Search of optimal realization matrix for filter implementation by using a genetic algorithm

It is known that numerical performance of a discrete filter is strictly dependant to the state space realization adopted. In particular realization matrix plays an important role into determination of a good representation when a finite length arithmetic is used. The optimal representation problem has not in general a solution in closed-form and numerical optimization algorithms must be adopted. Genetic Algorithms, a new class of optimization procedure that simulate natural adaptation, have shown to be a valid alternative respect to a classical approach. This procedure allows to find near-optimal solution by avoiding to deep in local minima.