Real-time algorithms for estimating jerk signals from noisy acceleration data

Abstract

The time derivative of acceleration, sometimes called a jerk, plays an important role in vibration control, ride comfort evaluation and so on. It is usually required to estimate a jerk from a noisy acceleration signal. Simple numerical differentiation of a filtered acceleration signal often fails to give satisfactory result. A model-based estimation technique is proposed in this paper. Behavior of a jerk signal is modeled by a random walk process. Then the problem of estimating the jerk is formulated as a state estimation problem. On the assumption that the observation noise has a colored noise component, a Kalman filter and an filter are derived for estimating the jerk. The performances of both estimators are compared by simulation. Although both filters provide comparable performances for ideal case, the filter with appropriate use of a priori information has better robustness than the Kalman filter and is more useful in real applications.