Battery life is the specification customers notice most and teams estimate worst. A device promised three years that lasts eight months is not a minor miss — it changes the maintenance model, the support cost, and often whether the product makes commercial sense at all.
The gap almost never comes from the battery. It comes from an energy budget built in a spreadsheet from datasheet averages, when real consumption is dominated by short bursts, retries, cold weather, and a handful of components that never actually sleep.
This guide covers how to calculate an energy budget properly, where the power really goes, and the techniques that turn months of battery life into years.
How to Calculate a Real Energy Budget
Battery life is not an average current. It is the sum of everything the device does in one cycle, multiplied by how often that cycle repeats.
Break the cycle into states
List every state with its current draw and duration: deep sleep, wake-up, sensor stabilisation, measurement, processing, radio transmit, waiting for acknowledgement, and back to sleep.
Energy for each state is current × time. Summing those gives the energy per cycle, usually expressed in microamp-hours or millijoules.
Multiply by real frequency, not planned frequency
A device reporting every 15 minutes runs 35,040 cycles a year. But it also retries failed transmissions, wakes on interrupts, and performs periodic housekeeping. Retries in particular are routinely left out and can double radio energy on a marginal link.
Subtract what the battery will not give you
Usable capacity is always lower than the number on the cell:
- Self-discharge — a few percent per year even if the device does nothing
- Temperature — capacity falls sharply in the cold, and internal resistance rises
- Pulse behaviour — a coin cell can supply tiny currents for years but may sag badly under a radio transmission burst
- Cut-off voltage — the device stops working before the cell is empty
A practical rule is to plan on roughly 70–80 percent of nominal capacity being usable, and to validate the worst case at the lowest operating temperature.
Measure, then trust the number
A spreadsheet is a hypothesis. Put a current probe on real hardware, capture a full cycle, and compare. The difference is usually caused by something nobody modelled: a sensor that stays powered, a pull-up resistor leaking, a regulator with high quiescent current, or firmware waiting in a busy loop instead of sleeping.
Every multi-year battery claim that survives the field was measured on hardware. Every one that fails was calculated from a datasheet.
Where the Power Actually Goes
For battery-powered connected devices the ordering is remarkably consistent:
| Consumer | Relative cost | What to do about it |
|---|---|---|
| Radio transmission | Dominant | Send less, less often; shrink payloads; avoid retries |
| GNSS fixes | Brutal when used | Fix rarely; use cached almanac; consider alternatives |
| Sensor warm-up | Often underestimated | Choose fast-settling parts; power-gate properly |
| Processing | Modest | Compute locally to avoid transmitting — usually a net win |
| Sleep current | Small but constant | Dominates over years; audit every leakage path |
The last row deserves attention. A device awake for 200 ms every 15 minutes spends over 99.9 percent of its life asleep, so a few microamps of unnecessary quiescent current can outweigh everything else combined over three years.
The dominance of transmission is also why on-device intelligence saves energy rather than costing it: running a small model to send a conclusion instead of a raw stream is almost always a net gain, which is the practical case for TinyML and data reduction at the edge.
Design Techniques That Buy Years
Choosing the Right Power Source
- Coin cells (CR2032) — tiny and cheap, but limited pulse current. Suitable for BLE beacons and very low duty cycles; marginal for anything with a demanding radio burst.
- Alkaline AA/AAA — inexpensive and replaceable, good capacity, but voltage declines steadily and performance drops in cold conditions.
- Lithium primary (LiSOCl₂) — the workhorse for multi-year industrial sensors: stable voltage, excellent temperature range, very low self-discharge. Often paired with a hybrid capacitor to handle transmission pulses.
- Rechargeable Li-ion — right where charging or solar is available; needs protection circuitry and ages with cycles.
- Energy harvesting — solar, thermal, or kinetic. Attractive, but design for the worst week of the year, not the average.
The power source and the duty cycle are one decision, not two. Choosing a cell before the energy budget is modelled is how products end up with an enclosure that cannot fit the battery they turn out to need — and hardware, unlike firmware, cannot be revised quickly.
Frequently Asked Questions
Conclusion
Long battery life is an engineering result, not a component choice. It comes from an honest energy budget built state by state, validated with a current probe on real hardware at the lowest operating temperature.
Sleep aggressively, power-gate everything, transmit rarely and briefly, let cheap sensors decide when expensive work happens, and report battery health so depletion is planned rather than discovered. Do that and multi-year operation stops being a claim on a datasheet and becomes something the fleet actually delivers.