| 摘要 |
<p>PROBLEM TO BE SOLVED: To suppress an operation error caused by 16-bit integer operation without increasing operational quantity at the time of realizing (8 points×8 points) two-dimensional inverted discrete cosine transformation at a high speed by executing the 16-bit integer operation in parallel of four lines. SOLUTION: Row operation parts 1, 2 respectively execute 8-point one- dimensional inverted discrete cosine transformation in a row direction. Column operation parts 3, 4 respectively input results obtained by substituting the operation results of the row operation parts 1, 2 by a substitution part 5 and execute 8-point one-dimensional inverted discrete cosine transformation in a column direction. Each of the operation parts 1 to 4 is provided with a 16-bit four- parallel adder/subtracter for executing the addition/subtraction of 16-bit data in parallel of four lines and a positive/negative symmetrical rounding 16-bit four-parallel product sum computing element for executing product sum operation for adding 0x400 when the multiplied result of 16 bits×16 bits is positive, adding 0x3fff when the multiplied result is negative, segmenting upper 16 bits including one code bit, and adding 16-bit data to the segmented 16 bits.</p> |
