Alastair M Glass,
COLOR="#00553b" size="3"> director of Photonics Research at Lucent Technologies’ Bell
Laboratories.
Â
The huge bandwidth of optical fibre and the
potential of multiterabit communication was recognized back in the Seventies during the
early development of fibre optic technology. For the last two decades, the capacity of
experimental and deployed systems has been increasing at a rate of 100-fold each
decade–a rate exceeding the increase of integrated circuit speeds. By 1996, the first
terabit systems were demonstrated in a number of research laboratories leading to the
expectation, based on historical trends, that commercial terabit systems will be deployed
at the turn of the next century. Because of the electronic bottleneck, these systems will
be achieved by wavelength multiplexing on the optical fibre. Greater functionality will be
introduced at the optical layer, transforming conventional point-to-point transmission
systems into flexible optical networks. Great demands will be placed on new photonic
components to meet the needs with such systems. The development of new photonic devices is
growing at a tremendous rate, and the integration of photonics components and photonics
with electronics will continue to drive down the costs and increase the functionality of
photonic devices.
The first-generation Dense Wavelength Division Multiplexed (DWDM)
systems developed in the early 1990s were point-to-point systems designed to increase the
capacity of installed fibre. Eight-wavelength systems have been deployed since 1995; 16-
and 32-channel systems and higher are now entering the market. In the research lab,
concepts and components for 100 channels and beyond are under development.
With the evolution of such high-capacity, long-haul systems, came the
concept of the wavelength add-drop in which individual wavelength channels are dropped and
added at amplifier nodes while allowing express traffic to pass through the node. This
minimizes the need for high-speed electronics at the nodes. The next step is to full
optical networking, including reconfigurable add-drop nodes and optical cross-connects in
which data traffic enters the network at any node and leaves the network at any other
node. Such networks will permit rapid restoration at the optical layer, flexible bandwidth
management in a multivendor environment, and cost savings.
Around the world, a number of operational testbeds are examining the
potential of optical networking. One of these, the Multiwavelength Optical NETwork
(MONET), consists of a fully functional testbed with three parts: a 2000-km long-haul
testbed, an optical cross-connect testbed, and a local exchange network. This testbed was
designed with multiple access points, spans with multiple originations, and terminations,
interoperability, scaleability, and transparency. This testbed has been fully operational
for the last 10 months and met performance targets for SONET, ATM, and analog FM
transmission. The economic study performed under this programme showed evident cost
savings of up to 20 percent offered by the optical network architecture.
Demand for DWDM systems has been from the long-distance operator.
However, this market will soon expand to include Metropolitan Area Networks (MANs) and
eventually Local Area Networks (LANs) and residential access. In these markets, the full
benefits of optical networking may become more evident. For penetration in LAN and access
markets, the price of photonic components must fall considerably from the present values
and new manufacturing technologies must be introduced for high-volume, low-cost component
fabrication.
Opportunities presented in the components area by the evolution of
optical networking are dramatic. New fibre designs have been developed to transport
multiple wavelengths without cross-channel interference. Demands on optical amplifier
performance include increasing spectrum and gain flatness across the entire bandwidth.
Routers are required to handle more wavelengths with low cross talk, low loss, and flat
transmission. Wavelength selectable lasers will be required both because of inventory
issues and for maximum networking flexibility (for instance, selecting a wavelength at the
edge of the network or interchanging wavelengths with the network). Reconfigurable
add-drop filters and high-count cross-connect fabrics will be essential, and network
management devices such as in-line wavelength monitors will be needed for network
management and control.
From the early development of low-loss optical fibre and reliable
high-performance lasers, the evolution of photonic systems has been paced by the
development of photonic components. System advances have been enabled by advances in
components. The invention of the optical amplifier revolutionized network design and
enabled the concept of DWDM and optical networking. In this paper, the major focus is on
the current research in network elements which enable further progress towards the vision
of terabit optical networking.
Â
align="right" hspace="3" vspace="3">Optical Fibre
Prior to the invention of the optical amplifier, the primary optical
fibre parameters of importance were optical attenuation (loss) and chromatic dispersion.
Installed fibre was optimized for use at 1.3 mm and 1.55 mm (dispersion-shifted fibre)
with low dispersion and losses approaching theoretical limits set by Rayleigh scattering.
The widespread introduction of erbium-doped fibre amplifiers and the large transmission
distance attainable before regeneration changed the focus on the fibre properties to
optical nonlinearity. Multiple wavelength channels propagating with the same velocity in
zero-dispersion fibre are seriously degraded by four-wave mixing of nearby channels. New
fibre designs which incorporate a small amount of dispersion at the signal wavelength
(i.e. Truewave fibre) eliminated this problem, allowing long-haul transmission of multiple
wavelength channels. At high frequencies (OC 192 and above) and long distances, dispersion
compensation is necessary. For longer distances or high bit rates, compensation,
etc., can be accomplished by appropriate cable design (dispersion managed cables) or by
dispersion compensation at amplifier modules.
Thus the three existing fibre alternatives are now: (1) dispersion
shifted fibre which supports high-bit-rate transmission but does not support DWDM;
polarization mode dispersion in early installed fibre may hinder upgrading some embedded
fibre to OC 192; (2) conventional unshifted fibre which can be used for DWDM but requires
too much dispersion compensation at 1.55 mm; and (3) Truewave fibre which is optimized for
WDM 1550 nm and permits high bit rates without dispersion compensation at distances up to
400 km at OC 192.
Beyond four-wave mixing, other optical fibre nonlinearities include
Stimulated Brillouin Scattering (SBS), Stimulated Raman Scattering (SRS), Self Phase
Modulation (SPM), Cross Phase Modulation (CPM), and Modulation Instability (MI). The
majority of these effects can be suppressed by techniques such as frequency dithering,
broadening, unequal channel spacing, or dispersion management and do not present a major
limitation in the foreseeable future. However, in future generation, ultra high capacity
systems utilizing a large number of wavelength channels and high total optical power, the
effects of these nonlinearities will have to be considered in systems design.
Â
align="left" hspace="3" vspace="3">Optical
Amplifiers
The first-generation Erbium Doped Fibre Amplifiers (EDFA) had a
wavelength-dependent gain profile determined by the erbium ion emission spectrum. When
multiple amplifiers are cascaded in a long transmission span, the gain slope of individual
amplifiers leads to large gain variations across the span. By incorporating fibre grating
filters with carefully tailored transmission profiles which compensate for the gain slope
into silica-based EDFA, flat-gain bandwidths of 40 nm with a gain variation ±1 dB have
been achieved with low noise amplifiers. Alternative approaches based on novel silicate
glass composition or fluoride-based glasses have been used to broaden amplifier gain
spectra, but for maximum bandwidth gain equalizing filters will be required.
SIZE="2">
The importance of these gain flattened amplifiers was demonstrated in a
system experiment consisting of a concatenated chain of eight EDFAs each separated by 80
km of Truewave fibre and carrying 32 wavelength channels separated by 100 GHz at 10 Gbps.
The received optical power after 640 km showed a variation of only 4 to 9 dB across all 32
channels when gain equalization is employed. This compares with a variation of over 30 dB
without gain equalization. In a recent breakthrough, a new amplifier design having a
bandwidth of 80 nm was demonstrated. Such an amplifier is capable of supporting 100
wavelength channels having 100 GHz spacing. This amplifier is based on silica EDFA with
two sections: one optimized for long wavelength channels beyond 1565 nm (L band) and the
other optimized for conventional channels 1525-1565 nm (C band), as previously
demonstrated. At this point, the gain spectrum in the L-band varies by several dB, but
simple addition of a second gain equalizing fibre grating filter in this section will
provide uniform gain over the entire 80-nm spectrum.
Â
align="right" hspace="3" vspace="3">Passive Optical
Components
Perhaps the greatest diversity of network element options is in the
area of passive components. For instance, a broad array of optical filters are available
from multiple vendors based on various technologies such as UV-induced optical fibre
gratings, thin dielectric films, Fabry Perot device, and bulk optic gratings. Many of
these meet the specifications of current WDM systems, but the future acceptance will
depend on scaleability of cost and performance as the number of wavelengths increases. The
gain flattening filters described earlier for optical amplifiers were based on
temperature-independent, long-period fibre gratings in which the transmission spectrum can
be precisely tailored by careful control of UV illumination of a specially designed fibre.
This same technology has been used to fabricate a number of other devices including
optical taps, add-drop filters, dispersion compensators, and reflectors.
SIZE="2">
A route to low-cost integration of photonic components uses planar
silica-on-silicon technology. Complex waveguide structures are readily fabricated by
photolithography in silicon oxides deposited on the silicon wafer. Waveguide grating
routers capable of routing up to 64 wavelengths are readily fabricated in this manner.
This technology is well suited to the low-cost integration of active photonic devices
(lasers, switches, and detectors), electronic chips and passive components on silicon
substrate, and the interconnection with optical fibre using techniques developed in the
silicon industry.
Â
align="left" hspace="3" vspace="3">Wavelength
Selectable Lasers
Commercial lasers are currently designed to operate at a single
precisely controlled wavelength, at frequencies up to 10 Gbps. High-performance and
low-frequency chirp is achieved by monolithic integration of a continuously operating
distributed feedback laser and an external cavity electro absorption modulator (EML
laser). This device represents the beginning of photonic integration of semiconductor
devices and incorporates many of the incredible advances in compound semiconductor
technology: multiple quantum well structure, chemical vapour deposition with atomic layer
control, and selective area growth. An alternative approach to EML is to use a
continuously working laser and an external Lithium Niobate modulator. Currently modulators
are commercially available for OC 384 systems. The next generation source for optical
networking presents even greater challenges. While it is feasible to construct
multichannel WDM systems with fixed frequency lasers, wavelength programmable sources will
be desirable for greatest flexibility and wavelength management in
future systems.
The simplest extension of the EML concept is the integration of eight
individual DFB lasers monolithically integrated with a combiner and electro absorption
modulator (or LiNbO3 modular). While this device has been used successfully in
experimental testbeds, it is not yet commercially available. Such an approach will meet
the needs of a modest number of wavelengths, but what about sources for 32, 64, and more
wavelengths?
For such applications, other approaches may be appropriate. A
completely novel approach to multiwavelength and wavelength programmable sources is the
chirped pulse laser. In this case, all the wavelength channels are derived from a single
ultrashort pulse laser which has broad spectral content. These pulses are dispersed in a
dispersive element (such as an optical fibre) to spread the spectrum in time. By using a
single high-speed modulator, individual time slots representing individual wavelengths can
be modulated. Experimental demonstration of 300 wavelength channels each carrying 37 Mbps
was achieved with a single modulator. By interleaving pulse trains from a modulator array
(TMD), 128 wavelengths were transmitted over 80 km fibre, each modulated at 2.4
Gbps–a total throughput of 0.3 Tbps. This device is still in the research stage and
its ultimate utility remains to be proven, but the attractive feature of wavelength
control (via modulator phase control) and the small number of components make such a
source an interesting option for many channel applications.
align="right" hspace="3" vspace="3">Optical
Cross-Connect
Reconfigurable add-drop multiplexers and optical cross-connects require
first, an optical filter for wavelength selection and second, an optical switch to
dynamically interconnect optical fibres. A number of mechanical switches are commercially
available which may be appropriate for slow (>ms) bulk-optic 1 x n demonstrations.
Integrated thermo-optic switches based on silica-on-silicon technology are commercially
available, but these too are relatively slow and are in their infancy from the point of
view of systems demonstrations.
The first 32 x 32 cross-connect incorporated in a WDM system was based
on Lithium Niobate technology. LiNbO3 switch technology uses a manufacturing platform
similar to that used to fabricate optical modulators. Four optical fibres, each carrying
eight wavelength channels, were demultiplexed and coupled to an array of 352 electro-optic
switches capable of directing any input fibre into any output fibre. The entire switch
array is reconfigurable at sub-microsecond speeds suitable for restoration at the optical
layer. At 2.5 Gbps per channel, this cross-connect throughput was 80 Gbps in its first
implementation—a rate comparable to current-day large ATM switches. Clearly, this
device can be scaled to higher capacity, but the eventual practical limitations are
unclear.
align="left" hspace="3" vspace="3">Optical Spectrum
Tap
Techniques already exist for management of DWDM systems, but as optical
networking evolves, optical monitoring is desirable to augment network management systems
and control systems, fault protections, etc. For instance, the gain of an optical
amplifier depends on the optical power in the signal channel. To control the gain and gain
transients, it is necessary to know how many wavelength channels are present at the node
and the signal levels. Local intelligence and feedback reduces the complexity of software
control.
A simple in-line optical spectrum analysis provides a simple technique
for in-line monitoring of multiple wavelengths which can be scaled to be arbitrarily
suitable to a large number of wavelength channels. Using a chirped and blazed fibre
grating, a small fraction of the signal is tapped out of the fibre and directed onto a
detector array for real-time monitoring of the spectrum and signal levels. These signals
in turn can be used to control the gain module locally.
This paper has outlined some of the progress made on network elements
which will be necessary to realize this vision, but the eventual choice of architecture
will be determined by the costs of the elements and the complexity of the software
required to manage the networks. For terabit systems, these issues remain major
challenges. As the development of network elements proceeds at an accelerating rate around
the world, similar rapid progress is being made on network architectures which meet
diverse custom and standardized interfaces requirements. As fibre continues to penetrate
LANs and moves closer to the home, the concept of optical networking discussed here for
WANs will eventually apply to these short-reach networks. The rate at which this proceeds
will depend on the development of optical networks elements which meet the criteria of
both performance and cost.