Model of computation accuracy in modular digital filters

Abstract

We show that hardware implementation of digital filters is accompanied by the problem connected with using a finite number of bits for representing coefficients, input data and interim results. This problem leads to the deterioration of magnitude response and computation accuracy. We provide grounds for using modular coding in digital filters in order to exclude errors caused by finite capacity when computing output samples. Thus, it becomes possible to meet the requirements to computational accuracy and magnitude response without increasing hardware costs and computation time.