| Extremal Optimization is a recent method for solving hard optimization problems. It has been successfully applied on many optimization problems. Extremal optimization does not share the disadvantage of most of the other evolutionary algorithms, which is the tendency to converge into local minima. Design of finite word length FIR filters using deterministic techniques can guarantee optimality at the expense of exponential increase in computational complexity. Alternatively, Evolutionary Algorithms are capable of converging very fast to a minimum, but have higher chances of failure if the ratio of feasible solutions is very less in the search space. In this thesis, a set of feasible solutions are determined by linear programming. In the second step, Extremal Optimization is used to further refine these results. This strategy helps by reducing the search space for the EO algorithm and is able to find good solutions in much shorter time than the existing methods. |
