PARALLEL DESIGN OF DISCRETE-COEFFICIENT FIR FILTERS

Abstract

Among all existing algorithms for designing discrete-coefficient FIR filters, the mixed integer linear programming (MILP) algorithm is the only known approach that can guarantee global optimality in the minimax sense. However, the high computation cost of MILP prohibits its application to the design of long length filters. In this thesis, we propose a new approach of parallel MILP algorithm to design discrete-coefficient FIR filters. The basic idea is to divide the original problem into several independent parts and distribute them to a cluster of workstations, which solve the respective subproblems simultaneously. The master slave model is employed in our algorithm as the processor cooperation control mechanism, and some techniques for work distribution, tree search, interprocess communication, etc., are discussed. Furthermore, various design examples are presented and the analytical results show that the filter design time can be reduced greatly when multiple processors are used, thus demonstrating the success and effectiveness of our algorithm.