Digital HDTV: Why Bits are Not Just Bits
In its quest to assimilate other industries, the computer industry has adopted the rallying cry, “Bits are bits!” At face value the expression is tautological, but it is intended to imply that once a signal is digitized it enters the domain of computing. Sadly, the expression has come to imply that the technical parameters of the original domain can be ignored! I will give three examples. • According to “Bits are bits,” the proposal is made by the computer industry that the pixels of an image should be represented and processed independent of each other. However, this notion violates Shannon's sampling theorem: Without a certain degree of interdependence among samples—otherwise known as filtering!—a sequence of samples cannot be accurately reconstructed. • The computer industry represents pixels in RGB components, generally no reference to transfer function, no reference to primary chromaticities, and no reference to white point. Accurate reproduction of color is impossible without knowledge of these parameters, and inaccurate reproduction of color is certainly the norm in computing. • Finally, “Bits are bits” seems to imply that it is desirable to generalize the raster to dimensions of m by n, without allowing specification of any numerical values. In particular, the proposal is made by some participants in the computer industry to adopt an HDTV transmission system that makes no reference to an underlying raster standard. While this is a desirable goal for the design of a software system, it is hardly feasible to design a CCD camera or a plasma display panel without knowledge of the dimensions of the array! — The television industry must take some responsibility for the gap between computer image representation and digital video representation: The television industry has, on many occassions, adopted technical parameters that are antagonistic to computing. Examples from the past include the choice of asymmetrical coding for luma in Rec. 601 instead of the obvious symmetrical coding, and cositing of subsampled chroma instead of the interstitial subsampling that was chosen for JPEG. A very recent example is the choice of non-square pixels in the consumer digital video cassette (DVC) media and its DV interface standard. — A new dichotomy looms on the ATV front. The current ATSC standard calls for luma to be coded according to the Rec. 709 coefficients, instead of the conventional, widely used Rec. 601 coefficients. No compelling scientific study has proven any performance advantage over the Rec. 601 coefficients. However, adopting new coefficients would introduce a huge disadvantage: The Y'C
- Published
- 1997-02
- Content type
- Original Research
- DOI
- 10.5594/M00216
- ISBN
- 978-1-61482-926-3