Learn / Sensors and actuators / Sensor basics

Calibrating Sensors and Estimating Uncertainty

ReviewedAdvanced ProAbout 6 min
BME280
In short

Calibrating Sensors and Estimating Uncertainty: the short answer

Every sensor reading has an error, and calibration measures and corrects it. Compare the sensor with a reference that is clearly better (at least three to four times more accurate) at several points across the range you care about.

Level
Advanced
Time
About 6 min
Topic
Sensors and actuators
You will learn to

Calibrate a sensor against a reference, correct offset, gain and non-linearity, and state how much a reading can be trusted.

Before this: Interpreting Sensor Data, Accurate ADC Readings: Calibration and Filtering

Every sensor reading has an error, and calibration measures and corrects it. Compare the sensor with a reference that is clearly better (at least three to four times more accurate) at several points across the range you care about. The simplest correction is an offset (add a constant). A two-point calibration corrects offset and gain, a straight line through two known points. If the residuals after a straight-line fit show a curve, fit a second-order polynomial or use a lookup table with interpolation.

Some references are easy to get. Ice water that is properly stirred sits at 0 °C, and boiling water is about 100 °C at sea level (lower at altitude, for example about 95 °C in Shimla). Known weights calibrate a load cell with an HX711. Saturated salt solutions in a sealed jar hold fixed humidity: table salt (sodium chloride) gives about 75 % RH. A good multimeter calibrates voltage readings. Record the date, reference, conditions and resulting constants with each sensor.

State the uncertainty, not just the value. Repeat a reading many times and compute the standard deviation to measure noise. Add the reference's own uncertainty, the sensor's resolution and any drift with temperature. A reading of 24.6 ± 0.3 °C is honest; 24.613 °C from a ±0.5 °C sensor is not. For a lab record, plot sensor reading against reference, show the fitted line and residuals, and quote the largest residual as the calibration error.

Accuracy, precision and resolution are different

Resolution is the smallest change a reading can show. Precision is how close repeated readings are to each other, and accuracy is how close they are to the true value. A sensor can show 0.01 °C resolution and still be 1 °C wrong.

Calibration compares the sensor with a better reference at known points and corrects the difference. For a temperature sensor, ice water at about 0 °C and a trusted thermometer give two points, from which you work out offset and slope.

Estimating uncertainty

Take many readings under the same conditions and calculate the average and the standard deviation. The standard deviation shows the spread, and the uncertainty of the average is that value divided by the square root of the number of readings. Combine the sources of error, such as the reference's own uncertainty and the sensor's drift, by adding them in quadrature when they are independent.

State a result with its uncertainty, for example 21.4 ± 0.3 °C, and say how it was found.

Drift and conditions

Sensors drift with age and with temperature, and a calibration is valid only for the range you tested. Check against the reference at intervals, and record the date. Keep the self-heating of the sensor in mind, since a sensor driven hard heats itself and reads high.

Common mistakes
  • Treating a high-resolution reading as an accurate one.
  • Calibrating at one point and using the sensor over a wide range.
  • Quoting a result with no uncertainty.
Advanced lesson · Pro plan

Practise this lesson with Pro

The explanation above is free for everyone. The practice for Advanced lessons in every domain are part of the Pro plan at ₹999 a month.

With Pro you also get: a worked example, a hands-on task, a 3-question check.

Common questions

What will I learn in "Calibrating Sensors and Estimating Uncertainty"?

Calibrate a sensor against a reference, correct offset, gain and non-linearity, and state how much a reading can be trusted.