Dithering for Floating-Point Number Representation
Rezső Dunay, István Kollár, Bernard Carlos Widrow · 2005
Dithering is widely used for decreasing the bias in fixed-point quantization and rounding. Since floating-point digital signal processors (DSP's) and floating-point arithmetic are becoming widely used, it is timely to investigate the necessity and possibilities of dithering for floating-point numbers. The paper introduces a simple model of dithers for floating-point, and discusses its practical use. Keywords: Dither, quantization, DSP, digital signal processor, floatingpoint, roundoff.