biasdrift

Bias drift is the slow, time-varying change in the bias (offset) of a measurement channel, producing a systematic error that evolves over time. Unlike random noise, drift alters the baseline of the signal and can accumulate, degrading long-term accuracy in sensors and instrumentation.

Commonly observed in inertial sensors (gyroscopes and accelerometers), optical sensors, temperature sensors, and electronic amplifiers, bias drift is driven by temperature dependence, component aging, supply-voltage fluctuations, mechanical stress, humidity, and radiation. In many systems it manifests as a low-frequency or quasi-static trend rather than high-frequency noise.

Drift is often modeled as a bias term that follows a stochastic process, such as a random walk or first-order Markov process. In Kalman filters and other Bayesian estimators, bias can be treated as an augmented state to be estimated alongside the primary quantity, enabling online correction.

Mitigation strategies include regular calibration against reference standards, temperature compensation, and the use of drift-stable components. Hardware measures such as precision references, thermal design, and proper shielding help. Sensor fusion that combines multiple modalities or redundant sensors can reduce the impact of bias drift, while detrending and post-processing can remove slow trends in data when re-calibration is not possible.

Understanding and managing bias drift is important in aerospace, robotics, metrology, and automotive systems, where long-term accuracy of orientation, position, and measurements is critical.