Error feedback for noise reduction in 2-D digital filters with quadrantally symmetric or antisymmetric coefficients

摘要

A novel method of applying the error feedback with quadrantally symmetric or antisymmetric coefficients in two-dimensional recursive digital filters is proposed. In order for the error feedback coefficients to be represented by the set of integers, the output noise variance of the filter is discretely optimized by using the depth-first-search strategy in the branch and bound mixed integer linear programming (BBMILP). The proposed error feedback leads to significant improvement in the roundoff noise characteristics and results in significant savings in hardware as well as speed of implementation.