Despite turmoil on the business side of the industry during 2001, newly emerging technologies and enhancements to existing technologies advanced both the capabilities and capacities of WDM systems.
Jeff Hecht,Contributing Editor
This was the year the fiberoptic industry ran full-speed off the cliff. Investors got a queasy feeling when they looked down and saw the ground had gone away, and their portfolios wound up flat as pancakes at the bottom. Yet the engineers kept on innovating, and impressive advances in optical technology belie the battering that component and system makers have taken in the stock market.
A close look shows that the technology is making some subtle but important changes in direction as the industry matures. When the revolution in wavelength-division multiplexing (WDM) first multiplied the capacity of single-mode fiber transmission, developers sought to build high-throughput systems around readily available technologies. That pushed the state of the art more quantitatively than qualitatively. To slice the optical spectrum into thinner slices, for example, many engineers simply refined existing optical-filter technology. Although that approach has paid some handsome dividends, it is pushing its limits in many areas.
Now engineers are working on new optical technologies that promise hitherto unavailable capabilities that are needed for new generations of systems. For example, passive techniques based on balancing cumulative chromatic dispersion over the length of a system work well enough to transmit 10-Gbit/s signals through hundreds of kilometers of fiber, and with care can stretch transmission distances to a few thousand kilometers. Yet active compensation systems are needed for 40-Gbit/s transmission, where polarization-mode dispersion poses serious problems, and chromatic dispersion must be compensated much more accurately.
Market trends are also pushing new technology in other ways. In the current financial climate, telecommunications carriers cannot afford bandwidth at any price. They want new technology that will cut operating costs and increase flexibility, as well as adding transmission capacity. At the system and subsystem level, they want equipment that can adapt to changing situations automatically, instead of requiring technicians to travel to the site every time they have to provision new services. At the component level, this means more flexible components, such as tunable lasers and wavelength converters, and a new emphasis on integrating devices and subsystems to cut manufacturing costs.
Developers no longer see tunable lasers as universal replacements that simplify the logistics of operating WDM networks. They now envision tunable lasers as tools that can greatly enhance the flexibility of optical networks. Like the ability to switch lanes on a freeway, the ability to dynamically tune laser wavelengths gives network operators the vital capability to go around obstacles that might block one optical channel but leave others open. More generally, tunable lasers provide a new degree of freedom to manage optical signals, a vital capability in future all-optical networks.
Several types of tunable lasers are based on external cavities, using movable components such as MEMS mirrors to select wavelengths in both edge-emitting lasers and vertical-cavity surface-emitting lasers (VCSELs). One new approach is tuning fiber Bragg gratings that serve as cavity mirrors. At the European Conference on Optical Communications (ECOC 2001), a team from Micron Optics (Atlanta, GA) the University of Southampton (Southampton, England), and the University of Tokyo (Tokyo, Japan) described continuous tuning of a distributed-feedback fiber laser across 27 nm by bending fiber Bragg gratings that serve as cavity mirrors.1 Tokyo researchers introduced the bend-tuning technique at the Optical Fiber Communications (OFC) conference 2000. Bending the fiber changes the spacing of high-index zones in the fiber grating (see Fig. 1).2
Another class of tunable lasers is semiconductor diode lasers with distributed Bragg reflection (DBR) gratings fabricated at the ends of the laser's active region. The wavelength resonant in the laser cavity depends on the wavelengths reflected by the DBR gratings on each end. Either changing the grating's temperature or changing the current flowing through it can vary the refractive index, modifying the resonant wavelength. The change is small, but tuning can be extended over a broad range by independently tuning a pair of gratings, each "chirped" with spacing varying along their length. Chirped gratings reflect a comb of wavelengths, and small relative shifts between the combs of wavelengths reflected by the two ends produces a Vernier effect—shifting the resonant wavelength of the entire cavity by a larger amount (see Fig. 2).
FIGURE 2. Small changes in chirped distributed Bragg reflection gratings on opposite ends of a tunable laser diode can produce a large shift in its output wavelength. The cavity resonates at a wavelength where peaks in the reflection curves of the gratings on opposite ends of the cavity overlap. That wavelength is at the center in the top curve, but a small shift in one curve (solid line) shifts the peak reflectivity to the left in the bottom curve.
A different family of tunable diode lasers includes multiple laser stripes integrated on a single chip, with each stripe emitting at a different wavelength. Transmitter electronics select one stripe, with fine tuning by modifying chip properties. NetTest's Photonics division reported a new version with 16 emitting stripes at ECOC 2001.3
WAVELENGTH CONVERSION AND OPTICAL REGENERATION
Tunable lasers also play important roles in many new schemes for wavelength conversion. Current systems put a receiver and transmitter back-to-back, so the receiver detects the input optical signal and produces an electronic signal that drives a transmitter emitting a different wavelength, forming an optical-electro-optical (OEO) converter. The next logical step is to use a tunable laser in the transmitter, so the output can automatically be set to the desired wavelength.
In the long term, most developers would prefer all-optical wavelength conversion. The most promising approach in development appears to be the optically controlled gate, in which a weak input signal controls gain of a semiconductor optical amplifier that is amplifying a continuous beam at a different wavelength. The input signal essentially modulates the phase of the amplified wavelength, and this phase modulation is converted into amplitude modulation in various ways.4 The output wavelength can be tuned if a tunable laser is the source of the continuous beam being amplified.
Moving to all-optical regeneration clearly poses formidable challenges, but could be a critical tool in extending the reach of high-speed fiber systems. Electro-optical regeneration is expensive, and today is usually buried inside big electronic switches at major system nodes. Developers hope that all-optical regeneration will be simpler and cheaper, which could make it an important element in future all-optical networks. All-optical regeneration might prove essential in helping combat the serious dispersion problems present at 40 Gbit/s.
The technology has a long way to go, but encouraging results have been demonstrated. At ECOC 2001, a team from the University of Bristol (Bristol, England) reported demonstrating 2-R regeneration (reamplification and reshaping) at 10 Gbit/s for the first time in an integrated distributed-feedback laser and semiconductor optical amplifier.5 Their device simultaneously converted the signal to another wavelength.
VCSELS AND HYBRID SEMICONDUCTORS
Long-wavelength VCSELs look good in the laboratory but have yet to enjoy the practical success of their shorter-wavelength cousins. At ECOC 2001, the University of Ulm (Ulm, Germany) reported transmitting up to 10 Gbit/s at room temperature using VCSELs made of indium gallium arsenide nitride (InGaAsN) on gallium arsenide substrates; single-mode power reached 700 µW at 1304 nm.6 A team from Bandwidth9 reported electrically pumped VCSELs that could be continuously tuned between 1530 and 1620 nm, and could be directly modulated at 2.5 Gbit./s.7 Yet limited power remains a big question mark, which could limit applications to metro networks—assuming the devices can be mass-produced economically.
In September, Motorola Labs (Tempe, AZ) reported cracking a long-standing problem in semiconductor devices—fabricating GaAs devices on silicon substrates without high defect levels that compromise device performance. The key was depositing a thin layer of strontium titanate (SrTiO3) on the silicon as a buffer. An amorphous layer forms at the interface between the strontium titanate and the silicon, accommodating the strain arising from a 4% lattice mismatch between GaAs and silicon. Previously accommodating the large mismatch had required epitaxial deposition of thick layers of GaAs on silicon, a time-consuming and expensive process.
Motorola fabricated high-speed GaAs devices on silicon for use in cellular telephones, but company officials expect the real payoff to be in optoelectronics. A big advantage is cost: silicon wafers can be grown in sizes to 12 in. and cost a small fraction of the price of GaAs wafers, which are fragile and limited to 6-in. diameters. Another is the prospect of integrating silicon electronics with GaAs emitters.
Some important steps remain. Motorola has yet to demonstrate either optical or electronic links between the GaAs and silicon parts of a chip—a key step in actual integration of devices fabricated from the two materials. The technology has yet to be extended to other semiconductors—particularly the indium phosphide (InP) materials used in long-wavelength laser diodes. The lattice mismatch between InP and silicon is much larger—about 8%—and Motorola officials say it would require a different material than strontium titanate. However, the stakes are also higher. Indium phosphide is both a faster and a more difficult material than GaAs; the largest InP substrates are 4-in. wafers.
Raman amplification has gained acceptance as a way to complement erbium-doped fiber amplifiers (EDFAs), with lower noise and a gain profile that peaks at longer wavelength. Hybrid amplifiers have found a home in long-haul terrestrial systems.
Now Raman amplifiers are looking at new frontiers, such as amplification at other wavelengths, either in standalone devices or to supplement other amplifiers. Because Raman amplification peaks at a fixed offset from the pump wavelength —13 THz in silica fibers—changing the pump wavelength shifts the amplified band. Developers are looking at Raman amplifiers in the 1310-nm band, and in the S-band from 1460 to 1530 nm.
With several closely spaced pump lasers, a Raman amplifier can have gain uniform across a much wider range than with a single pump (see Fig. 3). This fact has led to development of a family of "14xx" pump lasers through the 1400- to 1500-nm range, based on the InGaAsP pump lasers used for EDFAs. This technology could be used for Raman-only amplification, with no EDFAs. One such system was described by a group from the Heinrich Hertz Institute (Berlin, Germany) at ECOC 2001; it used 45 dB of distributed Raman gain to transmit 10 Gbit/s through 1800 km of fiber.8 An alternative approach is the discrete Raman amplifier, which can use special fiber with higher Raman gain than standard silica transmission fibers. Phosphate glass is being investigated because its peak Raman gain is considerably higher than silica fibers.
Developers are also trying to make EDFAs more flexible devices by adding features that make them "smart." A major goal is to make EDFAs flexible enough to adapt automatically when input signals change, instead of requiring onsite modification by a technician. Dynamic gain equalization is one approach, in which input signals are monitored, and the results are used to control the attenuation of variable optical filters so the output of the amplifier remains constant. This capability will become increasingly important as optical networks become more dynamic. For example, if an add/drop switch changes its configuration, the power on the affected channel is likely to change because the signal would originate at a different point in the network. A smart amplifier would detect his change and automatically adjust internal components so the output was constant.
The big news in fibers this year has been rapid improvements in photonic-crystal or "holey" fibers, in which internal microstructures confine light. Some types have a few holes running the length of a fiber that is mostly solid glass; others have a thin lattice of glass essentially suspended in air. One class of holey fibers functions as if it were conventional fiber that has air holes in the cladding layer, reducing its refractive index below that of a solid core, so light acts as if it is confined by total internal reflection. In true "photonic band-gap" fibers, light is confined in a central region—which can be hollow—by layers that have internal microstructures that prevent the propagation of certain wavelengths.
Developers have made a variety of holey fibers, some with special dispersion properties, but until this year all had very high attenuation. At OFC 2001, a team from Sumitomo Electric (Yokohama, Japan) and Hokkaido University (Sapporo, Japan) surprised other developers by reporting a holey fiber with record low attenuation of 0.82 dB/km at 1550 nm.9 Four holes guided the light through a high-index single-mode silica core, with effective area of 58 µm2; anomalous dispersion was +34.4 ps/nm*km.
Another important milestone was a holey fiber with zero-dispersion wavelength shifted to 810 nm, in the range of GaAs light sources. A team led by researchers at NTT Network Innovation Laboratories (Kanagawa, Japan) drew a 2-km length of the fiber, in which an array of holes surrounded a central solid region. In a postdeadline paper at the Conference on Lasers and Electro-Optics (CLEO 2001), they reported loss of 3.2 dB/km at 1550 nm and 7.1 dB at 850 nm.10 Further extensions of the technology could lead to fibers with low loss and dispersion in the 850-nm region, a window where high dispersion has limited long-distance and high-speed applications.
A big surprise came in September, when separate groups in Korea and Australia succeeded in making the first plastic holey fibers, a material that other developers had considered unpromising. Groups at the University of Sydney in Australia and the Kwangju Institute of Technology in South Korea both made holey fibers from polymethyl methacrylate (PMMA), a material long used in plastic optical fibers. Sources say photonic crystal fibers proved surprisingly easy to make from plastic. The development could create new opportunities for plastic fibers, long confined to niche applications because of their high attenuation.
Researchers scored a breakthrough on the hero-experiment frontier when two groups reported at OFC 2001 that they had squeezed more than 10 Tbit/s through a single fiber. NEC's computer and communication media research group (Kawasaki, Japan) set the record of 10.92 Tbit/s by transmitting 273 channels at 40 Gbit/s, spaced 50 GHz apart in the S-, C- and L-bands.11 Alcatel Research and Innovation (Marcoussis, France) came in a close second by sending 10.2 Tbit/s in the form of 256 channels at 40 Gbit/s. By combining polarization-division multiplexing with WDM, Alcatel squeezed all channels into the C- and L-bands, achieving spectral efficiency of 1.28 bits per hertz.12 Both experiments were limited to 100-km distances.
The trick to making such high capacities practical is to develop reasonable-cost hardware that can survive the rigors of an operational environment. Components for transmitting optical-data rates of 40 Gbit/s have begun to appear, but systems are further off. The impact of dispersion increases with the square of the data rate, so a fiber able to transmit 10 Gbit/s for 400 km can send 40 Gbit/s a mere 25 km. The requirements are so tough that passive chromatic-dispersion compensation techniques developed for 10 Gbit/s will not suffice at 40 Gbit/s. Active compensation will be needed to cope with operational variations such as thermal cycling, as well as to limit the variation of chromatic dispersion across the spectrum.
Polarization-mode dispersion poses tough problems because it varies over time in a seemingly random manner, like statistical noise, causing outages when it exceeds a threshold, like trucks rumbling by an open window can obscure telephone conversations. Compensation requires first measuring PMD, then using those measurements to adjust polarizing and dispersion-balancing components. Measurements of PMD must be continuous and the compensation optics must adjust quickly. Further complications arise because PMD varies independently for each wavelength in the system.
A few companies now offer PMD compensators and measurement equipment, and two sessions at ECOC 2001 described a wide range of developments. A group from the Chalmers University of Technology (Göteborg, Sweden) found that PMD creates dispersive waves that degrade soliton transmission when they interact with soliton pulses, but that dispersion-managed solitons can reduce the interaction.13 A group from Lucent Technologies Bell Labs (Crawford Hill Lab, Holmdel, NJ) warned that overall polarization of a WDM signal should be minimized to reduce nonlinear interactions between optical channels that can reduce the effectiveness of PMD compensation.14
New alternatives are emerging for reaching high capacities. Essex (Columbia, MD) is developing hyperfine WDM, with channel spacing as close as 50 MHz. In August, Essex announced that four companies would field-test its equipment to transmit 2.5 Gbit/s with channel spacing of 6.25 GHz. Essex says that 16 channels at that speed could fit into the 100-GHz bandwidth of a single 40-Gbit/s channel.
Optical time-domain multiplexing (OTDM) generates signals at channel rates higher than available from electronic transmitters by optically interleaving streams of very short pulses. At Tohoku University, Masataka Nakazawa used OTDM to interleave pulses of 300 to 400 fs, generating data rates of 1.28 Tbit/s on a single optical channel, which he has transmitted up to 70 km. At ECOC 2001, he described key technologies needed for this feat.15 Other groups in Germany and Japan described OTDM to data rates of 160 Gbit/s at ECOC 2001.
Analysts generally consider metro networks to be the next frontier for high-speed optical systems because they pose a practical bottleneck. Corporate networks and long-distance networks both operate at gigabit speeds, but most metro systems transmit megabits. At the peak of the optical boom, operators of regional networks were ripping up the streets at record rates to install new fiber. Some cities had to limit construction projects to keep traffic jams from getting out of control.
Today the most aggressive competitive telecommunications carriers are gone, and the incumbent regional telephone companies have cut their construction budgets. Nonetheless, a real need remains for expanding metro capacity as businesses and residential customers continue to use more bandwidth. Verizon says that fiber reaches or is close to most big office buildings in densely developed areas such as downtown New York and Boston. However, elsewhere most office buildings are not served by fiber; one widely-quoted estimate is that only 5% of business buildings have their own fiberoptic links.
- S. Y. Set et al, ECOC 2001, paper TuF3.4.
- S. Y. Set et al, Tech. Digest. OFC 2001, paper MC4.
- G. Souhaite et al, ECOC 2001, paper TuF3.2.
- J. Hecht, Laser Focus World, April 2001, 159.
- M. Webster et al, ECOC 2001, paper ThF2.6.
- F. Mederer et al, ECOC 2001, paper TuB3.2.
- G. S. Li, ECOC 2001, paper TuB3.3.
- E. Schulze et al, ECOC 2001, paper TuA2.3.
- T. Hasegawa et al, OFC 2001, postdeadline Paper PD5.
- H. Kubota et al, CLEO 2001, postdeadline paper CPD3.
- K. Fukuchi et al, Tech. Digest OFC 2001, postdeadline paper PD24.
- S. Bigo et al, OFC 2001, postdeadline paper PD25.
- C. Xie et al, ECOC 2001, paper TuA3.3.
- L. Mšller et al, ECOC 2001, paper TuA3.6.
- M. Nakazawa, ECOC 2001, paper TuL2.3.