ÐÓ°ÉÔ­´´

Later this year, a modified CD player, connected to a TV set, will deliver
up to 74 minutes of moving video pictures from a 12-centimetre disc that
looks just like a conventional audio CD. Next year, Hughes Aircraft, the
aerospace manufacturer, plans to launch two satellites that will deliver
150 TV channels directly into North American homes. The following year,
the satellite broadcaster Astra aims to be doing the same in Europe. By
then, American cable networks will be offering a choice of 500 channels
– and we will all have video recorders able to tape them perfectly.

The key to this revolution is the switch from analogue to digital technology,
and the consequent ease with which information can be transferred and stored
reliably. And behind the success of the switch lies an unprecedented agreement
among all sections of the electronics industry on a common standard for
coding TV and video pictures.

The standard is known as MPEG, which derives from the name of the committee,
the Moving Picture Experts Group, that agrees the standard under the joint
direction of the International Standards Organization, ISO, and the International
Electrotechnical Commission, IEC.

MPEG, the committee, was created in May 1988. It grew out of another
ISO/IEC committee, the Joint Photographic Experts Group, which was set
up to establish a standard for converting still pictures into digital code.
The system that JPEG came up with, known as the Discrete Cosine Transform,
divides the image into square blocks and encodes only those blocks that
are dissimilar to adjacent ones. So successful has the DCT been that the
system is also used to encode moving video pictures. However, the 25 or
30 images a second needed to convey realistic motion generate many tens
of millions of bits of digital code a second. While this has not been a
problem for professional video producers with the money to spend on expensive
equipment, it has made JPEG compression unsuitable both for domestic CD
players, which cannot read data faster than 1.5 megabits per second, and
for broadcasting programmes in the narrow bandwidths allocated for analogue
TV services.

Imagine the industry’s surprise when, in early 1987, researchers at
the RCA Research Center in Princeton, New Jersey, used less than 1.5 megabits
a second to generate a video sequence. They simply extended the idea of
comparing adjacent pixels in a picture and ignoring those that were similar
to comparing consecutive images in a sequence and encoding only the differences.
Though RCA’s results were crude, and it took many minutes of computer time
to process each second’s worth of motion, the conceptual breakthrough galvanised
researchers in every major electronics company into looking for better ways
of encoding what became known as Full Motion Video. They developed many
variants of the new DCT technique to improve pictures in a moving sequence.
This raised the awful prospect of a crippling standards battle, with different
companies marketing their own proprietary systems as the best.

Birth of common sense

The computer industry, with an appalling track record on compatibility
issues, might well have gone down this path. But the consumer electronics
companies had bitter memories of their heavy losses from the standards battles
between three video disc systems, four surround sound processes and half
a dozen video cassette tape formats. They also saw the commercial potential
of putting over an hour of moving video on the single side of a 12-centimetre
CD, or transmitting several channels of digital video in a single TV channel.
In a remarkable display of industry-wide common sense, the MPEG committee
was born.

For the first two years the meetings dealt only with code rates of up
to 1.5 megabits a second because an agreed standard in this range would
have the maximum impact on the biggest market, consumer electronics, with
the minimum disruption. With this in mind, the committee decided to reduce
the amount of picture information in a video to 1.2 megabits a second,
and leave 0.3 megabits a second of capacity in the data stream for stereo
sound to accompany the images. Its standard, agreed last November, is due
to be published next year.

In 1991, MPEG-2 was formed to devise a standard for encoding video signals
at higher rates, between 2 and 10 megabits a second, that are necessary
for broadcasting and video recording. A third group, known as MPEG-3, was
planned to consider rates between 10 and 20 megabits a second, which are
necessary for transmitting high-definition TV, but its brief was soon subsumed
within that of MPEG-2. In the meantime, MPEG-4 was formed to devise a standard
for very low bit rates (only tens of kilobits a second) for use with future
videophones and mobile TV systems, such as video cellular phones. So, while
there is an MPEG-1, MPEG-2 and MPEG-4, there is no MPEG-3.

The decision to consider low bit rates followed the invention in the
late 1980s of a technique, known as fractal compression, that could squeeze
the information needed to represent an image to what were once considered
unrealistic levels (see Technology, 1 May). Whereas the DCT technique used
to encode still pictures to the JPEG standard reduces the digital information
needed to store an image to one-tenth of what it would be without any compression,
fractal compression is reckoned to cut it to one-hundredth – without causing
any noticeable loss in realism. Invented by Michael Barnsley, a British
scientist working in the US for the Defense Advanced Research Project Agency
(DARPA, and now known as the Advanced Research Project Agency), fractal
compression reduces a picture to its basic, or component, shapes, and encodes
these. Barnsley had watched artists build large pictures from small component
shapes, and reasoned that all images, including photographs, could be broken
down into a limited number of shapes. The resulting images compare favourably
with those obtained by using the DCT adapted to reduce detail to a similar
level. Whereas the extreme DCT compression needed for videophones, for instance,
creates a coarse mosaic effect, called blocking, fractal compression produces
an image that looks like a broad-brush oil painting. Visually this is less
offensive than blocking.

Moving target

Fractal compression already works well with still images, and is used
to store the pictures on Microsoft’s new multimedia CD-ROM encyclopedia,
Encarta. But it is not yet ready for use with moving images. Work on MPEG-4
begins officially in September, and the target date for a standard is 1997.

At all MPEG meetings, delegates bring video tapes of the results they
are getting from computer simulations of compression coding. To make comparison
easier, the group agreed on some test sequences, some containing fine detail
and others fast motion, all deliberately chosen to show up any deficiencies
of a compression system. Usually these show up as loss of detail or artefacts
of motion, such as blur.

The agreement on MPEG-1 gave the manufacturers of microchips the all
clear they had been waiting for. Both C-Cube in California, and Motorola
in Illinois, immediately finalised work on chips that can decode moving
pictures and sound from a CD. The Japanese consumer electronics company
JVC is already selling a karaoke system that uses MPEG CDs. Later this
year, Philips will sell an adaptor that slots into its interactive CD players
to let them play Full Motion Video CDs. British record company Nimbus proposes
a ‘black box’ adaptor that will convert a conventional audio CD player into
a video CD player. All these systems rely on MPEG-1 decoding to provide
Full Motion Video. In June, Philips, Sony, Matsushita and JVC, agreed an
addition to the standard for the CD system that seeks to pull all these
strands together, to create a common format called Video CD (see Technology,
10 July).

C-Cube is making encoding chips too, and Nimbus will use these to offer
the film industry a service. It will charge around $100 a minute of screen
time to encode feature films for release on Video CD. Once the material
is coded, discs can be pressed in bulk for less than $1 each. By comparison,
film companies must pay around $3 each to duplicate video cassettes.

There is another advantage of releasing film material on MPEG discs.
The standard covers both the North American NTSC and European PAL TV systems.
One disc will play back in any country, provided it is used on a player
and TV that conforms to the local standard. However, although MPEG-1 and
MPEG-2 both use the DCT, the fact that equipment designed to the first standard
operates at a lower bit rate than that designed to the second means there
is a difference in quality in the pictures produced.

Existing TV and video systems display 25 (or 30 in the US and Japan)
images or ‘frames’ a second. To avoid visible flicker on screen, each frame
is broken down into two halves or ‘fields’, each containing half the total
number of horizontal lines for one full picture; the number is 625 in Europe
and 525 in the US and Japan. The odd lines appear in one field, even lines
in the next, odd in the next, and so on. On screen the two fields ‘interlace’
to give the illusion of full pictures being displayed at the rate of 50
(or 60 in the US and Japan) a second, which is sufficient to create a realistic
effect.

In sequence

To keep the number of bits a second as low as possible, equipment based
on MPEG-1 takes each full picture frame, reduces the number of lines that
make up the visible picture to 280, and then encodes only those lines. Rather
than interlacing the reduced number of lines, it repeats all of them sequentially
50 (or 60) times a second. Though there is no flicker, some clarity is lost
and there may be some motion artefacts, such as ‘judder’ when the camera
pans past vertical objects like lampposts.

For equipment based on MPEG-2, each field is encoded separately to
provide an interlaced display. This, and the larger number of bits available
from the data stream for coding detail, gives pictures that match or sometimes
better today’s TV and video images.

There are already plans to use MPEG-2, at a data rate of 3 megabits
a second, to provide better picture quality from a CD. This can be done
by recording twice the number of data pits on the disc surface. There are
also plans to take the idea a stage further, by making the pits one-quarter
their current size, to quadruple the density of data on the disc and deliver
over two hours of MPEG-2 quality video from a single-sided CD. But this
will require new laser optics in the player to focus on the smaller pits.
Philips has so far vetoed the idea because it breaks the existing CD standard,
which rigidly defines the pit size and player optics.

After a meeting held at Columbia University in New York in July, the
MPEG-2 members announced that they were on schedule to recommend a standard
by November, covering data rates between 2 and 15 megabits a second. Once
this agreement is reached, C-Cube, Motorola and other chip manufacturers
will finalise the design of MPEG-2 decoder chips. These should be in mass
production by spring 1994, ready to be built into satellite receivers and
cable TV decoders.

On the face of things, the way is now clear for a smooth transition
to digital video and TV, and eventually digital HDTV, with one decoder working
for everything. In practice, the one-decoder dream is unlikely to come
true. Broadcasters will add their own proprietary scrambling or encryption
systems to the MPEG coding to ensure that people can watch only those programmes
they have paid to unscramble. Viewers will thus very likely need to have
several different decoders, all containing compatible MPEG chips, but each
with its own extra circuitry.