Battery Management for IoT Devices: Protection, Charging and Fuel Gauging Without a Full BMS

The dashboard says 100 percent every morning. The node at the far end of the pond bund has said 100 percent for six weeks, through monsoon cloud and cool nights, and on a Thursday it stops reporting. When someone walks out to it, the cell is flat.
Nothing in that story needed a battery management system. A device that carries a single lithium cell needs three jobs done well: protection, charging and knowing what is left. Protection and charging are solved problems, with a pair of small parts per cell, microamps of quiescent current and behaviour fixed by the datasheet. Knowing what is left is not solved by a part. It is solved by a decision about what to measure, what to count and what to distrust, and that decision is the one the dashboard above got wrong.
So the verdict comes first. Protect with a dedicated part that never runs software, charge with a part that understands temperature, and estimate remaining charge by counting what your own firmware does, correcting it with voltage only when the cell is at rest.
Three Jobs, Not One Product
The phrase battery management system covers everything from a three-part circuit on a wearable to a liquid-cooled controller on a vehicle pack. Treating them as one thing is how a 100 mAh sensor ends up with a gauge that draws a tenth of its own battery.
For a single cell, which engineers write as 1S, the three jobs separate cleanly.
Protection keeps the cell inside the envelope its maker guarantees: not above the maximum charge voltage, not below the minimum discharge voltage, not beyond the maximum current, not short-circuited, not charged when too cold or too hot. This job has no room for firmware. A protection IC is a comparator, a reference and a gate driver, and it works when the microcontroller has crashed, which is the moment you need it most.
Charging moves energy into the cell in the shape the chemistry demands. For lithium-ion that is constant current, then constant voltage, then a termination, with temperature limits on the whole process. A charger IC does this autonomously and tells you its state on a pin or a register.
Gauging answers the user's question, how much is left. It is the only one of the three that is an estimate rather than a control, and so the only one where being approximately right and honest about the error is the goal.
When is a full BMS actually needed? When cells are in series, because then the cells need individual voltage monitoring and balancing. When the current is high enough that a protection MOSFET pair becomes a design problem of its own. When the pack is large enough that a fault is a fire, and the regulators in the market you sell into expect a documented safety architecture. A 1S node in a pond-side enclosure is none of these.
A single-cell IoT device needs three small jobs done correctly, not one large product. Buy the cell's guarantees with a protection IC, buy the charge profile with a charger IC, and spend your engineering effort on the estimate.
What Each Chemistry Asks of You
Chemistry decides the voltages, the temperatures and, importantly for gauging, how much information the terminal voltage carries. The table lists typical values from cell datasheets. They are starting points; the datasheet of the exact cell you buy is the only authority.
Aspect | Lithium-ion (NMC or LCO) | Lithium iron phosphate | Lithium thionyl chloride (primary) |
Nominal voltage | 3.6 to 3.7 V | 3.2 V | 3.6 V |
Charge limit | 4.2 V per cell (some 4.35 V) | 3.65 V per cell | Not rechargeable |
Discharge cutoff | About 3.0 V, cell dependent | About 2.5 V | About 2.0 V |
Charge temperature | Roughly 0 to 45 °C | Roughly 0 to 45 °C | Not applicable |
Voltage as a charge indicator | Usable, slope about 6 mV per percent mid-range | Poor, nearly flat for most of the range | Very poor, flat until near the end |
Typical self-discharge | A few percent a month | A few percent a month | Under one percent a year |
Two lessons sit in that table.
The charge window of 0 to 45 °C is not a suggestion. Charging a lithium-ion cell below about 0 °C can plate metallic lithium onto the anode, which permanently loses capacity and can create an internal short. A solar node in a cold-climate site, or a vehicle tracker parked outside in winter, will be asked to charge when it must not. The charger needs a thermistor, and the firmware needs to know the charger refused.
The slope row decides your gauging strategy. A lithium-ion cell in the middle of its range moves about 6 mV for each percent of charge, which a 12-bit ADC can resolve. A lithium iron phosphate cell moves under 1 mV per percent over most of its range. Whatever works for one will mislead you on the other, and the later sections put numbers on it.
The primary lithium thionyl chloride cell used in many long-life LoRa nodes is a different world again, with a passivation layer that sags the voltage at the first radio pulse after a long sleep. The earlier post comparing LiSOCl2, LiMnO2 and LiFePO4 for LoRa nodes covers that cell; this one concentrates on rechargeable single cells.
Protection: The Part That Must Never Be Clever
A 1S protection circuit is a small IC and two back-to-back MOSFETs in the negative leg of the cell. The IC watches the cell voltage, the voltage across the MOSFET pair, which reads as current, and, in better parts, a thermistor. When anything goes outside its window, it turns the MOSFETs off.
Typical thresholds for lithium-ion are shown here as a guide only. The exact values live in the protection IC datasheet, and the cell maker specifies the limits the IC must honour.
Fault | Typical trip point | What the IC does | What the firmware sees |
Overcharge | About 4.25 to 4.30 V | Opens the charge path | Charger input present, no charge current |
Over-discharge | About 2.4 to 3.0 V | Opens the discharge path | Power loss, a reset on recovery |
Over-current | Set by the sense resistance and a threshold | Opens after a delay of milliseconds | Brown-out during a transmit burst |
Short circuit | A higher current threshold, microseconds to react | Opens fast, latches until the load is removed | Dead device |
Over-temperature | Needs a thermistor input | Opens both paths | Often nothing |
Three practical points.
First, the quiescent current of the protection IC is a standing drain, typically a few microamps, and it counts. We include it in the budget below.
Second, the MOSFET pair has a resistance, commonly tens of milliohms for a small device. At a 150 mA radio burst that is a few millivolts, which is nothing, but at the 2 A peaks of a cellular modem it matters, and the over-current trip must be set high enough that your own transmit burst does not trigger it. A node that resets on the radio's first packet at low temperature, when the cell resistance has doubled, is often a protection IC tripping on a sag it mistakes for a fault.
Third, the protection IC is the last line, not the first. Normal operation should never reach its thresholds. If your device regularly gets within 50 mV of the over-discharge trip, the firmware's low-battery logic is too late.
The protection IC is the one component you want to be stupid. It should have no registers, no firmware and no clever modes. When a test shows it tripping in normal use, the fault is upstream of it.
Charging in the Field: Temperature, Solar and the 0 °C Rule
A lithium-ion charger runs three phases. A trickle phase gently raises a deeply discharged cell. A constant-current phase pushes a fixed current, typically 0.5C or below for long life, until the cell reaches 4.2 V. A constant-voltage phase holds 4.2 V while the current falls, and the charge ends when the current drops to a termination level, commonly one tenth or one twentieth of the charge current.
For an IoT device, four details matter more than the textbook.
Temperature gating. A charger with a thermistor input suspends charging outside its window, and a good one resumes when the temperature recovers. Make sure the thermistor is thermally coupled to the cell, not to the charger or the board, because a solar node's board and cell can differ by 20 °C at noon.
Power path. If the device runs from the same cell it charges, the load and the charger fight. A power-path charger gives the system its own supply from the input while it charges the cell separately, so a reset or a brown-out does not follow every cloud. Without it, a node on a marginal panel can sit at partial charge forever, because the radio burst pulls the cell down faster than the charger fills it.
Float voltage. A cell held at 4.2 V in a hot enclosure ages faster than one held at 4.1 V. Many designs deliberately end charge lower, at 4.1 or 4.15 V, giving up around ten percent of the capacity to gain cycle and calendar life. That is the right choice for a node expected to run for years in a sheet-metal box under the Indian sun.
Charger quiescent current. A charger with no input present still draws microamps from the cell. We include it in the budget.
A solar-assisted node adds one more consideration: the charger's input is a panel, not a supply. A panel with a poor maximum-power-point match wastes much of its output, and in monsoon weeks the panel may deliver a small fraction of its rating. The earlier post on sizing solar and battery for pond and shed nodes covers the energy side; here the point is that charging must degrade gracefully, so a day with no sun leaves the system exactly where it was.
The Quiescent Budget Nobody Draws
Before choosing a gauge, add up what will drain the cell while the device does nothing. This is the always-on current, and it is where battery life is quietly lost.
The table below uses an illustrative node that averages 85 µA from its own sleep and duty cycle, such as a LoRa node reporting every fifteen minutes, on a 2,600 mAh cell with 80 percent usable.
Always-on drain | Current | Share of the total |
Node sleep and duty cycle | 85.0 µA | 68.4 percent |
Protection IC | 3.0 µA | 2.4 percent |
Charger IC with no input | 8.0 µA | 6.4 percent |
Fuel gauge IC | 25.0 µA | 20.1 percent |
ADC divider, 3.7 V across 2 MΩ | 1.9 µA | 1.5 percent |
Regulator quiescent current | 1.5 µA | 1.2 percent |
Total | 124.3 µA | 100 percent |
Now the arithmetic that matters. On 2,080 mAh usable, the node alone would last 1,020 days. With every drain in the table it lasts 697 days. Remove only the gauge, and it lasts 872 days. The gauge, by itself, costs 175 days, about 20 percent of the battery life, to deliver a percentage number.
A gauge that draws 25 µA is not badly designed. It was designed for a phone that draws 100 mA with the screen on, where 25 µA is invisible. IoT inverts the ratio: the thing you are measuring draws a hundred microamps on average, and the instrument draws a quarter of that.
Two cheaper habits help before any other work. Gate the resistor divider that feeds the ADC with a transistor or a load switch, so it conducts only during the measurement, and read the thermistor the same way. Both are a handful of components and save a few microamps each.
Add the quiescent currents of every part on the battery rail before choosing any of them. A part that is invisible in a phone design can be a fifth of an IoT node's battery.
Four Ways to Know What Is Left
There are four families of estimate. Each has a place, and each fails in a predictable way.
Aspect | Voltage lookup | Voltage with IR compensation | Coulomb counting | State-based energy accounting |
What it measures | Terminal voltage, matched to a curve | Terminal voltage plus load current times resistance | Current through a sense resistor, integrated | Time spent in each firmware state times a known current |
Hardware needed | An ADC channel | ADC plus a current estimate and temperature | A sense resistor and an amplifier | Nothing beyond the firmware |
Standing current | Microamps if the divider is gated | Same | 20 to 30 µA for a typical gauge IC | Zero |
Typical failure | Load, temperature and curve flatness | Wrong resistance model when cold or aged | Offset error accumulates, no reference to full | Model error if a state is mis-measured |
Best for | Coarse alarms | Steep-curve chemistries | Larger loads, long runtime between full charges | Duty-cycled IoT nodes |
The strongest practical combination for a duty-cycled node is the last column with the first as a correction: count what the firmware does, then pull the count toward what the rested voltage says.
Why a Voltage Curve Lies, With Numbers
The terminal voltage of a loaded cell is the open-circuit voltage minus the current times the internal resistance. The internal resistance is not small, and it rises as the cell gets cold or old.
The example uses an illustrative open-circuit curve for a lithium-ion cell. The cell reads 3.78 V while the radio draws 150 mA, which is a quite ordinary measurement taken at the wrong moment. A raw lookup says 46 percent. Correcting for the voltage drop across the internal resistance, assumed to be 0.15 Ω at 25 °C and 0.35 Ω at 0 °C, gives 50.4 percent at 25 °C, 53.1 percent at 10 °C and 54.9 percent at 0 °C. The raw reading is 4 to 9 points low, and the error is largest in the cold, when the user most wants the number.
Now the flat chemistry. With a lithium-ion cell the mid-range slope is about 5.7 mV per percent. With a lithium iron phosphate cell, the slope between 20 and 90 percent is about 0.7 mV per percent. An ADC error of 10 mV, quite ordinary for a modest reference and divider, is a gauging error of 1.8 percent on the first and 14 percent on the second. The same hardware that gives a trustworthy percentage on one chemistry gives a coin flip on the other.
import bisect
# Illustrative open-circuit curve, 1S NMC-type cell. Measure your own cell.
OCV = [(3.00, 0), (3.45, 5), (3.60, 15), (3.70, 30), (3.80, 50),
(3.90, 65), (4.00, 80), (4.10, 92), (4.20, 100)]
V = [p[0] for p in OCV]; S = [p[1] for p in OCV]
def soc_from_ocv(v):
if v <= V[0]: return 0.0
if v >= V[-1]: return 100.0
i = bisect.bisect_right(V, v)
f = (v - V[i-1]) / (V[i] - V[i-1])
return S[i-1] + f * (S[i] - S[i-1])
def internal_resistance_ohm(temp_c):
"""illustrative: 0.15 ohm at 25 C rising to 0.35 ohm at 0 C"""
if temp_c >= 25: return 0.15
if temp_c <= 0: return 0.35
return 0.35 - (0.35 - 0.15) * temp_c / 25.0
def soc_ir_compensated(v_meas, i_load_a, temp_c):
return soc_from_ocv(v_meas + i_load_a * internal_resistance_ohm(temp_c))
def coulomb_offset_error_ua(offset_uv, sense_mohm):
return offset_uv / sense_mohm * 1000.0
def life_days(capacity_mah, avg_ua, usable=0.8):
return capacity_mah * usable * 1000.0 / avg_ua / 24.0
# 3.78 V at 150 mA: raw 46.0 %, compensated 50.4 (25 C), 53.1 (10 C), 54.9 (0 C)
# 2 uV offset: 10 mohm -> 200 uA, 20 mohm -> 100 uA, 100 mohm -> 20 uA, 1 ohm -> 2 uA
# slope NMC 5.71 mV/%, LFP 0.71 mV/%; 10 mV error -> 1.8 % and 14.0 %
# 2,600 mAh, 124.3 uA all drains -> 697 days; without the 25 uA gauge -> 872 days
The code ran and its asserts passed: the open-circuit lookup returns exactly 50 percent at 3.80 V, the offset conversion returns 20 µA for 2 µV across 100 mΩ, and the life function returns the expected value for a unit case.
The cure for load error is simple in principle: do not read the voltage while the radio is on. Wait until the load is under a milliamp for tens of seconds and the cell has relaxed, and read then. For a node that sleeps 99 percent of the time, that is the natural moment.
Why Coulomb Counters Built for Phones Struggle at 50 µA
A coulomb counting gauge measures the voltage across a small sense resistor in the battery path and integrates it over time. The amplifier has an input offset, a few microvolts in good parts. Across a sense resistor, an offset voltage is indistinguishable from a current. The conversion is Ohm's law: 2 µV across 10 mΩ looks like 200 µA.
Sense resistor | Error from 2 µV offset | Voltage drop at 150 mA |
10 mΩ | 200 µA | 1.5 mV |
20 mΩ | 100 µA | 3.0 mV |
100 mΩ | 20 µA | 15 mV |
1,000 mΩ | 2 µA | 150 mV |
The table shows a collision of two requirements. A phone uses a small sense resistor because it wants to waste no voltage at an ampere of load, and the offset error of 200 µA is lost in that load. An IoT node that averages 85 µA cannot tolerate a counter that invents 200 µA of phantom current. Raise the sense resistor to 100 mΩ and the phantom falls to 20 µA, but now the radio burst drops 15 mV in the sense path and the noise on the amplifier input is no easier. Raise it to 1 Ω and the burst drops 150 mV, which can brown out a radio.
Some gauge ICs correct offset by periodically shorting the input, and good ones do it well. The practical point is not that these parts are poor. It is that the datasheet figure for offset has to be converted to microamps at your sense resistor and compared with your average current before you commit.
That conversion is one line of arithmetic, and it shows up as a missed line in many reviews.
Counting States Instead of Amperes
A firmware knows what it is doing. It knows how long the radio was on, how long the sensor was powered, how long the processor ran, and, from the power-budget measurements you made at the bench, roughly what each state costs in current. Multiply, accumulate, and you have a coulomb counter with no hardware and no standing current.
The accuracy of this estimate is set by how well you measured each state's current, which you can do with a source-measure unit or a current analyser on the bench. It is not set by an offset that drifts with temperature. The weak spots are states you did not model and states whose duration varies, such as a radio retry. Count those by timing, not by assumption.
The C code below implements the idea. It stores charge as microcoulombs, a 64-bit integer, adds each state as current times duration, reports the percentage left, and corrects the count toward the open-circuit voltage only after the cell has rested. It was compiled and run.
#include <stdint.h>
typedef struct {
int64_t capacity_uc; /* usable capacity, microcoulombs */
int64_t used_uc; /* drawn since the last full charge */
} gauge_t;
static void gauge_init(gauge_t *g, uint32_t capacity_mah, uint32_t usable_percent)
{
g->capacity_uc = (int64_t)capacity_mah * 3600000LL * usable_percent / 100;
g->used_uc = 0;
}
/* once per load state: current in uA for duration in ms */
static void gauge_add(gauge_t *g, uint32_t current_ua, uint32_t duration_ms)
{
g->used_uc += (int64_t)current_ua * duration_ms / 1000;
}
static int gauge_soc_percent(const gauge_t *g)
{
int64_t left = g->capacity_uc - g->used_uc;
if (left < 0) left = 0;
return (int)(left * 100 / g->capacity_uc);
}
/* only after the cell has rested: load under 1 mA for 30 s or more.
Pull the counter a quarter of the way toward what the voltage says. */
static void gauge_correct(gauge_t *g, int rest_mv)
{
int64_t target_used = g->capacity_uc * (100 - soc_from_mv(rest_mv)) / 100;
g->used_uc += (target_used - g->used_uc) / 4;
}
/* Tested: one hour of 3,590 s at 25 uA + 5 s at 3 mA + 5 s at 120 mA
= 704,750 uC, average 195.8 uA; 24 hours = 16,914,000 uC = 4.70 mAh.
A counter that has drifted to 95 % while the rested cell reads 3,800 mV
(50 %) moves to 64 % after four corrections. */
The test confirmed the arithmetic: one hour of the stated load profile is 704,750 µC, an average of 195.8 µA, and after 24 hours the counter has used 16,914,000 µC, which is 4.70 mAh. The correction behaves as designed: a counter that had drifted to 95 percent, with the cell resting at 3,800 mV (50 percent on the curve), moved to 64 percent after four corrections, converging without jumping.
The blend matters. A hard reset to the voltage estimate makes the displayed percentage jump, and users read a jump as a fault. A slow pull toward the voltage makes the display move smoothly and still removes the drift in days.
One more rule makes the display trustworthy: never let the percentage rise while the device is discharging. Clamp it so that it only goes down between full charges, and let it rise only when the charger reports charge in progress.
Where the Numbers Go Wrong: Cold, Age and Self-Discharge
The usable capacity you feed the gauge is not the number on the cell label.
Cold. A lithium-ion cell delivers less capacity at low temperature because its internal resistance rises and the cutoff voltage is reached sooner. The loss at 0 °C and a moderate current can reach tens of percent on some cells. Read the discharge curves in the cell datasheet at your lowest temperature and your load current, and use that capacity for the alarm threshold, not for the display.
Age. Capacity fades with cycles and with time, and faster when the cell sits hot and full. A common rule of thumb is that chemical ageing rates roughly double for each ten degrees Celsius of temperature rise. It is a rule of thumb and not a law, but it is a good reason to keep the cell out of the sun and to end charge below 4.2 V for devices that live outdoors. A gauge with a fixed capacity will overstate remaining charge as the cell ages. Reduce the stored capacity each year, or learn it from full discharge events when they occur.
Self-discharge. A few percent a month for lithium-ion at room temperature, more in heat. For a node with a 700-day battery life, that is a significant additional drain that no load model includes. Add it as a constant in the state model, as a current: a 2 percent monthly loss on 2,600 mAh is about 52 mAh a month, or 72 µA, which is a large fraction of an 85 µA node.
Check that last figure, because it surprises people. Two percent of 2,600 mAh is 52 mAh. A month is 720 hours. Fifty-two milliamp-hours over 720 hours is 72 µA. Self-discharge in a hot enclosure can rival the node's own consumption, which is one more reason to use a cell of low self-discharge and to store spares cool.
The Percentage That Went Up and Down
This is a representative composite, genericised, with illustrative numbers.
A team shipped a solar-assisted outdoor node with a single lithium-ion cell. The firmware read the cell voltage through a divider every ten minutes and mapped it to a percentage with a lookup table. On the bench it looked right. In the field, the dashboard showed a percentage that rose to 100 every sunny day, fell to 40 overnight, and sometimes dropped by several points when the radio sent a packet in the cold.
None of that was a fault in the cell. Three effects stacked.
During daylight the charger held the terminals near 4.2 V, so the reading was the charger's voltage, not the cell's open-circuit voltage, and the lookup said 100. After sunset the terminal voltage fell to the cell's true rested level, which was lower. Whenever a reading was taken during a transmit burst, the voltage drop across the internal resistance, larger in the cold, read as lost charge, a few points at 25 °C and as many as nine in the cold, as in the 3.78 V example above.
The team's first reaction was to add a gauge IC. It would have cost about 25 µA, which on an 85 µA node reduces life by about a fifth, from roughly 872 days to 697 days in the budget table.
The fix was in the firmware. Count states for the energy used, using bench-measured currents for sleep, sensing and each radio mode. Read the voltage only when the charger reports not charging and the load has been under a milliamp for thirty seconds. Blend the rested reading into the count by a quarter at a time. Clamp the display so it falls monotonically between full charges. The dashboard became boring, which is what a battery gauge should be, and no hardware was added.
Certification and Shipping: BIS, IEC 62133 and UN 38.3
A lithium cell is a regulated article. Three rules touch an Indian IoT maker.
Safety registration. BIS lists sealed secondary lithium cells and batteries for portable applications under the Compulsory Registration Scheme, tested to IS 16046 (Part 2):2018, which corresponds to IEC 62133-2:2017. BIS pages describe the scheme as registration with self-declaration of conformity, notified by the Ministry of Electronics and Information Technology, with the Electronics and Information Technology Goods (Requirements for Compulsory Registration) Order, 2021 as the governing order. Trade sources reported in February 2026 a requirement to verify the rated capacity of such cells and batteries during registration. These sources are secondary, so confirm the current text with your test laboratory.
Scope. Whether a cell built into a fixed outdoor sensor counts as a portable application is a classification question. As with the certification of wireless modules, the safe course is to ask the laboratory for a written classification before you plan a launch around it.
Transport. Lithium cells and batteries must pass the UN 38.3 tests before they are shipped by air, sea or road, and the packaging and labelling rules depend on the energy content. The cell vendor's UN 38.3 test summary is part of your paperwork, and so is the summary for a pack you assemble yourself.
None of this is legal advice. The practical consequence for design is to choose cells and packs that already carry the registrations and test summaries you need, and to avoid assembling your own packs from loose cells unless you plan to test them.
Tests Before You Trust a Battery Design
Test | How to run it | A pass looks like |
Quiescent current | Remove the load, measure with a current analyser for ten minutes at three temperatures | Matches the budget table within 10 percent |
Radio sag | Log cell voltage during the worst transmit burst at the lowest temperature | Cell stays above the protection and brown-out thresholds with margin |
Cold charge refusal | Chill the whole assembly below 0 °C with the charger connected | No charge current, a clean status flag, no error loops |
Hot enclosure | Hold the enclosure at its maximum temperature for a week | Charge ends at the intended voltage, no thermistor trips |
Gauge drift | Run the load profile for a week against a bench coulomb measurement | Percentage within five points, never rising on discharge |
Brown-out recovery | Drop the supply under the cutoff and restore it | The device restarts, restores its count, no corruption |
Run the first two before the enclosure exists. They are cheap, and they decide whether the protection thresholds and the load model are right.
Measure the cold radio burst before you finalise the protection IC. A cell that sags 0.35 V at 0 °C can trip an over-current or under-voltage threshold that never fires on the bench at 25 °C.
When You Do Need a Real Battery Management System
The three-job approach has limits, and it is worth stating them plainly.
If cells are in series, add cell-level monitoring and balancing. Two cells in series that drift apart will overcharge the weaker, and a protection IC that sees only the stack voltage cannot prevent it.
If the load is amperes rather than milliamps, a gauge IC with a proper sense resistor and a learning algorithm earns its quiescent current. A vehicle tracker with a cellular modem and a heated enclosure is closer to that than to a pond node.
If the cell is large, or the application is one where a fault is a safety event, use a pack with a certified protection and monitoring design, and keep the documentation.
If you need a percentage accurate to a couple of points across the cell's whole life, including when it is cold and old, the model-based gauge ICs that track impedance are the right tool, and their 20 to 30 µA is the price. Then budget for it from the start, and size the battery accordingly.
Where to Start on Monday Morning
List every part on the battery rail and write its quiescent current next to it. Add them.
Compute the battery life from the total, with 80 percent usable capacity, and compare it with the requirement.
Choose a protection IC whose thresholds fit the cell datasheet, with over-current set above your worst radio burst.
Choose a charger with a thermistor input and a power path, and couple the thermistor to the cell.
Decide the end-of-charge voltage for the enclosure temperature, and accept the lost capacity.
Measure the current of every firmware state on the bench and enter it in a table.
Implement the state counter, a rested-voltage correction and a display that never rises on discharge.
Gate the divider and the thermistor so they conduct only while measured.
Run the cold radio burst and the cold charge refusal tests.
Ask the laboratory about registration and UN 38.3 for the exact cell or pack.
One Hundred Percent and Dead
The dashboard that read 100 every afternoon was not lying on purpose. It was reading the charger. A battery gauge is a small piece of honesty built from three things, a model of what the firmware does, a voltage taken at a moment when it means something, and a refusal to let the display jump.
Protect with a part that cannot crash, charge with a part that watches temperature, and count your own states. The gauge chip can wait until the arithmetic says it has earned its microamps.
Related service: Embedded Firmware Development.
Related reading: LiSOCl2 vs LiMnO2 vs LiFePO4 for LoRa Nodes, Sizing Solar and Battery for Pond and Shed IoT Nodes in India.
EurthTech is an engineering-first product design and embedded systems company with a development center in Hyderabad. To discuss a project, write to connect@eurthtech.com or see https://www.eurthtech.com/case-studies.




Comments