Investigation of trellis based filters for tracking

The authors seek to extend the work carried out by Demirbas (1981-1987) on the use of trellis diagrams for state estimation. Linear scalar examples are initially used to improve the accuracy of the algorithm and to highlight a lack of estimator robustness to modelling errors. This is subsequently overcome by using the measurement equation to propagate the trellis diagram. Nonlinear examples are then used to demonstrate estimator performance which is superior to that of the EKF. The extension to higher order systems uncovers further difficulties with the algorithm. These are examined using a linear tracking example, and overcome by quantising both the system and measurement noise sources. Finally, a nonlinear tracking example is considered and while, for this case, the trellis based algorithm does not perform as well as the EKF, the example helps to establish what conditions favour the use of the trellis based approach.<>