In the analog video world, the picture resolution is expressed in lines per picture height (LPH) and reflects the losses caused by the vertical and horizontal picture sampling processes. The choice of the transmitted bandwidth ensures near-equal horizontal and vertical resolution in the two contemporary scanning standards (525/59.94 and 625/50). The advent of digital processing of video signals has introduced a new twist in the concept of picture resolution. This article will analyze some of the implications.
Sampling frequency considerations
The ITU-R BT.601 Recommendation (Rec 601) is the first international agreement on how to migrate from two incompatible analog composite television standards (525/59.94 and 625/50) to a common component digital sampling concept. The dominant digital coding is based on the use of one luminance (E' Y ) and two scaled color-difference (E' CB and E' CR ). Early proposals for the sampling frequency of the E' Y signal specified a multiple of the subcarrier frequency (F sc ) of the associated composite video signal. This resulted in the 4:2:2 strategy of sampling E' Y at 4F sc and each of the color-difference signals at 2F sc , hence 4:2:2.
The major achievement of Rec 601 is choosing a set of sampling frequencies common to both the 525/59.95 and the 625/50 scanning standards. The selected frequencies are common multiples of 3.375MHz, as well as the line (horizontal) scanning frequencies (F H ) of both standards. A family of sampling rates based on the reference frequency of 3.375MHz has evolved, resulting in the well-known 4:1:1, 4:2:2 and 4:4:4 sampling strategies. Table 1 shows how the sampling frequencies are derived from 3.375MHz. Table 2 shows the relationship between the 4:2:2 component digital format sampling frequencies and F H in both scanning formats.
The sampling frequency imposes Nyquist constraints on the maximum sampled analog video frequency, which has to be lower than half the sampling frequency to avoid the occurrence of aliasing. It has therefore a direct bearing on the frequency response and the number of horizontal picture elements (pixels) that the system can handle. Table 3 lists significant parameters of the Rec 601 4:2:2 format. Sampling E' Y at 13.5MHz results in 858 pixels in the 525/59.94 scanning standard and 885 in the 625/50 standard. The digital active line accommodates 720 Y active pixels in both standards. Under ideal conditions, given the Nyquist frequency of 6.75MHz, 720 pixels per active line is equivalent to a horizontal resolution of 3/4 × 720 = 540LPH. Rec 601 specifies an anti-aliasing and reconstruction filter cutoff of 5.75MHz, which reduces the E' Y analog horizontal resolution to 455LPH (449LPH in 625/50). The E' CB and E' CR signals are subsampled by a factor of two at 6.75MHz. This results in 429 pixels in the 525/59.94 scanning standard and 432 pixels in the 625/50 scanning standard. The digital active line accommodates 360 C B and 360 C R active pixels in both standards. Under ideal conditions, given the Nyquist frequency of 3.375MHz, 360 pixels per active line is equivalent to 3/4 × 360 = 270LPH. Rec 601 specifies an anti-aliasing and reconstruction filter cutoff of 2.75MHz, resulting in an E' CB and E' CR signal analog horizontal resolution on the order of 218LPH (215 in 625/50).
The abovementioned analog resolution figures assume ideal brickwall low-pass filters. Such filters don't exist in practice, so the actual resolution could be worse.
Under the influence of the computer industry, various bodies have started referring to the number of samples (or pixels) per active line as horizontal resolution and to the number of active lines per picture as vertical resolution. This is misleading. For example, compare 720 pixels (horizontal resolution in the computer world) with the actual horizontal resolution of 455LPH. The figures for vertical resolution are also misleading. Compare 483 (number of active lines) with 338LPH (using a 0.7 Kell factor). Table 3 summarizes the situation for the two SDTV digital studio standards.
Figure 1 shows the spatial representation of several sampling structures along a scanning line. The 4:4:4 sampling structure shows that for every Y pixel there is a time-coincident C R and C B pixel. The 4:2:2 sampling structure shows that for every group of four Y pixels there are two C B and C R pixels. The 4:1:1 sampling structure shows that for every group of four Y pixels there is one C B and one C R pixel. The 4:2:2 and 4:1:1 strategies subsample C B and C R horizontally while keeping the vertical resolution intact. An MPEG subsampling strategy, known as 4:2:0, subsamples C B and C R vertically as well as horizontally using a process of interpolation.
The 4:2:2 sampling concept has undergone an evolution and, presently, it refers to a digital system where for every four Y pixels, there are two C B and C R pixels. There is no mathematical relationship with the line scanning frequency.
ATSC considerations
The members of ATSC could not agree on a single picture format concept. As a consequence the ATSC standard supports a range of program materials originating in different picture formats. Two program format levels are represented, namely HDTV and SDTV. There are two 16:9 aspect ratio HDTV production formats: SMPTE 296M (720 active lines progressively scanned) and SMPTE 274M (1080 active lines interlace scanned). There are several SDTV formats with a choice of 16:9 or 4:3 aspect ratio, interlaced or progressively scanned, as well as a 4:3 VGA format. Accounting for all picture scanning formats and frame rates, there are 18 picture formats supported by the ATSC standard, based on the nominal frame rates of 60Hz, 50Hz and 24Hz. If we take into consideration the NTSC-friendly rates of 59.94Hz, 29.97Hz and 23.976Hz formats, we end up with 36 picture formats. Table 4 presents the significant parameters of three basic 16:9 aspect ratio production formats — SMPTE 293M, SMPTE 296M and SMPTE 274M. The listed horizontal and vertical resolution figures, in LPH, are shown for comparison only. As shown, the progressive scanned formats are superior in terms of flicker but require considerable analog baseband bandwidths. All formats are compressed prior to transmission using MPEG-2 methods. Since progressive scanned signals are easier to compress, it is predictable that in the long run interlacing will be abandoned.
Michael Robin, a fellow of the Society of Motion Picture and Television Engineers and a former engineer with the Canadian Broadcasting Corp.'s engineering headquarters, is an independent broadcast consultant located in Montreal, Canada. He is co-author of Digital Television Fundamentals, published by McGraw-Hill .
Send questions and comments to: michael_robin@primediabusiness.com
