paper

The challenges of digital H m V Complex compression coding and motion compensation algorithms are key techniques used by system designers

In a sudden series of mas- terstrokes, U S . engineers have propelled high-de- finition television (HDTV) into the digital era. Until last J u n e no all-digital HDTV systems were can- didates for a terrestrial

HDTV broadcasting standard in the United States. Now four are scheduled to be test- ed bv the Federal Communications Com-

cal developments that might arise. Sikes said: “. . .the Commission wishes

to review carefully but quickly in early 1992 any. . .new advanced television develop- ments. If in the course of that examination the Commission finds that a new advanced television technology is sufficiently concrete and developed to the point that it can be test- ed, the Commission will supplement the testing procedures and timetable to accom- modate it.” He gave as an example the de- velopment of a fully digital system.

But over and above whatever the FCC may have intended, digital HDTV offers certain advantages over an analog or hybrid system. First and foremost, a broadcast digital signal should in theory supply home HDTV receivers with pictures that are free from atmospheric noise, let alone inter- ference from motors. car irmition svstems.

I mission (FCC). I

which deteriorate slowly as distance from the transmitter increases, but are still watch- able, pictures from digital transmissions sim- ply vanish. For example, one home receiv- er might get an excellent picture and the receiver just down the block, a little farther away from the transmitter, might get noth- ing, Carnes said.

The biggest issue for digital transmission in the terrestrial broadcast environment is how large the coverage area will be, Cames told IEEE Spectrum. Since spectrum is limited to the present very high-frequency (VHF) and ultxahigh-frequency (UHF) chan- nels, currently unused channels must be har- nessed for simulcast signals. Since these sig- nals cannot significantly degrade existing NTSC service, the transmitter power must be limited-especially in heavily populated areas where spectrum is crowded. The

The move to digital began in June when General Instrument Corp. an- nounced its DigiCipher all-digital sys- tem. In November, the Advanced Tele- vision Research Consortium-National Broadcasting Co. (NBC), David Sarnoff Research Center, Philips Consumer Electronics Co., and Thomson Con- sumer Electronics Inc.-gave notice that a digital system would be devel- oped jointly by the Sarnoff Center and Philips Laboratories.

In December, Zenith Electronics Corp., AT&T Bell Laboratories, and AT&T Microelectronics announced their joint development of an all-digital sys- tem (Zenith’s previous entry had been a hy- brid analog-digital design). And in January, General Instrument was again in the news with the revelation that it had joined with the Massachusetts Institute of Technology (MIT) to form the American Television Al- liance, which would submit two systems for testing: the DigiCipher system and a system to be developed with MIT. WHY DIGITAL? Back in March 1990, the FCC announced that it would consider only simul- cast systems for a US. terrestrial broadcast- ing standard-systems where one broadcast channel carries a standard National Televi- sion System Committee (NTSC) signal for reception on existing television receivers, and a second channel carries the HDTV sig- nal for reception on HDTV sets. But FCC chairman Al Sikes made it clear that the door was being left open for any new technologi-

Ronald K. Jurgen Senior Editor

28

One home receiver might get an excellent

picture and the receiver next door might get nothing

citizens band radios, and the like. According to James E. Cames, head of the

David Sarnoff Research Center, digital HDTV has additional advantages. The very sophisticated compression techniques that can be implemented digitally should produce superior image resolution and sharpness; and a digital approach provides more flexi- bility for channel utilization or to meet new scanning standards in the future.

‘‘Fu1fi11ing the digital promise, however, is no easy task,” Carnes told those attend- ing the IEEE Media Briefing in New York City in November. Any simulcast signal must peacefully coexist with a rather hostile NTSC broadcast environment. The digital signals must not interfere with or degrade existing NTSC channels. Further, the digi- tal signals have to be resistant to interfer- ence from existing high-power NTSC signals.

Digital service also naturally tends to de- grade abruptly. Unlike NTSC pictures,

6018-9235/91/004-0028 $1.00 1991 IEEE

issue, then, is whether the resultant coverage area will be large enough to make good business sense.

Despite these formidable challenges, all the system proponents plan to have their digital systems ready for ’iesting in accordance with the schedule set by the FCC’s Advisory Committee on Ad- vanced Television Service, together with the Advanced Television Test Cen- ter, Alexandria, Va. That schedule calls for a September start to the testing of the DigiCipher system from San Diego, Calif.-based General Instrument; an October start for the system from Ze- nith, Glenview, Ill., and two AT&T units

in New Jersey (the Murray Hill laboratories plus AT&T Microelectronics in Berkeley

~~ -

Defining Terms Aspect ratio: the ratio of frame width to frame height as defined by the active picture Chrominance: the colorimetric difference between any color and a reference color of an equal lu minance and a specified chromaticity interlaced scanning: a process in which half the horizontal scan Iines-every other one-of a tele vision picture are transmitted first, and then the other half are transmitted and woven between the first set of lines Luminance: the luminance intensity of a surface in a given direction per unit of projected area Progressive Iseuuenliail scanning: a process in which the entire picture is transmitted at once Ouantization: a process in which the continuous range of values of an input signal is divided into nonoverlapping subranges, and to each subrange a dircrete value of the output is uniquely assigned

it k F SPFC r f u i A P K I I 1991

Wayne C. Luplow, executive director of advanced research and development at Zenith Elec- tronics Corp. (standing left), and Richard Citta, manager of electronic systems R&D, evalu- ate the Glenview, Ill., company’s low-power, inte$erence-free HDTV system that uses taboo channels (top row of monitors). Conventional television broadcasts using the taboo channels

~ cause interference (bottom row of monitors).

Heights); a January beginning for the sys- tem from the Sarnoff center and Philips Laboratories, Briarcliff, N.Y.; and the follow- ing March for the systems from General In- strument and MIT in Cambridge. Until those systems are tested in Alexandria, Va., and at least one is subsequently tested in the field, no one can be sure how successfully the challenges can be overcome.

Two other systems will also be tested: an enhanced compatible system from the Ad- vanced Television Research Consortium and an analog simulcast system from Japan Broadcasting Corp. (”K), Tokyo. The Sar- noff ACTV system would improve existing NTSC transmissions to provide a picture with an aspect ratio of 16:9, so it is not in competition with the four digital simulcast HDTV systems. The NHK approach, how- ever, is a competitor for a US. simulcast standard but, at this writing, there had been no announcement that it would be convert- ed to an all-digital system. SYSTEM SIMILARITIES. All four proposed dig- ital HDTV systems have many design fea- tures in common. Besides transmitting dig-

Jurgen-The challenges of digital HDTV

itally in a simulcast mode, they depend on the use of so-called taboo VHF or UHF channels.

For the simulcast approach to be widely used, every existing television broadcast- er-and there are now over 1400-must have a second 6-megahertz channel in the existing television bands. No “new” spec- trum will be available, so simulcast signals must be able to use the currently unoccupied and unavailable taboo channels.

Theoretically, all channels could be oc- cupied if tuner performance was adequate, but in the interests of affordability, the FCC relaxed the requirements on tuner design by leaving channels vacant.

The taboo channels are those not present- ly used for NTSC broadcasts because their nearness to other assigned channels in a specific locality would cause interference. With UHF channels, for example, the mini- mum separation allowed by the FCC be- tween co-channel transmitters (those with the same channel allocation) varies between 250 and 355 kilometers (155-220 miles), de- pending on geographical location. For adja-

cent channels, the minimum specified dis- tance is 90 km (55 miles). Minimum distances are also required between chan- nel transmitters to avoid interference from sound images, picture images, local oscilla- tors, intermediate frequency beat, and in- termodulation. Those distances may be as little as 32 km (20 miles).

The four digital-system proponents are ad- vocating use of the taboo channels because it can take far less power to transmit canier- suppressed digital signals than to transmit analog signals. The required average trans- mitted power for the digital portion of an HDTV signal can be less than 10 percent of that required by an NTSC transmitter with the same service area. Reducing the power of the digital signal, however, makes it more susceptible to interference from the much stronger NTSC signals on the same or other channels.

All proposed digital systems-as well as analog systems-also use some form of video bandwidth compression after basic analog-to-digital conversion [see “Analog in, digital out,” p. 71 1. These include interlace scanning, removal of source redundancy, utilizing human perception limitations, three- dimensional processing (the third dimension is time), and entropy coding.

According to Kerns Powers, a television consultant in Princeton, N.J., although inter- lace scanning gives a 2 : l reduction factor in the total compression, a growing body of en- gineers feel that interlace is a poor method for 2:l compression as compared, for e m - ple, with diagonal prefiltering and diagonal subsampling.

Video bandwidth compression invariably involves compromises. For example, one compression technique is to transmit some of the detail in a picture over a longer peri- od of time. If motion is taking place in the scene, however, this approach causes smearing and ragged edges of that portion of the picture in motion. So some form of motion compensation must be used to off- set that effect.

All systems also use some form of error correction coding so that the digital signal input to the source decoder portion of HDTV receivers wdl be accurate. With video bandwidth compression, a single error-a 0 for a 1 or vice versa-can cause havoc. A BIT MUCH. There are two reasons why dig- ital HDTV signals occupy-before video compression-much wider bandwidths than do conventional NTSC broadcast signals.

The first reason, which also applies to ana- log HDTV signals, is that an HDTV image has about twice as much luminance defini- tion both horizontally and vertically (four times as many luminance picture elements or pixels) as do conventional television sys- tems, and further additional pixels are need- ed for the wider screen (16:9 aspect ratio in- stead of 4:3). This increase in luminance detail requires about five times the video bandwidth of conventional television sys- tems. Since extra bandwidth is needed for

2

the separate transmission of the color values, or chrominance, the total bandwidth required is six to eight times that of conven- tional systems.

As the baseband of NTSC video for VHF and UHF channels is 4.2 MHz wide, six times that is 25.2 MHz and eight times is 33.6 MHz. So to handle the increased lu- minance detail and chrominance information in HDTV simals. additional channel mace

Prediction

Quantization

ing a pixel until the trend of the signal can be more accurately observed and then coded accordingly.

In transform coding, blocks of intensities of pixels are linearly transformed into blocks of frequency data called coefficients. Select- ed coefficients are then quantized for trans- mission. Adaptive transform coding is done in one of two ways: by changing the trans- formation to match Dicture statistics. or bv

Karhunen- Transformation Subsampling: Huffman Vector Loeve spatial and temporal quantization Hadamard Coefficient Adaptive Shannon- Contour

selection Fano

, ,

of 21.0-29.LMHz is needed-in other’ I

I Delayed (tree) I

dimensional images. Alternatively, they may estimate velocities on a point-by-point or region-by-region basis without considering the overall motion. ERROR CORRECTION. To minimize the effects of inaccurate digital transmissions, methods have been developed to conceal or correct the errors. These techniques often entail adding from 1 to 4 bits to an 8-bit word. The word Dlus its extra bits is called a block.

I Bit plane

words, the space occupied by four to five additional 6-MHz channels.

The second reason is that when the analog output from an HDTV studio camera-one luminance and two chro- minance signals-is converted into a digital output, the analog luminance and chrominance signals are sampled at high rates, about 120 megasamples per second. Then the sampled values are quantized or converted into digital sam- ple words of 8 bits, for a total of 8 x 120 or nearly 1 gigabit per second be- fore compression.

The HDTV signal’s resolution deter- mines the bandwidth required, which in turn determines the minimum sampling frequency-twice the bandwidth, or Nyquist rate. CODIN6 OPTIONS. Because bandwidths oc- cupied by digital signals are much too wide to be practical, given that the FCC has speci- fied no new spectrum space for such HDTV signals, a variety of techniques has been de- vised to compress them.

The basic compression codes [see table] are outlined in the book Digital Pictures by Arun N. Netravali and Barry G. Haskell [see To probe further]. A practical coding system might be a combination of some of these basic codes.

One code-pulse code modulation-is described in “Analog in, digital out,” p. 7 l . Predictive coding-differential pulse code modulation (DPCM)-relies on predicting the value of the pixel to be encoded. The encoded values of the previously transmit- ted pixels are used for the prediction, and only the prediction error or differential sig- nal is quantized for transmission.

Predictive coding becomes adaptive if the prediction is based on local picture statistics, if the quantizer is ‘‘coarser” in visual terms, or if the prediction error is not transmitted whenever it is below a certain threshold. Predictive coding involves a delay in encod-

changing the criteria for selecting and quan- tizing the coefficients to match the subjec- tive quality requirements.

Interpolative and extrapolative coding techniques send a subset of the pixels, and the pixels in this subset are then extrapo- lated or interpolated in the receiver to ob- tain the untransmitted pixels. Adaptive cod- ing involves varying the criteria for selection of the samples to be sent and also varying the strategy for interpolating or extrapolat- ing the remaining samples.

Statistical coding techniques are used to assign bits to the quantizer outputs of DPCM or transform coded signals to minimize the average bit rate. MOTION COMPENSATION. Since bandwidth compression smears the portions of a pic- ture in motion and makes their edges rag- ged, as noted earlier, some form of motion- compensated prediction must be used.

Motion compensation can be performed on either luminance or chrominance com- ponents or both. It is simpler to estimate motion from just luminance and then use the same estimate for chrominance. But as with bandwidth compression, a variety of tech- niques can be used.

Motion estimation algorithms may focus primarily on extracting three-dimensional motion parameters from a sequence of two-

Classification of picture-coding techniques

When an error is detected in the word, the decoder in the receiver deletes the word and forms another in its place by interpolation from the preceding andlor following words. A word representing a picture element can be repeated to fd the gap formed by the deleted word. Alternatively, the preceding and follow- ing words can be averaged.

In fixed-bandwidth channels like those for terrestrial broadcasting, bits used for e m r correction are taken from bits available for data transmission.

More elaborate error correction is needed if several successive bits in a word are lost or changed. One approach

uses a forward error correction (FEC) code such as the Bose, Chaudhuri, or Hocquen- ghem (BCH) code. One FEC code can cor- rect as many a s four bit errors in a E - b i t block through use of word storage from which predictions are made about words not yet transmitted. QUADRATURE MODULATION. A method that fits two distinct signals into the same fre- quency space is quadrature (90 degrees out of phase) modulation. This technique has been used, for ewmple, in the NTSC system.

The NTSC color subcamer is chosen to be an odd multiple of one-half the line fre- quency so that the subcarrier spectrum in- terleaves with the baseband luminance spec- trum. The subcarrier is then quadrature- modulated by two independent color differ- ence signals, which together with the decod- ed luminance signal permit deriving the three color components: red, green, and blue. The spectral interleaving minimizes t h e luminance-chrominance crosstalk, w h e r e a s t h e q u a d r a t u r e modulation separates the two chrominance components.

Some techniques described are in all four proposed digital HDTV systems. DI6ICIPHER SYSTEM. General Instrument’s DigiCipher system uses 1050-line interlaced scanning at a 59.94 field rate and a 16:9 as-

(Continued on p . 7l)

Conditional Discrete I Quantization I I I reolenishment I cosine I Arithmetic I I R u n length I -

30 IEEE SPECTRUM APRIL 1991

Analog in, digital out

- l l i

A basic method for converting the analog signal outputs from a high-definition television camera into digital signals is pulse code modulation (PCM). An analog-to-digital converter [see illustra tion] is preceded by a l w p a s s filter that confines the signal to a limited spectrum to prevent spuri- ous components from appearing in the recorivert- ed analog output.

The converter samples the analog input, meas. uring its amplitude at regular intervals of time. The sampled amplitudes are then matched in the quan- tizer against 256 levels of amplitude so that each of these lmls can be represented by an 8-bit word. (Eight-bit words are most commonly used in tele- vision, and since each bit can be either a 1 or a 0, there are 28 or 256 different ways that the 1s and Os can be arranged.) The error in amplitude is about 0.4 percent after quantization, which is equivalent to 48 decibels below the maximum sig- nal level.

The process of matching each sampled ampli- tude with its unique digital word is the job of the encoder. It scans the list of words and picks out the one that matches the sampled amplitude at any time. The encoder then outputs the series of digital words in the same sequence as the one in which the analog signal was sampled. The list of digital words corresponding to the sampled am- plitudes is known as a code.

I 40 00101000

I Sampler 1

Pulse Quantizer code

level modulation

255 1 1 1 1 1 1 1 1

232 11101000

(Continued from p. 30) pect ratio. To fit the digital HDTV signal into a 6-MHz channel, it uses an algorithm based on transform coding. The particular type of transform used is called a discrete cosine transform (DCT). It transforms an 8-by-8 block of pixel intensities into an 8-by-8 block of frequency transform coefficients. The transform is applied in turn to new blocks until the entire image has been transformed. At the decoder in the receiver, the inverse transformation is applied to recover the original image.

The property of the DCT that makes it possible to compress the image is this: for typical images, just a few transform coeffi- cients suffice to compact much of the signal energy. For example, one transform coeffi- cient in an 8-by-8 block represents the dc energy of the entire block.

With a process called normalization, the number of transform coefficients that must be used is reduced. First, an algorithm is used to assign a variable number of bits to those coefficients, and then Huffman statisti- cal coding is used to assign relatively short code words to events with the highest prob- ability of occurrence.

’ In yet another step in the DigiCipher sys- ’ tem, the signal is compressed yet again by

first predicting how the next frame will ap- pear and then sending the difference be- tween the prediction and the actual image. The previous frame is often a reasonable predictor. This DPCM procedure works best

i Jurgen-The challenges of digital HDTV

with pictures containing only little motion. Motion compensation is applied by deter-

mining what has moved where since the previous frame. Once this information is known to the decoder in the receiver, a por- tion of the previous frame can be shifted or displaced in order to obtain a more accurate prediction of the next frame that has yet to be transmitted. The decoder reproduces the same prediction as the encoder at the trans- mitter and then determines the difference between the prediction and the actual image.

A chrominance preprocessor reduces the resolution of chrominance information rela- tive to luminance resolution. ‘First it separates the signal into luminance and chro- minance components. The luminance signal maintains full resolution, but a prefilter is used on the chrominance components to average pixels in groups of four horizontal- ly and groups of two vertically. The chro- minance components are then multiplexed with the luminance component. At the decoder in the receiver, the components are again separated and reproduced.

The DigiCipher system uses a sampling frequency of 51.80 MHz for the chrominance and luminance signals. A forward error-cor- recting encoder adds error correction bits to the 16 quadrature-amplitude-modulated (16-QAM) data stream (which includes four audio signals as well as the video signals). ZENITH-AT&T SYSTEM. The Zenith-AT&T Digital-Compatible HDTV System (DSC- HDTV) is an extension of Zenith’s previous

analog-digital system. DSC-HDTV is said to reject NTSC cochannel interference into the DSC-HDTV channel, providing an HDTV service area equal to that of an NTSC broad- cast station while radiating at least E decibels less power. The previous system’s simulcast feature in a 6-MHz band, its NTSC-like transmission signal timing, and the low power are all retained.

DSC-HDTV uses progressively scanned video signals of 787.5 lines per frame and 59.94 frames per second. That corresponds to 1575 horizontal picture lines 30 times per second to create a completely new picture with a 16:9 aspect ratio 60 times a second. Square pixels allow easy graphical interfac- ing with computer workstations a s well as production of special effects

A video compression algorithm developed by AT&T Bell Laboratories enables 34 MHz of picture information to be squeezed into a single 6-MHz channel. The algorithm is em- bodied in high-speed digital signal processors developed by AT&T Microelectronics.

Using only the luminance frames, motion from frame to frame is estimated by a hier- archical block-matching motion estimator. It produces motion vectors, which are com- pressed and sent to the output buffer for transmission. Using predictive encoding, the small differences between the new image data and the motion-compensated predicted image data are encoded through adaptive transform coding.

(Continued on p 73) 71

The displaced frame difference is encod- ed using a spatial transform, and the result- ing coefficients are quantized. Luminance and the two chrominance difference pixels are encoded separately. The encoded video is packed into a special format before trans- mission, which maximizes immunity to transmission errors by masking the loss of data in the receiver decoder. Thus, the du- ration and extent of picture degradation due to any one error or group of errors is limited.

With the video encoder generating about 17 megabits per second, all data bytes are protected by the Reed-Solomon codes ex- cept the sync interval bytes-because sync detection must take place before error correction-and spare data bytes, which are not currently assigned.

Output from the video encoder is fed to a data formatter and error control system where the video bit stream is combined with digital audio, ancillary data, and Reed-

- where preprocessing takes place, a

signal to aid c a m e r recovery at the re- cewer, and encoded synchronization signals are also added.

The DSC-HDTV system uses pulse amplitude modulation with vestigial sidebands. The number of possible am- plitudes is limited to four. Compared to continuous amditude modulation. this

pilot signal is added to the transmission

ence of transmission errors. ADTV uses quadmture amplitude modula-

tion with spectral-shaping techniques, togeth- e r with Reed-Solomon forward error correc- tion channel coding, to minimize interference from and to any co-channel NTSC signals. The correction codes are applied to the data bytes before the carrier-modulation stage. Depending on the priority, different codes are applied to the data. In addition, data in- terleaving is performed as part of the chan- nel-coding operations to ensure that bursts of channel bit errors can be treated as uncor- related random bit errors, which can often be corrected by the Reed-Solomon codes.

ADTV provides for up to four digital CD- quality-sound audio channels that nominal- ly make up two stereo pairs. Audio compres- sion closely follows Masking-Pattern- Adapted Universal Subband Integrated Cod- ing and Multiplexing (Musicam), the indus- try standard for broadcast digital audio.

is made available for transmission of closed captions or other digtal data. Sync bits are inserted into the transmt- ted bit stream to mark the frame boundaries.

Reed-Solomon codmg is used to cor- rect transmission errors caused by noise and interference. The system threshold is 19-dB carrier-to-noise rat~o. At that threshold, there is one un-

Important decisions are being ma& with- Out adequate thought

lyzed by a transform-subband analysis filter that divides the residual into 8-by-8 bands. To exploit the variation in sensitivity of the human visual system, t h e transform- subband coefficients are weighted accord- ing to the frequency band and luminance- chrominance components. The weighted coefficients are selected on the basis of their energy; those with the highest energy are selected until the required number of bits to encode reaches the limit available (0.24956 Mb). The video bit rate required is this num- ber times 60 frames per second, or 14.99 Mbis.

The location and amplitude of each cho- sen transform-subband coefficient are en- coded jointly using a Huffman encoding method .

The digital video information is mul- tiplexed together with four digital audio channels, an auxiliary 0.126-Mbls data stream, and 0.126-Mbis access control to

form the composite 19.143-Mbk digi- tal stream. The auxiliarv data stream

Solomon error control bytes. After the resulting data stream eoes to a mefilter

provides immunity to thermal noise and other impairments. PHILIPS-SARNOFF SYSTEM. Advanced Digital Television (ADTV) is the name of the entry from the Advanced Television Research Consortium. The system uses 1050-line in- terlaced scanning at a 59.94 field rate with a 16:9 aspect ratio.

The video compression technique used, called MPEG", upgrades the compression approach to HDTV performance level of the Moving Pictures Expert Group (MPEG)-a committee within the International Stan- dards Organization-to provide high-quality HDTV pictures at 20 Mbis. MPEG is a col- lection of compression methods, including motion estimation, motion-compensated predictive coding, adaptive DCT quantiza- tion, and variable-length coding-decoding. The comDression techniaue also incor-

1 porates a video data prioritization layer that allows the most important video data to be transmitted with the greatest reliability.

1 Another feature, Prioritized Data Trans- port layer, is a fast-packet cell transport for- mat in which information bits are c a m e d in cells consisting of fixed-size data, header, and trader. T h s format permits relative ease in switching and routing-even at high speeds. A well-known example of such a format is the asynchronous transfer mode protocol in the broadband integrated-ser- vices digital network. Cell relay provides rugged logical synchronization that is es- sential for reliable delivery of variable- length coded compressed video in the pres-

ADTV is-designed to provide flexibie sup: port of a wide range of services and future media formats. AMERICAN TELEVISION ALLIANCE SYSTEM. The ATVA-Progressive System from MIT on behalf of the American Television Alliance provides a video signal with 720 by E80 p k - els, a 16:9 aspect ratio, progressively scanned at 59.94 frames per second.

The encoding part of the system consists of conversion of red, green, and blue com- ponents into Y (luminance) and U, V (chro- minance) components. It also uses source adaptive encoding, motion estimation and compensation, transform-subband analysis of motion-compensated residuals, adaptive selection of high-energy transform-subband coefficients, quantization of the selected coefficients, entropy coding of the quantized coefficients, and data multiplexing-modu- lation. The audio signals are also digitally encoded.

For motion compensation, a prediction is made for the current frame to be encoded from the previously encoded frame and the motion vectors. The difference between the current frame to be encoded and the predic- tion (the motion-compensated residual) is then computed for Y, U, and V. When the motion-compensated residual has sufficient- ly large energy relative to the image frame, motion compensation is disabled, and the image frame itself is encoded. This is the case with scene changes, for example.

Each of the Y, U, and V components is ana-

For digital transmission, a single carrier with double-sideband, suppressed-carrier quadrature modulation is used. The quad- rature signal is generated at a sampling fre- quency of 4.86 MHz. Spectral shaping filters in the transmitter limit the signal to a 6-MHz double-sided bandwidth. A digital filter is used for spectral shaping. TO PROBE FURTHER. The technical discussion in this article is based in part on two books. The first, HDTV-Advanced Television for the 1990s, by K . Blair Benson and Donald G. Fink (Intertext Publications, McGraw- Hill, New York, 1991), provides excellent background on advanced television systems of all kinds, with limited material on digital HDTV. The second, Digital Pictures- Representation and Compression, by Arun N. Netravali and Barry G. Haskell (Plenum Press, New York, 1988), gwes detailed tech- nical information on the various basic com- pression techniques.

A third book, Signal Processingof HDTI! ZZ, edited by L. Chiariglione (Elsevier, Am- sterdam, the Netherlands, 1990), contains papers from the Third International Work- shop on HDTV held in Turin, Italy, Aug. 30- Sept. 1, 1989. Many of the papers address motion estimation, preprocessing, and HDTV transmission on digital channels.

A new report on video compression tech- niques and their commercial applications is available from IGI Consulting Inc., 214 Har- vard Ave., Suite 200, Boston, Mass. 02134; 617-738-8088. +

Jurgen-The challenges of digital HDTV I