In this paper we present a novel approach to the realization of floating-point computations, which are typical of various digital signal processing tasks. The proposed idea, referred to as "the two-accumulator concept", overcomes the phenomenon of accuracy reduction of floating-point computations in case of a long series of additions of small numbers with relatively large intermediate results. This phenomenon is a substantial drawback of the classic floating-point arithmetic, particularly in signal processing applications.
展开▼
