We are using the ADV7619 in 4:2:2 16-bit mode output and we came across some kind of chroma subsampling issue.
The ADV7619 is working at 1080p60 with a RGB-FULL computer source and here are some screenshots taken from our board while using a pattern with column width red, pink, blue, green colorbar with 2x pixel width that begin on even pixel position.
First of all, with the TEST-3-bars-2pix-with-pink.png, we can see that we have some blurry pixels around the 2x-pixel width columns. This picture was taken while the DS_WITHOUT_FILTER=0, where the chroma filter is active.
After, if we disabled the filter by setting DS_WITHOUT_FILTER=1, this is the result we get in the TEST-3-bars-2pix-with-pix-filter-disabled.png file, there is no transition between color column and the pixel value is the same as the one we put in the pattern, which is good.
If we change the test pattern by moving the pattern by one pixel to the right, the color bars will begin on odd pixel position, this is what we get when we enable or disable DS_WITHOUT_FILTER bit. See TEST-red-green-blue-odd.png and TEST-red-green-blue-odd-no-filter.png. As you can see, it seems that the result is worse when the pattern is on odd pixel position.
It seems that there is some kind of delay between luma and chroma when this filter is enabled.
Is that the case?
How this filter works exactly?
Is it doing some kind of mean value between Cb and Cr pixel values?
Is there some Cb or Cr values that are lost?
Do we really need to output from ADV7619 in 4:4:4 in order to get rid of this problem?
It seems that disabling this filter will improve performance but for some reasons, when we capture black text over a white background, some pink shadow will come up. In order to avoid this pink shadow around text, we need to enable this chroma filter.
Do you have any ideas what is going on?