Switching-Mode Power Supplies

I've always liked inductors and magnets and stuff, so this is somewhat a natural extension of that. But it's so much more useful! My goal is to create a basic SMPS design which just about anyone can build using scrap (or cheaply available) parts, is easy to understand, and provides cheap power conversion. It can be 12V unregulated to a regulated 5, 6.3, 15, 30...300...500V, or any combination thereof; my primary goal is 12V input to 300V 100mA so I can run a tube amp from a car. (A tertiary 6.3V or 12.6V for heater power would be a plus.) A secondary (and more dangerous) goal would be same output from 160V input, that is, rectified line voltage. No bulky power transformer!

So?

The theory being a DC input (be it from a battery or rectified source) is fed to a fast switch (since mechanical switches are notoriously incapable of 200,000 switch closures per second, we'll be using transistors) connected to an inductor or transformer. On the other side of the coil is a rectifier and filter to convert the power back to DC. I'll skip the represenative diagram (tons of informative stuff to be searched for!) and go to my first design.

Of course there's the basic self-excited oscillators, but I don't imagine them as easy to fiddle with. I prefer something a little more open-loop, driven by an external oscillator. Voltage or current-controlled frequency or duty cycle is a plus, allowing regulation. So after I got an oscilloscope and more theory and experience with electronics (the above circuit is like...old) and older and.. I threw together something like this:

This is a little more advanced than my first similar design (to which I can't find the schematic) but not by much. It goes like this: +15V supply powers all, except for the oscillator whose power is smoothed by the 100 ohm resistor and 220μF cap. Oscillator is your classic multivibrator, set for around erm... 20 to 50kHz in this one I guess. (I remember most of these breadboarded beasties using 47-100pF timing caps going up around 100kHz.) Bias to one of the transistors is controlled by a seperate, variably-biased transistor to crudely control width of one half of the square wave (idea at the time being, more on-time charges the inductor more, producing more output voltage, as opposed to pure PWM which towards the high end (>80%) cuts into the flyback pulse, degrading efficiency). This allows feedback later on, but at the moment it's open-loop and user controlled. Driver transistor is controlled by osc's emitter current (plus a 180 ohm resistor to improve turnoff characteristics). A 680 ohm collector load powers the output transistor, which then switches the inductor on and off.

Yeah...working theory at the time was to charge up as much magnetic juice in the inductor as possible by clamping it between V+ and GND, then getting the transistor to turn off in nanoseconds. What then happens is the magnetic field collapses, creating a big dV/dt (calculus talk for fast change in voltage) which acts in the opposite direction as wherever it was going before.. since the field was growing around the coil before, and the voltage was towards ground, it's now shrinking around the coil, and going to spring up above +V. If it's allowed to do this quickly, it can go *very* high. Like, 500V high. With no other damping, it'll easily be absorbed by the Vce of your transistor.. so pick one with 400V at least! Oh, and as long as I've brought up dV/dt, that's exactly what causes capacitors to draw current. So if there's some C across the coil, it's going to limit the speed of the flyback pulse by drawing current away. Sources such as winding capacitance and the transistor itself are big contributors.

Now, that's all fine and dandy, so what's stopping me from 300V 100mA? Reality, for one thing... The above theory is all wrong! By which I mean only partially... ;) For instance, 300V * .1A = 30W. At 12V, this is 2.5A - more like 3 or 4A counting inevitable losses. If I assume a 75% duty cycle (i.e., if f=100kHz, on time = 7.5μS, off time = 2.5μS), average inductor current during the on period will be around 4A. Physical laws tell us that, in a perfect inductor, current rises linearly with time, rate being determined jointly by inductance and applied voltage. Graphing current vs. time, it forms a triangle, whose area is energy stored in the inductor. (Note that when current is rising, it's storing energy; when falling, it is being released.) Thus peak current (when the transistor turns off) will be twice the average, or 8A. Other factors contribute further losses. What comes up must come down, and current through an inductor is no exception; that means releasing those 8A (peak) through a high-speed diode into a capacitor. You can't tell me it's going to be easy finding suitable parts here! That's pretty hard on the rectifier diode and filter capacitor. Not only that but if they don't respond fast enough, the pulse will be over anyway due to parasitic losses caused by the high dV/dt (parasitic capacitance, eddy current and hysteresis loss in the core, etc.). This method is starting to look...very...very....ugly.

But I'm not done yet...

Enter the transformer. By putting an additional winding on the above coil, I multiply the dV/dt output without affecting that of the magnetic field. Thus losses stay low while high voltage ratios are possible - and as an added bonus, I can use an entirely seperate winding to provide isolation!

This is actually an older drawing, predating the one above. I would've never guessed it would come back to bite me in the ass, but here we are. It's basically the same as above, except for using a transformer (and a lower F, but oh well). I should add I never did build it!

I did, however, build this one, as you can see! In the above photo I'm using an IRFZ44 instead of the 2SC3164 on stilts (er.. wires) I usually use. Easier to drive with the B&K signal generator. Useful information about it: sine-triangle-square wave output, ranging some ungodly low frequency up to 5.5MHz, with various tone-burst, duty cycle and voltage controls. In this project, the frequency, symmetry, output offset and amplitude are the most useful (but they are anyway :). To be more specific, I can vary the saturation of the transistor with a combination of amplitude and offset controls. Frequency and symmetry control on and off times.

Anyways, about the circuit. Sig gen feeds transistor feeds transformer. There's still a bit of flyback pulse and ringing on the primary so I put on a zobel network (the .001μF + 39 ohm) to dampen it. This is a useful consideration since the IFRZ44's Vdsmax is all of 40V, something I'd never be able to use with the previous circuits!

The secondary has no apparent junk, it's just a rounded square wave. So despite the windings being on top of one another, there's still a reasonable bit of leakage inductance in the thing. Oh well. Secondary is tapped with a PR1005 (600V 2A highspeed diode pulled from a computer power supply) which feeds a filter cap which feeds the load resistor. At the moment the most voltage I've measured is 70 or 80VDC, all of four measly watts - but hey, I've got half the turns ratio I'm shooting for here.