|
How exact works the implemented DCT?
The DCT use a cosine table with a width of 9 bit, the internal data path is 13 bit width and will be rounded to 11 bit coefficients at the end. The error, compared to a floating point DCT, is maximal 2. The following
quantization stage with typical values, reduce the 11 bit coefficients to quantized coefficients with only 7 significant bits. So we see that the DCT errors are covered by the quantization stage completely.
How long are the pipeline delays?
The pipeline delays can not be determinate exactly. They are depended from the image samples and can vary up to 64 clocks.
JE1xx: Number of delay clock samples, after the last sample from a block is written:
Minimal: 22
Maximal: 86
JE2xx: Number of delay clock samples, after the last sample of the last row is written:
In mono chrome mode:
Minimal: number_of_pixel_per row * 8 + 26
Maximal: number_of_pixel_per row * 8 + 90
In color mode:
Minimal: 2 * number_of_pixel_per row * 8 + 26
Maximal: 2 * number_of_pixel_per row * 8 + 90
JE3xx: Number of delay clock samples, after the last sample of the last row is written:
In mono chrome mode:
Minimal: number_of_pixel_per row * 8 + 32
Maximal: number_of_pixel_per row * 8 + 96
In color mode:
Minimal: 2 * number_of_pixel_per row * 8 + 32
Maximal: 2 * number_of_pixel_per row * 8 + 96
|
