On fixed-point quantization schemes

A fixed-point quantization scheme, named magnitude-dependent rounding is introduced. It permits the description and implementation of several quantization rules that are obtained according to a parameter value. This quantization scheme allows a unified approach to magnitude truncation and rounding, which are extreme cases. The implementation of digital filters that are free from zero-input granular limit cycles when implemented with magnitude truncation is considered as an application example. It is well known that magnitude truncation yields some frequency distortion and larger quantization noise when compared with rounding. Magnitude dependent rounding makes a compromise concerning these effects possible. To a large extent it allows the elimination of granular limit cycles without increasing the quantization noise and without distorting the frequency response of a digital filter