Efficient algorithms for high resolution fuzzy controllers with piecewise linearities

Classical fuzzy controller implementations become very slow and memory intensive, when high internal or output resolution is required, because membership functions are stored as lookup tables. High resolution fuzzy controllers represent membership functions by their characteristics and thus provide low memory effort and very fast inference, composition and defuzzification algorithms. We propose a high resolution fuzzy controller with trapezoidal membership functions stored in point lists and corresponding algorithms for minimum inference, maximum composition, and centroid defuzzification. The implementation shows a 7 to 10 times acceleration and a 20 to 1000 times memory reduction. For resolutions of more than 12 bit even higher improvements can be achieved