A method of designing M-D finite-precision FIR (finite-impulse response) digital filters is proposed. The method operates by predistorting the frequency response prior to quantization. The predistortion is designed to counteract degradation caused by the finite wordlength. The error measure used is a convex function of the filter coefficients. Such a property enables the authors to replace the time-consuming optimization techniques using direct search with the very fast techniques utilizing the first and second derivatives. Design examples are given for both the 1-D and the 2-D case. The 1-D results are compared to a method used by J.J. Nielsen (1989).
展开▼
