Difference between revisions of "Talk:10359 DWT"

Revision as of 07:12, 18 November 2008

2008/11/18

1. Solved. Datapath is ok now. The mistake was in programming tile's x-sizes and shifts for a large frame (and in tests everithing was correct).

2008/11/13

1. Made datapath to/from memory for the tranform (added 2 channels) Currently get a set of 3 frames:

step 1: frame from channel 1 goes directly to 353 and to the memory at the same time.

step 2: line by line frame is read from the memory and is written back to another location line by line: "read_line - write_line - read_line - write_line ..."

step 3: line by line frame is read from the memory.

step 1, direct	step 2, 1st SDRAM read	step 3, 2nd SDRAM read

Comments:

1. Feels like locations overlap somehow - this should be checked

2. No mistakes in the simulation stage (frames are 1620x4).

2008/07/31

Solved - 1. "right part is lost" - solved - mistake was in ch2_N_fp - line size in pixels

2. Made *.bit - board doesn't see sensors

2008/07/29

1. Simulation: output image is incorrect - right part is lost

2008/07/25

1. adjusted mcontr to read/write blocks to DDR

a. verified and corrected mistakes

TODO: adjust frame size more properly

2008/07/15

mcontr for block read/write is under work.

useful note:

1. SA[24:23]= Bank[1:0]
2. SA[22:10]= Row[12:0]
3. SA[9:0]  = Col[9:0]

4. SA[12:8] - ...

2008/07/08

Basic 1D DWT for the whole frame is done. For each pixel in the Bayer Pattern should be calculated a separate DWT (because the difference can be significant?) - when I calculate mean value between neighbours (e.g., R & G) - the resulting picture is strange.