Engineering Blog
Sensor Signal-Chain Calibration and Scaling Mistakes
A sensor result is created by the complete chain: measurand, transducer, excitation, wiring, analog conditioning, reference, ADC, calibration model, and environment. Correct scaling at one nominal point is not the same as trustworthy measurement.
- Reading Time
- 17 min
- Difficulty
- Intermediate
- Last Updated
- October 3, 2026
Separate Transfer Function From Error
A linear transfer function converts the measured electrical signal into engineering units. Calibration estimates its coefficients; uncertainty describes what remains unknown.
Formula reference
Linear calibration and ADC scaling
y = mx + bm = (y₂ - y₁) / (x₂ - x₁)b = y₁ - mx₁V_code ≈ Code · V_ref / (2^N - 1)Variable definitions
- x
- measured signal or ADC code
- y
- calibrated engineering value
- m
- calibrated sensitivity or slope
- b
- calibrated zero offset
The Sensor Signal Conditioning Guide develops bridge, calibration, ADC, and scaling fundamentals.
Ten Common Measurement Mistakes
1. Calibrating the sensor but not the complete channel
Excitation, wiring, amplifier gain, offset, ADC reference, conversion math, and mechanical installation all contribute. Calibrate the assembled measurement path when system accuracy matters.
2. Treating sensitivity as a universal constant
Sensitivity can depend on excitation, supply, temperature, frequency, load, orientation, wavelength, pressure medium, aging, and operating point.
3. Ignoring zero offset
A gain-only correction cannot remove bridge imbalance, amplifier offset, mechanical preload, dark current, or zero-g bias. At least two calibration points are needed for slope and offset.
4. Using nominal excitation in a ratiometric system
Bridge and resistive sensor output often scales with excitation. Measuring with an ADC reference derived from the same source can reject excitation variation; inserting a nominal value can reintroduce error.
5. Scaling ADC codes without headroom
A signal chain that reaches exactly zero and full scale at nominal conditions clips with offset, tolerance, overrange, drift, or transient input. Reserve analog and digital headroom.
6. Confusing resolution with accuracy
Extra ADC bits provide smaller code steps, not guaranteed absolute accuracy. Noise, reference error, INL, offset, gain error, and sensor uncertainty remain.
7. Adding error percentages without defining the model
Some errors are correlated or systematic, others independent or random. Worst-case sums, root-sum-square estimates, and calibrated residuals answer different questions.
8. Filtering before understanding bandwidth
Filtering can reduce noise but also attenuate the measurand, delay transients, or hide instability. Define sensor bandwidth, sampling, aliasing, and response-time requirements first.
9. Using room-temperature calibration everywhere
Offset, sensitivity, bridge resistance, amplifier parameters, references, speed of sound, photodiode current, and mechanical structures can drift with temperature.
10. Reporting more digits than the uncertainty supports
A calculator can display precise arithmetic while the measurement remains limited by calibration traceability, noise, tolerance, repeatability, hysteresis, and environmental effects.
Practical Examples
Load-cell bridge
A 2 mV/V load cell at 5 V produces only 10 mV full scale. Bridge offset, excitation error, amplifier offset, gain tolerance, noise, and mechanical preload can all be significant relative to that span.
Pressure sensor scaling
Mapping nominal 0.5–4.5 V to pressure is a useful start. Production accuracy needs actual endpoint calibration, supply and temperature behavior, ADC reference accuracy, and overrange handling.
Accelerometer tilt
Static tilt equations assume gravity dominates. Linear acceleration, vibration, cross-axis sensitivity, offset, scale error, and mounting alignment corrupt the inferred angle.
Ultrasonic range
Distance uses round-trip time and speed of sound. Temperature, target geometry, threshold detection, airflow, humidity, ringing, and blanking time create errors beyond timing resolution.
Measurement Review Workflow
- 1. Define measurand range, bandwidth, response time, accuracy, resolution, and uncertainty target.
- 2. Document sensor interface, excitation, output impedance, offset, sensitivity, nonlinearity, and environmental limits.
- 3. Allocate amplifier gain and ADC range with headroom for tolerance, drift, faults, and overrange.
- 4. Choose ratiometric or absolute referencing intentionally and model reference error.
- 5. Build an error budget separating systematic, random, correlated, and calibratable terms.
- 6. Select calibration points, standards, fixtures, temperature range, and recalibration interval.
- 7. Filter and sample according to bandwidth, aliasing, settling, latency, and noise requirements.
- 8. Validate the assembled channel over production, temperature, supply, mounting, and aging conditions.
Summary
Calibrate the complete measurement path, preserve headroom, separate resolution from accuracy, and state the uncertainty model. Sensor scaling becomes dependable only when excitation, offset, gain, reference, noise, environment, and physical installation are included.
Support reference
FAQ
What is the difference between sensor calibration and scaling?
Scaling converts signal units into engineering units. Calibration compares the actual system with known references and derives corrections for observed errors.
Why is two-point calibration commonly used?
Two distinct reference points determine both linear slope and offset. One point can correct only one of those terms unless the other is independently known.
What does ratiometric measurement mean?
The sensor output and ADC reference track the same excitation source, allowing source variation to cancel in the code ratio within the system's limits.
Does a higher-resolution ADC improve sensor accuracy?
Not automatically. It improves nominal code size, while total accuracy still depends on noise, reference, ADC errors, analog conditioning, sensor error, and calibration.
How much ADC headroom should I leave?
Base it on worst-case offset, gain, overrange, transients, tolerance, drift, and fault behavior. There is no universal percentage.
How should sensor errors be combined?
Use worst-case sums for guaranteed bounded terms when appropriate, statistical methods for justified independent random terms, and keep systematic errors separate from noise.
Why does a calibrated sensor drift later?
Temperature, aging, mechanical stress, humidity, contamination, reference drift, mounting changes, and recalibration interval can move the response.
Can a calculator provide measurement uncertainty?
It can propagate a defined model, but credible uncertainty also needs source specifications, distributions, correlations, calibration data, and traceability.