Electronic Design

Optical Networks Will Come To The Rescue As Bandwidth Demands Increase

New terabit optical technologies and components promise to help solve the bandwidth crises.

Bandwidth, bandwidth, bandwidth. As the telecommunications industry experiences unexpected exponential growth in data traffic, the demand for this commodity has become insatiable. Not only has there been an increase in the number of telecommunications subscribers, but also a tremendous increase in the online time used by existing subscribers, as well as a demand for greater speeds for attached devices and their applications. Over 50% of all traffic consists of data, rather than voice and video signals, a percentage that will continue to increase dramatically over the coming years as streaming video and voice-over IP (VoIP) signals emerge.

What's the answer to this dilemma? Telecommunications vendors say that the solution is to expand or retrofit existing networks and build new networks. A key element in this plan is the move to optical networking, which inherently carries much greater bandwidth capacity than the existing infrastructure.

Almost everyone in the networking industry sees the primary goal as scaling the optical backbone infrastructure by a factor of a thousand over the coming years. Will that produce a bandwidth glut? It doesn't appear likely. But with technological innovations occurring at a rapid pace, it's possible that bandwidth could ultimately become a tradable commodity, claims John Ryan of RHS, an optical-networking market-research firm. Instead of selling a fiber-optic cable or a wavelength on a cable, carriers and network service providers will simply sell bandwidth.

Before all of this can happen, though, cost considerations must be carefully addressed. The deregulation of the telecommunications industry in 1996 increased competition among the existing carriers and stimulated the establishment of many new independent carriers. Investments in network expansions have brought in higher revenues, but significantly increased operational costs and lower prices have placed a strain on profitability.

Other problems include the long provisioning time involved in establishing a connection for a customer and the need for improved quality of service (QoS), the measure of a carrier's ability to meet the customer's needs. QoS refers to the availability of the network with the needed bandwidth and the assurance of reliability, low error and jitter rates, and security.

As mentioned earlier, the two basic strategies for increasing the available bandwidth while controlling costs are to expand or retrofit existing networks and to build new networks. Both strategies are widely implemented on Synchronous optical networks, or Sonet (see "A Sonet Primer," p. 90).

Expanding and retrofitting existing Sonet backbone systems can be performed by three fundamental methods: by lighting up dark (unused) fibers, increasing transmission speeds in existing fibers, or by adding dense wavelength-division multiplexing (DWDM). The third approach is a way of multiplexing multiple data systems on a single fiber by employing laser transmitters on different frequencies. In most networks, unused or dark fibers can be called into service by adding additional equipment.

Higher speeds also are possible by installing new equipment. Data rates of 155 Mbits/s (OC-3) and 622 Mbits/s (OC-12) are still the most common today. Even the oldest of fiber-optic cables, though, is capable of handling data rates up to approximately 10 Gbits/s. Currently, there's a rapid in-crease in the adoption of OC-48 equipment (2.5 Gbits/s) with a trend toward 10 Gbits/s (OC-192). It will take a few more years to get to the OC-768 level of 40 Gbits/s or beyond, but we will surely arrive there.

The same basic strategies apply when building new networks, where communications providers are implementing DWDM and pushing for the highest speeds possible per fiber and per wavelength. Additionally, newer, greatly improved fiber-optic cables provided by suppliers, such as Corning, Lucent Technologies, Pirelli (now Cisco), and others, are easing the bandwidth crutch.

These newer fiber cables have lower losses and dispersion distortion than older fiber cables. This means higher speeds can be achieved over longer distances. The lower loss enables longer cable runs to be achieved before signal regeneration is required. Signal regeneration is typically necessary every 40 km or so. This optical/electrical/optical (OEO) conversion is expensive. New systems need fewer regenerations and, in many cases, none at all.

Another approach is to implement more optical and fewer electronic components. The development of erbium-doped fiber amplifiers (EDFAs) and Raman amplifiers has already begun to minimize the OEO expense and increase a fiber's span distance without regeneration. The use of optical switches eliminates OEO conversion too, and it offers the benefit of being speed and protocol independent. Also, by automating the switches, provisioning and network reconfiguration can occur faster and at lower cost.

In addition to the above solutions, three other trends are emerging. First, there's significant growth in metropolitan-area networks (MANs). While the long-distance backbone networks are undergoing expansion and upgrading, an astonishing number of MANs are being constructed in major cities to provide companies better accessibility to the high-speed backbone.

Plus, many expect the ubiquitous ring topology of Sonet to gradually be replaced with a packet-carrying mesh network that's more efficient and provides alternate paths for the signal. While that trend will no doubt develop in the future, the installed legacy Sonet networks can easily carry packet data associated with ATM, Ethernet, or other formats.

Each year, many more analysts predict the ultimate demise of Sonet. But with such a massive investment in these networks, they're not likely to be quickly replaced. Aside from that, the sales of new Sonet networks are actually increasing. Optical networks will obviously evolve, but newer non-Sonet architectures will show up first in MANs.

Finally, where does 10-Gbit Ethernet (10GE) fit into all of this? As soon as the IEEE's working group on 10GE (802.3ae) completes its work on this standard, it will become a primary alternative to the MAN.

Nothing has done more to increase the capacity of existing fiber-optic networks than DWDM, which permits multiple data streams to be combined on a single fiber. Current systems allow as many as 160 individual data channels to be carried simultaneously on a single fiber at data rates up to 40 Gbits/s, giving an overall capacity of 160 by 40 or 6400 Gbits/s (6.4 Tbits/s). The potential for future systems is over 200 channels per fiber at a data rate of 80 Gbits/s.

Wavelength-division multiplexing (WDM), another name for frequency-division multiplexing, has been widely used in radio, TV, and telephone systems. The best example today is the multiplexing of dozens of TV signals on a common coaxial cable coming into the home.

In DWDM, different frequencies or "colors" of light are employed to carry individual data streams. These are combined and carried on a single fiber. While frequency as a parameter is more widely used to distinguish the location of wireless signals below 300 GHz, at light frequencies the wavelength parameter is the preferred measure.

Remember that the relationship between wavelength in meters (λ) and frequency (f) is f = C /λ, where C is the speed of light in a vacuum or 2.998 × 108 m/s. The speed of light in fiber cable is a bit less than that or about 2.99 × 108 m/s. Wavelength is expressed in nanometers (nm) or microns (µm). Frequencies are expressed in terahertz (THz), or 1012 Hz.

Data to be transmitted in a fiber-optic network is utilized to modulate (by on-off keying or OOK) a laser-generating infrared light. Infrared signals best match the light-carrying characteristics of fiber-optic cable, which has an attenuation response to infrared light such that the lowest attenuation (about 0.2 dB/km) occurs in two narrow bands of frequencies. One is centered at approximately 1310 nm and the other at approximately 1550 nm.

The first coarse-WDM systems used two channels operating on 1310 and 1550 nm. Later, four channels of data were multiplexed. DWDM refers to the use of 8, 16, 32, 64, or more data channels on a single fiber (Fig. 1). Standard channel wavelengths have been defined by the International Telecommunications Union (ITU) as between 1525 and 1565 nm with a 100-GHz (approximately 0.8-nm) channel spacing.

The block of channels between about 1525 and 1565 nm is called the C or conventional band. Most DWDM activity currently occurs in the C band. Another block of wavelengths from 1570 to 1610 nm is referred to as the long-wavelength band, or L band. Wavelengths in the 1525- to 1538-nm range make up the S band.

Key components in a DWDM system are the multiplexer and demultiplexer. Numerous methods have been developed over the years to add and separate optical signals. Optical couplers can be utilized for multiplexing, and optical filters like fiber Bragg gratings or thin-films can be implemented for the demultiplexing. But the one method that appears to be emerging as the most popular is the arrayed waveguide grating (AWG). This device is an array of optical waveguides of different lengths made with silica (SiO2) on a silicon chip, and it can be employed for both multiplexing and demultiplexing (Fig. 2).

The multiple inputs are fed into a cavity or coupler region that acts as a lens to equally divide the inputs to each of the waveguides. Every waveguide in the grating has a length, L, that differs from its neighbor by ΔL. This produces a phase difference in the beams coming out of the gratings into an output cavity or coupler.

The output cavity acts as a lens to refocus the beams from all of the grating waveguides onto the output waveguide array. Any output contains the multiplexed inputs. For demultiplexing, the single multiwavelength input signal is applied to any of the inputs where the signals of different wavelengths propagate through the grating. The grating acts like multiple filters to separate the signals into individual paths that appear at the multiple outputs.

AWGs are popular because they are relatively easy to produce with standard semiconductor processes, making them very inexpensive. They have very low insertion loss and low crosstalk. A typical unit has 32 or 40 inputs and 32 or 40 outputs with 100-GHz spacing on ITU grid wavelengths. Some products have up to 64 or 80 input and output channels, and devices with higher channel counts are under development.

One such device, made by Kymata Inc., has a 3.5-dB insertion loss and better than 25 dB of adjacent-channel crosstalk. Other AWG manufacturers include Bookham Technology Inc., NEL Photonic Integration Research Inc., and Wavesplitter Technologies Inc. About the only disadvantage of an AWG device is that it's highly sensitive to temperature. Most manufacturers overcome this, though, with an on-chip cooling (Peltier) element to provide the necessary temperature stabilization.

A newer technique for expanding the number of channels per fiber involves an optical device called an optical interleaver or splitter. In most DWDM equipment, the standard wavelength spacing is 100 GHz. But spacing the signal-carrying wavelengths every 50 or even 25 GHz can double or even quadruple the number of channels per fiber. This job is accomplished by an interleaver.

Such a device takes two multiplexed signals with 100-GHz spacing and interleaves them, creating a DWDM signal with channels spaced 50 GHz apart. The process can be repeated, creating even denser composite signals with 25-GHz or 12.5-GHz spacing. The signals at the receiving end are recovered with the same devices used as splitters or de-interleavers.

Wavesplitter Technologies makes this type of device. The company's F3T Interleaver comes in 40- or 80-channel multiplexer or demultiplexer models for 100- or 50-GHz spacings. Data rates up to 40 Gbits/s are accommodated. At the recent NFOEC Conference in Denver, Colo., Wavesplitter demonstrated a 160-channel demultiplexer with 25-GHz spacing. With a data rate of 10 Gbits/s per channel, that translates to an aggregate data rate of 1.6 Tbits/s per fiber.

Figure 3 shows the big picture. The outputs of two eight-channel AWG 100-GHz-spacing multiplexers are combined in an interleaver, and the 50-GHz-spaced output is sent to a single fiber. An EDFA amplifies the composite signal. An add-drop mulitplexer (ADM) permits data extraction from one or more wavelengths from the fiber while new data is added on the same outgoing wavelengths. Another optical amplifier extends the distance of the fiber. Finally, a de-interleaver separates the signal into two 100-GHz-spaced signals which are then demultiplexed by the two AWGs.

Optical amplifiers boost the optical signal level over a wide bandwidth without electronic regeneration, making them ideal in multiwavelength DWDM systems. Because it's required for each wavelength, regeneration is expensive, but optical amplifiers solve this problem. In fact, if it weren't for the development of optical amplifiers, DWDM wouldn't be as practical.

In an optical amplifier, the input signal is fed through an optical isolator to prevent signal reflection (Fig. 4). The input signal is then combined with the signal from a laser diode pump in an optical combiner or two-channel WDM. The laser pump, operating at a wavelength lower than that of the signals to be amplified, provides the input power to produce the amplification. If C-band signals (1550 nm) are being amplified, then the laser pump runs at 980 or 1480 nm. The combined signal is sent to a length of special optical fiber that has been heavily doped with erbium ions.

The laser pump excites the erbium atoms to a higher energy state. When the photons produced by the input signal pass through the erbium fiber, they interact with the excited erbium atoms, causing them to relax to their normal state. In this process, additional photons are released at the same wavelengths of the input signals, thereby producing amplification.

Typical amplification gains are in the 15- to 20-dB range. But by using two laser pumps and erbium fiber coils, gains in excess of 35 dB are possible. Minimum input is typically −30 dBm, with maximum output up to 30 dBm. Noise figures run in the 5- to 6-dB range. Most EDFAs operate in the C band, although L-band amplifiers are now becoming available. Ciena, Lucent, Cisco, and SDL Inc. are examples of EDFA vendors.

The biggest problem with optical amplifiers is that the gain is a function of wavelength. Therefore, the wavelengths in a DWDM signal are amplified by differing amounts. Special gain-flattening filters (GFFs) are usually used to achieve a gain uniformity of ±1 to ±2 dB or less.

A newer form of the optical amplifier is the Raman amplifier (RA), with a similar configuration to that of an EDFA. The RA differs in that some systems have the laser pump placed after the doped fiber, so that the laser-pumped signal counter-propagates with the signals to be amplified. The high-power laser pump produces stimulated Raman scattering (SRS), which causes greater power transfers to the longer-wavelength signals.

If amplification is desired in the 1550-nm region, the laser pump is usually set approximately 100 nm lower. As with the EDFA, the gain is proportional to wavelength, meaning that some form of gain-flattening filters must be employed to accomplish a uniform response over the bandwidth. RAs introduce considerably less noise than EDFAs, so with equivalent gain they are useable over long spans of fiber up to 250 km. With the product just out of research labs, only a few suppliers of Raman amplifiers exist at this time. One source is SDL Inc.

Crosspoint switches are used throughout optical networking for reconfiguring the network, directing traffic, add-drop multiplexing, and protection switching. Most switching today is achieved electronically, but the trend is toward all-optical switching that eliminates the expensive OEO conversion processes. Furthermore, as higher data rates and different protocols are carried, all-optical switching also becomes more attractive because it's speed and protocol transparent.

Typical electronic crosspoint switches are available from many vendors, including AMCC, TriQuint, and Vitesse. Some are made with SiGe and others from GaAs to get the desired switching speed. The most common configuration is a 32- by 32-switch matrix where 32 inputs can be switched to 32 outputs. Switches with a 64-by-64 matrix are available too. These can be connected in arrays to handle an even larger number of inputs and outputs.

An example of a modern electronic crosspoint switch is the OXC family from I-Cube Inc. This line of switches is made with low-power 0.25-µm CMOS and implements a true crossbar matrix rather than a digital logic approach with FPGAs or ASICs (Fig. 5).

The OCX160 is an 80-by-80 switch, while the OCX256 is a 128-by-128 switch. Both operate at up to 667 Mbits/s, permitting the handling of OC-12 signals. While higher data rates of OC-48 and above occur in the long-haul part of the network, most switching is performed at a subrate, at slower speeds like OC-3 or even OC-1.

Key features of the OCX switches are a low-voltage differential-signaling (LVDS) interface, outputs that can be configured to flow through or be registered (clocked), and a nonblocking architecture. Nonblocking means that there will be a guaranteed connection between the desired input and output regardless of what other connections already exist. The chips also have multicast and broadcast modes where any input can be connected to desired multiple outputs or all outputs respectively. Finally, the OCX switches have an on-chip SRAM that allows a user to quickly reconfigure the switch within nanoseconds as the use demands.

On the downside of optical switches are their slightly higher switch costs and slower switching speeds. The higher cost is, of course, more than compensated for by the elimination of the OEO conversion equipment. The slower switching speeds (more than 1 ms) aren't necessarily a deterrent as most switches don't have to respond to data on-the-fly.

Numerous Techniques Available
A wide range of optical-switching methods have been developed in the past years using thermo-optic techniques, liquid crystals, and lithium-niobate (LiNO3) materials. The newest types showing the most promise, however, are those that employ microelectromechanical systems (MEMS) and bubble ink-jet printer technology.

Standard semiconductor manufacturing methods are used to make the MEMS devices that feature an array of tiny mirrors. These mirrors can be tilted by the application of a static voltage, and by tilting a mirror, light can be rerouted to a desired output port.

Small arrays of 4-by-4, 8-by-8, 16-by-16, and 32-by-32 matrices are already in limited use. But larger arrays, like 512 by 512 and 1024 by 1024, are under development. These all-optical switches have a typical loss in the 2- to 3-dB range and a switching time between 10 and 15 ms. Companies making these devices include Optical Micro-Machines Inc., Calient Networks, Nortel Networks (Xros), and a recent startup called Iolon.

One of the most promising optical switches is Agilent's N3565A 32-by-32 Photonic Switching Platform (see the opening illustration). This device uses planar lightwave circuits with intersecting silica waveguides. At each intersection of the waveguides, a trench is etched and filled with a clear liquid whose light index matches that of the waveguide so that light will pass directly through. Beneath the trench at every intersection is a tiny thermal heater that creates a bubble of the liquid when current is applied. That bubble reflects the light from the waveguide to an intersecting one. This is the same thermal technology that's used in Hewlett-Packard's ink-jet printers.

The photonic switch is fully transparent over the 1270- to 1650-nm range, permitting it to transmit light in any format. It has an insertion loss in the 2.5- to 7.5-dB range, depending on switching conditions. Switching time is less than 10 ms. The 32- by 32-switch elements can be combined to form larger arrays, such as 512 by 512. In addition to the 32 switchable inputs and 32 outputs, the device has 32 add ports and 32 drop ports. A dual 16- by 32-switch variation is available too.

Another example is the PhotonX-8 thermo-optical switch from Lynx Photonics Networks (Fig. 6). This 8-by-8 switch uses planar optic waveguides and Mach-Zehnder interferometers to achieve nonblocking operation with a rate up to OC-192. When a switching signal is applied, a heating element causes a change in the index of refraction on one of the optical waveguides, thereby introducing a phase shift that causes an input signal to be cancelled or blocked. Kymata also offers 1-by-2 or 2-by-2 thermo-optical switches with a switching speed of less than 2 ms and an insertion loss of less than 2 dB.

Chorum Technologies offers 1-by-2, 1-by-8, and 2-by-2 optical switches using liquid-crystal technology. By applying a voltage across the liquid, the polarization of the light can be changed, permitting light to be blocked or passed. This technique can be adapted to variable attenuation.

Aside from switching time and insertion loss, another key specification to watch for in photonic switches is extinction ratio. This ratio of the output power in the on-state to the output power in the off-state should be as great as possible. MEMS switches have the best performance with figures in the 40- to 50-dB range. Other optical switches can achieve extinction ratios of 30 dB or more. And, crosstalk between channels should be greater than −40 to −50 dB.

One of the primary building blocks of an optical network is the transceiver. This is a combination of an infrared laser-diode transmitter that converts the digital data to light pulses, and an optical receiver that converts the light pulses back to a digital format. In the receivers, either a positive-intrinsic-negative (PIN) diode or an avalanche-photodiode (APD) optical detector is used with an amplifier to convert light to the digital signal. The APD is more sensitive, making it the preferred detector in long-haul systems. The PIN detector, however, has a higher frequency range, so it's the best suited for use in high-speed applications.

In long-haul networks where higher power is needed, separate laser transmitters are employed. For years, the Fabry-Perot cavity edge-emitting laser was the primary transmitter. The distributed feedback (DFB) laser with its built-in corrugated waveguide filter is more widely used today because its narrower output spectrum makes it superior in DWDM applications. A variation of the DFB laser is the distributed Bragg reflector (DBR) laser that allows easier adjustment of gain and frequency. Most lasers have built-in Peltier cooling capability to prevent them from burning up and to maintain them on frequency.

Moreover, two primary trends exist with laser transmitters—tunable lasers and external modulators. Most lasers are fixed for a precise wavelength. But with the number of distinct DWDM channels pushing upwards of 200, producing and stocking so many different and very expensive laser components is impractical. Luckily, laser-diode manufacturers have recognized this inventory control problem and responded with tunable lasers. Most of these are only tunable over a 10- to 20-nm range, permitting them to cover only about ten to twenty wavelengths. This means that several different ranges of tunable laser are needed to cover the full-DWDM range. Lasers with a tuning range of 50 to 100 nm are in the works.

Recently, progress has been made in vertical-cavity surface-emitting lasers (VCSELs). While these devices haven't been a factor in long-haul or MAN applications, recent breakthroughs provide a new alternative. A short while ago, Novalux Inc. announced its NECSEL device with sufficient power for this market. VCSELs also offer the potential of wavelength tunability when included in MEMS structures.

Another major trend in lasers pertains to how a laser is modulated. In most systems, the laser diode is kept on just above the threshold of lasing, but with little or no light output. The laser current is then quickly switched to a higher level to transmit a pulse.

But at gigabit rates, switching the laser current results in chirp, a slight variation in wavelength. This widens the output bandwidth, making it undesirable in DWDM systems. To overcome this problem, an external modulator is used. While maintained at full output, the laser's output is passed through an electronic-optical shutter.

Two popular shutter types exist: the electro-absorption modulator (EAM) and the lithium-niobate modulator. Both kinds are widely used. They're both available as separate devices, or they may be incorporated into a standard laser assembly. Such laser modulators can modulate a light beam at rates up to 40 Gbits/s. An example is CyOptics' indium-phosphide EAM.

Pulsed lasers are yet another recent development (Fig. 7). These devices generate short pulses, called solitons, that are used in a modified return-to-zero (RZ) modulation scheme. A soliton is a very short pulse (about 5 ns for 10 Gbits/s and 2 ps for 40 Gbits/s) that has been especially shaped to take advantage of the nonlinear phase-modulation effects of fiber cable. Such pulses aren't subject to the usual pulse dispersion (spreading), meaning that the pulses can be propagated over very long distances at extreme data rates. These short pulses are a necessity in very long-haul (greater than 1000 km) 10-Gbit/s networks, and especially for future 40- and 80-Gbit/s networks.

Finally, optical techniques also are increasingly being used in LANs. The popular Fibre Channel (FC) architecture is obtaining wide adoption for LANs as well as newer storage-area networks (SANs). FC is even showing up in some of the smaller MANs due to its low cost for 1- and 2-Gbit/s operation over MAN distances. Even 10- and 100-Mbit/s (100BaseSX) Ethernet networks have optional optical transmission modes to extend a LAN's range. Furthermore, optical technology is gaining ground as 1- and 10-Gbit/s Ethernet networks are deployed. Now the question is when the carriers will extend the optical connection to the home and office.

Some Optical Networking Equipment Manufacturers
Agilent Technologies Inc.
(800) 432-4844

Bookham Technology Inc.
(408) 451-3940

Chorum Technologies
(214) 570-3532

Calient Networks
(805) 562-5501

CyOptics Inc.
(781) 229-5820

I-Cube Inc.
(408) 341-1888

JDS Uniphase Corp.
(408) 434-1800

Kymata Inc.
(925) 251-0900

Lucent Technologies
(908) 582-8500

Lynx Photonic Networks Inc.
(310) 456-7587

Nortel Networks
(800) 466-7835

Novalux Inc.
(408) 736-0707

NTT Electronics Corp. (NEL)

Optical Micro Machines Inc.
(858) 320-2800

Photonic Integration Research Inc.
(614) 876-5655

SDL Inc.
(408) 943-9411

Wavesplitter Technologies Inc.
(510) 580-1395

General Web sites with fiber-optic information:

TAGS: Components
Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.