SFILTEX: toward an expert system for time-series filter design

This paper describes an attempt to develop a statistical expert system (FILTEX) as an intelligent aid for time-series filter design. To this end the knowledge of the filter design strategy is represented in Prolog and coupled with numerical routines of a general purpose signal processing package. This knowledge-based system is conceived as a set of independent knowledge sources integrated into a system by a blackboard mechanism which embodies overall control of the filter design process. Modularity and flexibility of knowledge representation in such a framework preserve usability of the evolving system during its development from the original numerical package to an expert system for filter design. This approach seems to be more flexible than the use of shells and less time consuming than building from scratch. A novel method for incorporating classical statistical information into an uncertainty management mechanism is presented. Experimental results confirm the feasibility of the approach and set directions for further research.