Glowing mercury thyratrons: inside a 1940s Teletype switching power supply
righto.com
righto.com
As someone pointed out, that's a switching voltage regulator, not a switching power supply. The transformers there are all upstream of the switching.
I've restored five Teletype machines like the OP's Model 19 [1], so I've needed similar 120VDC 60mA power supplies. So I designed my own switching power supply.[2] This has a USB port for input, and a 120VDC 60mA output for directly driving the Teletype machine. It's powered entirely from the USB port.
This seemed impossible to some people. There's only 5V at less than 500mA coming in, and 120VDC 60mA out. But it's not impossible, because the load is inductive and intermittent. The selector magnet in old Teletypes has a huge inductance, about 5.5 Henries. (Not mH, H.). The 120VDC is only needed for about the first 1ms of each bit time, to force current through that huge inductance. By 5ms or so, you only need about 6V. So you can charge up a capacitor to get the initial 120V, then let a sustain supply take over.
My design is totally modern, built from surface mount components and in a small case. Here's the schematic.[3] There's an explanation in [2].
It's been amusing to see the reaction of the Teletype community. They like it, but most can't solder surface mount. One hobbyist is making these things for others. I put the design on Github as open source and made a few for myself, and I've sold some board kits. Not enough potential volume to have it manufactured.
Informally, here's how a switching power supply works. Everywhere else in electronics, you try to get rid of spikes. In switching power supplies, you make and use big ones. You start with a source of DC power, and you hook that to the primary winding of a transformer, with a switch so you can turn the power on and off. You turn the switch on, and current flows into the transformer. The magnetics in the transformer charge up, storing energy. After a while (milliseconds) the magnetics will saturate, and can't store any more energy. You now have a short circuit, DC going through a low-resistance transformer. But you turn off the switch before that happens. (Switching power supplies are always milliseconds from burnout, which is why they burn up if the switching fails.)
When you turn the switch off, you now have an open circuited inductor. The energy in that inductor has to go someplace. It comes out as a huge spike, in theory infinite voltage if the transformer resistance was zero, and in practice it can be a few hundred volts. It can't come out the primary, because the switch is open. So it comes out the transformer's secondary winding, where it's fed through a diode into a capacitor. There's the output.
It's simple. An old-style auto ignition with a coil and breaker points works this way. The problems come in as you make it well-behaved. First, controlling the switch is complicated. You want to open the switch before the transformer hits saturation. Failure to do this will burn something out. So there's usually current sensing. Then you want to turn the switch back on when the output voltage from the inductor drops below the voltage in the output capacitor, because no more current will flow through the diode after that.
That just makes it output power. Then you need output voltage sensing, which shortens the charging time to reduce output to maintain the desired voltage. You need protection to shut everything down if the switch gets stuck. (MOSFETs tend to fail in the ON state, and lack of good protection circuitry causes fires.)
This thing works by making big spikes at a few hundred kilohertz. That makes it a radio transmitter. You need inductors and bypass caps to prevent it from blithering all over the RF spectrum. Or sending spiky noise to its output or input. The bypass caps and inductors need to be close to the source of the spikes, so PC board layout really matters. These things will not work on a breadboard.
All this is why switching power supplies have so many small parts. Once you get it right, they work beautifully. Very high efficiency and low heat.
[1] http://www.aetherltd.com [2] https://github.com/John-Nagle/ttyloopdriver [3] https://raw.githubusercontent.com/John-Nagle/ttyloopdriver/m...
Can't vacuum tubes such as krytrons still provide the ability to switch vast amount of current in the order of kiloamperes at kilovolts, which would be hard to use semiconductors for?
Apparently they were used in atomic bombs for delivering power to exploding-bridgewire detonators iirc.
Their main claim to fame is switching speed. I imagine that GaN FETs have finally rendered them obsolete, or will soon. Measuring these things is not entirely trivial ( https://www.edn.com/design/test-and-measurement/4437360/How-... ).
For krytron's specifically it says 'These devices have maximum voltage ratings from 3 to 10KV, and peak current rating of 300-3000 amps' which would have been useful for detonating the exploding-bridgewire, which requires a hefty current.
Also I believe modern versions such as the following Thyratron https://www2.l3t.com/edd/pdfs/datasheets/L4906A.pdf are used with lasers. Peak current 20,000A. Peak voltage 35kV.
I figured it would be controversial to call the power supply a switching power supply, but I haven't seen a solid reason yet to exclude it. Putting a transformer upstream of the switching makes it an on-line power supply as opposed to an off-line power supply (dumb names, but I didn't make them). I'm not sure how you're distinguishing between a "switching voltage regulator" and a "switching power supply".
A switching power supply has actual power conversion driven by the switching. See The Art of Electronics, by Horowitz and Hill, 3rd ed., section 9.6.
That thyratron circuit is a lamp dimmer circuit repurposed as a voltage regulator. Like SCRs and triacs, a thyratron is a switch you can't turn off. You have to wait for the input power to turn off. Usually that's the next cycle of AC. Because you can't turn off the power, you can't generate inductive spikes, so you can't pull the basic trick that make a switching power supply go.
There were some real switchers in the tube era. The most common one was used to generate the high voltage (10KV-20KV) for CRTs. Those used the horizontal oscillator, running at 15KHz and yanking the beam back at the end of each scan line, as a spike generator. That approach used a real vacuum tube, not a cold-cathode gas filled tube like a thyratron. So those were high voltage but low current devices.
Switching power supplies with serious current output had to wait for a component that could turn off fast under load. Power MOSFETS, etc. If you could do that with a thyratron, we would have had switching power supplies by 1950.
So what's a charge pump? Chopped liver?
Finding the right MOSFET was a huge headache. I kept trying reasonable ones in LTSpice, then on a real board. Gate capacitance of the MOSFET really matters. You normally think of MOSFET gates as drawing nearly zero power, but in a switcher, you want to turn them on fast, which means pumping in almost an amp for a few nanoseconds.
That IC is intended for charging up photoflash units. I'm using it at a lower voltage with a lower capacitance but a faster cycle time. The examples show charge times around 1 second; I only have 22ms. I'm only charging 2uF. It's two 1uF ceramic caps; none of the current limits of electrolytic caps. Not surface mount, though; I tried those new ceramic multilayer surface mount capacitors, but they have some very strange properties; the capacitance declines with voltage. OK in filters, terrible for energy storage.
It took me seven boards to get this working. The first few were a much simpler design with a 555 timer running the switcher. It worked, but it turned out I needed 2uF instead of 1uF because the Teletype selector magnet has an inductance larger than the ham community thought it did. The simpler design couldn't charge up 2uF in 22ms (one bit time); it took about 30-35ms. I had to start over with a more efficient design. This is all running off a USB port, so there's a limited power budget.
Running off USB port power added of complexity. There are strict rules about drawing power from USB ports, and if you violate them, even for a microsecond, the USB port turns off, and on many devices stays off until power cycled. That's a good thing; it's why hot-plugging works. The AP2553W6 manages startup current draw and has comparators checking for abnormal situations. It has a reverse current flow detector. Spikes from the switcher made it back into the power input and tripped the AP2553W6's protection. Had to add another surface mount ferrite bead, L1, to damp out current spikes. A lot of switcher design is about putting small capacitors and inductors in the right places to damp out trouble. Most of this you can see in LTSpice.
What you don't see in LTSpice is the effect of board layout. LTSpice assumes idealized wires with zero resistance, capacitance, inductance, and coupling. The layout around U1 really matters. First time around, I didn't follow the recommended layout, and the system would not oscillate.
The LTSpice model is on Github, along with the KiCAD files, so you can play with the parameters.
[1] http://www.analog.com/media/en/technical-documentation/data-...
But a windmill is mostly coils and magnets and when you short it it actually will charge the battery, briefly robbing the blades of some momentum. So, if you periodically short the coils using a bunch of powerfets and an oscillator you can charge a windmill in low wind conditions that would otherwise do nothing.
So far so good. Built it, tested it, worked like a charm. And then one day I decided to temporarily decouple the windmill from the switchboard, but I had forgotten about that little booster. FOOM, instant fire on the booster board after I pulled the switch. The FET circuitry and the oscillator had happily continued to work on the power provided by the windmill, and had allowed the end stage of the circuit to reach a very large multiple of the voltage that it normally dealt with because the battery kept the voltage pegged to a maximum of about 48V!
Needless to say that led to a somewhat more robust V2...
I was shocked when a docent walked over and turned it on. The rectifiers lit the whole room up with a beautiful purple glow. Beautiful.
Doing a quick web search it looks like the museum is no longer. Too bad - even though I couldn't read the plaques, it was still one of the most fascinating museum experiences I've ever had.
I rarely see this in common expression. Nice choice.
Old tech has great appeal. Scale is our size, and features forces, artifacts we can see, hear and feel easily. I can imagine that glow. Damn cool.
It is not all that widely used in the US either, in my experience anyway.
There is also a particle "do" which means "into" and also "up/until". In Czech, the opposite of this sense is "přes", meaning "over/beyond".
So this is the basis of an old joke that "docent" means someone weighing up to 100 kg; above that they become a "přescent":
https://cestina20.cz/slovnik/prescent/
:)
[1] https://www.flickr.com/photos/trained_4_life/10978575914
See here for some background of their early days and photos : http://www.bbceng.info/Operations/transmitter_ops/Reminiscen...
No glow from the mercury rectifiers though - they're steel cased thanks to being 600kW
They operate many of the transmitters each weekend. They applied to the FCC for a commercial coast station license, probably the last one, and got it. So they can send at quite high power. They talk to museum ships around the world.
I wonder what other EM ranges it emits. Glowing purple makes me think of UV but maybe also some X-Rays?
As for x-rays the typical voltage drop across a long tube is like 100 volts, the whole point of a ballast is matching the normal hundred or so volt drop to whatever your local line voltage is using at a constant controlled current. Anyway the penetrating power of an xray is linked to the voltage drop, so it would be difficult even theoretically to generate xrays stronger than a hundred volts or so. So given that it takes 50 KV to 100 KV for a dental xray source to beam completely thru a head, you're looking at like a thousandth the penetrating power. Very hand wavy estimate that 100 volt xray source will likely not penetrate the glass and almost certainly not penetrate clothing or the surface layer of dead skin and skin oils.
The risk assessment for being nearby a heavy cloud of vaporized hot mercury would seem to imply the main risks would be biochemical in nature not physics in nature. Actually the most realistic risk is an economic death penalty if that leaks and requires hazmat cleanup.
Common beam power tubes and power pentodes sometimes glow blue from residual gas in the tube even with plate potentials ~500V.
[1] https://www.e2v.com/shared/content/resources/File/documents/...
You can see the idea is to run a timer that is phase locked to the incoming mains frequency. Now you can control when the switches turn on during a half cycle. If you turn the switch on at the beginning of a half cycle, you have full power. If you turn the switch on midway through the cycle, you get half power and so on. So your command signal asks the controller for more and more of the complete wave cycle until you get the full RMS value of the line voltage plus the load current (minus the losses in the rectifiers of course). The resultant output is a chopped up EMI laden mess but it does the job quite nicely after some filtering.
They were also used in early motor drives to control the speed of a brushed DC motor from single or three phase AC source. I know some lathes from the 50's or 60's had thyratron motor drives in them.
At my work we have an Electron beam welder which uses an SCR controller for the high voltage power supply. It's interesting: the controller is directly fed 480V three phase. From there in comes in through a breaker, a contactor, and two current sensing transformers.Like so (dammit variable width fonts...):
A--~~--||--S--^^^^^^--/--SCR---\---)
B--~~--||-----^^^^^^--|-BRIDGE-| )Inductor
C--~~--||--S--^^^^^^--\----------/---)
Key: ~~ fuse, || contactor, S current sensor, ^^^^^^ transformer primary
The three phases then run to the power three supply transformers in series and then off to a three phase SCR bridge for a total of six SCR's. The output of the bridge has a huge 200 pound inductor across it. The idea is the bridge is phase fire controlled and the inductor is so high in value that the controller can slowly watch the current ramp when the SCR's are turned on and wait until the feedback from the power supply matches the command from the potentiometer and adjust the phase angle firing accordingly. It's creating a controlled short circuit using the series transformers as the load. It's a primitive solid state method of varying an AC voltage. Before the SCR system they used a motor generator with an op-amp PID loop watching the feedback and control pot who's output controlled a small phase fired SCR bridge that delivered a varying DC voltage to the generator field winding. You effectively had a motor generator who's output varied from 0-480V AC three phase. Today you'd have a small metal oil tank containing an entire SMPS which is smaller than the control cabinets for our old linear supplies.
It turned out that mercury rectifiers make a lot of electrical noise when they switch, so there was an annoying 120 Hz buzzing noise. But it did look cool.
Cool project! Do you say you can't bring yourself to work with them for safety reasons (as in, I imagine mercury vapor is very dangerous if it escapes the tube) or is there another reason?
Using SCRs or triacs to pre-regulate the voltage of a linear regulator was a very common technique well into the late 80s for high power, precision power supplies. The power supply shown here works very much the same, except there being no linear post-regulation stage.
Contrast this with the HV generation in devices using CRTs, which early on (~50s) started to use high-frequency converters, usually of the resonant kind.
https://youtu.be/k15pWPBNAUE?t=45s
https://www.garagejournal.com/forum/showthread.php?t=264474
These were made from 1939 through at least the 1990s, although the vacuum tubes went away in the mid-1980s. Amazingly long-lived product.
This is more like a vacuum analog of the SCR.
Switching on and off based on phase angle isn't the same thing as the working principle in SMPS's.
An SMPS, in a nutshell, uses pulses of current to "charge" an inductor, which continues to source current during the off periods when the current pulse is cut off and the magnetic field is collapsing. (Inductors oppose changes in current.)
There is also non-inductive SMPS using capacitors only: the charge pump.
That's exactly what happens in this power supply. The tubes charge up a giant (grapefruit-sized, if grapefruit were cubes) inductor, which supplies the current when the thyratrons aren't.
I don't see a good reason not to consider this a switching power supply. Phase angle is just PWM at a lower frequency.
You certainly couldn't call it a linear regulator, because the thyratrons are either fully on or fully off at all times. From the input's point of view it will exhibit negative resistance, with a reduction in the current required as the input voltage rises. That makes the classification especially hard to refute.
It's true that energy is being stored in the smoothing choke, but that alone doesn't make the difference since the choke could be used to accomplish the same thing in a traditional linear supply. The choke isn't inside the feedback loop, though, and that is a point in favor of the "It's not a switcher" camp. But it's the only one. In any event, charge-pump supplies don't need an inductor at all, and nobody questions whether they operate in linear or switched-mode.
(Edit: looking at the schematic, the filter choke is indeed inside the feedback loop. Case closed, it's a switching supply.)
https://www.youtube.com/watch?v=QY6V2syGnZA
Fortunately he didn't "pop" that one "for science" unlike most other things he covers, it's too valuable :D
One of the cool things about that era of Teletype was the mechanical serial to parallel converter. This was a multi-lobe camshaft that as the pulses came in and the electromagnet selector engaged/released, would (if your timing was set right) select five bars (for your Baudot code) that controlled the position of the type box where all the letters and numbers were.
That's partly why I built a switching supply interface. Far less heat. Small box, no ventilation required.
I saw something like this in a railway museum in Yorkshire or thereabout when I was a kid - I recall its label included 'rectifier' and I think it contained mercury, and was flashing or sparking in some way. I guess it must have been one of the mercury rectifiers as already mentioned, though I don't remember it glowing quite so beautifully. Certainly had me intrigued.
This seems like an odd statement, given the image directly next to that sentence (and all evidence from any electronics device I've ever taken apart). The transformer is the largest piece, followed possibly by the inductors. IANA electrician, though, so maybe there's some context that I'm missing (i.e. a transform isn't technically a single component)?