Quantized FIR filter design using compensating zeros

This article presents a design method for translating a finite impulse response (FIR) floatingpoint multiplierless filter design. Conventional wisdom dictates that finite word-length (i.e., quantization) effects can be minimized by dividing a filter into smaller, cascaded sections. In this design method, it is shown how to quantize the cascaded sections so that the finite wordlength effects in one section are guaranteed to compensate for the finite word-length effects in the other section. This simple method called, "compensating zeros," ensures that: (1) the quantized filterʹs frequency response closely matches the unquantized filterʹs frequency response (in both magnitude and phase); and (2) the required hardware remains small and fast.