| 摘要 |
PURPOSE:To assemble a convolution integration device with high accuracy and low cost by assigning limited entire bit number within a narrowed definition range and applying prescribed processing to a filter function at the outside of definition. CONSTITUTION:A representative filter function used for a convolution integration operation has a function within a range of + or -0.5 other than a central point because of a numerical limitation of itself, a digit of 1/2 is assigned to a sign S bit in the format and the data within the range of -0.5<x<0.5 is expressed with accuracy of n bits below the decimal point. As for the middle point F0 of the filter function, the operation F0XDi with corresponding data Di is executed by (1/4)F0XDi+(1/4)F0XDi+(1/4)F0XDi to handle the value F0(=1.0) of the middle point as a coefficient (1/4). |
