Genetic Programming based multichannel identification of nonlinear systems by Volterra filters

Genetic Programming (GP) is utilized to search the optimal structure of Volterra filter in this paper. The Volterra filter with high order and large memories contains great amount of cross product terms. In stead of applying GP to search all cross products, GP is utilized to search a smaller set of primary signals which evolve to the whole set of cross products. With GP's optimization capability, the important primary signals and the associated cross products of input signals attributing most to the outputs will be chosen while the primary signals and their associated cross products of input signals which are trivial to the outputs will be excluded from the possible candidate primary signals.