Originally Posted by janos666
You may find bigger deviance if you measure a 10% gray patch with Default and Adaptive HiRes modes (also note the variation between the repeated measurements).
I have the feeling you may understand this ADC part a bit better.
Besids what looks best subjectivel, do you know how could we set up the ADC parameters to minimize the processing on the ADC side?
You say it's 10-bit, and I want to preseve the full range of my 0-1023 input until the more precise 3D video processing.
How do I do that?
The problem is that ADC can both over- and under-shoot the VP input range and 10-bit is enough to keep relatively smooth gradients even after some processing (the solution would be trivial with a fully <=8bit chain or easy if it could only over- or undershoor the VP range.)
The gray gradient seems smooth after the manual tuning but I can't do anything about a slight near-white anomaly: like R=G=B= 255 lives it's own little life at the upper end of the gradient which follows the rules much heavily (100% white is less affected by the white balance controls than other near-white shades and near-white is also more affected than the rest of the gradient).
I guess the chain isnt't really optimized for full range input.
Sorry if I couldn't explain it. I am using my smartphone and my limited English. I will try to draw it when I got home.
May be I should compress the level with my GPU or see if it helps to convet the RGB to YCC with the GPU (the latter would be still full range but offload the VP a bit.)
This legacy range compression should have been eliminated by Rec709, just like the chroma subsampling.
I'm still working on this procedure but it appears promising. I haven't been able to fully evaluate the end result yet because I'm still optimizing and will have to do a recalibration because it throws off grayscale and gamma quite a bit. I found that manually editing the ADC Results values is not the way to go. With these you are adjusting the gain/offset of the input signal but there is another step which appears to be setting the actual mapping of the input range to digitizer levels when you perform an automatic calibration. It's the combination of setting gain/offset and then having the display autocal that got me the best results. I also found that setting gain/offset with subcontrast and brightness worked better than editing the results table.
So for example, this is what I did yesterday that so far looks like it gives good results in near black reproduction but I haven't measured yet.
For the low end:
1. At my desired contrast and gamma, my MLL limits proper display of black levels to video level 23.
So I used the brightness control to set level 9 as the first flashing bar in the black clipping pattern.
2. Performed HDMI autocalibration.
3. Check black clipping patterns and set first flashing bar back to 17-18.
This has the effect of mapping input level 16 back to the VP processor at where level 23 used to be, so that I should get betting gamma tracking as I approach black as MLL instead of true black. Need to measure to verify.
For the high end:
1. My input is YCbCr 16-255. Other discussions in this forum have shown allowing video levels up to 240 is sufficient and that allowing all Superwhite levels is not needed. The autocalibration used for the low end maps input white 235 to level 940 to allow headroom for superwhite up to 255. So I used the white clipping pattern and increased my subcontrast to clip at 240 at my desired contrast and cell level. Now I know that input level 240 is at the top of the display range. The immediate effect this has is that you have to lower light output using the CELL control not contrast. This is because CELL is after the digitizer and contrast is before so it won't screw up your input range. The other thing I noticed was a rather amazing high end peak white of 58 ftL.
So now I have something like contrast 76 and CELL 8 to get 28 ftL.
@janos666: For your full range input using HDMI you might want to try clipping peak white to 235 using subscontrast, performing the autocal, and then setting subcontrast back to 128.