Guillaume Gastebois schrieb: > Hello, > > I try both {0x00, 0x3f, 0x03, 0x26}, and {0x00, 0x3f, 0x00, 0x26}, > you can find result under : > http://ggastebois.free.fr/lide90_snoop/17_test1.tar > and http://ggastebois.free.fr/lide90_snoop/17_test2.tar
Looks a lot better. The offset*.pnm actually show a black image, coarse.pnm is gray. That is good. I suspect you are still running with the change in thresholds?: @@ -4545,9 +4546,9 @@ val = first_line[i * 2 * channels + 2 * j + 1] * 256 + first_line[i * 2 * channels + 2 * j]; - if (val < 10) + if (val < 1000) cmin[j]++; - if (val > 65525) + if (val > 40000) cmax[j]++; } Please undo that change. Should give you a nice 2-3-step offset calibration, that actually works(at least i hope so). Regarding the output format of the AFE, stay with {0x00, 0x3f, 0x03, 0x26} for now. This does not seem to make any difference, but there are suspicously many 16 bit words with the binary pattern ".... .fgh .fgh ...."(that is, the two middle nibbles share the lower 3 bits). We may be sampling the digital image data at the wrong times. As the most significant byte seems to come through correctly, this does not need immediate fixing. (On a second thought, this may affect the offset calibration. See the thesholds. We'll see.) Regards, Pierre