Digital Filter Structures and Quantization Error Analysis

Abstract

Practical implementations of digital filters introduce errors due to finite-precision data and arithmetic. Many different structures, both for FIR and IIR filters, offer different trade-offs between computational complexity, memory use, precision, and error. Approximating the errors as additive noise provides fairly accurate estimates of the resulting quantization noise levels, which can be used both to predict the performance of a chosen implementation and to determine the precision needed to meet design requirements.