"Slayer exciter" analysis

[acmq]

Looking in the Internet I found many incorrect descriptions about how this circuit works, and so made a precise analysis to see what happens.
First, solving the confusion about winding directions of the coils, it's obvious that any combination of directions can be used, if the primary coil is correctly connected:



A simulation of the circuit, with some idealizations to simplify the interpretation, results in the following ideal waveforms:



It's assumed that the output voltage is at 1000 V initially, and that the current through the resistor (that serves just for startup) is negligible. The transistor is an ideal switch.
The output voltage is practically a cosinusoid at the resonance frequency of L2 with its distributed capacitance, of the coil and of the top load.
The base voltage is a square wave, positive when the transistor conducts.
The collector voltage is almost zero when the transistor conducts, has a large peak when it ceases to conduct (this is what heats the transistor), and then becomes negative and rises a bit when the transistor is not conducting, due to the action of the transformer.
The current in the secondary coil L2 is almost sinusoidal, and controls when the transistor conducts or not. When the transistor is conducting it flows to the base of the transistor. When it is not conducting it flows through the LED. Has a small step when the transistor ceases to conduct.
The current in the primary coil is a ramp added to a sinusoid when the transistor conducts.

A (very) detailed analysis, with the oscillator used as transmitter for "wireless energy transfer" experiments: http://www.coe.ufrj.br/~acmq/tesla/WirelessEnergy.pdf

[Weston]

Nice write up! A lot of bad information about the slayer exciter online.

To improve efficiency I don't think you need something as complicated quasi-resonant snubber. You should be able to tune the coil as in Class-E, where an added capacitor across C-E of the BJT resonates with the primary leakage inductance to recycle the energy and give zero voltage turn on.

[acmq]

Nice write up! A lot of bad information about the slayer exciter online.

To improve efficiency I don't think you need something as complicated quasi-resonant snubber. You should be able to tune the coil as in Class-E, where an added capacitor across C-E of the BJT resonates with the primary leakage inductance to recycle the energy and give zero voltage turn on.

The snubber studied in the document gave uncertain results, mostly due to the approximations used. I tried the capacitor tuning the primary. An ideal simulation says that the idea should work, but in practice it doesn't. I prepared a setup using the largest coil shown in the document, using a neon lamp suspended at some distance to verify if there is some change in the output. What happened is that it's possible to observe a slight resonance looking at the collector waveform, similar to the simulation, but the output falls dramatically when any significant capacitance is put there. Investigating why.

[T3sl4co1l]

Can clamp the collector voltage by using a CT winding, well coupled (bifilar or better), with a diode up from ground to the free end of the winding.  Like the top left part of this, https://www.seventransistorlabs.com/Images/Mag_Amp2.png but self-excited, of course.

This reduces the problem from a matter of L1's leakage (considerable, because it can't be very well coupled to an L2 of any useful size), to a matter of the leakage between the ends of the winding (which can be quite good, say k > 0.98 or so?).

Alternately, if the transistors can be arranged in series symmetrically, then when both of them switch off, each end of the coil flies back towards the opposite rail, where a single catch diode can be used.  This is called a 2-switch converter, and is commonly seen in modern computer PSUs (with PFC, "80 PLUS" rating and such).  Unfortunately, driving the high side transistor requires additional circuitry.

So it's better to just use two transistors symmetrically, in push-pull -- the usual "ZVS" driver circuit.  Often erroneously called Royer, but that is commutated by magnetic saturation, not resonance; it's actually a Baxandall oscillator (yes, the one famous for an audio tone control, among many other things).  In this case, the feedback is magnetic, which is fine (just a few turn sense winding).

The snubber studied in the document gave uncertain results, mostly due to the approximations used. I tried the capacitor tuning the primary. An ideal simulation says that the idea should work, but in practice it doesn't. I prepared a setup using the largest coil shown in the document, using a neon lamp suspended at some distance to verify if there is some change in the output. What happened is that it's possible to observe a slight resonance looking at the collector waveform, similar to the simulation, but the output falls dramatically when any significant capacitance is put there. Investigating why.

It's a tricky situation, because you're trying to double-tune a well-coupled network, while modulating the impedance on one end of it (the transistor is alternately saturated (~0Ω impedance) and open).

Double-tuning is alright if you have well defined impedances; you can peak the response at critical coupling.  You still might not need critical coupling, as long as the secondary gets enough voltage at one or the other resonant peak (overcoupled).  But what does it mean, to have a source that's PWM'd between zero and infinite impedance?  You don't magically get a real impedance out of that.  Maybe you do?  I don't know, I'd have to think about it a lot.  I think it's at a point where you fundamentally must approach the hard problem of a nonlinear system, i.e. where LTI analysis is utterly insufficient.


Hmm...

So, if you make the constraints that:
1. For it to be called "class E operation", the switch load must be:
a. Resistive to inductive at the fundamental frequency,
b. Resonant, so that harmonics are reflected back (i.e., as negative current during part of the switching phase, thus recycling that reactive energy into the supply); and so, consequently,
c. Capacitive in parallel, which allows us to use semiconductor switches like BJTs and MOSFETs, with antiparallel diodes and relatively high output capacitance (Ccb / Coss).
2. Since the waveform is fixed, there isn't much plasticity, i.e., it has a low impedance.  At least near the fundamental.  Whatever's left will dominate the equivalent impedance.

That must be where the key is; the fundamental might not have much room to move, but it's all squeezed into the harmonics.  The resonant peak can be shorter and taller, or longer and flatter; it can be more or less convex on its return.  I've not actually read up on harmonic balance analysis, but this must be germ of the method, huh.

So you'd have to figure that out, and that's your source impedance.  You need the source impedance to figure out how resonant-tank-ly the primary circuit behaves.  If it's got quite a low impedance, it may not act much at all as a double-tuned network, which should be convenient for tuning purposes.

Tim

[acmq]

Can clamp the collector voltage by using a CT winding, well coupled (bifilar or better), with a diode up from ground to the free end of the winding.  Like the top left part of this, https://www.seventransistorlabs.com/Images/Mag_Amp2.png but self-excited, of course.

Again, a simulation says that the idea should work, but in practice little difference is seen, and there is a reduction in the output, worse with slow transistors. I had made a bifilar coil for other purpose, but it seems adequate for this use. The coupling is not perfect but seems good enough. The consumed power is slightly smaller with the snubber, depending on the transistor used, but it appears that the gain is at the cost of smaller output power, as the transistor continues hot. The problem with all snubbers appear to be related to the presence of the transformer. When the primary current is interrupted and diverted from the transistor some energy is drained from the secondary. I am simulating the circuits with LTSPICE, but the transistor models there have some strange behaviors, as not showing this.

With snubber:


Without snubber: