Looking at STMicroelectronics’s appnote AN392, “Doc ID 1863”. It’s a little bit shocking, though hopefully not literally; it contains a very cost-optimized, compact power supply circuit for microcontroller projects that can be kept safely away from the delicate bodies of the humans.
Quoting Horowitz and Hill:
Never build an instrument to run off the powerline without an isolating transformer! To do so is to flirt with disaster. Transformerless power supplies, which have been popular in some consumer electronics (radios and televisions, particularly) because they’re inexpensive, put the circuit at high voltage with respect to external ground (water pipes, etc.). This has no place in instruments intended to interconnect with any other equipment and should always be avoided. And use extreme caution when servicing any such equipment; just connecting your oscilloscope probe to the chassis can be a shocking experience.
Consequently they do not explain how to build such consumer electronics (though they promise that chapter 9x will.) This appnote is about how to do it, a topic also occasionally discussed in the YouTube channel of bigclivedotcom. So how do they suggest doing it?
They’re running an ST6210 8-bit microcontroller directly off the hot side of mains current, connected to the touch sensor at the human’s finger through two or three 4.7-MΩ resistors (thus at 340V peak you get 36 microamps or less, not enough to feel). The 5-volt rail is literally the hot side of the power line (though fused), but the “ground” is floating; a simple regulated capacitor dropper produces about 5 volts on the ground rail as follows. The neutral side of the power line is connected to an 820Ω half-watt resistor, which is connected to a 220-nF 400-V capacitor, which is connected to the low side of a 5.6-volt zener diode, whose positive end is connected to the 5-volt rail. A 1N4148 clamps the ground rail to be no more than 0.7 volts above the low end of the zener, and there’s a 100-μF 10-volt energy storage capacitor connected between the ground and the 5-volt rail.
(This appnote is from 1998, so this is not capacitive touch sensing; instead, the human’s finger forms part of a voltage divider between the hot and neutral powerline rails, protected by the megohms.)
This is not quite a standard capacitive voltage divider circuit.
The 220-nF capacitor produces a 12–15-kΩ reactance at 50–60 Hz, losslessly limiting the current to 20 mA by itself; I assume the 820-Ω resistor is to limit inrush current, because it has ≪1% effect at powerline frequencies. At steady state, with the storage capacitor charged, this current sloshes back and forth across the zener (remember that the high end of the zener is directly attached to the hot side of the power line), but when the zener is reverse-biased and the capacitor is discharged, there’s an easier return circuit path through the capacitor and 1N4148, which initially only has a 600-mV voltage drop before starting to conduct heavily rather than the zener’s 5.6 volts. Once the capacitor is charged, though, the zener becomes the easier current return path, and the return current flows through it instead.
20 mA through 820 Ω gives you 330 mW, which is why you can’t use a ¼-W resistor. The zener is also going to dissipate 55 mW. An ideal dropper capacitor wouldn’t dissipate anything but real dropper capacitors will have some dielectric heating from the constant 20 milliamps, or 10 milliamps at 120Vrms; this will probably be on the order of 1 mW. The rest of the circuit should have much smaller power consumption.
They say their board uses 3 mA; the 100-μF storage capacitor will then discharge at 30 V/s, or 0.6 V in the 20 ms between 50-Hz peaks.
So with five simple passive components and no electromagnetics you get a regulated 5 volts directly out of the power line. It’s just not safe to touch the circuit while it’s turned on, or to connect it to any other circuits, and the dropping capacitor is a bit of a beast. In Capacitors: some notes on tradeoffs I have some pricing info on smaller, totally inappropriate capacitors; the 19.4¢ Nichicon UWT1H470MCL1GS is 47μF and 50V, so you would need strings of 8 of them to get the required 400 volts, but each such string would only have 6 μF, so you would need 38 such strings in parallel for a total of 304 capacitors costing a total of US$60. Capacitors designed for such applications would probably be smaller, cheaper, and cooler.
You probably don’t need a bleeder resistor across the dropper cap (and the ST appnote doesn’t show one); if you unplug it at the wrong part of the cycle, it could have 340 volts across it, but since it’s only 220 nF, that’s only 18 mJ, not enough to be dangerous to a human. It’ll blow holes in your MOSFETs, though.
Unlike a standard capacitive voltage divider, the output voltage barely depends at all on the input voltage or the dropper capacitance. The dropper capacitor just serves to limit current and thus power dissipation. The output voltage is determined by the zener, or more precisely the zener minus the rectifier diode’s forward voltage; the only relationship to the input voltage is that the output can’t be more than the peak-to-peak input voltage (minus a couple of diode drops) and if it gets close you will have less current draw. Moreover, this power supply always draws very nearly the same current, whether anything is running from it or not, unless the current load is so high as to substantially drop the voltage in the power-storage capacitor.
This means, of course, that its efficiency is always terrible: as bad as 0% (when the load is turned off) and never much more than 30% or so.
The appnote isn’t about regulated capacitor droppers; it just mentions in passing that “the board supply comes from the mains through a simple RCD circuit”. The appnote is actually about controlling triacs (“the least expensive power switch to operate directly on the 110/240 V mains”); it recommends using a microcontroller to inject a turn-on pulse at the appropriate point in the cycle. In the appnote ST recommended a BTA 16-600CW triac for motor control so as to need no snubber, but it needs 60 mA to trigger it. The more common (?) 95¢ T405Q-600B-TR I mentioned in My attempt to learn about jellybean electronic components would work and only needs 5 mA, but would presumably require a snubber.
So in ST’s appnote the inefficiency of this power supply is insignificant: if you’re controlling a 700-watt vacuum cleaner with a 15-mW microcontroller, it hardly matters that you’re burning 200 or 400 mW to get a regulated power supply, as long as you don’t have dozens and dozens of vacuum cleaners plugged in for every one you’re using.
In Notes on the STM32 microcontroller family I calculated datasheet power consumption for a number of STM32 processors; even without using power-down modes, a number of them would run at under 50 μA at 131 kHz, and with power-down modes you could reasonably reduce power consumption by another factor of 1000, although as mentioned in Can you bitbang wireless communication between AVRs? How about AM-radio energy harvesting?, it’s easy to leak multiple microamps through your bypass capacitors. This suggests that 10 μA might be a reasonable current to design an embeddable powerline-powered IoT device for. (As long as it doesn’t have to be controlling a triac or something, anyway.)
Adapting the above design for the lower current level, though, the dropper capacitor could be 220 pF instead of 220 nF, the storage capacitor could be 0.1 μF instead of 100 μF, and the input resistor would, I think, be unnecessary. Using the C = εA/d formula for capacitance, εr of 3 and an egregiously large plate separation of 1 mm, you could get the capacitance you need from 83 cm² of foil “shielding” wrapped around a plastic-insulated electrical line. (3 is a reasonable guess for many plastics.) That’s a rather large chunk of foil, though, and a better option might be to run a high-efficiency buck regulator off a smaller piece of foil instead of just regulating with a zener.
If you’re going to try to go with such a capacitive connection, you might want to do it on both the hot wire and the neutral wire, despite this requiring four times as much foil. That way, you have no dc connection to the powerline at all.
In theory at these current levels you could use reeally thin wires. In Balcony battery I estimated that 142-μm copper wire would probably work for five-amp fuse wire. Suppose that’s correct. Here the wire needs to carry ten thousand times less current, so it can have ten thousand times less surface area per resistance, which means it could be 21 times narrower and thus have 21 times less surface area and 441 times more resistance. But that would be 7-μm copper wire, which is going to be hard to find and maybe even a bit dangerous to handle. If we take copper’s resistivity to be 16.78 nΩ·m, as in Executable scholarship, or algorithmic scholarly communication, that wire is 436 Ω/m.
I’m pretty sure a 220-pF 400-V X7R ceramic capacitor would be a few millimeters in size and cost under a dollar, and that’s a much better option than meters of foil snaking around your conduits and junction boxes. But it means that you need a direct non-galvanically-isolated electrical connection to the mains power and thus a fuse.