in parallel for the primary winding, and the single fourth wire is the feedback winding. In this blog I used regular telephone wire and connected the three together for the primary. And I later reduced the turns to increase the current to the LEDs.

5th Q.

5) I thought a small capacitor in parallel with the resistor would improve switching efficiency (particularly as my resistors available were nearer 5k than 1k) and AC performance, I could not convince myself it helped. Is that easy to explain?

I measure the LED current by putting a 1 ohm resistor in series with the LED and measure the voltage across it. If I get 20 milllivolts, then I know the LED is getting 20 milliamps. This method may not give accurate current readings, but it allows one to make comparisons when he is making changes to the circuit. I have seen other JTers claim that the capacitor in parallel with the resistor helped to speed up the switching and gave more light without increasing the supply current. But I have not been able to get this to happen when I tried it. I think those who get their JT circuit to do this have some other deficiency in their circuit that may cause the capacitor to help.

6th Q.

6) Compared to a varied input voltage say 0 – 4v and either a pure series resistive + LED load or a JT. How does a JT output compare? At what available input voltage is a JT superfluous? My string of parallel white LEDs conduct dimly even at 2.1v, so even two NiMH in series would light them a bit. What is the Joule Thief’s justifying voltage range?

The JT is designed to boost a lower voltage up to the voltage of its output, usually a 2 to 3V LED. The JT’s supply voltage must be less than the LED forward voltage, because the current would then go right through the coil and the LED would start to light when the forward voltage is reached, and the LED would light without the JT operating. If the supply is two AA cells in series, the 3V would be enough to light the LED, but if two LEDs were in series across the output, then they would not light and the JT would do its job. However as I said earlier, the 1:1 turns ratio would mean the negative voltage on the base would be more than the 5 or 6 volts maximum for most transistors. This is when the feedback winding would have fewer turns, to reduce the peak negative voltage.

So with a 1:1 winding ratio, the JT is designed to have a supply voltage of less than the LED’s forward voltage, which is about 2 volts for a red LED, on up to 3.6V for an ultraviolet LED. White and blue LEDs have about a 3.2V forward voltage.

Some people make the mistake of thinking the JT will do all sorts of magical things. The JT does not need to be used if the voltage of the supply can be increased, such as by putting multiple AA or AAA cells in series. Three AA cells will give enough voltage to light the LED brightly, with a 33 ohm current limiting resistor in series with each LED. I’ve built many of this kind of light. And the efficiency is usually better than the JT.

RFI Suppressor Sleeves

Paul asked how to avoid these ‘lossy’ toroids. The explanation is simple. Use the toroid at the Joule Thief frequency of 100 thousand Hz or so, and its losses are minimal. Use the toroid at ten million Hz, a hundred times higher frequency, and the losses become much greater. I buy the Fair-Rite 2673002402 “RFI suppressor sleeves” which look just like a toroid, and use them for my high efficiency Supercharged Joule Thief. They do a very good and efficient job at 250 kHz. I started out using some high permeability ferrite toroids that were made with the “75 mix” which is the ferrite material that is used to make the toroid. This is high permeability, several thousand depending on the size of the toroid. I got these from an “RFI suppressor kit” sold by Palomar Engineering. I then used the RFI suppressor sleeves that I got from the ends of keyboard and mouse cables that were defunct. These made excellent Joule Thief coils. The RFI and regular toroids are all made from the same types of ferrite, so it’s all in the range of frequencies at which they are used. The high permeability allows the windings to be made with fewer turns of heavier wire which has very low resistance and low loss.

One other thing: the core is often color coded, which seems to vary by maker. Lower mu cores are often iron powder, while higher mu cores are typically ferrite. Some experimenters like to use the cores they pull out of PC power supplies.. One core, a dual winding core that is yellow, seems to be a popular one since it has two windings that look the same, so all one has to do is hook it up and it works.The Youtube videos are full of Joule Thief experimenters building esoteric coils such as Tesla pancake and Rodin coils. As far as I can tell, the electrons see them as another inductor with properties much like the POTC – plain old toroid coil with bifilar windings.

The Transistor

So far, I’ve talked about the coil mostly. But the coil and all of the other parts of a Joule Thief can’t do anything without the transistor, and without a transistor that is suitable for the job. The transistor’s job has to deal with switching a very low voltage, and very high current. And it has to do this with very low losses. This is a heavy burden and not every transistor can handle it well. If you use a BC337, PN2222A, 2N4401 you will get a reasonably bright LED with a current of between 15 and 20 milliamps. If you use a 2N3904, BC547, 2SC1815, the LED current will be substandard, about 8 to 10 milliamps. And if you use the transistors that were made for this kind of use, you may get much more than 20 milliamps. These are the 2SC2500, 2SD5041, BC639, SS8050, NTE11. All of the above are NPN; there are also PNP transistors.

Paul asked what makes a transistor suitable or unsuitable for a JT. One important factor is good beta holdup at high currents. A datasheet for transistors usually has a graph showing the current gain (beta) on the vertical axis and collector current on the horizontal axis. The graph for the 2N3904 shows a rapid falloff of current gain as the current approaches 100 milliamps. The graph for the 2N4401 shows that the gain falls off well above 100 milliamps, more like 200 or 300 mA. The BC337 does even better. That’s one reason why the 2N3904 will give you 8 to 10 milliamps of LED current, while the others will give you 15 to 23 milliamps.

Another important factor is the Vce(sat) at high currents. The better transistors will have a Vce(sat) of 1/4 volt or less at high currents, 200 to 400 milliamps or more. The lower the voltage, the less wasted power there is in the transistor. Also, the transistors I mentioned above have high current gains, some more than 500 and could even be much higher. This reduces the losses due to the current needed to drive the transistor’s base.

Paul said his JT is drawing 20 mA from the supply. The typical JT has a resistor that is 1000 ohms, and if it is changed, it will change the base bias and hence the current consumption from the supply. I recently ran into a Youtube video of a JT which used a transistor that was clearly not made for this job. The transistor should be chosen to handle the full supply current needed to drive the LED with the 20 mA that would be considered ‘full brightness’. Then if the circuit has to supply less than 20 mA to the LED, the 1000 ohm resistor can be increased to reduce the drive to the LED.

One must remember that the efficiency of the typical JT is only about 50 percent. If you want your LED to get 20 mA average current, at 3.3 volts, that’s 66 milliwatts. At 50% efficiency, the supply power should be 132 milliwatts. Divide that by 1.5V, and you get 88 milliamps supply current. The typical JT draws about 70 to 90 mA, and puts out about 15 to 20 mA to the LED. Most JTs will do this with a PN2222A, 2N4401, BC337-25 transistor. If a JT draws less current and calculates to lower efficiency, I would say that the transistor is not up to the job and should be replaced with one that is.

Paul is concerned that the JT’s transistor is not being fully turned off. The LED has about 4 or 4.5 volts across it, so the 1:1 winding ratio gives about 4 to 4.5 volts peak negative voltage at the base bias resistor, and since the base is reverse biased and effectively an open circuit, most of that voltage appears across the base to emitter junction. I’ve put an o’scope on mine and I see the negative voltage on the base, indicating that the base to emitter junction is reverse biased part of the off time.

If one measures the coil primary winding resistance and finds that it is low, say 1/4 ohm or less, than it can be assumed that the resistance is low enough to not influence the JT to its detriment. If the circuit still underperforms, I would consider putting another transistor in parallel with the existing one and see what happens. If this helps the circuit put out more current to the LED, then I would assume the transistor is the limiting factor in the JT’s performance, and try a different transistor that is more capable of doing the job.

I may add some to the above, when I find the blogs that I’ve done about these subjects.