Optimal design of finite precision FIR filters using linear programming with reduced constraints

An algorithm for the design of optimal cue-dimensional (1-D) and two-dimensional (2-D) FIR filters over a discrete coefficient space is proposed. The algorithm is based on the observation that the equiripple frequencies of a subproblem (SP) in the branch and bound (BaB) algorithm are closely related to those of neighboring SP's. By using the relationship among the SP's, the proposed algorithm reduces the number of constraints required for solving each SP. Thus, the overall computational load for the design of FIR filters with discrete coefficients is significantly alleviated, compared with the conventional BaB algorithm.