Joule Thief Information Page

I am not going to write about how to make a Joule Thief but rather I simply want to add a little more information for everyone. As I come across more, I will add it to this page.

A list of reference web articles/pages will be at the bottom of this page. After I got my joule theif working I did more research and some of my own experiments. The references below are some of the better articles I've come across which better explain how to build a joule thief as well as what you can expect duriung its operation.


To read the wiki article, Click Here        


* The first thing I want to show is a small table of the inductance values of the toroid bifilar inductor vs. the frequency of the pulses that I measured.

NOTE: Some of the inductances will show 0/## this is because my inductance meter could not measure an inductance of only one winding. In these cases I also measured the inductance of the two windings.
Inductance
(μsH)
Frequency
(KHz)
0/<1 2,000 (=2 MHz)
0/42 1,000 (=1 MHz)
69 17.5
123 17.2
223 4.9
616 5.0
The two inductors with 0/## were air core different diameters of the same length of cat 5 pair wire about 42" long. I simply wanted to see if such a bifilar inductor would work.

All of the ferrite toroids were considerably different. It is rather interesting that I have an extremely small inductor at 616 μH, the OD of which is ~6mm and it is ~ 3mm tall, and the 223 μH inductor has an OD of 18mm and is 12mm tall. These are considerably different physically and very different inductances and yet the frequencies are very close.

The red listing is, IMHO, where we want to be, if possible, even a lower frequency would be better still. I'd like to see if I can get it to ~ 500Hz! The reason for this is that the fewer the number of pulses per second, the longer the battery should last. Assuming all pulses are the same height/voltage and width/duration.

Transistor Tests @ 1.28V, 495-3849-ND @ 800 μH, 44 Loops
Transistor
Freq
(KHz)
On
(μs)
1 Hz
(μs)
On Time
( % )
Peak
Volts
2N2222 2.6 120 384 31 1.5
2N3904 1.2+ 320 820 39 2.2
2N4401 1.2- 330 840 39 2.8


Inductor and LED Tests @ 1.28V & 2N3904
From
Core Order #
O.D.
(mm)
Wire
Guage
( # of )
Loops
mA
μH
Freq
KHz
On
(μs)
1 Hz
(μs)
On Time
( % )
Peak
Volts
LED
Color
Digi-Key 495-3851-ND 12.5 24 11 38.7 527 4.2 90 240 37 1.8 Blue
Digi-Key 495-3851-ND 12.5 30 25 23.1 570 1.4 280 690 30 1.8 Blue
Digi-Key 495-3851-ND 12.5 30 25 22.6 ? ? ? ? ? ? White
Digi-Key 495-3849-ND 10 32 11 56.5 131 8.3 120 235 54 1.2 Amber
Digi-Key 495-3849-ND 10 32 44 39.2 800 2.2 280 460 54 1.2 Amber
Digi-Key 495-3849-ND 10 32 11 56.6 138 7.84 40 127.5 31 2.8 White
Digi-Key 495-3849-ND 10 30 12 54.2 158 8.16 42.5 122.5 34 2.7 White
Digi-Key 495-3849-ND 10 32 22 41.6 497 3.70 100 270 37 2.8 White
Digi-Key 495-3849-ND 10 32 44 30.6 800 1.78 220 560 39 2.8 White

The human eye cannot tell that something is pulsing when the pulsing is fast enough. Here in the USA we use AC current of 60 Hz and in Europe they use 50 Hz. So, if there was some advantage to it, and it could be done easily, getting the frequency down to 100Hz, would also be acceptable.

You're probably thinking that we'd notice a difference in how bright the LED appeared. All I can say is that I didn't notice any difference between the LED going at 5KHz with the same LED going at 1MHz so I doubt we would notice any difference between 5KHz and 500 or even 100 Hz.

I ran two tests - one using the 5KHz inductor and another using the 1MHz inductor - and there was a substantial difference in how long the 1.1V batteries lasted. I initially thought this was because the 1MHz inductor has 200 times more pulses per second. However, after thinking about it, I decided that it is only the percentage of time that the LED is on during one cycle that determines how much total energy is used and so determines how long the battery will last. Thinking about it a bit more, I thought almost the exact opposite, that it might be due to the length of time the inductor is "charging" that determines the life of the battery. The shorter the LED is off, the longer the battery will last. Or, put another way, the less current used, the more efficient the circuit is. I ran a few more tests on Feb. 22, 2014, and discovered that this looks correct. The longer the LED is off, the lower the frequency and the lower the amount of current is used by the circuit. All of this should mean that this is more efficient and should cause the battery to last longer. This does use more windings and so more wire. I believe that the items that determine the on vs total time are: the core, the wire size, the resistor value, and the LED. I'll have to run a few tests to see how each of these affects the ratio of LED on time to the time of one cycle. We get some indication that different inductances with the same core and the same wire do not change the on-time/cycle-time ratio in the above grid testing the Inductor and LED.

The failure mode for the 1MHz inductor was very interesting: the LED would light for about 10 seconds then turn off for about the same length of time while the battery recovered some of its voltage. It would rise to a voltage of about 0.62V and this would turn on the LED again until it hit a low of about 0.51V, at which point the LED would turn off. This cycling continued throughout the day, another eight hours, when I finally replaced the 1MHz inductor with the 5KHz inductor and ran the battery down to about 0.37V.

NOTE 1: I have noticed similar frequency changes depending on the inductor value. However, when R3UK recommends that the more windings the better, I'm not sure he's correct. I'm not really sure what he means by "better" since when I've added too many windings, the LED doesn't light at all. Look at the tests done and listed here we see that the more windings of the same kind of wire increases the inductance which decreases the frequency but the on-time/cycle-time ratio does not change and so the battery life remains the same. So, what is "better" about this arrangement? On the other hand, more wire was used but this did not change either the performance of the LED, making it brighter, or increase the life of the battery, so there was no improvement there, nothing better about that. On the other hand, using more wire takes more time to wind the inductor. That's not an improvement, and then there is less wire left over to build another unit. So far, I find nothing better about using more windings yet.

Date: 02/22/2014 - Today I ran a few more tests and this time I was also measuring the current the circuits used. I used the three LEDs: Blue, amber,and white. It was very clear that the more windings, the lower the frequency, and the less current the circuits used. This, then, is a very good reason to use more wire to lower the frequency, within reason.

NOTE 2: When I used so much wire that the inductances were around 3.5 mH, the LED did not light up at all, so that is definitely not better. Too high an inductance as well as too low an inductance will cause the LED to not light up. So far I know it works from ~< 1μH to > 800μH. In the transistor test grid above, we see that 800 μH still worked fine so the LED fails to light up somewhere between 820 μH and 3.5 mH. I don't think it is worthwhile to do more tests to determine the inductance value that stops the LED from lighting up. However, since it now turns out that the more windings the less current is used, perhaps more tests to see how low a freqency is possible with a reasonable number of windings would be worthwhile.

Other tests will need to be done to determine how to change the on-time/cycle-time ratio to light the LED for the shortest time possible while having the longest cycle time and still have an acceptable brightness over the life of the battery.

NOTE 3: LEDs that can be used: Many other articles mention that blue or white LEDs should be used. I have used Amber LEDs to make a few Christmas "candles" for windows which have difficult or no access to the AC. These work very well with dead batteries of around 1 Volt. Any LED that is listed as a 3V LED or a 20mA LED should work with the Joule Thief, so you don't have to limit yourself to blue or white LEDs.

So, what is it we want? This all depends on what we want to do. If we just want to have a little night light or flashlight, then making the unit as efficient as possible may not be all that important. After all, we're using "dead" batteries. If, however, we want to put up a light that lasts as long as possible, then using the best transistor, ferrite core, number of windings, and LED all must be considered.

Amber LEDs were used in our Christmas window-lights to better simulate candles. The ferrite cores were salvaged from a dead CFL bulb plus one from something else. No doubt they are not the most efficient. But, they do get the job done.

If we want the most efficient joule thief, assuming this means using the least current, then we want:
  1. A core that needs the fewest loops and produces the lowest frequency.*
  2. The LED that causes the circuit to use the least current. This seems to be the white LEDs.
  3. The best transistor, which so far is the 2N4401.
  4. The most windings for the lowest frequency, within reason.
  5. The best size wire to reduce the current used. It seems that larger diameter wire is better.*
* I'll have to run more tests using the same wire and number of loops on several different cores to give you those results. So far, the larger core, Digi-Key 495-3851-ND with 25 loops, has a lower frequency and uses less current than the 495-3849-ND core using 44 loops.



Core vs Current Tests @ 1.28V, 2N4401, White LED
From
Core Order #
O.D.
(mm)
Wire
Guage
( # of )
Loops
mA
μH
Freq
KHz
On
(μs)
1 Hz
(μs)
On Time
( % )
Peak
Volts
LED
Color
Digi-Key 495-3848-ND 10.0 30 11 55.9 207 9.52 50 105 48 2.4 White
Digi-Key 495-3849-ND 10.0 30 11 54.2 158 8.16 42.5 122.5 34 2.7* White
Digi-Key 495-3849-ND 10.0 30 11 53.6 149 8.33 40.0 120.0 33 2.5 White
Digi-Key 495-3874-ND 6.3 30 11 40.3 190 11.4 32.5 87.5 37 2.4 White
Digi-Key 495-3847-ND 10.0 30 11 36.9 466 5.26 75 190 39 2.4 White
Digi-Key 495-3851-ND 12.5 30 11 35.8 624 4.88 120 205 59 2.4 White

*
 
This was the original reading as seen in the "Inductor and LED Tests" grid above. All other measurements were made on the same day with the same Volt/current meter, power supply, Inductance meter, and Oscilloscope.




The non-Simple Jouel Thief

The non-simple Joule Thief uses different numbers of windings/loops for the collector and base of the transistor. I'll do a few experiments to find the best winding ratio in order to find the lowest current used for one of the ferrite cores I have. In the image below I show that I will start with 20 loops for both sections and go down from there until I find the lowest currect used. The second test I'll start with about half as many loops, 12, and I'll start with the base using 8 loops.





Non-Simple Jouel Thief, 1.20V, 2N4401, White LED, 495-3847-ND, 30 gauge Wire
#
Loops
Short
μH
Long
μH
Total
mH
On
Time
Off
Time
% on
-off
Freq.
KHz
 
mA
Peak
Volts
20 701.1 700.5 2.675 110 225 48.9 2.99 42.1 2.2
15 413.8 702.7 2.125 110 220 50.0 3.08 39.5 2.1
10 196.6 703.0 1.603 105 205 51.2 3.28 36.3 2.1
8 130.5 702.8 1.415 100 195 51.3 3.39 35.6 2.0
5 54.3 697.7 1.150 65 150 43.3 4.65 40.4 2.0
------ ------ ------ ------ ------ ------ ------ ------ ------ ------
8 130.5 702.8 1.415 100 195 51.3 3.39 35.6 2.0
8 130.5 702.8 1.415 35 90 38.9 8.16 75.2 2.4

With these tests I started using a new Inductance/Capacitance meter I picked up at: AADE in kit form. He also sells it completely built. I put it together in less than two hours one afternoon.

The simple joule thief, 20 X 20 loops, took up the complete circumference of the toroid. Loops were then removed from the short side (the end going to the base of the transistor), but the loops were not compressed until the end when the best number of fewer-to-maximum number of loops, 8 X 20, was discovered. This caused the joule thief to use the lowest current.

Only the data for the 8-loop version with the excess wire removed was kept.

When searching for the best version of the joule thief inductor, I measured just the currents for 9 loops, 36.1 mA, then 8 loops 35.5 mA, and last for 7 loops which was 36.0 mA. I also switched which number of loops went to the collector and base of the transistor which is colored in red to show how bad it is to get the wiring wrong. Getting the wiring backwards more than doubled the average current used by the circuit, shown in the  RED  row.

Conclusion: While the "simple" joule thief is easier to make, one can make the joule thief more efficient (use less current), by removing some windings from the inductor that goes to the base of the transistor. The best in this case turned out to be 8 loops by 20 loops, with the eight loop end going to the base of the transistor. With this reversed, the 20-loop end going to the base of the transistor, the circuit used far more current. The ratio is 8:20 or 4:10, or 40%.

We now know two things:
  1. the more loops used the lower the current
  2. a ratio of about 4:10, base to collector, number of loops lowers the current even more.






Next I tried 12 loops and then removed loops for the base of the transistor section to see if the current could be lowered any more while maintaining the LED brightness.

Non-Simple Jouel Thief, 1.20V, 2N4401, White LED, 495-3847-ND, 30 gauge Wire
Using 12 loops and starting with 8 loops for the base of the transistor.
#
Loops
Short
μH
Long
μH
Total
mH
On
Time
Off
Time
% on
-off
Freq.
KHz
 
mA
Peak
Volts
8 352.8 785.5 2.160 75.0 135 55.6 47.6 31.1 2.2
7 263.0 788.8 1.984 75.0 130 57.7 4.88 30.4 2.2
6 175.1 808.5 1.839 75.0 130 57.7 4.88 29.6 2.2
5 109.2 839.2 1.729 72.5 127.5 56.9 5.00 28.5 2.2
4 55.73 813.2 1.517 62.5 115 54.4 5.63 30.3 2.2


Looking at the ratio of loops to the base to loops to the collector gives around 40%, 0.40 for the 8:20 loops, and 0.417 or 41.7% for the 5:12 loops. The lowest current use is when the loops are in a ratio of 40 - 42%. Much above or below that is out of the sweet spot. On the low end, 33 - 35% or 1:3, 4:12, and 7:20 gets worse and on the upper end, 45 - 50% or 9:20, 6:12 or 1:2 gets worse.

Looking at the ratio of loops of the base to collector gives around 40%, 0.40 for the 8:20 loops, and 0.417 or 41.7% for the 5:12 loops. So far, the lowest current used is when the loops are in a ratio of 40 - 42%. On the low end, 33 - 35% or 1:3, 4:12, and 7:20 gets worse and on the upper end, 45 - 50% or 9:20, 6:12 or 1:2 gets worse.

If the lowest current is wanted, it looks like we'll need to increase the collector number of loops to the max we can put on any particular toroid, perhaps 40 for the collector and then around 16 loops for the base of the transistor for a 10mm diameter toroid.

Since it was easier to try fewer loops rather than more, I added two tests with 10 and 5 loops for the collector. I also thought I noticed an anomaly with one of the things I thought I'd just learned, namely, that more loops lowers the current and fewer loops increases the current. Looking at the grid of data below, we can see that, in fact, there is an anomaly around 10 - 12 collector loops. These were both lower than using 20 loops and, in fact, the 5/12 loop test used about 20% less current. It looks like I'll have to try 2/5 loops and then I'll try 12/30 and 16/40.





Non-Simple Jouel Thief, 1.20V, 2N4401, White LED, 495-3847-ND, 30 gauge Wire
Comparing collector loops from 20 to 5 and base loops ~40% the number of collector loops.
#
Loops
Short
μH
Long
m/μH
Total
mH
On
Time
Off
Time
% on
-off
Freq.
KHz
 
mA
Peak
Volts
14/30 ------ ------ ------ ------ ------ ------ ------ 19.0 2.2
13/30 ------ ------ ------ ------ ------ ------ ------ 19.2 2.2
12/30 745.3 3.646 7.073 18.5 29.5 62.7 2.08 19.8 2.2
8/20 374.5 2.002 3.896 125 210 59.5 2.96 22.1 2.1
5/12 109.2 839.2 1.729 72.5 127.5 56.9 5.00 28.5 2.2
4/10 51.9 561.0 1.186 55.0 100.0 55.0 6.45 32.0 2.1
2/5 11.0 88.0 336.8 22.0 45.0 48.9 14.9 43.8 2.0
------- ------- ------- μH ------- ------- ------- ------- ------- -------
5/5 85.6 85.8 648.3 26.0 51.0 50.9 13.0 49.1 2.2
4/5 46.7 87.2 538.5 25.0 49.0 51.0 13.5 47.7 2.2
3/5 23.4 87.2 438.4 24.0 47.0 51.1 14.1 45.9 2.2
2/5 11.0 88.0 336.8 22.0 45.0 48.9 14.9 43.0 2.0
1/5 4.21 86.5 242.6 17.0 36.0 47.2 18.9 46.4 1.95


After doing the 12/30 test, I found that there wasn't enough room on the toroid for the 16/40 loops test and so I decided to end the tests at 30/12. The LED looked as bright and was only using ~20mA. I also tried 11/30, ~21.3 mA and 13/30, 19.2mA to confirm that 12/30 was the best. As can be seen, it turned out that one more loop 13, gave a slightly lower value so we now see that the best ratio is a bit above 40%. Right now the best data suggests about 43.3% or 13/30. I didn't leave enough wire to do another complete loop but did manage almost a complete loop, and the current went down a bit more to 19.0mA so I suspect that 14/30 would be the best and the ratio is 46.7%.

It seems that the numbers are not perfect for whatever reason. In the original data using 20 loops and removing loops, I found that 8/20 or a ratio of 40% was the best, and even one more at 9/20 or 45% was worse. Yet, using 30 loops, I found that 14/30, a ratio of 46.7%, was the best, and just a little better than 13/30 at 43.3% which was better than the expected best at 12/30 or 40%.





How to wind the toroid for the lowest current usage

As in antenna building, it is always easier to remove some wire to make the antenna shorter than it is to put more wire on to make it longer. So, since we now know that the lowest current is achieved by using 40 - 47% as many loops for the base coil as for the collector coil, this should be easy. Starting with 50%, the number of collector coil loops for the base coil loops can be removed from the base coil until the lowest current usage is found. Start by measuring the current, then remove loops from the base coil one at a time until the lowest current is passed. Next, add one loop back to the base coil to get back to the lowest current. Now cut off the excess wire. Example: we want to make the best inductor using 30 loops for the collector. This means we start with a piece of wire long enough, with some extra, to wind 45 loops on the toroid. We first start with some extra extending out and wind 15 loops, then fold the remaining wire to be wound and leave some extra for the transistor base coil; now loop the folded wire 15 more times.

Now, cut the folded end and connect the end of the transistor base section to the beginning of the transistor's collector coil section. Starting with 15 loops for the base section and removing one loop at a time, measure the current each time until the lowest current usage has been found and passed. Then, add the one loop back to get back to the lowest current. Remove all the excess wire and the coil is done.





Here's a practical application of the simple joule thief. I needed three simulated candles for three widows which had no easy way to get AC power to the "candles". Here you can see the three simulated LED candles I made with salvaged ferrite cores.

Three Christmas Window Lights, Insides & Working

Also see the LED Tester built from this basic circuit.


Low Voltage Battery Traces: Batt~0.57V, 0.2V/Div & Batt~0.41V, 50mV/Div @ 10μs, Amber LED


The above traces were, on the LEFT, for battery voltage of 0.57V and the peak voltage looks to be ~ 1.05V - the on-time/cycle-time ratio is ~36%. The RIGHT trace is for a battery voltage of 0.41V and the peak voltage looks to be ~ 0.95V and the on-time/cycle-time ratio is now ~23%.


I also made a replacement Piano Light "bulb" out of two LED modules with three LEDs in each of them, the base of a dead LED light bulb, and the circuit from inside a 12V, 500mA switching power supply.

Here are a few of the better articles I've come across explaining the Joule Thief:

Title: Joule thief
Link: http://en.wikipedia.org/wiki/Joule_thief
Synopsis: A good explanation, schematic, and O-scope trace which shows what my scope showed.

Title: MAKE A JOULE THIEF
Link: http://www.bigclive.com/joule.htm
Synopsis: Clearly shows how to connect the windings of the ferrite cord inductors.

Title: Making A Simple Joule Thief (made easy), by ASCAS
Link: http://www.instructables.com/id/Making-A-Simple-Joule-Thief-made-easy/
Synopsis: The one I used to make mine, no explinations but it works.

Title: "Joule Thief" Circuits, crude to modern... by Dave Kruschke
Link: http://www.instructables.com/id/Joule-Thief-Circuits-crude-to-modern/
Synopsis: Here Dave shows how this circuit works using a mechanical switch!

Title: Bifilar coil
Link: http://en.wikipedia.org/wiki/Bifilar_coil
Synopsis: How to wind it & the different connections

Title: The Joule Thief!
Link: http://www.r3uk.com/index.php/tech-tips/43-electrical-tomfoolery/179-the-joule-thief
Synopsis: R3UK has some very good explinations, experiments, and results. Ideas on better transistors to use than the 2N3904



HOME Return to: [My Home page]