A single lithium D-cell browns out your Cat.1 modem mid-transmit
Battery-powered cellular sensors fail in a way the datasheets do not warn you about. The cell has enough energy for years of operation, the modem has a comfortable supply-voltage spec, and the bench prototype runs for weeks. Then the field units start missing uploads, and the logs show the modem resetting partway through a transmission.
The cause is the cell's internal impedance. A cell can hold years of energy and still be unable to deliver a two-amp pulse without its voltage collapsing.
The current the radio actually draws
A Cat.1 LTE module pulls a short, high-current spike every time it transmits. On the modules we use for low-power meter reading, a transmit burst can reach roughly 2 A for a few milliseconds, against an average draw measured in microamps for the rest of the cycle. The ratio between peak and average is enormous, and the peak is what breaks things.
An ER34615, the 3.6 V lithium-thionyl-chloride D cell that makes multi-year battery life possible, has high internal impedance by design. That chemistry trades current delivery for energy density and shelf life. When the modem demands 2 A, the cell cannot supply it without its terminal voltage sagging.
Where it fails
In our SPICE model of the supply, a bare ER34615 driving a 2 A transmit burst sags to about 3.22 V at the module pin. The module's documented minimum operating voltage is 3.3 V. So every transmit burst drives the rail below spec, and the module browns out and resets, often mid-transmission. That is why the symptom presents as flaky uplinks rather than a dead board.
The standard reflex is to add bulk capacitance, and on its own it does not solve this. A 1000 µF electrolytic plus a 100 µF ceramic looks reassuring on the schematic, but bulk electrolytics have their own ESR and cannot source a millisecond-scale 2 A edge fast enough. You end up buffering the wrong frequency band.
The fix that holds the rail
The component that matches the load is a hybrid pulse capacitor, placed across the cell. It is built to deliver exactly the short high-current pulse that the ER34615 cannot. With it in the loop, the same 2 A burst in simulation holds the rail at about 3.42 V, comfortably above the 3.3 V floor.
The design rule that falls out of this: size the energy source for average draw, and size a separate pulse buffer for peak draw. They are two different problems, and one part rarely solves both.
The cost of getting it right is small. The bare cell runs roughly ¥40–80 depending on grade; adding the hybrid buffer puts the pair around ¥60–90. That ¥20 difference is the line between a unit that runs for years and a unit that resets on every upload. On a battery sensor in the field, that failure is a truck roll, not a redeploy.
Why bench testing misses it
Brownout-on-transmit hides during development for two reasons. A fresh cell has lower impedance than an aged one, so early units carry more margin than the fleet will a year in. And bench supplies or USB power mask the problem entirely, because they can source the spike the battery cannot. The failure only appears on real cells, in the field, under marginal signal, which is also when the modem transmits longest and hardest.
To catch it before shipping, measure the rail under a real transmit burst on a real cell, with an instrument fast enough to see a millisecond sag (a Joulescope or PPK2, not a multimeter), and do it on an aged cell rather than a fresh one.
Where AgentKick fits
We design battery-powered edge and IoT hardware where the failure modes live in the gaps between datasheets: power integrity, brownout, multi-year duty cycles. If you are taking a cellular sensor from prototype to a field fleet and want the power path validated before it ships, that is the kind of work we do, usually as a short scoping engagement into a phased build.