← Journal Testing · · 7 min

Battery life testing that predicts real field runtime

A practical guide to battery life testing: measure real current, budget charge per event and derate the cell so your runtime claim holds up.

Milo Perović
Founder & lead engineer

  • Battery
  • Power
  • QA
Battery life testing that predicts real field runtime

Most battery estimates are written in a spreadsheet: sleep current from the datasheet, a guess at radio time, a nominal cell capacity, and a confident “two years”. Then the first units come back from the field after seven months. Battery life testing is how you replace that spreadsheet with numbers you have actually measured, on real hardware, under conditions that look like the field.

This is for founders and engineers with a working prototype who need to promise a runtime to customers, choose a cell, or decide whether the firmware still needs work. None of it needs an expensive lab, but all of it needs discipline about what you measure and how.

Why datasheet maths drifts from field runtime

A spreadsheet model is not useless. It is the right starting point, and we build one on every battery product. The problem is that it inherits every optimistic assumption you fed it, and the errors almost never cancel out. They stack in one direction.

  • Sleep current is higher than the MCU figure. The microcontroller may sleep at single-digit microamps, but regulators, sensor standby modes, pull-ups, leaky protection diodes and a forgotten LED driver all add their share.
  • Radio events take longer than planned. A BLE advertisement is brief, but a Wi-Fi reconnect after a router reboot, a LoRaWAN join retry or a cellular network attach can cost many times the energy of a normal transmission.
  • Nominal capacity is not usable capacity. Cell capacity is rated at a specific discharge current, temperature and cut-off voltage. Your device rarely matches all three.
  • Firmware changes after the estimate. A new logging feature or a slightly longer sensor warm-up quietly adds a few percent, and nobody updates the model.

We cover where the stray current tends to come from in our post on where the microamps really go. This post is about measuring it and turning the measurement into a runtime you can defend.

Measuring current with enough dynamic range

The difficult part of measuring a connected device is the range. A typical sensor node sleeps at a few microamps and then draws tens to a few hundred milliamps during a radio burst lasting milliseconds. That is five orders of magnitude, and both ends matter to the average.

A bench multimeter in series is fine for checking a steady sleep current, but it is the wrong tool for the full profile. Its shunt resistor creates a burden voltage that can brown out the device during a transmit peak, auto-ranging is too slow to follow bursts, and its reading is an average of something it never really saw.

What works better:

  • A dedicated power profiler such as the Nordic Power Profiler Kit II or a Joulescope. These switch ranges fast enough to capture both sleep and burst current in one trace and integrate charge over time.
  • A source-measure unit if you already have one, which also lets you emulate a battery’s voltage and, on some models, its internal resistance.
  • A low-value shunt and a differential probe on an oscilloscope for looking at the shape of a single event, such as inrush when a sensor powers up.

Whatever you use, power the device the way the battery will: same voltage, through the same connector and protection path. Disconnect the debugger, because an attached SWD or JTAG probe often keeps debug logic powered and inflates sleep current. Remove USB as well, since a USB-UART bridge can back-power the board through its I/O pins.

Building a battery life testing profile from real events

A single capture tells you little. What you want is a charge budget per event, multiplied by how often each event happens in the field. We break the device’s behaviour into a short list and measure each one separately, with enough repetitions to see the spread.

EventWhat to measureWhat makes it vary
Deep sleepAverage current over several minutesTemperature, sensor standby, RTC source
Sensor readCharge per reading, including warm-upSettling time, oversampling, bus speed
Normal transmissionCharge per uplinkPayload size, data rate, TX power
Reconnect or joinCharge per attempt and number of attemptsSignal quality, network behaviour, back-off logic
OTA updateTotal charge per updateImage size, link quality, flash erase time

Measure in microamp-hours or millicoulombs per event, not peak milliamps. Then build the daily budget: sleep charge plus each event’s charge multiplied by its daily count. The reconnect line is the one teams most often leave out, and it is frequently what separates a good estimate from a bad one. Test it deliberately: switch the access point off, move the device to a weak-signal spot, and measure what the firmware does when the network is unhelpful.

Derating the battery honestly

Once you have a daily charge budget, you divide usable capacity by it. The word “usable” is where most of the optimism hides.

Cut-off voltage and pulse loads

Your device stops working when the battery voltage under load falls below what the regulator or radio needs, not when the cell is empty. Small primary cells with high internal resistance, such as CR2032 coin cells, are the classic trap: they are rated at around 220 to 240 mAh at very light loads, but a radio pulse of tens of milliamps can drop the terminal voltage enough to reset the device long before that capacity is used. A reservoir capacitor helps, and it should be sized from a measured pulse, not a guess.

Temperature

Cold reduces available capacity and increases internal resistance, so a device that is fine on the bench can brown out outdoors in winter. If the product lives outside, repeat the key measurements in a cold chamber or, at minimum, a freezer with a thermocouple on the cell. Heat tends to increase self-discharge instead.

Self-discharge and ageing

For multi-year products, self-discharge matters as much as the load. Lithium thionyl chloride primary cells lose very little per year but can show voltage delay after storage. Rechargeable lithium-ion cells lose more each month, and the protection circuit adds its own quiescent draw. Use the cell manufacturer’s curves, and leave margin for capacity loss over the product’s life.

In practice, we usually apply a combined derating somewhere between 20 and 40 percent of nominal capacity, depending on chemistry, temperature range and how close the pulse loads come to the cell’s limits. That range is a starting point, not a rule, and the measurements should narrow it.

Validating with a real discharge run

A model built from measured events is far better than a datasheet sum, but it is still a model. Before committing to a runtime claim, we run at least a few units on real cells with an accelerated schedule: the same firmware with intervals shortened so that months of events happen in days or weeks.

  1. Log battery voltage and a timestamp from the device itself, and if possible current from an external profiler in parallel.
  2. Run until the device stops working, and record the voltage at which it failed and why: brown-out, failed transmission or watchdog reset.
  3. Compare total delivered charge with the model’s prediction. A gap of more than about 10 to 15 percent means an event is missing or mis-measured.
  4. Keep in mind that acceleration changes the duty cycle, so it tells you less about sleep-dominated behaviour and self-discharge. Use it to validate events and cut-off, not as the final runtime.

Write the result up in a form someone outside engineering can follow. We explain how we structure that in a bench report a founder can actually read.

A checklist before you promise a runtime

  • Sleep current measured with the debugger and USB disconnected, at the lowest and highest operating temperatures.
  • Charge per event measured for every event, including reconnects, retries and OTA updates.
  • Daily budget based on realistic field counts, including bad-network days.
  • Usable capacity derated for cut-off voltage, pulse loads, temperature and ageing.
  • At least one discharge run on real cells that agrees with the model.
  • Battery life testing repeated after every firmware release that touches power, radio or sensor timing.

If you cannot tick every line, quote a range rather than a single number, and say what it depends on. Customers forgive a conservative estimate far more easily than a device that dies early.

Working on something similar?Free scoping call →