In Update 13 we said the booster whined and guessed it was a compensation problem. Guessing twice is not a method, so this time we put an oscilloscope on it. The noise turned out to have a shape you can look at — and the board is quiet now.
Noise is data. A converter that sings is a converter doing something its designer did not intend, and on a hardware wallet the same power rail that drives the display also sits alongside everything else on a very small board.
What the scope showed
The yellow trace is the switching node at 10 V per division; the green one underneath is the output ripple, AC-coupled at 100 mV per division, both at 1 µs per division. You can see the converter is not switching smoothly. It fires a pulse, and then the node rings down in that decaying wobble before the next pulse arrives. Those rings show up on the output as ripple, and they are what you can hear.
The reason is that the display asks for very little current. A boost converter running that light stops conducting continuously and starts delivering energy in separate packets, and the gaps between them fall low enough to land in the range your ear picks up. Nothing was broken — the circuit was doing exactly what a boost converter does when it has almost nothing to push.
The circuit behind it
It is a deliberately plain design: an SX1308 boost converter with a 4.7 µH inductor and an SS34 Schottky diode, 18 µF in and 18 µF out, and the output voltage set by a 10 kΩ trimmer over a 200 Ω resistor. Few parts, all of them ordinary and easy to source. On a device with no battery, running only from the USB rail, that simplicity is the point.
Quiet, at 12.03 volts
The fix turned out to be the inductor itself. We had a bare wirewound part on the board — the winding exposed and free to move. At the frequencies the converter was ringing at, a winding like that flexes, and a coil that flexes is a small loudspeaker. We swapped it for a moulded inductor, where the winding is encapsulated in the core material and cannot move the same way. The noise went with it. The board is silent now, the display runs, and the rail measures 12.03 V on the meter — close enough to the 12 V the panel wants.
Why a faint noise was worth this much attention
Partly because an audible whine is a symptom rather than a nuisance: it told us the converter was operating in a mode we had not designed for, and modes you did not design for are where surprises live. Partly for a plainer reason — a device that buzzes in a quiet room is a device people unplug, and a hardware wallet that gets unplugged and left on a desk is not being stored the way it needs to be. Neither reason is glamorous, and both are the sort of thing that gets skipped when the goal is a demo rather than a device somebody keeps for ten years.
One down. The other problem from Update 13 is still open: the USB-C works with the cable one way up and not the other, and that is where the next stretch of work goes. The prototype is still headed for Bitcoin Thailand Conference 2026.