TN28 -- Notes on Square Pixels for 1125-Line HDTV Charles Poynton [now Poynton@inforamp.net] Sun Microsystems, Inc. 89/03/09 11:40 SCOPE This document outlines possible choices for detailed parameters (in particular, sampling frequency) for the proposed 1125-line HDTV system which, in addition to a picture aspect ratio of 16:9, achieve a sample aspect ratio of exactly unity (i.e. "square pixels", in computer graphics terminology). The property of samples having equal horizontal and vertical spacing will be important in HDTV applications related to computer graphics, and could contribute greatly to the acceptance of the detailed HDTV parameters in systems for printing and publishing, medical applications, simulator applications, and other non-broadcast industries. This document is an update of Poynton Vector Corporation Technical Note 28 dated June 27, 1987, which was submitted to the SMPTE AHG-HDSS committee as document N15.4/6 REF 34 REV 7. General information about HDTV is available in an associated document Current State of High Definition Television. GENERAL The basic parameters of the 1125-line HDTV system are now established in the U.S. by ANSI/SMPTE 240M, and are contained in Annex II of the international document CCIR Report 801-2: Number of scanning lines: 1125 Number of active lines: 1035 Field rate: 60.00 Hz Interlace: 2:1 Aspect ratio: 16:9 Samples per "active line": 1920 for luminance, 960 for colour difference With agreement on these basic analog interface parameters, the digital representation of the 1125-line system is in the process of being standardized. "ACTIVE" SAMPLES Note that although a line rate of 33.75 kHz is implied by these parameters (1125x60 Hz), and the count of "active" samples per line is directly specified, neither the total sample count per line or the sampling frequency are specified or implied, and discussion of values for these parameters is currently taking place. The figure of 1920 "active" digital samples is suggested in an appendix of ANSI/SMPTE 240M, and has agreement of the Japanese manufacturers' group BTA. This number is derived from the number of "active" samples in CCIR Rec. 601-1 (720), times the nominal doubling of horizontal resolution (2), times the increase in aspect ratio (4/3). The word "active" is in quotes because there is not agreement on whether this number refers to the number of samples per picture width (i.e. between the 50% points of a white flatfield), or the number of digital samples per line which are permitted to be above blanking level. In a system with finite bandwidth (and risetime), some number of non-blanking digital words are required to represent the tails of the blanking transitions. If the tails of blanking are truncated, then the picture width will be narrowed as the signal passes through successive pieces of equipment. CCIR Rec. 601-1 specifies the number of "active" samples per line as 720, for both 525-line and 625-line television systems. The definition of "active" is not made explicit in this document, but it is stated that the number 720 was chosen as being sufficient to accommodate all non-blanking samples at the widest picture width tolerance of both 525- and 625-line systems. The reference number of samples per picture width (i.e. the number of samples between the 50% points of a reference white flatfield) is somewhat less than 720, nominally 702 in 625-line systems (for blanking of 12 us), and between 702 and 712 in 525-line systems (for blanking between 11.6 us and 10.8 us). CHOICE OF 1920 "ACTIVE" SAMPLES The BTA group has indicated the desire to maximize the count of samples per picture width. The choice of 1920 samples per picture width has a particular advantage when combined with a sample rate of 74.25 MHz: this sample count represents an increase in picture width over existing (5:3 aspect ratio) HDTV hardware and software which is almost exactly equal to the increase in aspect ratio from 5:3 to 16:9. With this choice of analog picture width, existing 5:3 hardware and recorded material can be used in a 16:9 system, providing a slight cropping of picture width (about 6%) and a slight alteration of aspect ratio (less than 1%) is allowed. The blanking width resulting from this choice is about 3.77 us. It is not clear how the BTA proposal would deal with transition samples. Either the number of "active" digital samples should be somewhat greater than 1920, or the number of samples per picture width should be somewhat less than this. CHOICE OF 1840 SAMPLES PER PICTURE WIDTH For a total picture aspect ratio of 16:9, the aspect ratio of each sample (for a choice of 1920 samples per picture width) is 120:115, only 4% off-square. That is, the horizontal spacing of samples is 4% less than the vertical spacing between samples. This ratio must decrease slightly in order to accommodate transition samples within 1920 "active" digital samples; this will make the samples even closer to being square. With a choice of 1840 samples per picture width, the sample aspect ratio can be made exactly unity. (The value 1840 is simply the number of lines per picture height of 1035, times the picture aspect ratio of 16/9.) This choice of active samples can be made for a number of choices of total samples per line (i.e. a number of choices of sampling rate). CHOICE OF 2200 TOTAL SAMPLES The current BTA proposed sampling rate is 74.25 MHz, which is 11/2 times 13.5 MHz. This corresponds to 2200 samples per total line. Square samples are easily achieved with this sample rate, resulting in 1840 samples per picture width and a blanking time of 4.85 us almost identical the blanking time of current 5:3 equipment. However, with square samples at this sample rate a 4% picture aspect ratio error would result for existing 5:3 material displayed on new equipment. (1840 samples per picture width at 16:9 picture aspect ratio corresponds to 1725 samples at 5:3 aspect ratio.) CHOICE OF 2070 TOTAL SAMPLES A total sample count of 2070 simultaneously achieves square samples and exact preservation of the aspect ratio of existing recorded material. This sample count corresponds to a sampling rate of 69.8625 MHz, exactly 5.175 times the CCIR 601 sampling rate of 13.5 MHz. This is a somewhat inconvenient multiple [23x3x3/(5x2x2x2)]. The corresponding analog blanking value is about 3.29 us, which is somewhat short for current CRT technology. CHOICE OF 1053 LINES PER PICTURE HEIGHT If the narrower of the 525-line and 625-line values for samples per picture width is taken (625-line, at 702), then using the 2:1 horizontal sample relationship outlined above, the number of samples per picture width appropriate for the 1125-line system is 1872. With this count of samples per picture width, a count of 1053 lines per picture width obtains samples of exactly unity aspect ratio. This option is unappealing in that it would require a change to the SMPTE 240M analog interface standard. CHOICE OF 2100 TOTAL SAMPLES A choice of 2100 total samples per line results in a sample rate of 70.875 MHz, 5.25 (21/4) times 13.5 MHz. This rate has the following properties: 1. downconversion to 525-or 625-lines can use a horizontal subsampling ratio of exactly 1/2; 2. blanking time (3.67 us) is almost identical to the BTA proposal; 3. existing recorded material is reproduced with less than 1.4% aspect ratio error, an error generally agreed to be imperceptible; 4. the sample rate is related to 13.5 MHz by the simple integer ratio 21/4; 5. the 240M sync waveform can be generated exactly, using a pulse width basis of 21 clocks (the sync pulses can be generated exactly using 23.625, 13.5, or 10.125 MHz clocks); and 6. samples have exactly unity aspect ratio. This choice seems to be the best compromise between exactly square samples, preservation of existing material, and blanking time. RECOMMENDATION I recommend: 1. that the term "samples per total line" (S/TL) be defined as the number of sample periods per line period, and that this parameter be assigned the value 2100; 2. that the term "luminance samples per active line" (S/AL) be defined to be the maximum number of samples which may be assigned a value different from blanking level, and that this parameter be assigned the value 1920; 3. that the term "luminance samples per picture width" (S/PW) be defined to refer to the number of digital samples between the 50% points of the picture width, as referenced to a white flatfield, and that this parameter be assigned the value 1840; 4. that a note indicate that these sampling parameters result in a reference sampling frequency of 70.875 MHz, and a sample aspect ratio of exactly unity.