杏吧原创

IT WAS a challenge no self-respecting hacker could resist. 鈥淗ere鈥檚 an
invitation to show off your skills, make some money and help shape the future of
the online digital economy,鈥� it read. 鈥淎ttack the proposed technologies. Crack
them鈥�If you can remove the watermark or defeat the other technology on our
proposed copyright protection system, you may earn up to $10,000.鈥�

The gauntlet was thrown down last September by Leonardo Chiariglione, head of
the music industry鈥檚 Secure Digital Music Initiative. The industry was desperate
to find some way to stem the flood of commercial recordings that were being
illicitly posted on the Net, and downloaded for free by anyone with a computer
and a modem. Pirates had even begun to copy the songs and sell their
compilations at bargain prices. It was the ultimate industry nightmare: at best,
fear of piracy would prevent record companies putting their wares on the Web. At
worst, free downloads and pirate copies would mean no money to pay musicians,
the supply of new talent would dry up, and the music鈥攁nd the
profits鈥攚ould stop.

The way to halt this haemorrhage, the industry believed, was to indelibly
鈥渨atermark鈥� every song it released. While the mark would be undetectable by
legitimate users, it would spoil any attempt to make illegal copies. And the
SDMI thought it had a watermarking scheme that would do the job. Any hackers who
rose to its challenge and managed to evade the watermark would be paid to reveal
how they did it. The loophole would be plugged, the music would be secure and
the industry could enter a golden age of distributing and selling its music on
the Internet.

Yet just a few months on, many insiders believe the system is a lame duck.
They claim it offers little defence against determined pirates and, worse, it
disfigures the music it is designed to protect. Even the head of the SDMI admits
his system may not work. So is it really possible to regulate the music market
or is the mighty recording industry heading for oblivion?

All digital music is easy to copy. If your computer is fitted with a
CD-writer, you can put a music CD in one slot, a blank CD in another, and a few
minutes later you have a perfect copy. But the crisis didn鈥檛 really strike until
the highly compressed MP3 format came along a couple of years ago. Unlike a CD
track, MP3 files are small enough to be downloaded from the Net in a reasonable
time via a standard modem. Thanks to the Internet and a new generation of CD,
DVD and MP3 recorders, the floodgates were open.

The idea of marking valuable goods to prevent copying is as old as money. But
this won鈥檛 work with music, films and photographs. No one will want to buy music
blemished by audible marks, so any practical watermark for a digital recording
will have to be undetectable by legitimate users鈥攅ven to golden-eared
hi-fi buffs listening to high-quality audio DVDs. And as well as being
inaudible, the watermark has to be durable enough to survive transmission,
recording, repeated copying, digital coding and decoding鈥攁nd deliberate
attempts to remove it.

Adding watermarks to digital audio and video data is easy, you might think.
Secret authentication codes鈥攕treams of 1s and 0s鈥攂uried in the
digital data you want to mark could be used to identify the source, and control
copying on digital recorders. Unfortunately this isn鈥檛 good enough. The moment
you compress the digital data鈥攆or Internet transmission, for
example鈥攖he watermark codes disappear. Digital watermarks also vanish when
the music is converted to analogue form, for playing over the radio, for
example.

So the best way to make a watermark indelible is to impress it on the
original, analogue recording. But this, too, is easier said than done鈥攁s
more than 30 years of trying has demonstrated. Way back in the 1960s, The
Beatles鈥� electronics company, Apple Electronics, proposed a system that buried
an inaudible, ultrasonic tone in an LP recording. The idea was that it would
interfere with the high-frequency bias signal used by analogue tape recorders,
generating a lower-pitched audible tone on the tape that would spoil the
recording. In practice, the ultrasonic tone did not record or reproduce
reliably, and was all too easily filtered out.

In 1982, CBS Records came up with Copycode. Instead of adding a tone, CBS
used a notch filter to remove all sound in a narrow range of frequencies centred
around 3.8 kilohertz, almost four octaves above middle C. Sensors built into
every tape machine would prevent it recording whenever there was a notch in the
music. The notch was inaudible and could not be removed. Well again, that was
the theory, and the record industry lapped it up.

The International Federation of the Phonographic Industry, world trade body
for the record companies, announced in May 1986 that it was backing Copycode. So
did the Recording Industry Association of America. Everything looked rosy, until
some spoilsport audio engineers warned that the notch would be audible and ruin
the music. Then the US National Bureau of Standards ran independent tests on
Copycode, and reported in March 1988 that the system did not work reliably and
degraded the original sound. Copycode was quietly dropped.

Nine years later, the record companies tried again. The IFPI took on the job
of testing several new ways of embedding digital data in an analogue music
signal. By Christmas 1998, together with the Netherlands Organisation for
Applied Scientific Research, the IFPI had tested 11 systems using CD-quality
music. But no results were ever published and the project was soon overtaken by
events.

In autumn 1998, after months of legal wrangling, electronics company Diamond
Multimedia finally launched its Rio MP3 player, the first portable player
specifically designed to play music downloaded from the Web
(New 杏吧原创, 19 June 1999, p 32).
MP3 music files are small enough to be sent
across the Internet, yet the sound quality is almost up to CD standard.
Suddenly, anyone with a computer and a phone line could download tracks, listen
to them on an MP3 player or turn them into CDs and sell them on.

Realising that MP3 would hit its sales, the music industry quickly set up the
SDMI with the goal of finding a viable audio watermark. The SDMI asked a
consortium made up of Intel, IBM, Toshiba and Panasonic鈥攆our companies
with a long-standing interest in digital security鈥攖o choose the best
system. The consortium, called 4C Entity, picked a technology developed by
Californian company Verance.

Although Verance will not discuss details of how its technology works, some
information can be found in the patents filed by the inventors. The Verance
encoder analyses analogue music and alters or modulates the peaks and troughs of
the waveform with a string of digital bits that contains the copyright message.
Each watermark message consists of 72 bits that carry information on the artist
and tells a CD, MiniDisc or DVD recorder with a suitable decoder chip whether it
can make copies, how many and in what quality. It can even tell suitably chipped
MP3 players not to play pirated tracks. Meanwhile, crawler software can roam the
Internet looking for marks. If it finds them, it informs the recording company,
which can then check that the material is being used properly.

In mid-1999, 4C and the SDMI organised tests in the US to find out whether
the Verance mark was audible. They took the unusual step of guaranteeing
anonymity to the 50 golden-ear listeners who took part, in case their judgment
was later called into question. The SDMI also refused to publish details of the
test procedures. Leading audio engineers immediately objected to this. 鈥淚t鈥檚
unprofessional and unscientific,鈥� says Stefan Shibata, technical director of
Deutsche Grammophon鈥檚 Emil Berliner studios near Hanover. While the industry
will have to pay to get the security system up and running 鈥渨e have never been
invited to take part in any listening tests,鈥� says Shibata.

This criticism led the IFPI and Verance to stage further listening tests at
Sony鈥檚 studios in London last July. Half the testers failed to pick out the
watermark, and the IFPI argued that this proved it was inaudible. However,
several participants complained bitterly at the way the tests were
conducted.

Industry consultant Martin Colloms was furious: 鈥淚 withdrew from the tests.
The quality of the material was appalling. I couldn鈥檛 avoid wondering whether
they had chosen material that was bad to conceal what the system was doing.鈥�
Grammy-award winning British engineer and producer Tony Faulkner also had no
doubt that the watermark was spoiling the sound: 鈥淭his system is not transparent
and anyone who says so just isn鈥檛 telling the truth.鈥�

Finally, in September 2000, came the SDMI鈥檚 hacker challenge. The
organisation watermarked music samples and posted them on the Web, with an open
invitation for anyone to try and remove or disable the marks without degrading
the sound. In all, 447 people tried to break into the system. By November, the
SDMI had paid two hackers $5000 each for successful attacks. But the SDMI
refused to identify those it regards as successful hackers, or the methods they
used. 鈥淥ur definition of a successful attack is that the watermark is removed,
the removal is done without substantially affecting the quality of the music,
the hacker provides a detailed description of how the hack was done and the hack
is reproduced on more music samples,鈥� the SDMI says.

Then a team of nine researchers from Princeton University in New Jersey,
Xerox Palo Alto Research Center in California and Rice University in Texas, led
by Princeton computer scientist Edward Felton, went public claiming success at
hacking the system. Because they refused to sign the confidentiality agreement,
they were disqualified from winning a cash prize and the SDMI discounted their
work. 鈥淎s the Princeton hackers have elected not to comply, the Princeton group
submission cannot be considered a successful attack,鈥� says the SDMI.

But Felton insists that the hack was successful, and that his team defeated
the watermark. 鈥淥ur focus has always been on the scientific question of whether
the SDMI鈥檚 technologies, if deployed, could be defeated by pirates. We
demonstrated that they could be defeated by making small modifications to the
music files so that the watermarks were no longer detectable but the sound
quality was still acceptable.鈥� They brand the SDMI鈥檚 watermark technology as
鈥渇lawed鈥� and 鈥渋nherently vulnerable鈥�. They warn: 鈥淎s currently designed it would
not work.鈥�

The researchers plan to put details of how they hacked the system on the Web,
but have not yet done so because hacking is a criminal offence under the US
Digital Millennium Copyright Act. 鈥淲e are still discussing how much information
we can provide,鈥� Felten says. But even without this information, he thinks it
won鈥檛 take long for hackers to design a program to disrupt the watermark and
place it on the Internet for anyone to access. 鈥淲e had less information than a
real user and we were able to break the technology in less than three
飞别别办蝉.鈥�

Other observers point to what they see as a more fundamental flaw: the system
relies on decoders built into every recorder and player. The DVD Forum, which
sets the standards for all DVD formats, decreed that from October 2000 every
super-hi-fi DVD-Audio player must have a Verance sensor built in to control
playback and copying. Likewise, most new Internet download devices such as MP3
players are now 鈥淪DMI-compliant鈥�, and contain Verance chips to control copying.
Although the record companies have not yet started adding watermarks to their
tracks, having the chips built into players locks the industry into the system,
and gives hackers an easy target. 鈥淓very watermark will be hacked if you give
consumers a system they do not want and also give them a decoder to attack again
and again and again,鈥� says Christian Neubauer of the Fraunhofer Institute for
Integrated Circuits in Erlangen, Germany, where the MP3 and Advanced Audio
Coding were invented.

The safer way, Neubauer says, is to use watermarks to hide identity codes in
music and give decoders only to copyright owners who are trying to prove theft
of their work. That way the system can be changed if hackers compromise it. 鈥淭he
SDMI system may give us a few years, a few months, or a few days, but it is not
for eternity,鈥� says Karlheinz Brandenburg, director of the Erlangen
institute.

Verance, however, is standing by its system. Its chairman, David Liebowitz,
says that testing is 鈥渄one and complete鈥� and the company is pushing ahead to
integrate the system into hardware such as CD players.

The SDMI still refuses to say what loopholes the hackers revealed, and
Chiariglione has recently announced that he is leaving the organisation. But
early this year, at a conference hosted by a musicians group called the
Coalition for the Future of Music in Washington DC, he conceded that 10 or 15
per cent of the people who use online services would be able to defeat the
protection system. This did not invalidate it, he said, because avoiding the
protection system would be too 鈥渋nconvenient鈥� for most users. Whether
鈥渃onvenience鈥� is enough to protect music copyright remains to be seen.