White Papers

Performance comparison of extended and unscented Kalman filters 

TIAX
Suggested by J. Ashmore

In the four decades since Kalman first proposed his Kalman filtering has become a well-established method that is used to reduce noise in a wide variety of applications. More sophisticated approaches that build on the basic principles of the Kalman filter, included extended [2] and unscented [3,4] versions, have been developed in order to address nonlinear problems. While the flexibility of Kalman filters is a significant benefit, it is important to choose the filter design carefully for a given application in order that reliable and robust results are achieved.

TIAX is in the process of developing Kalman filters for orientation sensors that incorporate two types of sensors with different noise characteristics. The state equation is nonlinear and therefore either an extended or an unscented Kalman filter must be used. An important set of questions concern:

  • In the scenarios we are concerned with, is the performance of an unscented Kalman filter significantly superior to that of an extended Kalman filter?
  • If so, is the performance improvement general to a variety of different types of noise?
  • For given input noise levels, can we predict the reduction in noise that the different types of filter can provide?
  • What are the differences in the computational efficiency of the different types of filters?

TIAX will provide a sample data set to use as a benchmark for the performance of the different filters developed. The results will assist us in our efforts to design and implement the filter that best matches our performance requirements.