Design of finite memory filters
Date
1989
Authors
Xu, Hao
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis deals with the design of finite memory filters for both the one and the two dimensional cases.
Finite memory filters have many applications in real time control, estimation and identification, particularly in cases where the information about the system dynamics and the noise statistics is not precisely known. A recursive finite memory filter is developed to fit an nth order polynomial to equally spaced measurements. The parameters of the filter are determined by applying a weighted least-squares performance index, and therefore the stability of the recursive finite memory filter is guaranteed. The implementations of the nonrecursive and recursive finite memory filters are obtained and the performance of both filters is compared.
The performance of the finite memory filter is compared with the one of discrete Kalman filter. It is shown that the finite memory filter requires considerably less computations than the Kalman filter. The Kalman filter showed a better performance in the presence of white Gaussian noise while in the presence of non-gaussian noise both filter performed similarly. A 2-D finite memory filter is developed by a second order bivariant polynomial to a set of 2-D measurements. A weighted least-squares performance index is used to ensure stability and the performance of the recursive and nonrecursive implementation is compared.