Synthesis of low coefficient sensitivity digital filters using genetic programming

This paper proposes a new approach to the synthesis of low coefficient sensitivity digital filters using Genetic Programming (GP). GP is applied to the synthesis problem by establishing a mapping between the S-expressions and the filter structures. Genetic operators are then applied to the S-expressions in order to change the connections between the elements in the filter structures. The fitness measure that includes the coefficient sensitivity enables the selection operation to choose low sensitivity filter structures. In this paper, two coefficient sensitivity measures are used-the magnitude sensitivity and the relative sensitivity. A numerical example is presented to demonstrate that the sensitivity of the filter structure synthesized by GP is lower than that of other low coefficient sensitivity filter structures proposed so far.