RustyBolt.Info/wordpress

I watched this YouTube video.  I liked it, it’s a well made video.   

The author asks a question about the 9 min point.  He was not sure why the voltage peaked out at about 40 volts, and then stayed the same.

First, it is not wise to disconnect the LED load from the JT because it causes excessive voltages and can damage the transistor.  There is no information worth having by causing the collector voltage to go so high that it causes the collector to emitter breakdown to be exceeded.  During normal operation with a connected LED, that collector voltage will never exceed about 5 volts.

We must remember that the two windings are equal – their ratio is 1:1 or 1 to 1.  That means whatever voltage appears on one winding also appears on the other, and in this case it’s opposite polarity because the feedback winding is used to invert the signal.
The transistor’s emitter to base junction is not supposed to have a reverse voltage higher than 5 volts.  During normal operation with a LED this voltage should not be higher than 5 volts.  But in this case, without the LED, the voltage is much higher than that, or it tries to go higher.

At the same time the emitter to base junction reverse voltage is increasing and when it gets to about 8 volts, the emitter to base junction will start to break down.  If this happens even for a short while it will permanently damage the transistor, and its current gain will be permanently reduced.  The breakdown can be monitored with a current sensing resistor in the base lead.  Both A and B channels are connected to this resistor and the ‘scope’s A minus B mode is used.  Normally there should be current in the forward (conducting) direction *only*.  If there is a current spike in the reverse direction, this means the E-B junction is breaking down.

The reason that the no load voltage increases and then hits a plateau is that the E-B junction is breaking down, and wasting the power that would normally go to the LED.  That’s bad for the transistor.

A few other things that I have confirmed by building and experimenting with more than a hundred Joule Thiefs:

The higher the Q of the coil, the less loss there will be and more power will get to the LED.  A good point of compromise between inductance and wire resistance is about 100 microhenrys.  A ferrite gives the best performance because the coil has to only be around a dozen or two turns on a ferrite toroid core made with type 43 material.  A good, cheap core is the Fair-Rite 2643002402 “suppressor bead” available from Mouser for about 12 cents apiece.

The transistor makes a huge difference in the maximum power the JT will put out.  The 2N3904 and BC547 types have a maximum of 100 mA collector current, and are not capable of lighting a LED to its maximum 20 mA current.  Instead use these transistors.

BC337-25,  PN2222A,  2N4401,  SS8050

or any transistor that has a high current gain at 500 mA or more.  There are special transistors made for this switching purpose, such as 2SD5041, KSD5041, 2SC2500, NTE11(too expensive), and equivalents in the surface mount package.

Part 3 of YouTube video 

I watched Part 3 where he compares 5 different JTs.  I say different because he didn’t control for all the variables.  Here’s why.

Starting off with the concept of five different JTs, this whole comparison is inconclusive because there are too many differences in variables between JTs to be able to compare them.  The only thing he is changing is the number of turns on a toroid.  Therefore in order to control for the type of core, the transistor, the LED and most importantly the resistor, *all* of these should be the same identical part.

The toroid cores from the same batch may have as much as 20% tolerance in their inductance.  The same with transistors.  The LEDs are not critical because the actual brightness is not being measured.  But all 5 of the resistors should be the standard JT value of 1k ohms, and should be matched.  Using variable resistors makes the measurements meaningless – the battery current can be set or mis-set to any value that he chooses.

All 5 of the batteries have to be matched for their capacity in mAh.  But why is the battery discharge time factor measured?  If he used 5 batteries, and each of the 5 JT circuits were identical and drew the same current, he would have found there were different discharge times for each battery, because the capacity varies from cell to cell.  Instead the battery current should be measured, and the discharge time of the battery should be separated, and not a part of this experiment.  Also, other authoritative sources say that the Ni-MH cells should never be discharged below 1 volt because it damages the cell.

With the transistors, he matched their current gains.  But the transistors have a different current gain at the high currents found in a JT.  The JT does not operate in the linear (amplifying) region, it is used as a switch, it’s either fully on (saturated) or fully off most of the time.  The current gain is not as important as the Vce(sat) which should be as low as possible.  

He should have monitored the currents in the supply and LED.  You set up a single JT with a 1 ohm resistor in series with the positive battery lead and a 1 ohm resistor in series with the cathode of the LED.  Measuring the DC voltage across these resistors, 1 millivolt across 1 ohm gives 1 milliamp current.  So it’s easy to monitor the currents with a DMM on the millivolts range.

I’m very concerned about using 5 different cores.  What he should have done is wind ten turns center tapped, then bring the leads out to taps.  Then wind another ten turns, and bring the leads out to taps.  And so on.  So the number of turns on the same core can be changed by moving clip leads to different taps.  This eliminates differences in the cores because the same core is alway used.

After building more than a hundred JTs, I’ve found that a good compromise between inductance and wire resistance is a high mu ferrite core with enough turns to give about 100 microhenrys for each winding.  The frequency of oscillation isn’t critical, but will be about 70 kHz with a 1k resistor.

Using the standard JT circuit, I have never been able to get more than about 65% efficiency, and typical efficiency is about 50 to 60%.

I have a lot more information about the Joule Thief in my blog

rustybolt.info/wordpress/

My post to FB Aug 14

Does anyone know about the timelines on what I’m talking about below?

The early TRs (transistor radios) were AM only because the germanium transistors available were not capable of FM frequencies.  The first TR was the Regency, and it used the Texas Instruments transistors that were similar to the Raytheon CK722.  I’m guessing this was ’54.

The early germanium transistors had enough collector to base capacitance to make it necessary to use neutralization in the IF amplifier(s) to prevent oscillation and get enough gain.  Not sure when, but the transistors (Ge or Si??) became lower capacitance and didn’t need to have so much neutralization, so they could be more easily assembled.

The cheap epoxy silicon transistors replaced germanium, and they could work adequately on the FM BCB, so AM/FM radios finally hit the market.  National Semi and Fairchild Semi were leaders in low price ‘jelly bean’ transistors in the 1960s, but I’m not sure exactly when.

Sometime after that, maybe the ’70s, the SAW filters replaced IF transformers so the IFs didn’t require tuning, thus saving on assembly costs.  The radios still required some alignment in the front end.

Eventually the radios became more integrated, more functions were done by a single chip.  I don’t remember what year I started seeing chips in radios.   I think the IF was changed to where it and the detector were using modern technology, like synch detection.  Tuning became digital so only a few buttons were needed.

From  FB comment 

More than 3 decades ago I took programming classes using a mainframe and minicomputer.  The error messages were numerous and cryptic, and those were for just a single mistake, such a missing right parenthesis.

One would think that since then, the compilers would be more helpful, but it doesn’t seem so.  I’m glad my career was in the hardware side of computers!  😵  👎

I’m not a software geek.  But it seems to me that a compiler could do better than give an error message and quit.  With today’s “Artificial intelligence” and “Fuzzy logic”, it seems to me that the compiler could learn what mistakes are common, and when an error occurs, it could give a list of possible solutions.

From FB comment

I got an ultrasonic cockroach ‘repeller’ from someone who had it on the kitchen counter. I opened it up and found it was full of cockroach carcasses.  Looks like it attracted cockroaches instead of repelling them.  They probably sprayed real cockroach killer in it. Since it was obvious that it didn’t work, I took it apart and dumped the infested case, but kept the PC board. It was labeled…

PANASONY

stupid maker would’ve had not one, but TWO corporations on their back for copyright infringement. Duh!