Overview
This board is an experiment in making a wireless sensor actually behave like a low-power device. Putting a coin cell behind a Bluetooth MCU is easy. The design is low power only if it spends almost all of its time asleep, wakes for a real reason, and can still get a short packet on the air without the cell collapsing.
The radio is the part that is easy to name. Sleep current, the recharge of a local capacitor bank, the way the accelerometer wakes the system, and the boundary between power domains are the parts that decide whether the board is genuinely low power.
What I was trying to build
A small sensor on a CR2032. An nRF54L15-class Nordic part for the radio and the MCU. An ADXL362 as the accelerometer. The intended life of the board is: stay asleep, wake when there is something to notice, advertise a small amount of information over BLE, and go back to sleep.
The current direction is advertising, not a connection that stays open. The advertisement is roughly periodic, on the order of once a minute. That interval is a design direction, not a measured duty cycle from a finished unit.
The ADXL362 is here because its standby and activity-detection power are low enough that the MCU should not be the thing polling for motion. If the microcontroller has to stay awake to watch the sensor, the coin cell is already being spent on the wrong job.
The radio pulse
A CR2032 can support a small average current. It is a poor source for the short, sharp current a BLE advertisement draws. The cell’s internal resistance turns that pulse into a voltage sag, and a sag in the middle of an advertisement is how a board browns out while it is trying to talk.
The current approach puts a local capacitor bank on the rail that has to supply the pulse. The battery recharges the bank between events. It does not have to deliver the whole radio transient itself. That trades one problem for a few more specific ones, which is the point of doing it on purpose.
The bank has to recover before the next advertisement. The current that recharges it has to be something the cell will actually provide. The voltage during the pulse has to stay high enough that the MCU does not reset. A series resistor between the cell and the bank is one way to limit that recharge current. More resistance is easier on the cell and slower to refill the bank, and it drops voltage, so the next pulse starts from a lower point. I have not closed on a value, and this page does not invent one.
The bank also has to be kept from backfeeding circuitry that is not supposed to run from it. A charged capacitor is a supply. If that path is left open, parts that should be off stay biased, and the sleep current stops being the number the power budget assumed.
Still being checked, not finished: recharge time between advertisements, whether series resistance belongs between the cell and the bank, voltage sag during the radio event, and backfeed out of the capacitor bank. No sleep-current or battery-life figure is listed here, because I do not have one I trust yet.
Waking from the sensor
Motion is the accelerometer’s job. The ADXL362 can recognize activity and inactivity and interrupt the MCU, so the microcontroller can stay asleep until the sensor has already decided something happened. The MCU then does the small piece of work it was woken for, advertises if there is something to send, and sleeps again.
The periodic advertisement and the motion interrupt are different wakes. They are easy to tangle. A timer that brings the whole system up “just to check” will own the average current long before the radio pulse does. The reason to use this accelerometer is that most of those checks never need the MCU.
Around the chip
The schematic and PCB are being drawn around the Nordic requirements for this class of part, not around a black-box module pinout. That includes the external oscillator and the parts around it, the decoupling the reference design actually calls for, and SWD brought out with the reference voltage the debugger needs. Footprints are chosen because the board has to be assembled: parts that can be bought, land patterns that match them.
I am not publishing a measured sleep current, a radio range, a board outline, or a battery-life estimate. Those are the results that would say the experiment worked. They are not in hand.
Where it stands
In progress. The open questions are the ones above: asleep current, capacitor recharge, sag during the advertisement, and whether the accelerometer is really the thing watching. The ESP32 platform is the other board, and it is a different constraint. That one is about a wide input and connectors. This one is about a coin cell that cannot afford to be awake.