Originally Posted by zoyd
What I don't get is the companion document found here
that talks about high frequency luminance errors present in all
video signals based purely on the fact that video signals are encoded non-linearly. Gregr, can you explain this one?
Zoyd sent me a note asking that I comment on this, but I just don't have any time for a week or so to get into any detail. The basic issue is a constant vs non-constant luminance system. All of our video source systems (based on YPbPr or YCbCr encoding) are non-constant luminance. Basically a constant luminance system would encode luminance signals using linear RGB components, i.e. it would conform to the CIE concept of luminance. But instead we encode luminance as a combination of gamma corrected RGB signals, which Poynton (myself and most other video engineers) refer to as a luma signal to make that distinction clear. In a constant luminance system we would combine the linear RGB components into a luminance signal and then apply gamma correction to that luminance signal at the source - i.e. we would have Y' = Y(r,g,b) ^ 1/gamma rather than Y' = Y(r^1/gamma, g^1/gamma, b^1/gamma), and then in the display we would degamma the Y signal and recover the linear R,G,B components using a matrix with the degamma'd Pb,Pr signals. i.e. we would have Y = Y' ^gamma, Pb = Pb' ^ gamma, Pr = Pr' ^ gamma followed by YPbPr to RGB matrix conversion. But instead we perform Y'Pb'Pr' to R'G'B' matrix conversion followed by individual R = R' ^ gamma, G = G' ^ gamma, and B = B' ^ gamma. A basic reason this non-constant luminance system was adopted was because the CRT intrinsically performed the individual R',G',B' to R,G,B operations for free and that was important in 1953. So as a consequence of this compromised approach part of the luminance signal really appears in the color difference signals (Pb',Pr' or Cb',Cr' for digital systems). Hence, when the color difference signals are bandwidth limited (as they are in all video systems through analog filtering or digital subsampling/filtering such as 4:2:2 or 4:2:0 digital formats) some of the high frequency luminance information is lost. I don't have time to read and comment on the document you refer to above, but that is the origin of high frequency luminance errors in our video systems. (Actually there is an additional reason in the case of Rec 601 systems because the wrong luma coefficients are also used, but that is independent of a non-constant luminance system.)