134 IBEE TRANSACTIONS ON BROADCASTING, VOL. 37, N0.4, DECEMBER 1991

Source Coding for Digital HDTV Terrestrial Broadcasting

William Y. Zou Public Broadcasting Service

Alexandria, VA

1. Introduction

In June 1990, the first alldigital terrestrial broadcasting HDTV system was proposed and submitted to the Federal Communications Commission (FCC) for consideration as the United States standard. Since then, three more proponents have offered alldigital systems for evaluation. It is claimed that an alldigital system will provide true high definition quality and coverage area equivalent to NTSC without noise and interference. Less transmission power may be required and the signal is easy to encrypt. This document will review the proposed source coding algorithms and discuss how they are employed in the proposed digital HDTV terrestrial broadcasting systems.

2. Source coding

A major objective of source coding is to represent an image with as few bits as possible while preserving the level of quality and intelligibility required for the given application. Certain modulation techniques appear to be able to support approximately 20 Mbits/s of data for terrestrial broadcasting in a 6 MHz bandwidth channel. This places an apparent upper bound on source coding goals. Typical source coding elements include motion prediction/compensation, transform coding, quantization, and entropy coding. Each of these elements attempts to exploit the redundancy present in the source image and the limitations of the display device and the human visual system.

2.1 Motion prediction/compensation

In television pictures, there is a high degree of correlation between adjacent picture elements in one frame, or between elements in different fields or frames, leading to a high degree of redundancy. An efficient way

to reduce the redundancy of information is to transmit the difference between the adjacent frames. The value of the difference is usually smaller than the original values of the frame, and can be coded by fewer bits. The difference can be obtained by first generating an estimated frame using the information of a previous frame or a future frame and subtracting it from the current frame. Motion prediction/compensation is used to predict the current frame so as to make the difference between the current frame and the estimated current frame as small as possible.

Motion prediction methods may be classified broadly into two groups, that is, block matching and spatio-temporal constraint methods. The block matching methods involve considering a small region in an image frame and searching for the displacement which produces the "best match" among possible regions in an adjacent frame. Algorithms of the spatio-temporal constraint methods are based on the spatio-temporal constraint equation [Lim, 19901.

Block matching methods are used in the DigiCipher, DSC-HDTV, and ADTV systems. The ATVA-P system employs the spatio-temporal constraint method.

2.2 Transform coding

In transform coding, an image is transformed to a domain significantly different from the image intensity domain, and the transform coefficients are then coded. Transform coding techniques attempt to reduce the correlation that exists among image pixel intensities more fully than do waveform coding techniques. When the correlation is reduced, redundant information does not have to be coded repeatedly. Transform coding also exploits the observation that for typical images a large amount of energy is concentrated in a small fraction of

0018-9316/91$01.00 0 IEEE

the transform coefficients. This is called the energy compaction property. Because of this property, It Is possible to code only a fraction of the transform coefficients without seriously affecting the image. Another desirable property of a transform is the reduction of correlation among transform coefficients, called the correlation reduction property. By proper choice of the basis function, the correlation among the coefficients can be reduced. An example of a transform is the discrete cosine transform (DCT) which is the most widely used transform in image coding. In DCT coding, an image is divided into many sub-images or blocks (typically 8 pixels by 8 pixels), and each block is coded separately. By coding one sub-image at a time, the coder can be made adaptive to local image characteristics. For example, the method of choosing quantization and bit allocation methods may differ between uniform background regions and edge regions. To take advantage of the energy compaction property, two approaches are widely used to determine which transform coefficients to code; they are zonal coding and threshold coding. In zonal coding, only the coefficients within a specified region are coded. In threshold coding, transform coefficients are compared with a threshold, and those above the threshold are coded. The choice of which transform coefficients are to be coded depends on local image characteristics and can be controlled based on buffer status.

Subband transforms are generally computed by convolving the input signal with a set of bandpass filters and decimating (down-sampling) the results. Each decimated subband signal encodes a particular portion of the frequency spectrum, corresponding to information occurring at a particular spatial scale. An adaptive bit allocation algorithm that distributes the bits among the subbands will make the subband coding scheme suitable for a wide variety of images [Woods, 19911.

DCT coding is used in the DigiCipher, DSC- HDTV, and ADTV systems. Subband coding is used in the ATVA-P system.

2.3 Quantization

The process of assigning a specific continuous scalar quantity to one of several discrete levels is called quantization. If each scalar is quantized independently,

135 the procedure is called scalar quantization. If two or more scalars are quantized jointly, the procedure is called vector quantization (VQ). A major advantage of VQ is that it can exploit statistical dependence among the scalars in the Mock. VQ is used in DSC-HDTV system to represent the possible combinations or patterns of quantizers which can be applied to a given block of coefficients.

Adaptive quantization is used in all of the proposed systems to optimize the source coding.

2.4 Entropy coding

Statistically, certain values appear more frequently than others. It is possible to assign a shorter codeword to those values occurring frequently and a longer one to those occurring less frequently. From information theory, the entropy is the theoretically minimum possible average bit rate required in coding a message. One optimal codeword design method which is simple to use, which is uniquely decodable, and which results in the lowest average bit rate is Huffman coding (variable-length coding). Also, when the transform/subband coefficients are properly scanned, the coefficients tend to be ordered from high amplitude to low amplitude, with the result that zero coefficients typically appear as a single long run which can be coded by a short codeword.

Entropy coding is used in all of the proposed digital HDTV systems to code coefficients of the image itself and image related ancillary information.

3. Buff er

Due to the entropy coding (variable-length coding) and the adaptive quantization, the data rate from the source coding varies locally. A buffer is used to regulate the variable input bit rate into a fixed output bit rate for transmission. The state of the buffer is calculated periodically and relayed back to adjust the quantization level.

Buffers are employed in all of the proposed digital HDTV systems.

136 REFERENCES BIOGRAPHY

LIM, J.S. [1990] Two-Dimenslonal Signal and Image Processing. Prentice-Hall.

WOODS, J.W. 119911 Subband Image Coding. Kluwer Academic Publishers.

GENERAL INSTRUMENT [August 19911 DigiCipher H D N System Description. Submitted by General Instrument Corporation on behalf of the American Television Alliance to the Federal Communications Commission, Advisory Committee on Advanced Television Service. FCC ACATS Document SS/WP1-0193.

ZENITH [September 1991) Digital Spectrum-Compatible HDN: Technical Details. Monograph published by Zenith Electronics Corporation and AT&T Bell Laboratories. FCC ACATS Document SS/WP1-0193.

ATRC [February 19911 System Description - Advanced Digital Television. Submitted by the Advanced Television Research Consortium (ATRC) consisting of Thomson Consumer Electronics, Inc., Philips Consumer Electronics Company, NBC, and the David Sarnoff Research Center to the United States Federal Communications Commission Advisory Committee on Advanced Television Service. FCC ACATS Document SS/WP1-0184.

MIT [February 19911 ATVA-Progressive System. Submitted by Massachusetts Institute of Technology on behalf of The American Television Alliance to the Federal Communications Commission, Advisory Committee on Advanced Television. FCC ACATS Document SS/WPl- 01 82.

William Y. Zou received a B.S. degree in electrlcal e n g i n e e r i n g f r o m Shanghai university, Shanghai, China in 1984, and an M.S. degree in electrical engineering from the State University of New York at Binghamton, New York in 1988.

From 1984 to 1986 Mr. Zou was employed by Shanghai Television Broadcasting Technology Research Institute as a design engineer. In 1988 he joined the Public Broadcasting Service as a Communication Systems Engineer. His research activity includes high definition television (HDTV) and digital compression. He is actively involved in testing and evaluating digital transmission systems for PBS and its member stations.

Mr. Zou is a member of the Institute of Electrical and Electronic Engineers and the Society of Motion Picture and Television Engineers. He is actively involved in the FCC Advisory Committee on Advanced Television Service and the Advanced Television Systems Committee.