Design of N-D digital filters with finite wordlength coefficients

A design technique for N-dimensional (N-D) IIR (infinite impulse response) digital filters is developed. Instead of using optimal step size in jth iteration of Fletcher-Powell algorithm, a suboptimal step size is used to increase the overall computation efficiency. Criteria are developed for the suboptimal step size to keep the mapping matrix D/sub j/ positive definite and to ensure the stability of the transfer function. The approach does not require checking the positive definiteness of the mapping matrix D/sub j/ or the stability of the filter. A optimization algorithm with finite-wordlength coefficients is developed. The algorithm minimizes the performance error caused by truncating the filter coefficients to a specified number of bits and provides a means for finding the smallest number of bits that results in an acceptable error. Two examples of 2-D digital filters are designed by the proposed technique.<
>