Many embedded powertrain and vehicle controls rely on speed and acceleration information. These measurements, however, are often noisy. Good filters are needed to reduce the noise. In some vehicles, acceleration sensors are not used due to cost. For those vehicles, estimators are needed to estimate acceleration. The paper introduces a new process to design speed filters, acceleration filters, and acceleration estimators.
A physics-based discrete state equation is used to describe the relationship between speed, acceleration, and jerk. Then, a Kalman filter is developed to get the optimal estimates for speed and acceleration from available measurements. Vehicle test data shows that these filters are effective in reducing noise without introducing significant time lag.
The filter design process requires little iteration because there is only one design parameter, which is the ratio of one-by-one covariance matrices of process and measurement noise. The design process is fast and straight forward. Filters with the right frequency response can be quickly generated. This result is achieved by a unique assumption about the derivative of jerk.