paper

THE ~ GRAND ~

ALLIANCE

36

The US. HDTV standard

THE GRAND HE FUNDAMENTALS OF TELEVISION have been reinvented for the digital age. Home delivery of a digitally clean, pan- T oramic, high-definition image, accompa-

nied by digital surround sound audio, could be just around the comer. And it would be broadcast ter- restrially, as well as by cable and satellite.

Such a vision moved the Federal Communica- tions Commission (FCC) in 1987 t o spearhead a competitive testing and standardization process in the U.S. marketplace. Years of developing rival high-definition television (HDTV) system proto- types followed, but in 1993, the key U.S. conten- ders came together in a Grand Alliance aimed at developing a U.S. standard for terrestrially broad- cast HDTV. This spring, prototype hardware is being tested at the Advanced Television Test Cen- ter in Alexandria, Va. This summer, the proposed standard is t o be submitted for final certification t o the FCC.

Based as it is on the MPEG-2 video compression and transport protocol, multiple transmission for- mats, Dolby AC-3 digital audio, and vestigial side- band digital modulation, this HDTV system will supply the home with superb images and spectac- ular sound while also, as part of the National In- formation Infrastructure (NII), relaying digital data. In society at large, the digital technology involved is expected t o create new business activities and stimulate key industries, such as semiconductors, computers, communications, and displays, giving the United States a chance at once more dominat- ing consumer electronics. T h e impact of HDTV technology on the consumer and the national economy should be felt for decades.

An HDTV system employing multiple trans- mission formats, digital compression, packetiza- tion, and modulation techniques is a marked ad- vance over current world television standards such as NTSC, PAL, or Secam (in full, National Television System Committee, phase altemation line, and Sequentiel Couleur Avec Memoire).

001 8 - 9 2 3 5 / 9 5 / $ 4 0 0 0 1995 IEEE

This standard for digital h ig h-def i n i t ion television wi I I be submitted to the Federal Communications Commission this summer for its approval

These analog systems deliver noisier and lower- resolution video and audio, with less flexibility and more processing artifacts. Their success has been immense, but the technology they use is now more than 40 years old. In those decades, major advances have occurred in such key enabling technologies as signal processing, telecommunica- tions, digital compression and transmission, and very large-scale integration. In the past, for exam- ple, the idea of giving an ordinary television re- ceiver enough memory t o store an entire frame of image data was unthinkable. Today it is unthink- able not t o utilize digital memory for video com- pression technology.

Under the proposed standard, homes will re- ceive an HDTV picture composed of more than two million picture elements (pixels), approxi- mately six times the spatial resolution of present NTSC images. Progressive scanning, one of the format attributes supported in the proposed sys- tem, will provide a temporal resolution of 60 full frames per second, twice the 29.97 frames per sec- o n d of standard NTSC images. T h e motion seen will be smooth enough for sports fans and com- puter graphics experts, and details will be fine enough t o be viewed up close even on very large television screens. T h e picture will be presented in a panoramic horizontal-to-vertical aspect ratio of 16:9, as in the movies-a big improvement on the 4:3 aspect ratio of today's home receiver.

For viewers, perhaps the biggest bonus will be the absence of all the NTSC artifacts. Since the

IEEE SPECTRUM APRIL 1995

HDTV images will be digitally clean, there will be no ghosts or beats from unwanted modulation products in the transmission channel, no snow on weak channels, as the viewer will receive only clear images within the broadcast coverage areai no chrominance artifacts, which show up as false and artificial colors on detailed patterns; and no color tint errors. T h e signal will be either excel- lent or absent.

As for what is heard, the digital surround sound system will be on a par with compact discs for quality.

Within each 6-MHz channel bandwidth, the proposed

initiative, since it can carry not only HDTV data, but such digital data as electronic newspapers or stock prices to homes. T h e ability t o deliver high-resolution video and audio in a cost-effective manner is also expected t o influ- ence video communications applications, including pic- ture phones, video conferencing, and multimedia.

HDTV'S HISTORY HDTV research and development began in Japan back

in 1968. NHK, the government-sponsored television broadcaster, joined forces with Japanese electronics man-

ufacturers and the govemment there in a massive project that absorbed well over US $ 1 billion. T h e resulting analog sys- tem, called multiple sub-Nyquist encoding (MUSE), utilizes digital signal compres- ALLIANCE sion, analog transmission, and satellite

A Researchers in Zenith's laboratories in Glenview, Ill., evaluate the high- definition picture performance of digital HDTV technology developed by Zenith and AT&T. The comoanies are two of the nine member organizations of the Grand Alliance that'is working on a U.S. HDTV standard.

HDTV system will deliver a digital channel of approxi- mately 20 Mb/s transmitted using a modern digital trans- mission technology called digital vestigial sideband mod- ulation. This data channel will be important t o the NI1

delivery direct to consumers In the United States, the allocation of

the television broadcast spectrum came under attack in the mid- 1980s by mobile radio interests that wanted t o have part of the upper UHF band. T h e FCC agreed to reassign to them some of the unused spec- trum, but broadcasters counter-attacked by insisting that the spectrum was needed for advanced television services.

To lend substance t o the claim, demons- trations of MUSE-encoded TV were con- ducted in Washington, D.C., in early 1987. A modified UHF broadcast transmitter sent the signal over the air from a suburb t o the FCC, the National Association of Broadcasters, and the U.S. Capitol. T h e event was politically successful, although technical challenges remained: the MUSE signal occupied two 6 - M H z NTSC chan- nels, was quite susceptible t o ghosting, and contained motion artifacts.

MUSE and other HDTV proposals had t o use satellite delivery because of their large bandwidth requirements and were not directly applicable to broadcast television. In recognition of this fact, 58 U.S. broad- cast organizations in August, 1987, peti- tioned the FCC t o inquire into the setting of terrestrial broadcast standards for advanced television. So in September, the agency created the Advisory Committee on Advanced Television Service. Richard E. Wiley, a former FCC chairman, has chaired this committee for seven years, under four FCC chairmen.

Many people thought that the new broadcast standard would be either en- hanced-definition television (EDTV) or, i f HDTV, then some derivative of MUSE.

T h e y were wrong. By the end of 1988, the FCC had re- ceived 2 3 advanced television proposals.

In the same year, broadcasters and the Electronic In- dustries Association (EIA) established the Advanced

THE GRAND A L L I A N C E 37

Television Test Center (ATTC) in Alex- andria, Va., to test all proposed HDTV systems. CableLabs, Boulder, Colo., an RsiD group supported by the cable tele- vision industry, agreed to provide the ATTC with cable-related testing. T h e FCC advisory committee thereupon con- tracted with the two organizations to test the dozens of proposals. Subjective evaluations of picture quality and the ef- fects on it of transmission impairments were to be conducted in Ottawa at the Advanced Television Evaluation Labor- atory (ATEL), established for the pur- pose by the Canadian governmental Communications Research Centre

Key FCC decisions In March 1990, the FCC announced a

number of key decisions. First, it decided in favor of simultaneous broadcast rather than the augmentation of NTSC signals. The latter process combines the NTSC signal with an augmentation signal in a second channel to create HDTV, and its signals are backwardly compatible with standard NTSC signals. In the simulcast approach, the HDTV signal is fully self- contained and is transmitted in a single television channel. It does not depend on the NTSC signal nor is it compatible with normal NTSC reception.

Even so, the FCC realized that a simul- cast approach would allow a transition period during which NTSC and HDTV could both be on the air; thereafter, the NTSC spectrum could be recaptured and reassigned as that kind of broadcasting was phased out. (Zenith Electronics Corp., Glenview, I l l . , and the Massachusetts Institute of Technology had championed the use of a simulcast approach.)

Second, the agency announced its pre- ference for HDTV over the less demand- ing EDTV, saying that it would rule first of all on HDTV Defying a view widely held in the United States, Europe, and Japan, the agency challenged the con- tenders to deliver HDTV in a single 6 - M H z broadcast channel.

Each developer of an advanced televi- sion system was required to submit an ap- plication for A R C testing and to make an initial payment by May 31, 1990. Then, just ahead of the payment deadline, Chicago's General Instrument Corp. (GI) trumped the field by filing the first pro- posal for all-digital HDTV The technol- ogy was widely held to be many years in the future, but GI had learned how to apply frequency compression to fit multi- ple NTSC channels into a single satellite transponder channel.

Though there were skeptics, the news galvanized the competition. By year-end, another contender announced that it was switching over to an all-digital HDTV

system: the Advanced Television Research Consortium (ATRC, which consisted of the National Broadcasting Co. and Philips Electronics North America Corp., both of New York City, the David Sarnoff Research Center, Princeton, N.J., and Thomson Consumer Electronics Inc., Indianapolis, Ind.). Zenith, too, teamed up with ATsiT to create an all-digital system. Finally, in Januaryl991, MIT teamed with GI and made the MIT system all-digital.

Altogether four all-digital HDTV sys- tems had now been proposed for testing at the two test organizations. There were two other proposals as well. The NHK Narrow MUSE system, with analog trans- mission, was a 6 - M H z bandwidth tech- nology slimmed down from the original 30-MHz MUSE. An NTSC-compatible EDTV proposal put forward by the Ad- vanced Television Research Consortium was later withdrawn.

All six systems were tested successfully at the ATTC and ATEL during 1991-92. In response to an FCC advisory commit- tee request, and under intense competi- tive pressure, all remaining contenders expressed a desire to upgrade their digital audio systems from two-channel stereo to five-channel surround sound and to pro- vide flexible packetized transmission like that of the pioneering ATRC system.

T h e advisory committee then set about picking the winner, convening a special panel for the job in February 1993 . T h e panel decided on all-digital technology, but could not select a clear winner among the four all-digital systems. It recommen- ded further testing and also asked for improvements.

Enter the Grand Alliance T h e advisory committee not only ap-

proved the panel's recommendation but went a step further. Aware that the com- petitors had been discussing a merger, the committee went on record as supporting a Grand Alliance of the all-digital conten- ders. Accordingly, after months of negoti- ations, the Digital HDTV Grand Alliance was announced on May 24, 1993, by its seven members: ATsiT, GI, MIT, Philips, Sarnoff, Thomson, and Zenith. Its goal was to utilize the best features of the four earlier systems in a " best of the best" sys- tem and HDTV standard.

Most of the rest of 1993 was devoted to system definition. Technical specialist groups of the Grand Alliance resolved details and cooperated with the FCC committee's expert groups, and in O c - tober of 1993, that committee's Technical Subgroup approved the definition as sub- mitted. Approval of one element, the modulation subsystem, was deferred until early 1994, following a "bakeoff" compe- tition. The modulation subsystem was

then taken to Charlotte, N.C., where ex- tensive field testing clearly established the superiority of digital transmission.

Finally, last year, the Grand Alliance system was constructed. AT&T and GI jointly built the video encoder. Philips constructed the video decoder. Sarnoff and Thomson cooperated in building the transport subsystem, and Zenith built the modulation subsystem. Dolby Laborat- ories Inc., of San Francisco, is supplying its AC-3 digital sound system. Following system integration at Samoff, the Grand Alliance prototype HDTV system is to go to the ATTC for laboratory testing, which will again be conducted by ATTC, Cable- Labs, and ATEL. The full prototype will then be sent to Charlotte for field testing.

T h e advisory committee will complete its work this year by giving the FCC its recommendation on H D T V standards. The agency will open a Notice of Propos- ed Rulemaking en route to the Report and Order that will officially set the standard, perhaps before the end of the year.

T h e FCC will be supported in its stan- dards-setting by the Advanced Television Systems Committee (ATSC), which is writing the HDTV Broadcast Standard specification, and by organizations such as the EIA and the Society of Motion Pic- ture and Television Engineers, which are writing industry voluntary companion standards.

IN HDTV DESIGN ESlCNlNG A SIMULCAST HDTV system for terrestrial broadcast- ing in the United States must

balance a multitude of issues: how to uti- lize the limited broadcast spectrum effi- ciently, while delivering high-quality high-definition pictures and audio and ensuring the integrity of existing NTSC broadcasts. A transitional phase will have to be arranged for broadcasters and pro- gram providers, and the consumer elec- tronics industry will have to develop new receivers. T h e HDTV system must be interoperable with other media. Tech- nological hurdles associated with system deployment will require solutions.

In a simulcast approach, a stand-alone HDTV signal is transmitted in a single television channel. Because of the limited availability of new broadcast spectrum, the FCC decided that HDTV broadcasts should occupy no more than the standard 6 M H z allotted for NTSC broadcasts. Since NTSC broadcasts must remain in- tact, channels currently off limits for broadcasting will have to be used for the new HDTV sewice.

T h e taboo channels exist because, for a given broadcast area, the normal televi- sion spectrum has channels of 6-MHz bandwidth spaced 6 MHz apart. In two

38 I E E E S P E C T R U M A P R I L 1995

geographically adjacent broadcast areas, the spectrum occupied in one is empty in its neighbor, and vice versa; as a consequence, television viewers in adjacent service areas are exempt from interference from neighboring broadcasts.

NTSC is made available t o the public through a robust analog synchronization and audio signal. Its availability is generally measur- ed over a coverage area by what are called grade contours. T h e grade B contour for NTSC is de- fined by an "acceptable" picture performance of better than a 28- dB signal-to-noise ratio in half the locations in the area of broadcast coverage, 90 percent of the time. In an HDTV broadcast system, the same considerations apply, It should be understood that unlike analog systems, digital broadcast systems work almost perfectly to the edge of the service area but fail rapidly, with a sharp increase in bit error-rate, when operating below what is termed the threshold carri- er-to-noise ratio.

An HDTV simulcast system for the terrestrial U.S. environment must deal with random noise and co-channel interference. It must also provide coverage comparable to that of NTSC. For a grade B cov- erage area, co-channel broadcasters (neighboring service areas relying on the same 6-MHz channel) that are the minimum 1 55 miles apart should deliver acceptable pictures up t o a radius of 55.5 miles from the broadcast tower. They should be capable of withstanding high levels of interference from. and Droduce

Transmission 1

modulator

, transmitter

Receiver r - - - - - - - - - -

I

low levels of interfeience in, co-channel NTSC broadcasts.

Another key requirement for a U.S. HDTV system is flexibility enough t o support a broad scope of services and serve many applications, both current and future. Novel uses of video, audio, and data in new products and services will require interoperability among di- verse types of delivery media, as well as among a wide variety of consumer elec- tronics, telecommunications, and comput- ing equipment in t h e home and office.

Interoperability is needed t o make the system easy for consumers t o use, t o expand the marketplace, and to give the electronics industry an incentive t o em- brace HDTV technology. Among com- pressed video formats, in addition, inter- operability would encourage synergy between HDTV and other computer and multimedia applications. Adopting the same video and audio compression stan-

A A television system must interface to assorted source input formats, including various film standards, magnetic and optical media, and synthetic imagety. In the Grand Alliance HDTV system, different video source formats are covered by using multiple-transmission (scanning) formats.

The video (audio) source is compressed by the video (audio) encoder, whose output is a string of bits representing the video (audio) source. Full digital surround sound with compact-disc quality typically oc- cupies about 5 percent of the available channel capacity.

The transport system's output is a string of bits representing video and audio sources and other data. A subsystem reconstitutes the string for transmission over a channel through modulation; it even com- pensates for the noise likely t o be introduced by the channel.

At the receiver, the signal is demodulated and turned back into a string of bits, which are demulti- plexed into streams of video, audio, and other data. From these streams decoders or processors recon- struct image, sound, and formats appropriate to the digital data.

dards in both the computer and consumer industries will facilitate the direct ex- change of compressed video and audio, eliminating the cost and artifacts associat- e d with repeated compression and decompression.

T h e HDTV system must in addition be flexible enough to operate over cable and broadband optical-fiber networks and with various source and display scanning formats, so that filmed material can be used optimally in tandem with sporting events and typical broadcast material.

THE FIVE HDTV SYSTEM FUNCTIONS HE GRAND ALLIANCE HDTV system functions are format selec- T tion, video coding, audio coding,

transport, and transmission [see diagram above]. N o t every television application has the same performance requirements- hence the appeal of multiple formats, which permit tradeoffs specific to each

family of program material. Digital repre- sentation and processing in an HDTV sys- tem allow the support of more than one scanning format, while facilitating inter- operability among formats and different video services and applications.

T h e formats chosen for the Grand Alliance HDTV system support a range of program material, using two line struc- tures and several frame rates. T h e pro- posed specification includes two main for- mat variations, with 720 and 1080 active lines per frame. Active lines represent the picture elements that are coded and trans- mitted; some other lines, in the vertical blanking interval, are not sent. In NTSC, for example, there are about 481 active lines per frame, with 44 lines in vertical intervals (22 in each field), for a total of 525 lines. T h e new 720-line format uses 1280 active samples per line and the 1080-line format uses 1920. Those choic- es yield square pixels for all formats, with

THE GRAND A L L I A N C E 39

Format 720 x 1280 F', at 60 frames per second

720 x 1280 P, at 24 or 30 frames per second

1080 x 1920 I, at 30 frames per second

1080 x 1920 P, a t 24 or 30 frames per second

a picture aspect ratio of 16 by 9. These formats, plus simple conversions among them, can cope with broadcasting, com- puter multimedia, computer graphics, industrial imaging, and the NIL

For the progressively scanned 720-line format, the frame rate can be 60, 30, o r 24 H z . (Lines are scanned in succession, whereas the interlaced format, like that used for NTSC, scans every other line every other frame.) T h e 1080-active-line format includes an interlaced version at 6 0 fields per second, as well as a progres- sive-scan variation for film modes at 24- and 30-Hz frame rates. (All of the formats are also supported with the NTSC-related frame rates, namely, 59.94 Hz, 23.976 Hz, and 29.97 H z , on the grounds that the 59.94-Hz frame rate may simplify working with NTSC material during the intervening simulcast period.)

As for compression of a 1080-line pro- gressive-scan format at a 60-Hz frame rate, the Grand Alliance determined that at the present time this approach could

Applications Sports, concerts, animation, graphics, up-con- verted NTSC, commercials

Complex film scenes, graphics, animation, possible reduced data rate

Scenes shot with a n interlaced-scan camera

Films with highest spatial resolution

not provide adequate HDTV quality if it used available channel capacity-hence the inclusion of a 1080-line interlaced format. Exclusively progressive-scan transmission can b e supported as it becomes technically and economically feasible.

T h e pixel clock rate is 74.25 M H z for the 60.0-Hz frame or field rate and ap- proximately 74. 1 8 M H z for the 59.94-Hz frame or field rate, which is iOOO/iOO1 times the 60.0-Hz rate.

O n e consequence of a multiple-format capability is the need for transconversion before and after coding. A transconverter at the encoder input converts pictures for coding and transmission in a format other than the source's format. Similarly, a transconverter at the decoder output lets the format of the final display-the one the viewer sees-be different from the re- ceived format. It is highly unlikely, for instance, that television receivers will change their display scanning parameters t o display different formats; instead, they

720 x 1280 P, at 24 or 30 frames per second

1080 x 1920 1, at 30 frames per second

1080 x 1920 F! at 24 or 30 frames per second

will convert whatever is received into the format designed into their displays.

T h e ability t o decouple the transmis- sion or coding format from the source for- mat makes for flexibility. For instance, if interlaced video is presented t o the sys- tem, then the user has the option of dein- terlacing, followed by transconversion to a progressive format, typically 1280 active pixels by 720 active lines at 60 frames per second. In addition, the decoder's output must be converted t o drive a given display format when the decoder receives multiple formats [see tables on this page].

T h e display format is independent of the transmission formats, since the display itself is not expected t o switch between formats. A decoder will provide frame buffering for all formats, and transconver- sion will derive the display format. If interlacing continues to be used for trans- mission, a deinterlacer must be part of every receiver with a progressive display.

To tackle feature films, the Grand Alliance HDTV proposal includes 24- and 30-Hz frame rates, with the corre- sponding adjustments for NTSC-related frequencies. Consequently the encoder will be designed to encode the reduced frame rate from image sequences that or- iginated at 24 and 30 H z . For both the 720- and 1080-line formats the encoder will identify and exploit the lesser frame rate if the film-mode material is presented t o it at 59.94 H z or 60.0 H z .

O n c e again, the initial set of scanning formats can be extended for higher per- formance or t o satisfy the needs of new applications when those become techni- cally and economically possible. In fact,

I

Use 3 2 or 2 2 pulldown Not required Use 3 2 or 2 2 pulldown. spatial conversion, inter- lace recommended

' Spatial conversion, 24/30 frame rate change not

Deinterlace, spatial Not recommended Not required Not recommended conversion

Spatial conversion, 3.2 or Spatial conversion, 24/30 Use 3:2 or 2:2 pulldown, Not required 2:2 pulldown frame rate change not interlace

recommended

destination format at 60 frames per second

Useful for 6 0 - H ~ progres- sively scanned display or

tape recording

Not required

Convert FROM source format 720 x 1280 P, at 60 frames per second

at 24 or 30 frames per second

Not recommended

at 30 frames per secnd

Useful for interlaced display or 60-field-per- second tape recording

at 24 or 30 frames per second

Spatial conversion and Not recommended* l- interlace

I E E E SPECTRUM APRIL 1995 4 0

a generally agreed-upon goal is eventual- ly t o include the ability t o transmit a coded 1080-line 60.0-Hz progressively scanned HDTV video sequence. Such an accomplishment would be the highest- performance picture format for the HDTV system.

T h e Grand Alliance believes that one way t o achieve this aim is by sending en- hancement data (over and above the original system's coded data) that will complement the basic HDTV stream. A new television receiver would decode both of the data sets and combine them t o create a full 1080-line progressively scanned image.

Video encoder

Any o f the Grand Alliance formats could be enhanced in this way, that is, the enhancement is capable of building on a compressed progressive-scan 1080- line film mode (30 H z ) , 720-line video mode (60 Hz), or 1080-line interlaced scan transmission.

T h e additional capacity required for t h e enhancement data stream could have any of several sources. T h e y in- clude increasing compression efficien- cy, obtaining additional channel capac- ity from an alternative source (such as an extra channel or transport from another transmission medium), or, after NTSC broadcasting comes t o an end,

increasing signal power a n d data rate, or re-using NTSC channel capacity for enhancement and other data.

Compression to fit For compression of video signals, the

Grand Alliance uses a motion-compen- sated discrete cosine transform (DCT) algorithm that was employed by all of the contenders in the first round of test- ing. DCT exploits spatial redundancy, and motion compensation exploits tem- poral redundancy. DCT was chosen for its good energy-compaction properties, and the many fast algorithms available afford low-cost implementation. In addi-

4 e%;, processor A

I Video input

Video decoder

Compression data

Panel encoding processor , Original frame I F l n m - .. ,

Discrete cosine transform (DCT)/

fine-motion estimator

panel mux coarse-

motion

J simulator 'i

coefficients

. . .t.

Video out

4 The analog-to-digital converter and clock MOV- ery element turn the analog video input into digital signals for red, green, and blue. R also provides a 27-MHz clock for use in the transport system and the generation of the presentation time stamps. Programmable gamma correction compensates for the nonlinear transfer function of cathode-ray tubes, while color space conversion t o Y, U, and V uses the SMPTE-24OM specifications from the Society of Motion Picture and Television Engineers. It also performs horizontal subsampling of the chrominance color-difference signals. The film detect/reoder function detects 24-frame film mate rial, with a look-ahead feature for smooth transition t o and from the 24-frame processing mode. it also rearranges the order of frames for the B-frame pm- cessing. The panel multiplexer performs vertical subsampling of the chrominance colordifference signals, then splits the data into six panels for paw el encoding processors.

4 In the panel encoding processor, the coarse motion estimator calculates motion vectors for each macroblock with one-pixel accuracy over a horizon- tal range of *127 pixels for P pictures and *63 pix- els for 6 pictures and a vertical range of *31 and +I 5 pixels for frame and field vectors, respectively. This estimator also decides field or frame vectors and field or frame discrete coding transform (DCT) cod- ing on the basis of mean absolute errors. Custom- designed chips provide the wide motion-estimation ranges with the full search algorithm. Also, the DCT/fine-motion estimator calculates the half-pixel motion vectors on the basis of reconstructed frames, a decision also based on mean absolute errors. Certain other decisions for each macroblock are based on mean square errors. The panel quanti- zation processor quantifies the DCT coefficients using a threshold that figures in the characteristics of human vision. The macroblock layer data is merged with slice, picture, group of pictures, and sequence layers. The compressed data is then buffered, formatted into packets, and supplied to the transport system. The coded data merge func- tion also determines the global quantization level.

4 In the video decoder, the packet analyzer sorts out packets and directs video packets to the rate buffer. A sub-analyzer extracts timing and other control information and supplies them t o the sys- tem and display controller. The compressed video data are supplied to eight decode engines for vari- able-length decoding, inverse quantization, inverse DCr, and motion compensation. The decoded video passes t o the video buffer, which arranges it into video frames. The last function provides the analog video output.

THE GRAND ALLIANCE 41

tion, the Grand Alliance system employs source-adaptive coding and other tech- niques for greater coding efficiency.

MPEG-2 syntax from the Moving Picture Experts Group will be used. For one thing, the MPEG-2 tool kit supports most of the compression algorithms used by the system proponents. For another, it will promote worldwide acceptance of the Grand Alliance system, which con- forms t o t h e MPEG-2 main profile implemented at high level.

The video encoder [see top dia- gram, previous page] has been de- signed to support both 1080-line inter- laced and 720-line progressive formats. It also supports bidirectional-frame (B- frame) prediction, wide motion-esti- mation ranges, field and frame motion vectors, and adaptive field and frame DCT coding, as well as forward anal- ysis with localized quantization-level control and automatic film detection for high picture quality.

B-hame prediction improves com- pression efficiency by predicting the current frame from past and expected frames. T h e wide motion-estimation r a n g e s - u p to + I 2 7 horizontal and *31 vertical in the prototype hard- ware but fully compliant with MPEG- 2 requirements-are needed for track- ing fast motion. T h e field and frame motion vectors and the adaptive fieldframe DCT coding greatly im- prove the compression efficiency for the 1080-line interlaced format. T h e forward analysis with localized quanti- zation-level control further decreases the visibility of the compression arti- facts by exploiting the characteristics of the human visual system.

Of course, the video decoder [at bottom in diagram on previous page]

streams (such as video, audio, and auxil- iary data). Each packet contains only one type of data: video, audio, or ancillary. As there is no fixed mix of packet types, the transport mechanism can dynamically allocate the available channel bandwidth for complete flexibility.

Each transport packet consists of a 4- byte packet header followed by 184 bytes of payload [see diagram on below]. The header includes means for synchronizing

view, can be supplied by conditional ac- cess, which permits video, audio, and data t o be scrambled independently. In- formation in a packet's header indicates whether its payload has been scrambled.

Sometimes, extra header information is required-for periodic synchronization of audio and video timing, for example, or random entry into the compressed bit- stream, o r local program insertion. Then a variable-length field, the adaptation head-

Video Audio 1 Audio 2

- - _ - _ _ - - - _ _ - -

Video Audio 1 Audio 2

- - _ - _ _ - - - _ _ - - - - - _ - - _ - _ _ _ - - - - _ 188-bvte Dacket - - _ -

184-byte payload (includes optional adaptation header)

I

- -Adaptation header (variable length)- - - - -

Packet synchronization Type of data in packet Packet losdmisordering protection Encryption control

Time synchronization Media synchronization Random-access flag Bit-stream splice point flag

has been designed t o &pport a l l b f the features of the encoder.

Packet delivery T h e HDTV system owes much of its

power and flexibility to the packetized transport technology employed for the broadcast delivery of the multimedia ser- vice. T h e MPEG-2 systems standard, on which the Grand Alliance HDTV system is based, encapsulates the compressed video, audio, and auxiliary data bit-streams in information packets of a fixed length. These packets are well-suited for terrestri- al channels, where transmission errors and data loss are likely. Besides packetization, the transport technology provides two other important functions: multiplexing and synchronization of the services that comprise a program.

Overall, the transport technology cre- ates a stream of fixed-length information packets from a variety of elementary bit-

A The Grand Alliance system uses a switched packet approach, similar to the one used in today's voice phone network, to deliver its multimedia components. Each packet consists of a Cbyte header and a 184-byte payload. The payload consists of one type of service (either video, say, or audio or text). The Cbyte header allows the receiver to determine what kind of data will be found in the payload. Over 8OOO types of services can be specified in the header. Additional information assists the receiver in iden- tifying thos packets that have been corrupted in the transmission channel.

The payload may contain an adaptation header, which provides useful information specific to the type of data. For example, it might indicate that a video packet represents the start of a new picture for a videocassette recorder-like fast-forward feature.

packets and identifying payload service. T h e sync byte, always the first byte of the packet header, contains a fixed, pre- assigned value. A 13-bit field called the packet identifier, also found in the pack- et header, affords a way of multiplexing various elementary bit-streams. Since the location of the packet identifier field is always fixed, packets corresponding t o a particular elementary bit-stream are sim- ple t o extract once packet synchroniza- tion has been established.

T h e fixed length of the packets also helps with error detection and correction. When poor transmission conditions over- whelm the error protection scheme, the packet header supplies as a means of sig- naling t o elementary stream decoders that the packet is t o be ignored. T h e decoder is then able t o mask the errors, preventing the appearance of blemishes on the pic- ture and sound

Subscription services, such as pay per

er, is placed in the payload of the trans- port packet. For synchronization, the pro- gram clock reference in the transport stream maintains a common time base be- tween the encoder and decoder. For lip sync between audio and video, the ele- mentary streams carry presentation time stamps that instruct the decoder when, relative to the master clock, t o present the decoded information.

To create a program, the transport sys- tem multiplexes a control data stream with packetized elementary bit-streams sharing a time-base [see diagram on page opposite]. These programs, along with an overall system control data stream, can also be multiplexed t o form a multiplexed system. Being flexible, the transport sys- tem defines programs as any combination of elementary bit-streams, and multiple programs can be multiplexed for transmis- sion through the channel.

Obviously, services that cannot be

4 2 I E E E SPECTRUM APRIL 1995

anticipated may be implemented in the future. It is therefore crucial that the transport architecture be extensible.

In point of fact, new elementary bit- streams could be handled at the transport layer without hardware modifications if new packet identifiers were assigned at the transmitter and were filtered in the bit-stream at the receiver. Backward com- patibility would be ensured, since existing decoders would automatically ignore new

T h e AC-3 coding process involves con- verting audio signals from the time into the frequency domain. The audio spectrum is continuously analyzed, and significant features are encoded with enough bits t o achieve an adequate signal-to-noise ratio.

Some of the spectral components are less audible, either because of psycho- acoustic masking (where part of the audio spectrum is overshadowed by a signal component that i s near in frequency and

digital modulation methods, the signal has additional features that enhance it when reception is difficult. T h e essential VSB technique was chosen on the basis of laboratory tests of competing systems, but desirable features from these rivals- especially any operating unilaterally at the receiver-can be added.

For broadcast, the system uses a signal designated 8-VSB, comprising a four-lev- el AM vestigial sideband signal plus trellis

Audio 1 (PID 2) Audio 2 (PID 3) Data 1 (PID i) Data 2 (PID J) Data N (PID m)

map (PID n)

"Nova" --+ Elementary stream "This Old U - . . r - u

Program stream map ( P I D O ) -

PID = Program identifier

L The transport layer multiplexes fixed-length packets to create a single stream that may be composed of several individual programs. One uniquely identified packet type (PID 0) contains a map of which programs are to be found on this transport stream. Each program in turn may comprise several distinct services or pack- etized elementary streams. For example, a program might carry multiple audio services for different lan- guages to accompany a common video service. The mix of services that comprise a given program are de- scribed in an elementary stream map.

packet identifiers-possibly an opening for a compatible form of three-dimen- sional HDTV

Dolby-based audio T h e system the Grand Alliance is pro-

posing does as much for audio quality as for picture quality. Discrete multichannel digital audio is an ideal match for digital high-definition wide-screen pictures. T h e sound delivered suits everything from the inexpensive monophonic television with a built-in speaker t o the home theater pre- senting cinema-like pictures and sound.

T h e audio is supplied by Dolby Labs' AC- 3 digital audio compression system, which has beguiled cinema-goers since 1991. T h e AC-3 perceptual coding sys- tem encodes discrete multichannel sound (left, center, right, left surround, and right surround, and low-frequency enhance- ment channels) into a bit-stream at a rate of 384 kb/s. T h e sound is of better quali- ty than the matrix surround-sound of such current media as broadcast television and VHS Hi Fi.

THE GRAND ALLIANCE

louder) or by the frequency-dependent threshold of human hearing. Insignificant (because inaudible) spectral components are not encoded. Somewhat significant spectral components are partially funded with bits and reproduced with an imper- ceptibly reduced signal-to-noise ratio.

A number of features improve the audio system's usefulness by preventing unwelcome fluctuations in audio level between program segments o r channels, allowing the listener t o select the full dynamic range of the original audio pro- gram before the broadcaster imposed range compression, and providing a mix of audio channels for listeners with only mono or stereo receivers. Multiple audio bit-streams may be delivered simultane- ously for multiple languages or for ser- vices for the visually or hearing impaired.

Digital VSB excels T h e transmission system for the pro-

posed HDTV standard uses a vestigial sideband (VSB) technique. Although the spectrum of VSB resembles that of other

coding that turns its four-level input into eight-level output signals. T h e basic data trans- mission spectrum is flat over most of the channel, with a roll-off region at either edge, like a quadrature amplitude- modulated (QAM) spectrum. But unlike QAM, where the suppressed carrier is in the cen- ter of the band, VSBs suppres- sed carrier frequency is at the roll-off of the lower band edge.

Constant-value data-all Os, for example, when the trans- mitter input is disconnected- would ordinarily produce a nonuniform spectrum, increas- ing the interference into exist- ing NTSC stations. For this reason, input data is modified by a known pseudo-random scrambling sequence, which flattens the spectrum on aver- age [see diagram, p. 441. Be- cause the MPEG-compliant data generated by the Grand Alliance video compression sys- tem is segmented into 188- byte packets including a sync

byte, data is transmitted in similar seg- ments. Each contains 187 data bytes, 20 Reed-Solomon parity bytes for forward error correction, and one segment sync byte. This last byte replaces the MPEG sync, which can be re-inserted as re- quired at the receiver.

The one-dimensionality of the VSB signal allows the demodulated baseband multi-level signal t o be digitized with a single analog-to-digital converter. Digital circuitry handles all later processing.

Over cable, where the signal-to-noise ratio is controlled, a 16-level VSB modu- lation suffices without trellis coding. T h e increase in the number of levels does not alter the signali spectrum, but does dou- ble the available data rate vis-a-vis 8-VSB.

T h e special features added t o the VSB signal assist the receiver in acquiring and locking onto that input even under the occasionally extreme conditions of terres- trial broadcasting. T h e idea is t o provide a known and stable reference for the vari- ous carrier and clock recovery functions. T h e receiver is then sure to acquire the

4 3

signal whenever the data itself is usable. T h e principal special feature is the in-

clusion of a small pilot carrier instead of t h e totally suppressed carrier usual in QAM. T h e pilot is placed so that it falls on the Nyquist slope of NTSC receivers, minimizing co-channel interference into existing service. I f there are n o other channel impairments, this pilot can be acquired down t o a signal-to-noise ratio of 0 dB and provides a margin for signal acquisition under severe conditions, such as ghost) or co- channel interference.

lnteroperability ranks high Getting a diverse collection of

television, imaging, and informa- tion systems to work together is difficult because of the prolifera- tion of standards in some indus- tries and the lack of them in oth- ers. Still, interoperability has been an important design goal for a U.S. HDTV system since 1990, when the competing digital sys- tems were being developed.

A high degree of interoperabili- ty in an HDTV system indicates that a balance must be struck among many different and often conflicting considerations related t o broadcasting, cable television, and consumer electronics, as well as computing, telecommunications, and the NII. These systems are not especially interoperable with one another, yet there are good reasons for making them interoperable with HDTV. Although effortless interop- erability with all the systems is im- possible, it can always be provided

HDTV transmission standard must be easily interoperable with a wide variety of existing pixel formats, including those used in motion picture film, currently available HDTV production equipment, the NTSC television standard, and such computers as PCs and workstations. This goal is extremely demanding, since these widely accepted standards differ in pixel format, scanning approach (interlaced

lines and active 1280 pixels by 720 lines. Each has a 16:9 aspect ratio, with the square pixels crucial t o computer interop- erability. It is should be noted that this choice represents a preference for inter- operability with computers rather than with the current NTSC television stan- dard. The two Grand Alliance scan for- mats are deliberately related by a factor of 3:2, so that a simple interpolation filter

with appropriate interfacinglconver- sion electronics. In that sense, interoper- ability is a matter of degree.

Three fundamental principles of the Grand Alliance HDTV system provide a basis for interoperability. First, the layered digital system architecture is compatible with the O p e n System Interconnection (09) model of data communications- the reference model for virtually all mod- e m digital systems. Second, the HDTV system fully exploits the flexibility of a digital system by using a headeddescrip- tor approach, rather than a single, rigidly specified data format. Finally, each layer of the system is designed t o be interoper- able with other systems at corresponding layers, so that many applications are able t o make use of various layers of the HDTV architecture.

T h e picture formats of the Grand Al- liance system provide for square pixels, where the vertical and horizontal spatial resolutions are equivalent, and for pro- gressive scan. At the picture layer, the

Compared to the NTSC signal, the vestigial sideband NSB) spectrum is relatively flat across the entire 6- MHz channel. The suppressed carrier of the VSB signal is a small pilot tone, located at the rolloff of the lower band edge, where it can be hidden from today's NTSC television receivers. The VSB signal can be protected from interference from the strong energy of NTSC carriers by comb filters at the receiver.

and progressive), and frame rate. Whereas all but the cheapest computer displays use square pixels and progressive scanning, for instance, there is no one industrywide standard or even a predominant format. Further, no simple technique of convert- ing among text and graphics representa- tions is commonly available.

NTSC television, on the other hand, uses rectangular pixel resolution and inter- laced scanning, but its intemational stan- dard (CCIR 601) provides a format for 525-line component video of active 483 lines by 720 pixels. Fortunately, the televi- sion industry has developed techniques and equipment for converting images from one pixel format and frame rate t o anoth- er, achieving interoperability among film and the various world television standards. Indeed, such conversions are common in television and film production.

As noted earlier, the two pixel formats provided by t h e Grand Alliance HDTV system are active 1920 pixels by 1080

suffices to convert one into the other. The scan formats have a like 3:2 relationship t o the common VGA computer format of active 640 pixels by 480 lines-VGA has a narrower 4:3 aspect ratio-providing simple interoperability with text and graphics in VGA format .

For the sake of flexibility and interoper- ability with television, film, and computers, the Grand Alliance formats are progressive scan with one exception: the highest com- bination-active 1920 pixels by 1080 lines at 60.0 Hz-is interlaced, but retains square pixels. Progressive scanning, be- cause of its many advantages and its impor- tance for interoperability with computers, is used in five of the six combinations of scan formats and frame rates. The remain- ing interlaced format is preferred only for some types of picture material, such as those used for much entertainment televi- sion, and in addition is interoperable with existing interlaced sources. Just as conver- sions can be performed among various scan

44 I E E E SPECTRUM APRIL 1995

formats, an interlaced scan can be convert- ed to progressive scan by a deinterlacing filter that " fills in" the missing lines.

Thus, the Grand Alliance HDTV sys- tem provides for multiple formats and frame rates, decodable by any Grand Alliance HDTV receiver, and convertible to its particular display format. T h e flexi- ble resolution/frame rate transmission approach, by accommodating various production standards, will enable diverse industries, program producers, applica- tion developers, and users to make their own tradeoffs among resolution, frame rate, compression, and interlaced artifacts. In effect, they are free t o choose the for- mat/frame rate combination that provides the best picture quality for their particular use. Receiver manufacturers also benefit from the freedom to produce a wide vari- ety of products and consumers from the freedom t o choose among them.

T h e compression approach of the Grand Alliance HDTV system turns the raw video and audio samples into a cod- ed bit-stream-essentially a set of data and computer instructions, which the re- ceiver executes to recreate the original picture and sound. T h e system accord- ingly gains attributes that contribute to its interoperability with the compression syntax commonly used by computers and multimedia systems.

T h e Grand Alliance video compression syntax is based on the MPEG-2 video data compression draft international stan- dard from the International Standards Organization and MPEG. This relation- ship will permit HDTV devices t o inter- operate with MPEG-2 and - 1 computer multimedia applications directly at the compressed bit-stream level. For example, consumer HDTV VCRs will be able t o produce an output bit-stream that can be input t o a multimedia computer, and it will also be possible t o interface HDTV receivers with CD ROMs containing f u l l - motion video. Although appropriate inter- faces must be specified, this common com- pression standard clearly smooths the way t o interoperability. Even t h e header/ de- scriptor that specifies the HDTV system's multiple format and frame rate capabili- ties is owed t o a mechanism that is part of the MPEG-2 standard.

T h e transport subsystem, since it de- fines the basic format of data packets, can contribute a lot to interoperability. This packetization serves many purposes: it wraps up the data into fixed-size units suitable for forward error correction en- coding, it multiplexes the various ele- ments of the program (video, audio, data, and so forth); it provides time synchro- nization for those elements; and (through the use of packet identifiers in the head- eddescriptor) it ensures flexibility and

extendibility with backward compatibility. Again in conformity with international

standards, the system uses a fully compati- ble subset of the MPEG-2 transport system. The 188-byte packet has a 13-bit packet identifier with a powerful headeddemiptor that can combine video, audio, and ancillary data streams into a single program stream. No predetermined mix of data is required, and furthermore, the mix may change dynamically from moment to moment. This approach allows great flexibility in the services that can be provided.

While it is desirable to achieve inter- operability with other packetized com- munication systems, the current installed base of data communications networks falls far short of this ideal. Using Ethernet, fiber-distributed data interface (FDDI), and Internet standards, they all have quite different packet sizes, formats, and proto- cols, and require interface electronics t o " bridge" networks. In fact, the emerging asynchronous transport mode (ATM) standard uses packet sizes, formats, and protocols that differ from those of any of the above standards.

T h e Grand Alliance HDTV system has emphasized interoperability with ATM by using a packet size that is easily encapsu- lated within that transport mode. While the delivery of MPEG-2 applications over ATM links is likely to be the most usual interoperability scenario, the opposite possibilityy-delivering ATM data over an MPEG-2 link-is also provided for in the Grand Alliance system, plus full function- ality of both the MPEG-2 and the ATM transport streams. This approach preserves the full functionality of both methods.

Transmission interoperability is served by a serial bit-stream and aided at the trans- mission layer by the nature of digital sys- tems. Naturally, different modulation tech- niques are used for physically different channels (terrestrial simulcast, cable, satel- lite, and fiber), but demodulation of these signals into the serial bit-stream is the basis for transcoding among the modulation techniques and ensuring interoperability among physically different delivery media.

HDTV and the NI1 By deploying digital HDTV and creat-

ing a new consumer market, the United States can accelerate the development (and consumers' acceptance) of improve- ments to the nation's information infras- tructure. HDTV can be delivered by exis- ting cable television systems, and put on the air quickly by broadcasters. It will cre- ate a mass market for low-cost, high-reso- lution displays in agile receivers that use sophisticated digital circuitry t o process and display multiple pixel formats and frame rates. Adding a more powerful mic- roprocessor-a simple one is already

there for control function purposes-to an HDTV set will yield an NII-ready information appliance at a small cost.

In fact, the best way of introducing ad- vanced NI1 applications may well be t o employ entertainment as a catalyst for getting the necessary capabilities into t h e home. Even so, this move will succeed only i f HDTV receivers are affordable. Furthermore, some consumers will want NI1 capabilities and some may not.

In all likelihood, any HDTV receiver- even one with an interlaced display-will serve as an adequate NI1 terminal for a cor- nucopia of consumer applications. Not all information infrastructure applications are directed at the general public, but still, the acceptance of NI1 and its growing use by consumers will probably generate demand for more advanced HDTV receivers with ever more capable processors and displays. In this way, HDTV could be a catalyst to the NIli further development and the cre- ation of new NI1 applications. Conversely, as the NI1 advances, consumers will de- mand ever more sophisticated HDTVs and other information appliances. +

About the authors The Grand Alliance supplied all but one of

the 12 authors of this article from its seven member organizations. Carlo Basile is a distinguished fellow and research depart- ment head of the video communications department at Philips Laboratories, Briar- cliff Manor, N.Y. Alan P. Cavallerano is a principal member of the research staff at the same facility. Michael S. Deiss is a senior member of the technical staff at Thomson Consumer Electronics, Indiana- polis, Ind. Robert Keeler is a distinguish- ed member of the technical staff at AT&T Bell Laboratories, Holmdel, N.J. Jae S. Lim is a professor of electrical engineering and director of the Advanced Television Pro- gram a t the Massachusetts Institute of Technology. Wayne C. Luplow is vice pres- ident for consumer electronics engineer- ing and HDTV, for Zenith Electronics Corp., Glenview, 111. Woo H. Paik is senior vice president of new business and advanced development for General Instrument's (GI's) Communications Division, San Diego, Calif. Eric Petajan is supervisor of the HDTV research group a t AT&T Bell Lab- oratories, Murray Hill, N.J. Robert Rast is vice president of HDTV business develop- ment a t GI's Communications Division. Glenn Reitmeier is director of high de- finition imaging and multimedia at the David Sarnoff Research Center, Princeton, N.J., where Terrence R. Smith is director of television research. Craig Todd, the 12th author, i s a senior member of the technical staff with Dolby Laboratories Inc., San Francisco, which does not belong t o the Grand Alliance.

THE GRAND A L L I A N C E 4 5