Trade-offs in linear filter design (Q1902565)

The paper examines trade-offs arising in the design of linear filtes. It is known that there exist fundamental limits on filter performance -- limit that are independent of the method of design. These limits can be used to determine a priori whether desired design specifications are achievable. Analogous constraints are well-known in the area of feedback control systems, and in the paper they are articulated for the filtering problem. Design problems usually have multiple, conflicting objectives, necessitating trade-offs between these objectives. Here, low sensitivity to data which may not be precisely known, is considered as an important objective, which must be traded against other objectives. The filter design trade-offs developed here are expressed in terms of two sensitivity functions. They are defined to be the relative sensitivities to input and output errors respectively. Also the effect of model error on filter performance is considered. The fundamental constraints which hold for all linear unbiased estimators are presented. These constraints provide limits on achievable performance in filter design. An illustration of the results is given with a simple example. The results can be of interest for specialists in filter design.