Fundamentals of Optical Networking

The Bell system and other nations' telephony providers began deploying optical fiber for long distance voice trunks as early as 1977. Those first commercial deployments carried DS3 traffic (45Mbits/sec) a few dozen miles before requiring a repeater to regenerate the signal. Today's fiber optic systems support throughputs as high as 1Tbit/sec, with unassisted propagation distances of 3,000 miles and farther.

The story of the huge increases in capacity that optical networking enables is the story of a handful of technical breakthroughs, beginning with the laser in the late 1950s, followed by huge advances in the clarity of glass fibers, then by native optical amplification technology, and finally by Wavelength Division Multiplexing (WDM).

Any sort of light will travel along an optical fiber. The boundary between two light-conducting media with differing indices of refraction serves as an efficient mirror to light that hits the boundary at a sufficiently low anglethe same phenomenon accounts for the glare of the sun across a body of water in the late afternoon or early morning.

Optical fiber cabling encases a narrow thread of glass inside a layer, or "cladding," made of material with an index of refraction different from that of the fiber. Thus, properly introduced light will travel great distances, so long as the fiber itself doesn't absorb the light.

However, ordinary white light sources or even the single-wavelength light emitted by LEDs can only carry limited quantities of data. White light's many wavelengths each travel through the fiber at different speeds, limiting the resolution of a series of digital pulses . While the wavelengths of LED-generated light are bunched together around a single value, they are still not as tightly bunched as those produced by a laser.

The coherent light produced by lasers has a uniform phase, as well as a uniform wavelength. The upper throughput limit of LED-generated signals is about 300Mbits/sec, whereas lasers haven't shown any sign of pooping out at 10Gbit/sec (OC-192) and higher data rates. Lasers have also traditionally been capable of creating higher levels of power than LEDs.

A technology on the verge of real deployment is the tunable laser. Historic lasers, of course, produce a very pure, single bandwidth. In the short term, a laser capable of generating multiple wavelengths would cut the operating and inventory costs of carriers that operate networks using many wavelengths. More significantly for the long term , tunable lasers hold the promise of all-optical, high-speed switching and routing.

The second key to the development of optical networking was the development of low-loss cable. In the 1970s, fiber losses were on the order of 20 decibels (dB) per kilometer. State-of-the-art fiber optic cable nowadays has losses in the range of .2dB to .3dB per kilometer.

It turns out that silica or glass aren't the limiting factors, but that trace impurities, especially such fairly common elements as iron, copper , nickel, and chromium, are the biggest problem. In fact, one part per billion of these elements is the threshold of tolerability.

Furthermore, hydroxyl ions (OH), present in every water molecule , will absorb important wavelengths unacceptably if they are present in one part per 100 million. The development of fabrication techniques that could achieve these exceedingly high levels of purity were key to increasing the performance of optical fiber.

Electronics Plus Optics

The typical applications of the 1980s used fiber optics purely for transport. Inputs to cable were always electronic, and at the destination the data was converted back to an electronic signal. Furthermore, whenever the signal weakened along the way, an electronic repeater regenerated it. There was no purely optical regeneration capability until the development of the Erbium Doped Fiber Amplifier (EDFA) in the late 1980s.

An EDFA is a purely optical device in the sense that there is no conversion of optical signals to electronic signals, and all the relevant steps of its operation are the actions of photons no electron flows are required.

Erbium is special because it, like several other rare earth elements, has a complex system of electron shells or bands, including a metastable band whose energy level difference from the base configuration of erbium ions is close to the energy of a photon in the 1,550 nanometer (nm) range.

The 1,550nm area is central to most modern optical developments because optical signal losses are at their lowest values in this range. (The longest visible wavelength, where red turns into infrared, is approximately 770nm. The wavelengths that carry signals in optical fiber are not visible to the naked eye.)

A section of erbium-doped fiber is exposed to intense levels of light980nm and 1,480nm wavelengths are workable choicesand the erbium ions respond by absorbing photons, whereupon some electrons jump into higher-energy metastable orbits.

The arrival of a signal photon causes an electron in a metastable energy band to jump back to the base band and give up a photon with precisely the same wavelength and phase as the incoming signal photon. Thus a cascade of photons with the same direction, wavelength, and phase can result from a single photon, resulting in the purely optical amplification of the input signal.

Purely optical operations aren't desirable just for their technical elegance . For one thing, optical operations can take place faster than electronic ones. As the commonly encountered throughput rate of a backbone link reaches 10Gbits/sec and higher, there is some question whether electronic devices can keep up, especially when cost is taken into account.

EDFAs amplify rather than repeat, as traditional electronic regeneration devices do. Both amplification of the incident signal and digital repeating get the job done, but amplification has a strong flexibility advantage. For example, if you decide to change the line code of the signal, perhaps encoding two or three bits per symbol instead of one, an amplified link requires no change in configuration.

A repeater, on the other hand, has to be replaced or reconfigured to accommodate line code changes. Furthermore, as WDM signals become widespread, a link with EDFAs needs no upgradeit simply amplifies wavelengths as they arrive , cascading out appropriate flows of photons for each wavelength.

If a link with repeaters is converted to WDM, repeaters for each wavelength have to be installed at each regeneration point, along with splitting and recombining hardware to get all the signals off to the correct electronic processor. In general, an EDFA is likely to cost substantially less than an electronic repeater solution.

Passive Optical Networks

EDFAs make it possible to get optical signals to just about any destination without electronic processing. However, there are few purely optical solutions for routing or switching optical signals, whether the solution resembles circuit switching or packet switching, and whether the links are connection-oriented or connectionless.

One purely optical technique employs some of the characteristics of shared-medium LANs or cable-TV systems, however.

You can use optical couplers and splitters to distribute every wavelength to every node of the network, then you can filter all the inappropriate wavelengths at the end point. (This filtering step is analogous to shared Ethernet, where every node on the network has electronic access to each frame, but nodes process only the frames intended for them.) Thus, every end point has access to one or more wavelengths over the common medium.

On a hybrid fiber/coax cable-TV system, the final multiplexing step is performed electronically , but the network between the head end and the optical end nodes is a Passive Optical Network (PON). Access and metro-area networks that don't have to cover long link distances may be able to avoid amplification, and thus have more choices for usable wavelengths. PONs often employ wavelengths in the 1,310nm range.

WDM and Dense WDM (DWDM) have been the big news in optical networking for the last 10 years or so. The number of wavelengths that can be accommodated on a single fiber has recently doubled every couple of years. The first-to-be-standardized 1,530nm to 1,560nm basic range supports only 40 channels with a channel spacing of 100GHz, with another 40 channels defined in the 1,560nm to 1,600nm "extended" range that can be amplified with enhanced EDFAs. The number of channels doubles as the channel spacing is halved, but there is an upper limit eventually determined by various noise, distortion, and crosstalk factors.

AT&T Bell Labs demonstrated 1,022 wavelengths in a recent test, but not with an OC-192 throughput rate on each wavelength. Because EDFAs operate across a fairly narrow window in this wavelength range, using other wavelengths would necessitate either an alternative optical amplification technology or electronic regeneration, with the increased costs and complexity mentioned earlier.

Wavelength "routing," which might be more accurately described as wavelength circuit switching, is a crucial part of a DWDM network (see figure 1). Today's commercial products require an electronic conversion to move traffic off one wavelength and onto another. Given the latency and relative slowness of the electronic conversion, switching a wavelength using this system takes a millisecond or longer.

Figure 1: Wavelength Routing. A network of optical fibers, each of which carries multiple wavelengths (or lambdas), can be used much more effectively if data paths can be switched from one lambda to another at each node. Today's wavelength routers require an optical-to-electronic-to-optical conversion in order to switch to a different node.

Several solutions that more closely approach the pure-optical ideal have been demonstrated, however. One involves some form of tiny mirrors that are reconfigured to reflect demultiplexed wavelengths along their intended paths. Another system, recently demonstrated by Agilent Technologies, involves injecting inkjet-like bubbles into a pattern of junctions to steer optical traffic.

While both of these approaches avoid the dreaded optical-to-electronic conversion, they are still rather mechanical and slow. The estimated time for their operation also takes milliseconds , rather than the microseconds or nanoseconds required to create ATM- or frame-relay-like virtual circuits.

These technologies are suitable for manual provisioning and for the sort of protection switching that SONET provides, but the pure optically routed network is still years away.

Resources

A truly magnificent book that covers all the contributory technologies affecting optical networking is Understanding Optical Communications , by Harry J. R. Dutton, Prentice Hall PTR, 1998, ISBN 0130201413.

Cisco Systems acquires optical networking companies for breakfast . The white papers at the URL below are concerned with wavelength routing. Go to www.cisco.com/warp/public/cc/cisco/mkt/optical/wave/prodlit/index.shtml.

Alcatel has a collection of white papers on various optical networking topics at www.usa.alcatel.com/telecom/transpt/optical/techpaps/techpaps.htm.

Nortel has multiple classes of optical offerings, including end-to-end optical Ethernet. You can find several white papers on this topic at www.nortelnetworks.com/products/01/endtoendethernet/doclib.html.

This tutorial, number 142, by Steve Steinke, was originally published in the May 2000 issue of Network Magazine.