Integer, floating point, and block floating point data formats have been analysed in order to establish tools for selection of a data format offering large dynamic range and low quantization and saturation noise constrained by a limited number of bits and minimal processing demands. A (new) modified block floating point format applying fractional exponents seems to have attractive properties for the storage of Gaussian distributed data. (Related projects - see: www.dcrs.dtu.dk).
|Effective start/end date||01/01/1997 → 31/01/1999|
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.