I came across this document on scribd, and found it very interesting. The guy’s name is Fernando Garcia and it is dated May 2009. Unfortunately there is no contact information, for there are several unanswered questions I have that he could answer.
He discusses how he measured the efficiency of a Joule Thief and how he measured the actual light output of the LED without using a luxmeter. Then he compares the first to another JT. I especially liked the graph he made of the LED current versus the CdS photocell resistance. It shows that the CdS resistance is not linear in response to the current to the LED. But it really isn’t that important because once he finds the CdS resistance, he only needs to refer to the graph to find the current.
The important point about using this photocell is that it will remove the possibility that the pulses through the LED are causing an error in the measurement. In other words, once the photocell is graphed with the LED current from a DC power supply, the pulsating current causing light pulses from the LED will be filtered out by the photocell, presumably because the photocell cannot respond to the pulses, which is basically the same way our eye responds. Then if you get a photocell resistance of 100 ohms for example when it’s lit by the Joule Thief, the JT should be putting out the same amount of light as when it was lit with DC and the photocell resistance was 100 ohms.
Measuring the input current He discusses how he went about measuring the power supply current at 1.5 volts. He inserted a current sensing resistor in series with the positive supply line, and measured the voltage across it and calculated the current. But he went to elaborate lengths to filter out the “high frequency ripple” pulses in the supply line caused by the Joule Thief’s switching on and off. He used a low ESR 2.2 uF tantalum capacitor, two more capacitors and a 2.2 uH choke to remove the pulses. I highly recommend using a bypass capacitor across the supply leads. But instead of filtering out the pulses in the supply, I would simply put a low pass filter on the voltmeter’s input: a 1k or 10k resistor in series with the positive voltmeter lead, and a 1 uF capacitor across the voltmeter leads. The voltage drop across the 1k or 10k is almost zero because the typical digital multimeter has a very high input resistance. Also, the great thing about doing it this way is that you can move the filter with the meter leads to another part of the circuit and have the same benefits of the filter.
The Joule Thief He shows the circuit and admits, I’m happy to see, that the 2N3904 he used was not a good choice of transistor. I think that this choice may have influenced his measurements in a detrimental way.
He goes on to explain how he built the Joule Thief. He explains how he made the coil, which he mistakenly called a transformer; I quote:
The transformer was bifilar-wound on a Micrometal’s T30-26 core. Fully wound it gave about 8.5 µH, which is also a little on the lowside.
The T30 toroid core is only 0.3 inches or 7.6 mm outside diameter, which is a very small core. The 26 type ferrite material is low permeability, so the inductance is very low. He said fully wound, but what size wire did he use? He also didn’t say how he wound the two windings. He said they were bifilar wound. Were they both the same size wire?
He goes on to say that the inductance was 8.5 microhenrys. But I would say that that is a lot more than just a little on the low side. In my experimentation with dozens of JTs, I would say that the coil should be 30 microhenrys minimum with an optimum point between 100 and 300 microhenrys. A lot has to do with the size of the core and the core material, and the size of the wire. With the tiny core that he used and fine wire, he might be able to get 30 uH, but the wire will have to be very thin. And the wire will have higher resistance, which will cause more losses and lower efficiency.
I think he should have chosen a higher permeability core, or else larger to allow thicker wire. He later goes on to use a larger core, a T50-8 core. But then he “adjusts” the turns to give the much too low 8.5 microhenrys. Apparently he thought that was necessary to make the comparisons. I’m fairly certain that the result of using the low 8.5 uH inductance is that his Joule Thief ran at a frequency that was very much higher than a typical JT. A typical JT runs at frequencies between a few tens of kHz to a few hundred kHz, with a broad optimum point around a hundred kHz. He gave no information about what frequency at which his JT was running. I think it was running at as much as a megahertz, or maybe even more. I’ll explain below why this is important.
Further Measurements He then adds a 1N5817 Schottky diode and 1 uF filter capacitor to the output between the transistor and the LED, to light the LED with DC without the pulses. In his comparison, he showed that adding this diode and cap made a large increase in the “figure of merit”. Later he says that adding the diode and cap is a no brainer, and to always include it (see note at the bottom).
He didn’t say anything about what kind of LED he used. The schematic says “Wht LED” so I presume that that is what he used. I have my own theory as to why the diode and cap increase the figure of merit. This tends to get involved but it is important, so I hope you can bear with me.
All semiconductors have a junction capacitance, because the junction (or junctions) acts like the plates of a capacitor. In a transistor this is called the Miller capacitance. In a LED, there is a small amount of capacitance, which, I will have to try to find out what it is for a typical white LED. Also, diodes have a recovery time during the reversal of polarity that is when the hole and electrons are swept out of the junction. In a Schottky diode, the junction is half a diode, because half the junction is metal, not semiconductor, therefore it has faster recovery time. The 1N4004 rectifier is a “standard recovery” diode, because it operates at power line frequencies and has very little losses during the recovery time. If we operate a 1N4004 at tens of thousands of kilohertz, the losses are much greater, so these are not used at the frequencies found in switching power supplies. Instead the fast recovery and ultrafast recovery diodes are used, and also Schottky diodes. These have very low losses at a hundred kHz or more.
The LED is also a diode, and will have a similar recovery time. If this is slow, it will have an effect on the efficiency at high frequencies. The Schottky diode has a very fast recovery and this may be why the Schottky rectifier and filter gives better performance than the LED alone. One thing is certain: there is a half volt drop across the Schottky as it rectifies, and this has to be accounted for as far as loss goes. Assuming 30 milliamps peak current (just a guess) and a half volt drop, that is 15 milliwatts of loss that ends up wasted as heat in the Schottky diode. This is why I chose to eliminate the LED current from being rectified in my Supercharged Joule Thief.
Back to the document. Since the waveform at the LED is complex, I do not measure the LED voltage. I assume that the LED voltage is 3.3V for calculating the power to the LED.
In the formula he gave, he did not say current, which is confusing.
At the beginning of the document the author stated that power out divided by power in gives efficiency. Then he uses the figure of merit, which looks to me like he is saying it is the ratio of the (equivalent) LED currents (his formula doesn’t say it’s current) between the original circuit and the modified circuit. In making the graph the author did not measure the voltage across the LED for each point on the graph. Therefore the power to the LED is not known. With LED current but without the power, the efficiency cannot be calculated.
One could assume a constant voltage of 3.3V across the LED, which is how I calculate the LED power. I have attempted to read the LED voltage with a DMM, but I get the battery voltage, not the LED voltage. I have also read the voltage across the LED with an oscilloscope, which gives peak and RMS readings. The RMS reading applies only to sine waves, but the JT does hot have a pure sine wave, so RMS readings are not valid. I do not measure the LED voltage since the waveform at the LED is complex, and may cause the meter to give a false reading . I assume that the LED forward voltage is 3.3V for calculating the power to the LED. The LED forward voltage stays relatively constant as the current changes so this assumption is a reasonable one.
Difference of opinion I must state that I do not agree with the conclusions the author gave in the “Comments On The Results”. The reason I do not agree is that the coils (he mistakenly calls them transformers) in the JT he used have much lower inductance than a typical JT. In his 3rd comment he says (I quote) “for maximum efficiency it pays to use magnetic cores optimized for switchmode supply usage”. Ironically, the small micrometals cores he used are for radio frequency use, not for switching power supply use. He said the coil was 8.5 microhenrys; the typical JT may be 100 or more microhenrys. Therefore the frequency of his JTs is much, much higher than a typical JT, and it is my opinion that this causes the JT’s operation to be less than optimum and the results that he found were not typical of a conventional JT that the average experimenter might build. I think the author should change the ferrite core to a higher permeability core so that the coils measure at least 100 microhenrys, and then make the measurements with these coils.
The author stated that adding the capacitor across the base resistor was beneficial. My opinion is that as I said above, his Joule Thief was running at a frequency much, much higher than a typical JT. Therefore adding the capacitor to speed up the transistor’s switching was much more noticeable in his results because the transistor was running at too high a frequency, which caused the transistor’s switching times to be more influential on the circuit. I think if he was using the switching type of cores and the frequencies were much lower, the capacitor’s effect would be much less, and the improvement may not be worth adding the capacitor. In my own experiments, I have found that when I add the capacitor (depending on its value), it sometimes helps a small amount, or sometimes it reduces the LED light output. I have used values from a few tens of picofarads to over 1000 picofarads, and I don’t find that it is of much help.
Update 12-24 at 7 PM: I wound a low permeability core with 12 turns of 24 AWG wire and measured the inductance at 8.9 microhenrys. I connected this up to a 2N3904 (same as he used), a 1k resistor and a white LED. I measured the frequency and got 780kHz, which is ten times higher than a typical JT and is much too high, it’s in the middle of the AM broadcast band. The supply current at 1.5 volts was about 45 milliamps, which is low for a JT but typical of a 2N3904.
One other factor that I’ve thought about but found very little info on is the response of the phosphor used in white LEDs. The white LED is actually a blue LED but the chip is covered with the phosphor that converts part of the light from blue to the other colors, which then looks like white light. The blue LED can respond very quickly to the electric pulses, putting out similar pulses of blue light. But what happens when the blue light hits the phosphor? From what little I’ve read, the phosphor is much slower and cannot respond quick to the light pulses. Is it possible that this phosphor’s slow speed is responsible for the loss of efficiency as the JT’s frequency increases?
Note: The 1N5817 Schottky diode is an excellent 1 amp rectifier for high frequency power, but it is not commonly available at the local electronics stores. So the neophyte, not realizing how important the recovery time is, decides to substitute the 1N4002 1 amp rectifier, or the 1N4148 diode. The 1N4002 and similar has the slow recovery problem and this will cause excessive losses. The 1N4148 is not made for high currents and the forward voltage drop will be much higher and the losses will be much higher. In both cases, the benefits of using the 1N5817 are lost and it would most likely be better if the LED was used alone with no rectifier.
Back to experimenting…