Music on Demand

Bell Labs's Ken Thompson, the father of Unix, has invented a new technology that could mean never having to buy a CD again.

By Charles Platt

Four hours after quitting time, the utilitarian, vinyl-floored hallways at Bell Labs are dark and bare; the only sound comes from massive fans cooling the louvered black tower-cabinets of a Cray supercomputer working through the night on some unimaginably complex calculation.

But here at the far end of a hallway, in a large, open space like a student lounge are sounds of human life. There's a soda machine and some whimsical mementos - an old Indiana license plate with the word UNIX on it, a Beavis and Butt-head poster, and two pink plastic lawn flamingos perched incongruously atop a 4-foot room divider. Just below them sits their owner, a big, long-haired, balding man with a ferocious scraggly gray beard and gold-rimmed bifocals that reflect the monochrome 23-inch video monitor on the desk in front of him. While a couple of hyperactive young programmers stand 10 feet away, bickering about bandwidth and scribbling equations on a white board, the bearded man scrolls through a list of song titles from the 1950s. He clicks a mouse button. Almost instantly, the sentimental sound of "All I Have to Do Is Dream," by the Everly Brothers streams from a couple of bookshelf speakers wedged among stacks of computer documentation.

The bearded man's name is Ken Thompson, and he's been at Bell Labs for almost 30 years - his entire working life. He has a private office, but he uses it only as a mail drop. From around 1 p.m. each day till 10 or 11 at night, he sits on a contoured office chair in the communal area he shares with nine other computer scientists. This is where he writes computer code, oblivious to the conversations and the clicking of many keyboards around him.

Thompson has gained legendary status in computer science. He invented Unix, an operating system of such flexibility, portability, and power that it has dominated industry and academia for more than two decades and is now used worldwide on almost all hardware above the micro level - this includes systems that sustain the Internet. But Unix is only part of the story. Thompson has done original work in areas ranging from artificial intelligence to audio compression (his current interest). Throughout his career, he has been guided not by corporate policy or by profit, but by his own quixotic impulses.

In scuffed sneakers, old jeans, and a baggy red gingham shirt that hangs untucked like a Kmart kimono, he has the presence of a hippie Zen master - calm, detached, and totally self-sufficient. Moreover, despite his dilapidated garb and unkempt hair, he maintains a role of quiet authority. He's an elder statesman treated with deferential respect by his young co-workers.

Thompson's love affair with electronics dates back to his childhood. He recalls that when he was 12 years old, living with his family in southern Texas, he "started hanging around a local radio sales-and-service store. I just kind of lurked there. It was run by a guy named Fred Schultz. Pretty soon, Fred started taking me out to fix transmitters mounted on the local oil rigs. I would scramble up the rig and unbolt the transmitters and bring down pieces of them."

This was when the first transistors were rendering vacuum tubes obsolete. But semiconductors were US$5 or more apiece - a lot of money for a seventh-grader in the 1950s. "I would manage to save maybe a dollar," Thompson recalls. "And then Fred would pretend he'd found a special sale in one of his catalogs, and he'd give me the parts for whatever money I had, so I could build little radios and oscillators and things like that."

Thompson's father was in the Navy, so the family moved around a lot. When Thompson went to college at the University of California at Berkeley, he says it felt like "the first permanent home I'd ever had." He studied electrical engineering and earned a living in his spare time repairing jukeboxes and pinball machines.

Then a recruiter from Bell Labs came around. "I wasn't looking for a job at all," Thompson recalls. "I was just floating, trying to delay a decision on what I wanted to do." But Bell Labs was willing to pay for a round-trip air fare from San Francisco to New Jersey, so Thompson went and spent two days interviewing. "Most of the departments wanted to use my talents for menial tasks, like writing a program to wind coils - a bunch of garbage. But there were two departments where people were conducting pure research. I asked what I'd be doing if I worked there - they couldn't really tell me because I would be allowed to do whatever I wanted. I found that hard to imagine. I'd never thought it was possible."

So, when they offered him a job, Thompson readily accepted.

"It's hard to describe the state of computing in those days," says Thompson. "If you wanted to print something, you had to encode it on punch cards, then put them through a batch spooler. The operating systems back then required four hours of maintenance per day, just to keep them running."

He decided to develop something better. The Unix operating system was the result. Thompson wrote the first version in 1969 on a PDP-7, a closet-sized computer that arranged memory in 8,192 18-bit "words" (unlike the 8-bit bytes used in computers today). Computer memory was hideously expensive, fabricated from copper wires crisscrossing through little ferrite rings about a quarter-inch in diameter. It took one ring to memorize each binary digit, and 8 Kwords was considered a lavish, expensive quantity of RAM to work with.

Thompson says he wrote Unix mainly for his own use. He was never very interested in what happened to it outside of AT&T. "If bug reports came in," he recalls, "we sometimes listened to them and sometimes didn't. It wasn't our job to maintain it. After all, it was essentially free."

In keeping with this take-it-or-leave-it attitude, Unix has never been a very helpful kind of operating system. Critics complain that it's unnecessarily terse and unforgiving; if you delete a whole bunch of files, for instance, there's no way to get them back.

Thompson seems a bit terse himself - amiable, but not very interested in idle conversation. Likewise, he prefers software that does the job without making a fuss. "I don't like mundane applications that draw purple borders and highlight lines of text in orange," he explains. "It's annoying."

He picks up a copy of Wired that happens to be lying nearby. "There's a similar kind of problem here." He frowns at the multicolored text, then points to the page number. "Look at that. Why is every other numeral highlighted?" He shakes his head. "I'm convinced the only reason they do that is to annoy you. What other reason could there be?"

In the late 1970s, Thompson turned his attention from Unix to writing chess-playing software, which won the world computer chess championship in 1980, running on a computer he had modified for the job.

Today, at the age of 52, he's still hacking code at Bell Labs in much the same style as when he started as a young college graduate. Why has he stayed so long while others have come and gone?

"Some people are more ambitious," he says. "They want a management career, so they go where there's more meaningful management. Here the management is just an umbrella, not really a focus of power. Then there are people who need to be directed and only flounder if they're allowed to do whatever they want. When I'm interviewing someone for a job here, I sometimes ask them what they'd do if they had complete freedom and all the money in the world. Many of them have no idea."

But Thompson has flourished under the unconditional patronage of Bell Labs. He even met his wife here, in a department just down the hall. They live together in a big, modern house 15 minutes away in the hills of New Jersey, enjoying a lifestyle that conforms to Thompson's odd hours. On a typical evening, he returns home shortly before midnight and settles down to watch a movie on his big rear-projection TV with his wife and a couple of house guests: young computer scientists shyly enjoying the couple's informal hospitality.

It looks like a comfortable life, and Thompson makes the most of it. In 1994, he visited Russia and spent $12,000 to fly as a copilot in Soviet fighter aircraft, one of them a MiG-29. He drives to work each day in a red Corvette and also maintains a street-legal Model T hot rod, which he partially customized.

In 1992, he decided he wanted something more. Wouldn't it be good, he thought, if he could sit at home and use a computer to gain easier access to music - not just a limited selection, but almost everything recorded - and to arrange it in such a way that users could browse freely through the archives.

He saw no theoretical reason why this shouldn't be possible. In the same spirit that had motivated him to develop Unix for his own use, he began to study the possibilities.

Consider the music retailing system as it exists today. You go into a store, purchase a disc, take it home, play it, and put it on a shelf. It's no different from the way people bought records 70 years ago.

Now imagine something slightly more convenient: a music database on a computer, cross-indexed by artist, date, and song title. Imagine that when you click on a song, you hear it immediately, straight from your hard drive to your stereo.

There's only one snag in this scenario: Digitized music eats a huge amount of disk space. The sounds on just one CD require 600 to 700 Mbytes of storage. So Thompson looked for a way to compress music and conserve space.

There are two methods of compressing sound: lossless and lossy. In the lossless approach, digitally sampled music is coded in such a way that it can be turned back into its exact original sequence of bits. This approach allows a maximum compression ratio of about 3 to 1.

By contrast, the lossy system discards the pieces of sound the ear can't pick up. When you see the decompressed signal displayed on the screen, it looks completely different from the wave form of the original - but it sounds miraculously the same.

For the past 10 years, companies like Dolby and Philips (the European electronics giant) have been developing schemes along these lines based on analyzing sound waves, describing them mathematically, and storing the results as a sequence of numbers.

But Thompson found the answer right next door to his workspace at Bell Labs. "The acoustic research department here has been doing pure research on and off since the '20s," he says. "That was when Western Electric used to do the soundtracks for movies. They told me they had a good algorithm for compressing music. I looked at what they had, and -"

He shakes his head ruefully. "They're acoustics people, not computer people. They were using Fortran. Original, monstrous Fortran. I reduced the algorithm's size by a factor of five and sped it up by a factor of hundreds and then started encoding music with it."

Most of the original research work was done by Jim Johnston, under the guidance of Joe Hall and Jont Allen, two other Bell Labs scientists.

Johnston freely admits that his early efforts were less than elegant. "It took me two years to develop the basic algorithm;

I was hacking it 95 ways because I was working in the dark. Then the Fortran was translated into C language, 5,000 lines became 26,000 lines - and Ken Thompson came along and assassinated the whole mess."

Collaborating with a young programmer named Sean Dorward, Thompson rewrote the code - it still performed the same task, but it ran in real time. In other words, the decompression program no longer took an hour or more to unlock two minutes of music; it could keep pace with the music, running in the background while the music played. Without this development, the system would have been unmarketable.

Sitting at his workstation under the watchful eyes of the pink plastic flamingos, Thompson turns back to his video monitor and clicks the mouse button. The crooning of the Everly Brothers lapses into silence. He scrolls through a huge list of songs, clicks on another at random, and something by Enya starts to play.

The original CD has been compressed and stowed on a massive storage system in the next room - a stack of 50 12-inch laser-discs able to hold a total of 300 Gbytes.

But the song could have been stored just as easily on the hard drive of a laptop - using Thompson's system, it has been compressed to less than 8 percent of its original size. (A Sony MiniDisc compresses music to 20 percent of its original size.) Nevertheless, the sound is still fresh and clean, indistinguishable from the original.

Using this system, a mono recording can be compressed by a factor of 15 to 1; even a full symphony orchestra in stereo can withstand 10 to 1 compression without a noticeable loss of quality. This means that within a couple of years, it should be possible for consumers to buy up to 15 hours worth of oldies on one CD.

But this is just the beginning of some awesome possibilities. Thompson has a vision of how music compression could revolutionize our access to sounds - if the distribution system will ever allow it.

"What I would like to see," he says, "is a central library of every recorded work. I'd like to sit down, pick pieces and compare versions, and I'd like to do this from home. One way would be via cable TV. The cable companies want to do this with movies, but that won't be commercially viable for years. They're aligning before the medium is ready. But with compression, the medium is ready for music right now. You could get 50 audio streams in one TV channel."

In Thompson's scenario, the cable box on top of your set becomes the output device for a remote music library that functions like the biggest jukebox in the universe. You select any song, old or new, and hear it immediately.

"Cable is a three-layered structure," Thompson says. "The first layer of distribution is already via fiber. The middle level is being converted to fiber, partly in anticipation of advanced services and partly because fiber needs less maintenance. As soon as fiber completes that second jump, you could probably have one out of three homes pick up a different tune, actively, without impacting TV broadcasts. Another way to achieve this is through ISDN, but that's not real yet. The only real thing in the home right now is cable."Of course, this would destroy the current system for distributing music.

"It would be a nightmare," Thompson agrees. "Music publishers have fits over this kind of thing. Distributors have already panicked over digital audiotape and pretty much crushed it. They're so fearful of the future. They don't stop to wonder if they'd like to swap what they do now in return for a monthly subscription fee from every home that gets cable TV."

But what about piracy? With unlimited access to every piece of music ever recorded, there would be unlimited home taping.

Thompson doesn't see this as a problem. "Once you have all the music stored somewhere - it doesn't have to be central, so long as it seems central - this alone would protect it. To reproduce the entire service, someone would have to invest a huge amount of money. You could steal a song, but who could steal them all? And if the listening fee is low enough, no one would bother to make copies. You'd just pay another nickel to hear the song again.

I think it would sell 10 times as much music as people buy now. But the guys who are potential losers are scared to death of it, and the guys who are potential winners aren't even aware of it. You don't find cable TV companies thinking about audio."

So how is it going to happen?

"I have no idea how it's going to be funded," Thompson says. "I don't even care if it's going to be successful commercially," he shrugs. "I just want to use it."

He points out that some erosion of conventional music distribution has already begun. "There are people on the Net swapping 8-bit sampled music they've created themselves. Aerosmith has released cuts from an album on the Internet. The only thing that saves full CDs from being exchanged via BBSes or the Internet right now is their bulk. But 20 or 30 years from now, the data on a CD is going to be a drop in the bucket."

AT&T, however, isn't the only company with the technology to enable this type of distribution scheme. In fact, Dolby Laboratories's work could turn out to be a major rival. Dolby released its first audio-encoding scheme, AC-1, back in 1985. It was designed mainly for specialized applications in satellite broadcasting. Subsequently, an AC-2 system was used for the transmission of stereo signals from TV studios to transmitters, enabling a compression ratio of around 5 to 1.

In the '90s, Dolby developed AC-3 for movie theaters. This enabled multiple channels of audio - three front and two rear, as well as an additional low-frequency bass, band-limited channel - to be recorded on one strip of film. "AC-3 was our first foray into multichannel audio coding," says Stan Cossette, an engineering manager at Dolby. "The system takes advantage of redundancies between channels that are being listened to simultaneously. Also, if there's a little noise on the rear channel while an explosion sounds on a front channel, you'd never hear it in an open room, so you can omit it."

According to Cossette, AC-3 has now been accepted as a standard for high-definition TV and is a leading contender in the ongoing battle to establish MPEG-2, an international protocol for compressing both audio and video signals. (The original version of MPEG is already being used in applications such as RCA's Digital Satellite System for satellite TV.)

MPEG-2 was supposed to be backwardly compatible with MPEG-1, meaning that the new standard would still work with old decoders. Philips persuaded the standards committee to use its Musicam sound-compression technology for MPEG-2, but neither AT&T nor Dolby felt satisfied with Musicam's performance; the MPEG-2 contest was eventually reopened to allow their nonbackwardly compatible systems to compete.

In 1994, Dolby's AC-3 and AT&T's compression system, developed by Johnston and Thompson, went head to head with two other contenders in special listening tests in Europe. Judges heard various types of music and rated the quality of the compressed and decompressed signals on a five-point scale.

At a sampling rate of 320 Kbits per second, AT&T's compression scheme scored highest overall. Cossette feels that Dolby's AC-3 came so close, it was virtually a tie. But Johnston angrily disagrees. "Dolby doesn't have a leg to stand on," he says. "They had two signals above 4.0 on the standard scale. We had six signals above 4.0. That is a clear indication of superior quality."

Johnston also points out that although none of the compression schemes scored a perfect 5 on all signals, it's meaningless because in some cases the judges gave the original sound a score less than 5, thinking it was a processed signal instead.

Even among experts, human perception of sound is unreliable. Johnston, now 41, learned his first lesson in this area during college days. After he installed a new sound system in the auditorium for touring rock bands, audiences complained that it wasn't loud enough - although it was as powerful as the one that had been there before. "So I took a pair of back-to-back diodes," he recalls, "and a series potentiometer, and the lower I turned the pot, the louder people thought it was." In other words, using parts available from Radio Shack for a couple of dollars, he added distortion to the sound without increasing the power. "The original signal was too clean," he says. "The more distortion, the louder people thought it was, and the more they liked it."

Later, he fooled some hard-core audiophiles by putting a switch on the front of a broken tube amplifier that had a transistor amplifier hidden inside it. The switch added some distortion of a different kind, and once again, people preferred the subtly altered sound. "Distortion is not necessarily bad," he says. "It all depends on what you're used to and what you're listening to. When people saw the vacuum tubes, they heard what they expected to hear."

High-end audio is a priesthood with its own strange sacraments, such as monster cables with gold-plated connectors. Johnston is about as welcome in this world as a blasphemer at Catholic mass. "Their measurements are close to meaningless," he says. "To take just one example: Distortion of a sine wave at one frequency is basically irrelevant. The response of a real system into a real complex load - including the interactions of the crossover in the speaker cabinets - is relevant, but it's a lot harder to measure, so you hardly ever see it done."

Johnston describes himself with disarming frankness as "grouchy, graying, and slightly bald." He speaks in the exasperated tone of a maligned iconoclast when talking about his run-ins with music salespeople or the true believers he jousts with in the Usenet group

His feisty combativeness suits him well for the continuing struggle over the MPEG-2 standard, which will now almost certainly incorporate two compression schemes - one of them Philips's Musicam, the other a nonbackwardly compatible standard probably from Dolby or AT&T.

Despite some nasty international politics (the standards committee is dominated by Europeans who are loyal to Philips), Johnston seems hopeful that an objectively reasonable decision will be reached - and he believes it will favor AT&T. "Dolby will have to accept the fact that its performance simply does not stand up to our performance," he says.

Mavericks like Johnston and Thompson have more to consider, though, if the radical change in music distribution Thompson envisions is to ever reach consumers ¬ their work is owned by one of the biggest corporations in the world. Does AT&T have any plans to push for a scheme along the lines he describes?

For the corporate policy response to this, we must turn to a different source: Sandy Fraser, associate vice president of the Information Sciences Research division at AT&T Bell Labs.

"Right now," says Fraser, "we're in competition with four other companies to have our audio compression standard adopted for FM radio broadcasts. The Federal Communications Commission wants to replace FM with digital to bring higher quality to the listener." (Presumably, digital compression might also enable more stations to share the existing radio spectrum.)

Will compression and decompression technology be sold to the public? Will AT&T offer its own decoder hardware, the same way it once tried to build microcomputers?

"We're working with ... various companies," Fraser says evasively. "I can't announce anything yet."

What about distribution? Will the corporation start sending music over its long-distance telephone network?

"With efficient digital coding," he says, "you certainly could transmit music over telephone wires."

How do the record companies feel about this?

"There is an expectation that this will happen; it's a matter of when, not if. But I don't think anybody really wants to see it happen because there's a well-established industry doing well out there - why rock the boat? And there's a great deal of worry that the consumer will make unregistered copies. But we have experience dealing with encryption, so there are things we can do."

This sounds like copy protection all over again. It's nowhere near Thompson's vision of a system so big and cheap that encryption would be unnecessary.

When will decoders become available?

Fraser hesitates, then backs away from the question. "Shortly," he says.

Within two years?

"I expect you'll be able to buy it in several forms, in much less than two years."

Sometime this year?

"I can't comment on that."

Will it cost more than $1,000?

He laughs. "It wouldn't sell if it cost more than that."

So in the near future - maybe even by the time you read this - you can expect this level of music compression to become an everyday reality. You may also gain the power to take a piece of music and squish it yourself to one-tenth its size, without any audible loss of quality. In that case, you'll be able to fit a CD onto an 88-Mbyte SyQuest cartridge and make as many undegraded bit-for-bit copies as you want.

Or, you'll be able to swap high-quality digital files over the Net without tying up your modem for hours at a time.

And maybe one day, you'll sit at home and browse through the biggest jukebox in the universe. When that day dawns, once again Thompson will have gotten what he wanted - and we will have entered a new era in which music flows into our homes like water.

Charles Platt ([email protected]) writes science fiction books and science articles. He is a frequent contributor to Wired.


[wired] [wired]


Copyright © 1995 Wired Ventures Ltd.
Compilation copyright © 1995 HotWired Ventures LLC

All rights reserved.