State of the art fiber
September 04, 2001 12:00 AM ET
by Jeff Hecht
RELATED STORIES
• More by Jeff Hecht
--------------------------------------------------------------------------------
From the September 2001 issue of UPSIDE magazine
Anyone who grew up watching Saturday-morning cartoons knows that the law of gravity may not apply until you look down. High-flying telecommunications stocks plummeted over the winter, but, like Wile E. Coyote, the fiber optics industry kept going beyond the cliff's edge.
The annual Optical Fiber Communications (OFC) Conference and Exhibit in mid-March attracted a record 38,015 people, more than double last year's 16,934. The number of exhibitors doubled to 970 -- sprawling over 630,000 feet in the Anaheim Convention Center in Anaheim, Calif.
Too late
In reality, the "uh-oh" moment had already come, but booth space and airline tickets had already been bought. The new companies crowding the show floor had already raised their startup capital, the first layoffs had already been felt and at least a few companies had scaled back their delegations.
Just before the conference, leading fiber maker Corning (GLW) warned that its earnings might slip below expectations; one wag suggested that its immense exhibition booth might account for the shortfall.
Yet you couldn't sustain gloom and doom long in the crowded technical sessions or on the bustling show floor. New technology is zooming from laboratory to market at an amazing speed. In the technical sessions, Tony Kelly, vice president of Kamelian, gave an invited talk on cutting-edge developments in semiconductor optical amplifiers and then handed out preliminary data sheets at Kamelian's booth.
Genoa earned a coveted spot in the late-papers session to report on another novel optical amplifier, and hustled the concept on the show floor. Two separate late papers reported sending a record-setting 10 trillion bits per second (bps) through single optical fibers.
It's as if the industry is counting on another law of cartoon physics: If you're running fast enough when the ground disappears from underneath you, you don't go splat when you land; you hit the ground running.
A hangover of debt
The Internet boom created a tremendous demand for additional bandwidth -- huge global information pipelines to carry data and digitized telephone calls. The telecommunications industry turned to fiber optics to fill that need, but the Internet bust rushed through telecommunications and hit the fiber industry.
"I learned that the laws of gravity apply up and down," said Jozef Straus, CEO, president, and co-chairman of fiber-component maker JDS Uniphase (JDSU), in a May 9 plenary talk at the annual Conference on Lasers and Electro-Optics (CLEO) in Baltimore.Two weeks before the conference, after years of explosive growth and a cascade of acquisitions, JDS Uniphase had announced the elimination of 5,000 jobs, or about 20 percent of its workforce.
Despite the dotcom meltdown, data traffic grew 50 percent in the last quarter of 2000, says Peter Hankin, general partner of the Infrastructure Fund and co-founder of market-research firm RHK.
Yet telecommunications companies faced what Straus called "a classic traffic dilemma." The revenue from new telecommunications services did not keep up with the cost of providing them, so companies lost money by expanding their capacity.
Established telephone companies with deep pockets slowed expansion plans but kept laying new cable. Competitive carriers that built with borrowed money stomped on the brakes, but not soon enough to avoid what Lightwave Advisors President John Dexheimer called "a massive debt hangover."
Drowning in debt
At CLEO, Dexheimer estimated that telecommunications carriers are in default on some $300 billion in debt. Much of that sum went into ill-fated satellite and wireless ventures, but some went into data networks, such as the bankrupt NorthPoint Communications (NPNTQ.OB).
Burned by bankruptcies and defaults, investors stopped lending money to expand networks. The slowdown first hit system makers such as Lucent Technologies (LU), Nortel Networks (NT) and Cisco Systems (CSCO), which stopped stockpiling formerly scarce components and started working off large inventories. In his PowerPoint presentation at CLEO, Straus summed up JDS Uniphase's experience by showing a racing kayaker plunging over a waterfall.
Like the kayaker, the fiber industry is bobbing back to the surface, driven by the buoyancy of a fast-growing technology. "The industry remains very robust in terms of demand and growth," Straus said, adding that it also faces major challenges.
Straus wants to cut the costs of delivering more bandwidth and stretching transmission distances and enhance network flexibility, so telecommunications companies can add new equipment and services quickly and efficiently, without making huge investments. That will take new technology, and the good news is that optical technology is growing tremendously.
The bandwidth race
A single optical fiber can carry signals simultaneously, at many different wavelengths -- a practice called wavelength division multiplexing (WDM). The trick to maximizing capacity is to send data as fast as you can down each optical channel, pack the channels as close together as you can and use as much of the optical spectrum as possible.
Small armies of top-level engineers outfitted with millions of dollars in equipment have achieved impressive results. Earlier this year, a team from Japan's NEC (NIPNY) packed 273 channels at 40Gbps into a single fiber, reaching a record 10.9 trillion bps. Alcatel (ALA) was a close second, reaching 10.2 trillion bps but squeezing its channels into a smaller slice of the spectrum.
The NEC experiment was equivalent to some 130 million digitized phone lines -- enough so that a single fiber could handle simultaneous calls from everyone living in Japan. Real-world technology is a few paces behind. In March, WorldCom (WCOM) and Siemens (SI) sent 80 channels at 40Gbps -- a total of 3.2 trillion bps -- through fiber that WorldCom had installed in the Dallas area.
So far, systems in commercial service have been limited to 10Gbps per channel, which allows channels to be packed together twice as tightly, so a single fiber could carry up to 160 channels if transmitters, receivers and optics are installed for all of those channels. Forty gigabits per second is proving to be a tough challenge for transmitters and receivers, which need four times the power to deliver four times as many bits.
State of the art fiber
page 3: Ways of transmission
The ugly problems come from the fiber itself. At 40Gbps, "you're worried about the dispersion of everything," said Fred Leonberger, JDS Uniphase's chief technology officer, in an executive seminar sponsored by the OFC organizers.
Getting it just right
The ultrapure glass used in optical fibers is an exceptionally good transmission medium, but the pulses spread a bit with distance, and, at 40Gbps, each pulse has to fit into a time slot lasting only 25 picoseconds (trillionths of a second).
At 10Gbps, it's possible to patch together fibers with different dispersion properties, so the pulses remain within tolerance at the end of the cable, but this is not possible at 40Gbps. The impact of dispersion increases with the square of the data rate, so a fiber that can carry 10Gbps signals for 400 miles can only transmit 40Gbps signals for 25 miles.
Worse yet, at 40Gbps, the tolerances are so tight that minuscule fluctuations in fiber properties caused by stress and temperature build up dispersion over longer distances to levels that require active compensation with complex sensors and feedback loops.
These problems don't intimidate developers, who recall that 10Gbps was a serious challenge not too long ago. While speaking at the pre-OFC seminar, Benoit Fleury, director of solutions marketing for Nortel's optical Internet, predicted that the need to conserve power and space at network nodes would push system operators to 40Gbps channels. Fleury, Leonberger and other speakers agreed that the production of 40Gbps systems is likely to start around 2003 or 2004.
Just-in-time bandwidth
For the past couple of years, skeptical analysts have warned that expanding telecommunications capacity could overshoot demand by a wide margin, producing a bandwidth glut. In fact, present Internet-traffic volume is not the overwhelming deluge of Internet myth. Andrew Odlyzko, division manager at AT&T Labs, estimates the total U.S. Internet backbone traffic to be 27,000 trillion to 50,000 trillion bytes per month in the first half of 2001. Averaged over an entire month, that number comes to only 80 billion to 150 billion bps, which a single fiber could carry comfortably.
Such numbers can be misleading. Internet-traffic volume varies tremendously with time: It may be a trickle at 3 a.m. on Monday but a torrent a few hours later, during peak business hours. The same is true for telephone traffic. Like freeways, telecommunications backbones need extra capacity to keep peak traffic flowing smoothly and are built with expansion in mind.
Fiber is cheap compared with right-of-way and construction, so once a company decides to build a network, it installs plenty of fiber. Companies like Metromedia Fiber Network (MFNX) install cables containing 400 or 800 fibers, giving them plenty of room for future expansion. In theory, each fiber has slots to carry dozens of optical channels, but, in practice, most fibers are not lit, and those that are lit carry only one or a few wavelengths.
Transmitters, receivers and wavelength-management optics are expensive, so telecommunications companies buy this equipment as they need it, filling channel slots to get "just-in-time" bandwidth. Few, if any, optical fibers today are filled to the brim with every wavelength they can carry. You can think of them as freeways with only one lane open for traffic -- economics that work for fiber-optic cables, but not for highways.
A major attraction of transmitting signals at many different wavelengths is that it opens the door to modulator growth. JDS Uniphase's Straus calls the philosophy "pay as you grow," with companies able to add capacity by adding transmitters at extra wavelengths. Other optics can also be added incrementally, as the need arises, which cuts initial investments.
Once the transmitters and receivers are installed, optical networking (see "New pipelines promise unprecedented speed") gives telecommunications companies more flexibility in managing bandwidth. Most existing high-capacity fiber lines include protection switching -- equipment that senses a failure and automatically routes signals around it by diverting them to other fibers.
Changing service to new or existing customers requires sending technicians to adjust network equipment -- "truck rolls," in industry jargon. That leads to high labor costs and a long waiting period for new services.
Telecommunications companies want a new generation of switching equipment that automates the process, so network operators at a central control facility can change services remotely.
State of the art fiber
page 4: Tradition vs. the new stuff
Not only would customers get service faster, but they could rent extra bandwidth for short periods or special events, such as running streaming video of corporate annual meetings. Network operators could dynamically balance transmission loads.
Paying for what you want
Providing this vital network flexibility requires new switching capabilities at the ends of fiber optic pipelines. Switching operates on two levels: components that redirect signals and large systems that include those switching components, plus dedicated computers that control how signals are directed.
The need for automation is pushing upgrades to the big boxes, while the need for flexibility and more efficient signal processing is also pushing a new generation of switching components.
Traditional fiber-optic systems have relied on electronic switches, which include receivers that convert the optical signal to electronic form and transmitters that convert the electronic signal back to light. The hot new trend is all-optical, or "transparent," switches, which redirect the signal while it remains in the form of light. A simple example is a mirror that moves back and forth, reflecting the input light in different directions.
One of the hottest new all-optical switching technologies is MEMS, or microelectromechanical systems, in which tiny moving mirrors redirect signals. The photolithographic techniques used to make integrated circuits are modified to etch away parts of the wafer, leaving microscopic structures that flex freely and don't wear out like bulk mechanical switches. Already used in displays and sensors, MEMS look very attractive for optical switching.
Arrays of many mirrors can redirect signals from any of many input fibers to any of many outputs, a function called a "cross-connect." Early last year, Lucent Technologies' Bell Labs reported using two arrays of micromirrors that could tilt back and forth in two directions to reflect light from any of 112 inputs to any of 112 outputs.
In May, Global Crossing Holdings became the first customer to install big switching systems that Lucent built around its MEMS mirror arrays. Global Crossing (GX), a provider of telecommunications solutions over a global IP-based network, will pay millions of dollars for switches containing two arrays of 256 tilting mirrors, each an inch square, for installation on both ends of its network of transatlantic cables.
Other companies are developing their own optical MEMS switches, including giants such as JDS Uniphase and startups like OMM (originally Optical Micro Machines) and Integrated Micromachines. Some companies use a different approach, in which the MEMS structure shifts between two extremes where it latches in place.
Several other types of all-optical switches are in development, based on technologies including liquid crystals, bubbles moving in a liquid-filled channel, and thermal effects on light guided through stripes in thin-film layers.
Switching wavelengths
It won't be enough to shift light between fibers in future optical networks. As signals are added to the fiber at different wavelengths, some signals will have to shift to other wavelengths, similar to the way cars move between lanes on a freeway.
Today, wavelength conversion takes brute force. In an approach called optoelectro-optical, a receiver converts optical signals to electronic form, and those electronic signals drive a transmitter at the new wavelength. It works, but it isn't pretty. A new generation of devices emerging from the labs does the job entirely optically, by using the input light signal to control or generate light at another wavelength.
One promising approach focuses the input light signal at one wavelength onto a semiconductor device that is amplifying a steady beam at a different wavelength. The input signal changes the amount of amplification at the other wavelength, modulating it to reproduce the signal at the second wavelength.
An ideal wavelength converter would have an adjustable output that could be set to any standard transmission wavelength. That takes a technology that's further along the development pipeline: tunable lasers.
Standard lasers emit light at a wavelength that is nominally fixed but, in practice, varies slightly with changes in device temperature or the drive current passing through the laser. As data rates have increased, laser manufacturers have spent a lot of effort stabilizing output wavelengths.Tunable lasers require enhancing these effects so that they change the laser's wavelength over a broader range and in a predictable way. The goal is to produce lasers that could be switched from channel to channel like a television set.
Like turning on the TV
Research scientists have long used tunable lasers, but they are delicate and expensive devices suitable only for the laboratory. New Focus (NUFO), a developer of cutting-edge optical equipment, has developed more practical versions -- used in fiber-optic test equipment -- based on moving optical components to adjust the output across a range of wavelengths.
Tunable-laser technology is taking the next step to devices that can be switched across optical channels and plugged into transmitters for WDM.
Most companies use different techniques. One approach is to fabricate many parallel laser stripes on a single semiconductor chip, where each stripe is designed to emit a different wavelength into a common output port. The drive circuit powers only one stripe at a time, selecting the wavelength. An alternative is to fabricate complex mirrors on both ends of the laser stripe, each one reflecting a different set of wavelengths.
The laser operates at the wavelength where the combined reflectivity of the two mirrors is highest, and this wavelength can be shifted a large amount by very small changes in the mirror properties. A variety of new and established companies pursue these approaches, including Agility Communications, a spin-off of Larry Coldren's laser research group at the University of California, Santa Barbara, and Agere Systems (AGRA), Lucent's former microelectronics division.
Straus lists tunable lasers among high-priority technologies. "I don't think anybody has deployed any tunable lasers," but, along with other tunable optics, they promise to greatly simplify logistics, he said. With fixed-wavelength lasers, every new optical channel adds not only extra capacity but also an extra set of expensive lasers to keep in inventory.
That's a problem for network operators as well as system builders, because they want spare parts ready onsite to quickly repair failed components. Having to keep track of dozens of different wavelength lasers at dozens of different sites would be an invitation to organizational chaos. The ideal replacement laser would be a single model that could be switched to any standard wavelength.
Smart amplifiers and integration
Developers are also targeting flexible components that will ease the process of upgrading. Network operators and financial managers have grown tired of "forklift upgrades," which require hauling last year's equipment to the scrap heap or investing time and money in a major overhaul. They want to avoid expensive truck rolls to manually adjust equipment at distant sites. Instead, they want automated systems that can either sense changes in network configuration or be adjusted by remote control.
The optical amplifiers used to stretch transmission distances are one example. Their performance changes as more wavelengths are piped through them. A series of amplifiers needs to have the same gain on all optical channels, or some signals will fade away with distance. This currently requires manual adjustment every time a channel is added or dropped from the system. Straus envisions microprocessor-controlled "smart amplifiers," which would automatically sense changes in the transmission load and adjust the optics to balance performance across the spectrum.
Active controls will be needed in more parts of the network as increasing speeds tighten performance requirements. The pulse spreading caused by polarization-mode dispersion varies in a seemingly random way as conditions along a fiber change.
Compensation requires sensors to detect changing pulse spreading and uses microprocessor-based control systems to adjust optics to reduce the dispersion. More automatic performance monitoring and adjustments will be needed as data rates move to 40Gbps.
The light leading the light
As optical channels are squeezed together more closely, systems will need control loops to keep transmitters and optics at precisely the right wavelengths. No system operator wants wandering wavelengths to move the Playboy Channel into the Disney Channel's slot.
Another trend is toward optical integration, an idea first pushed in the 1960s by the late Stew Miller, who was one of the earliest pioneers of optical-communication research at Lucent's Bell Labs. His idea was to guide light along stripes on flat surfaces, with each thin stripe guiding light along its length the same way optical fibers guide light through their cores.It's been done in the lab for decades, but practical integration has been harder in optics than in electronics.
Electrons interact strongly with matter, so you can pack millions of transistors onto a tiny chip. Light waves interact much more weakly, so optical components may be 10,000 times longer than transistors.
The size problem won't go away, but the growth of the fiber optics market has tipped the balance toward another attraction of integration: mass production. The first integrated optical circuit is very expensive, but, once production equipment is set up, the second and third circuits are a lot cheaper.
Once you get things working smoothly, the circuits are also reproducible, coming off the production line like cookies from a mold. The first big success of integrated optics has been the array waveguide, a single thin-film optical device that can combine or separate 16, 24, 32 or 40 wavelengths in a single stage. Other integrated optical devices are in the works.
Specializing for customer requirements
Evolving network configurations are driving other changes. The fiber-optic boom started with long-distance transmission, where system operators could justify paying top dollar to squeeze more bandwidth through hundreds, or thousands, of miles of fiber.
It's now reached the metropolitan market, where the economics and environments are different. Companies can't justify buying hugely expensive equipment to transmit signals 10 to 50 miles. Nor can they put the large, sensitive systems, designed for use in climate-controlled operating centers, at the ends of long-distance lines in the basement-level communication centers of urban office buildings.
Metro systems have to be both cheaper and more durable. It's a bit like comparing fighter jets with commuter aircraft. From far away, they both look like airplanes, but, up close, they're quite different. They might use a few common components, but most of the hardware is different.
For example, special amplifiers are being introduced for metro systems, which have less gain than their more powerful cousins used for long-distance transmission.
These changes reflect a general trend toward specialization that is even visible in the fibers themselves. Just a few years ago, almost all of the fiber Corning manufactured was SMF-28, a longtime standard with a single inner core surrounded by a single layer of cladding glass.
Now Corning also makes several types of fiber with more complex core-cladding structures: MetroCor, which is specialized for metropolitan markets; LEAF for long-distance systems; Vascade fibers for undersea cables; an enhanced SMF-28e that is more transparent at certain wavelengths; and the venerable SMF-28 for general-purpose transmission.
Corning also makes a family of special-purpose fibers for applications such as compensating for chromatic dispersion, delivering light from pump lasers to fiber amplifiers, and delivering light from a transmitting laser to an optical connector.
Growing up
The slowdown is giving the fiber industry time to mature. The message comes through loud and clear when JDS Uniphase's Leonberger says, "The technologies we choose must be manufacturable."
Yet technologies that are potentially disruptive, but not yet possible to manufacture, continue to hold developers in thrall. The past few years have seen the emergence of a revolutionary class of "holey" fibers, with holes running along their lengths.
The pattern of glass and holes creates zones that trap light, guiding it along the fiber in ways that are impossible in ordinary glass. Different patterns give the fiber properties that are impossible in solid glass.
The point where dispersion is at a minimum can be shifted to wavelengths unobtainable in glass. Other structures can guide light through hollow cores, offering ways to avoid effects, such as dispersion, that limit how far and how fast signals can travel. And holey fibers are only one aspect of the exploding field of "photonic bandgap" materials, which manipulate light in radically new ways.
Beyond the edge
In short, fiber-optic technology carried considerable momentum when it ran past the edge of the cliff. Some companies are sure to wind up as flat on the ground as Wile E. Coyote after his "uh-oh" moment in midair. The OFC show floor had an oversupply of venture-funded "me too" companies with interchangeable business plans.
Many of them will go splat. Yet the laws of cartoon physics allow characters with fast-moving legs to hit the ground running and zoom off like Road Runner. The tremendous momentum behind fiber optics and the global need for bandwidth ensure that some companies will keep on going. The problem is that, right now, while they're in midair, we don't know who will go splat.
Jeff Hecht is the author of "Understanding Fiber Optics," the fourth edition of which is just out from Prentice Hall (see fiberhome.com).
Jeff Hecht is a technology writer and author of "Understanding Fiber Optics" and "City of Light: The Story of Fiber Optics." His fiber optics website is fiberhome.com. |
