Measurement uncertainty is the range within which the true value of a measurement is expected to lie, stated with a confidence level. For a data acquisition system, it combines everything between the physical quantity and the number in your file: the sensor, signal conditioning, the DAQ hardware, noise, resolution and calibration. A reading of 50.00 psi means little until you know whether it is ±0.05 psi or ±2 psi.
This article is part of our data acquisition guide. Here we go deeper on how to build an uncertainty budget for a DAQ measurement, where uncertainty usually comes from, and how to keep it under control once the system is running.
Accuracy, resolution and uncertainty are not the same
| Term | What it describes | Common mistake |
|---|---|---|
| Resolution | The smallest change the system can display, set mainly by ADC bits and range | Assuming more bits means a more accurate measurement |
| Accuracy (spec) | The manufacturer’s stated maximum error for one component under stated conditions | Using only the DAQ card’s accuracy and ignoring the sensor |
| Uncertainty | The combined doubt about the final result from every source, with a confidence level | Adding worst-case errors instead of combining them statistically |
| Precision (repeatability) | How closely repeated readings agree | Treating a stable reading as an accurate one |
Where DAQ uncertainty comes from
- Sensor: accuracy, nonlinearity, hysteresis, temperature effects and long-term drift. Often the largest source.
- Signal conditioning: amplifier gain and offset error, excitation stability, filter effects. See signal conditioning.
- DAQ hardware: gain error, offset error, nonlinearity, and their temperature coefficients.
- Noise: electrical noise, ground loops and pickup in cabling.
- Resolution: quantization of the analog-to-digital converter.
- Sampling: aliasing and timing errors if the sampling rate and anti-aliasing filter are wrong for the signal.
- Calibration: the uncertainty of the reference used to calibrate the chain, and how long ago it was done.
- Environment and installation: temperature, mounting, cable routing and, for thermocouples, cold-junction compensation.
How to build an uncertainty budget
- Define the measurand. What exactly is being measured, in what units, over what range and conditions.
- List every source in the chain from the list above.
- Convert each to a standard uncertainty in the same units. A datasheet limit of ±a with no other information is usually treated as a rectangular distribution: divide by √3. Noise given as an RMS value is already a standard uncertainty. Quantization is one LSB divided by √12.
- Combine independent sources by root-sum-square (the square root of the sum of the squares), not by simple addition.
- Expand the combined uncertainty with a coverage factor, typically k = 2 for about 95% confidence.
- Check the dominant terms. Improvements only matter where the big numbers are.
This follows the general approach of the internationally used Guide to the Expression of Uncertainty in Measurement (GUM). Regulated or accredited measurements may need a more formal treatment.
Worked example: a pressure measurement
A 0–100 psi pressure transducer with a 0–10 V output (10 psi per volt) is read by a 16-bit DAQ input on a ±10 V range. The figures below are illustrative; use the datasheets for your actual components.
| Source | Datasheet or measured value | Standard uncertainty (mV) |
|---|---|---|
| Transducer accuracy | ±0.25% of full scale = ±25 mV, rectangular | 25 ÷ √3 = 14.43 |
| DAQ absolute accuracy at this range | ±3 mV, rectangular | 3 ÷ √3 = 1.73 |
| Measured noise | 0.5 mV RMS | 0.50 |
| Quantization | 1 LSB = 20 V ÷ 65,536 = 0.305 mV | 0.305 ÷ √12 = 0.088 |
| Combined (root-sum-square) | 14.55 | |
| Expanded, k = 2 (about 95%) | 29.1 mV = ±0.29 psi |
The transducer contributes about 98% of the combined variance. Replacing the 16-bit DAQ with a 24-bit one would barely change the result; a better transducer, or calibrating the transducer and DAQ together as one system, would.
Common mistakes
- Buying bits instead of accuracy. Resolution is rarely the limit; sensor and gain errors usually are.
- Adding every worst case. Simple addition overstates uncertainty for independent sources; root-sum-square is the standard method.
- Ignoring temperature. Datasheet accuracy is often specified over a narrow temperature band or for a set time after calibration.
- Calibrating parts, not the chain. An end-to-end calibration of sensor plus DAQ removes several error terms at once.
- Forgetting the reference. Your calibration can be no better than the standard you calibrate against.
Keeping uncertainty under control in service
Uncertainty grows after installation as sensors drift and conditions change. A running system stays trustworthy when it has:
- A calibration interval set from the drift data of each sensor and channel, recorded per channel.
- Periodic checks against a reference, ideally automated so they actually happen.
- Software that stores calibration coefficients, dates and the uncertainty with the data, so results can be traced later.
- Alerts when a channel’s reference check moves outside its expected band.
For lab and engineering leaders
An uncertainty budget is what makes data defensible, whether for a customer specification, a regulator or a published result. Building it during design shows where money should go, often to a better sensor rather than more expensive DAQ hardware.
Need measurements you can defend?
Sciotex designs and builds custom data acquisition systems on any hardware and language, including NI and LabVIEW, with calibration and traceability built into the software. Designed, programmed, built and supported in the US since 1997.
Frequently asked questions
Measurement uncertainty is the range within which the true value of a measured quantity is expected to lie, at a stated confidence level, combining every source of error in the measurement chain.
Accuracy is usually a component specification for maximum error under stated conditions. Uncertainty is the combined doubt about the final result from all sources, including the sensor, conditioning, DAQ, noise, resolution and calibration.
Convert each source to a standard uncertainty in the same units, combine independent sources by root-sum-square, then multiply by a coverage factor, typically k = 2 for about 95% confidence.
Only if quantization is a significant term. In most systems the sensor, gain error or noise dominate, so a 24-bit DAQ changes the result very little compared with a better sensor or end-to-end calibration.
Set the interval from each channel’s observed drift, the sensor manufacturer’s recommendation and the consequences of an error, and verify between calibrations with reference checks.