Round-off Error of Floating-Point Digital Filters
The MIT Press eBooks · 1969
Abstract : This paper is concerned with the accuracy problem of digital filters. When a computer using fixed-point arithmetic is used for the realization of the filter, the problem has been studied extensively. The only result available for floating-point arithmetic realization of digital filters is an absolute upper bound on the accumulation of the round-off error using a deterministic approach. The bound is over pessimistic as to be expected. In this paper, a statistic approach is adopted to treat the round-off error problem for floating-point computer realization of filters. (Author)