Sensor noise and bias
Separate random variation from a persistent measurement offset.
Research note · November 14, 2026Research question
Does a controller respond differently to noise and to a small sustained bias?
Method
- Define a clean sensor trace and its unit and frame. Create separate perturbations for random noise and constant bias, keeping their parameters and random seeds explicit.
- Run the same controller against each trace with the same initial state. Compare task error, control effort and operating-limit crossings rather than relying on visual smoothness.
- Vary perturbation magnitude over a stated range and retain all cases. Do not tune the controller on the final evaluation traces and then present them as unseen tests.
Records to retain
- Unperturbed sensor trace
- Noise model and seed
- Bias value and duration
- Controller response and task metrics
Source basis
Measurement uncertainty can have multiple components, and repeated variation is not the only source of uncertainty. NIST’s distinction between Type A and Type B evaluation is useful when documenting how sensor assumptions were chosen.
Interpretation & limits
Synthetic noise is not a calibrated sensor model. A chosen distribution may omit drift, clipping and correlation. This proposed test does not establish field robustness.
Next experiments
- Add a slow-drift case after testing fixed bias.
- Compare synthetic assumptions with a small measured sensor record when available.
References
Technical reference
Implementation details supporting this note, drawn directly from the maintained documentation. These are technical references, not additional papers or new experimental results. Expand a topic to inspect its method, fields and limits.