Measuring Current Inside Your Own Product

A product that cannot measure its own current draw cannot tell you its battery percentage, cannot detect a stalled motor, cannot bill by energy consumed, and cannot report a failing heater before a customer does. Current sensing looks trivial on a block diagram — a resistor and an amplifier — and turns out to be where a surprising amount of a product's accuracy and field diagnostics actually live. Almost every mistake in it is made at schematic time and is unfixable in firmware.

Start from what the measurement is for

"We want to measure current" is not a specification. Four questions produce one:

• What range? A device that sleeps at 8 microamps and transmits at 180 milliamps spans more than four decades. Almost no single chain covers that well, which means two chains or switchable gain.
• What accuracy, over what temperature? Two percent at room temperature is easy. Two percent from 14°F to 140°F over a product's life is a different circuit.
• How fast? Coulomb counting needs continuous integration; motor stall detection needs milliseconds; overcurrent shutdown needs microseconds and should not go through software at all.
• Where in the circuit? On the battery, per rail, or per load? A single sense point tells you the product drew more than expected. Three tell you it was the radio.

Shunt resistors: value, power, and placement

The workhorse method is a precision low-value resistor in the current path with an amplifier across it, and the design comes down to choosing the value. Pick it so full-scale current produces a voltage the amplifier likes — commonly 25 to 100 millivolts. Too small and you amplify noise and offset; too large and you waste headroom and burn power. At 2 amps, a 25 milliohm shunt gives 50 millivolts and dissipates 100 milliwatts, needing margin in the rating and some copper around it.

Three placement details decide whether the measurement is any good:

• Use a four-terminal (Kelvin) connection. Sense traces connect to the shunt's own sense pads, or the inner edges of its pads — never to the fat current traces. Solder joint and trace resistance can be a large fraction of a 5 milliohm shunt, and if the sense taps see it your reading is wrong by a percentage that moves with temperature and assembly variation.
• Route the sense lines as a tight, equal-length pair away from switching nodes. A shunt amplifier has a differential input: noise picked up equally on both lines mostly cancels, noise on one does not.
• Choose a low-temperature-coefficient part. Thick-film resistors can drift hundreds of parts per million per degree; metal-element shunts drift a fraction of that for a dollar or two more. The shunt heats itself, so that drift is not something you calibrate away at one temperature.

These are layout-intent items to write on the schematic rather than assume, alongside the other annotations in common first PCB design mistakes.

High-side versus low-side sensing

Low-side is cheaper and fine for a small load whose ground reference does not matter. Its structural weakness: a load that shorts to chassis or ground bypasses the shunt entirely, so the fault you most want to catch is the one you cannot see. High-side sensing needs an amplifier whose inputs tolerate the full bus voltage — parts exist for 5 V, 36 V, 80 V and higher — but it sees every fault and leaves the load's ground undisturbed. For a battery product, a motor drive, or anything with a safety implication, high-side is normally right. Motor applications tend to want both, plus in-phase sensing, as covered in motor control electronics and feedback.

Current-sense amplifiers and integrated monitors

A dedicated current-sense amplifier has a fixed trimmed gain, low input offset, and a common-mode range well above its own supply. It exists because the equivalent built from a general-purpose op-amp and four resistors has a common-mode rejection ratio set by resistor matching, which in a 1 percent world is poor.

One level up sits the integrated power monitor — an INA-class part combining the amplifier, an ADC, a bus-voltage measurement, a multiplier, and an I2C or SPI interface. You wire a shunt and read current, voltage, and computed power from registers. For most products this is the right default:

• It removes the analog design problem and the ADC input problem in one part, typically for $1 to $4, and frees ADC channels on the microcontroller.
• Many parts accumulate energy in hardware, so firmware need not sample fast enough to catch every transient — a real advantage when the processor is asleep most of the time, as it should be under a low-power firmware architecture.
• Several include programmable alert thresholds that assert a pin, giving fast overcurrent response without polling.

Its limits are bandwidth and range. For microsecond fault response, use hardware comparison in the driver or converter itself; the rails feeding your logic have their own protection behavior, discussed in choosing regulators for a circuit.

Hall-effect and coreless sensors for higher currents

Above roughly 20 to 50 amps, a shunt becomes awkward: the power dissipated is significant, the resistor is physically large and expensive, and it self-heats enough to move its own value. Magnetic sensing sidesteps this.

• Hall-effect sensors with a magnetic core clamp around a conductor and measure its field. They are galvanically isolated by construction, which is why they dominate high-voltage and mains applications, and they insert almost no resistance.
• Coreless integrated sensors route the current through a low-resistance internal lead frame and measure the field on-die — isolation and a compact footprint at moderate cost, at some expense in accuracy.
• Rogowski coils measure rate of change, so they read AC and transients well and DC not at all.

The tradeoff is real: magnetic sensors bring larger offset and drift than a good shunt, must be kept away from motors and speakers, and cost more. They earn their place when the current is large, isolation is mandatory, or shunt loss would be unacceptable.

Calibration, offset, and drift

Every current measurement carries two errors. Gain error, from shunt tolerance and amplifier gain, scales with the reading and calibrates out at one point. Offset error, from amplifier input offset voltage, does not scale — it is a fixed number of microvolts referred to the input, invisible at full scale and potentially the entire reading at 1 percent of full scale. Practical handling:

1. Measure and store a zero-current offset at end of line, and re-measure in the field whenever the load is known to be off. This alone removes most of the low-end error.
2. Calibrate gain against a known load in production test. One point written to nonvolatile memory turns a 3 percent chain into a sub-1 percent chain for the price of a fixture step — exactly what a production test fixture exists to do.
3. Budget drift separately. Offset drift, shunt drift, and amplifier gain drift all move with temperature. If the product must hold accuracy across its full range, pick low-drift parts or measure board temperature and correct.
4. Log raw counts, not just computed amps. When a field unit reports something implausible, raw values plus stored calibration constants distinguish a broken sensor from a broken load — the argument for remote diagnostics and logging.

Why your fuel gauge is only as good as this chain

Battery percentage is usually coulomb counting: integrate current in and out, correct against voltage and a cell model, report a number. Every error in the current chain integrates too.

Consider a device that sleeps at 30 microamps for 23 hours a day and wakes briefly. With 20 microamps of uncorrected offset, the gauge is wrong by most of the sleep consumption — and since sleep dominates the daily energy budget, reported runtime can be off by half. The part's full-scale accuracy is irrelevant here; the offset at the bottom of the range is everything. That is why serious battery products use a dedicated fuel-gauge IC with an auto-zeroing front end or run two sense ranges, and why sensing belongs inside the conversation about battery pack design rather than after it.

Projects House designs sensing chains as part of the power architecture, with the calibration step planned into production test so the accuracy on the datasheet is the accuracy customers see. If your product needs to know what it is drawing — for a fuel gauge, for diagnostics, or for billing — reach out through our contact form.