PowerElecHub

Current Sense Resistor Calculator: Shunt Value, Power and Error Budget

Choose a shunt resistor for a current-sense amplifier and ADC: the value that uses the ADC range at full current (rounded to E24), dissipation and power rating, current per LSB, and an error budget from tolerance, TCR, gain error, offset and quantization, plotted against current. Also shows the frequency above which the shunt's inductance distorts the reading.

Current and measurement range

Bidirectional sensing puts zero current at mid-scale, so each direction gets half the ADC range. Headroom keeps a fault or overshoot from clipping the reading.

Error sources

Use maximum datasheet values for a worst-case budget. ΔT is the shunt's temperature change from where it was calibrated (or 25 °C), including its own heating.

Shunt resistance

3.6 mΩ

ideal 3.71 mΩ, rounded down to E24

Dissipation at RMS current

518 mW

choose a rating of at least 1.04 W

Error at full scale

1.71 %

worst case; RSS 1.12 %

Error at 10 % of full scale

2.28 %

worst case; RSS 1.20 %

  • The shunt's L/R corner is at 573 kHz. Above it the shunt voltage follows L·di/dt more than i·R, so switching edges show spikes; filter the amplifier input or use a lower-inductance shunt.

Signal

Shunt voltage at I_fs
72 mV
Amplifier output at I_fs
1.44 V
Share of the ADC range used
87.3 %
Current per ADC LSB
11.2 mA
Time constant L/R
278 ns
L/R corner frequency
573 kHz

Error budget

Tolerance (ratio)
1.00 %
TCR × ΔT (ratio)
0.45 %
Gain error (ratio)
0.20 %
V_os as current (fixed)
6.94 mA
Quantization ±½ LSB (fixed)
5.59 mA
Current where V_os error is 1 %
694 mA

Measurement error against current

Ratio errors set a floor that does not depend on current. Offset and quantization are fixed amounts of current, so their share grows as the current falls.

DC error budget with maximum values. Common-mode rejection, input bias current, thermal EMF at the shunt terminals, PCB copper in series with a non-Kelvin connection, and noise are not included. Tolerance and gain error can be removed by a one-point calibration; TCR and offset drift cannot.

How it works

Choosing a shunt is a compromise between two things that pull in opposite directions. A larger resistance gives a larger signal, so the amplifier's offset and the ADC's step size matter less. A smaller resistance dissipates less power and drops less voltage in the power path. The calculator starts from the measurement side: the amplifier output should reach the ADC's range, less some headroom, at the largest current you need to read. That fixes the ideal resistance R = (1 − headroom)·V_span/(G·I_fs), where V_span is the full ADC range for unidirectional sensing and half of it for bidirectional sensing around mid-scale.

Shunts are sold in discrete values, so the ideal value is rounded down to the E24 series. Rounding down means full-scale current can never push the amplifier past the ADC range; the cost is that a little of the range goes unused, which the page shows. Precision shunts are also made in values outside E24, so check the manufacturer's list before settling on a value.

The error terms fall into two groups, which application notes from ST and TI also separate. Ratio errors scale with the reading: the shunt's tolerance, its resistance change with temperature (TCR × ΔT), and the amplifier's gain error. They are a fixed percentage at every current. Offset-type errors are a fixed amount of current: the amplifier's input offset divided by R, and half an ADC step referred back to the shunt. At full scale they are small; at a tenth of full scale they are ten times larger as a percentage. The chart shows how the two combine. A worst-case sum assumes every error is at its limit in the same direction; the root-sum-square figure is a more typical expectation for parts that are not all at their limits.

Power is set by the RMS current, not the peak: P = I_rms²·R. Shunt power ratings assume a particular board temperature and copper area, so a margin of two is common, and the shunt's own heating is part of the ΔT that drives the TCR error. Finally, every shunt has some inductance, typically below a nanohenry to a few nanohenries for surface-mount metal-strip parts. Above the corner frequency R/(2π·L) the voltage across a milliohm shunt is dominated by L·di/dt, which shows up as spikes at switching edges. With milliohm values that corner can fall within a few hundred kilohertz.

Design procedure

  1. 1Fix the largest current you must read without clipping, including overcurrent and fault transients, and the continuous RMS current.
  2. 2Enter the amplifier gain and the ADC range, choose bidirectional or unidirectional sensing, and leave some headroom.
  3. 3Look at the dissipation; if it is too high, a higher-gain amplifier allows a smaller shunt.
  4. 4Enter tolerance, TCR, temperature rise, offset and gain error from the datasheets and check the error at the lowest current that matters, not only at full scale.
  5. 5Lay out the shunt with a Kelvin connection, keep the sense traces together, and filter the amplifier input if the L/R corner is close to your switching edges.

Formulas

Shunt value from the ADC range
R = (1 − h)·V_span/(G·I_fs), V_span = V_ADC (unidirectional) or V_ADC/2 (bidirectional), rounded down to E24
Dissipation
P = I_rms²·R, rating ≥ 2·P
Current per LSB
ΔI = V_ADC/(2^N·G·R)
Ratio errors (fraction of reading)
ε_ratio = tol + TCR·ΔT + ε_gain
Offset errors (fixed current)
I_os = V_os/R + ΔI/2
Error at current I
worst: ε_ratio + I_os/I; RSS: √(tol² + (TCR·ΔT)² + ε_gain² + (V_os/(R·I))² + (ΔI/(2I))²)
Shunt inductance corner
τ = L/R, f_L = R/(2π·L)

Worked example

The defaults describe phase-current sensing in a 48 V motor drive: ±20 A to measure, 12 A rms continuous, a gain-of-20 amplifier into a 3.3 V, 12-bit ADC centred at mid-scale, and 10 % headroom. Each direction has 1.65 V of range, so the ideal shunt is 3.71 mΩ; the E24 value below it is 3.6 mΩ, which gives 72 mV at 20 A and uses 87 % of the range. At 12 A rms it dissipates 0.52 W, so a part rated 1 W or more is needed, and a 2 W 2512 shunt leaves comfortable margin. One ADC step corresponds to 11.2 mA. With a 1 % shunt, 75 ppm/°C and a 60 °C rise, plus 0.2 % gain error, the ratio errors add up to 1.65 %. A 25 µV offset is worth 6.9 mA and half a step 5.6 mA. At 20 A the worst-case error is 1.71 % (1.12 % RSS); at 2 A it is 2.28 % (1.20 % RSS), and below 0.69 A the offset alone exceeds 1 %. Raising the gain to 50 lets the shunt shrink to 1.3 mΩ and the dissipation to 0.19 W, but the 1 % offset point moves up to 1.9 A and the worst-case error at 2 A grows to 2.9 %. A 0.1 %, 25 ppm/°C shunt instead brings the full-scale worst case down to 0.51 %. The shunt's 1 nH is assumed for illustration; with 3.6 mΩ it gives an L/R corner at 573 kHz. The numbers were checked against a separate Python implementation.

Common mistakes

  • Checking accuracy only at full scale; offset errors dominate at light load, where efficiency or current-mode control often needs accurate readings.
  • Rounding the shunt value up so the full-scale current clips the ADC.
  • Picking a shunt power rating from average current instead of RMS current.
  • Using the TCR at 25 °C with no temperature rise, when the shunt runs 50 °C or more above ambient at full load.
  • Routing the sense traces from the shunt's power pads, which adds copper resistance with a TCR near 3900 ppm/°C.

Frequently asked questions

Why round the shunt value down instead of to the nearest value?

Rounding up would make the amplifier output exceed the ADC range at full-scale current, so the largest currents would clip, which is usually the worst place to lose the reading because that is where overcurrent protection acts. Rounding down only costs a few percent of resolution. If you round up anyway, reduce the headroom figure to check that you still stay inside the range.

Should I use the worst-case or the RSS error?

Worst case if the measurement guards something that must not fail, such as an overcurrent trip threshold. RSS is a better estimate of what a typical unit will show, because it is unlikely that tolerance, TCR, gain error and offset are all at their limits in the same direction. Datasheet maximum values are often several standard deviations from typical, so RSS of maximum values is still conservative.

Can calibration remove these errors?

A one-point gain calibration at a known current removes the shunt tolerance and the amplifier gain error at the calibration temperature. A zero-current reading removes most of the offset. What remains is drift: TCR times the temperature change after calibration, offset drift, and gain drift. That is why the ΔT input should be measured from the calibration temperature, not from 25 °C, if you calibrate.

What is the shunt inductance for, and where do I find it?

A low-value shunt's own inductance, plus the inductance of the sense traces, adds L·di/dt to the voltage the amplifier sees. With a few milliohms and a nanohenry, the L/R corner is in the hundreds of kilohertz, so fast current edges appear as spikes. Many metal-strip shunt datasheets quote inductance below 1 to a few nanohenries. An RC filter at the amplifier input with a time constant equal to L/R cancels the effect to first order.

How were the numbers checked?

A separate Python implementation computes the same quantities and finds the E24 value by listing every E24 value across sixteen decades rather than by the decade arithmetic used in the calculator. All values in the example match to the displayed digits. The split into ratio and offset errors follows ST's application note AN5849 and the shunt-selection example in TI's INA302/INA303 datasheet.

References

  1. STMicroelectronics, application note AN5849, "Expected output error for a current sensing amplifier" (offset, gain and application-dependent errors including shunt tolerance and tempco).
  2. Texas Instruments, INA302/INA303 datasheet (SBOS775), section "Selecting a current-sensing resistor" (splitting the error budget between offset and gain error and deriving the shunt value and its power).
  3. STMicroelectronics, "Accuracy of current sensing" (chapter 4 of ST's current-sensing guide): offset-type errors dominate at low current, ratio-type errors at high current.
  4. The values on this page were recomputed independently in Python.

Results are engineering estimates. Verify your design with datasheets, simulation and measurement.

Related calculators