espite all the talk of optical networks, a significant portion of these deployments is only partially optical–chiefly for the long haul–and when key operations like amplification, switching and routing come into play, a costly optical-electronic-optical (OEO) conversion must take place.

Minimizing the number of OEO conversions in a link seems to many a logical approach. Steve Georgis, chief executive officer of Network Photonics (Boulder, Colo.), agrees, calling OEO conversions "the most expensive portion of optical networks, both because of the speed hits and the cost involved."

"In the transmission portion, lasers and modulators are very expensive and every conversion requires new modulation, a new laser source," he said.

Georgis and others note that the best place for optical switching is closest to the core and that electrical conversion functions are best accomplished closer to and at the edge. "That’s the goal," he said, "keeping in mind that eventually IP [Internet Protocol] traffic will be converted to electrical."

Among the main obstacles to an "all" optical network, notes Georgis, is the practice of using "active" backplane technologies, or ones that do OEO conversions on the backplane, routing converted optical signals from line cards through the backplane and on to other destinations. "There are some ways to use lower-cost components like VCSELs [vertical-cavity surface-emitting lasers] in active backplanes, but you need a passive approach, without OEO conversions on the backplane, to truly connect light-to-light," he said.

Such a passive approach would work well for redirecting light pulses, obviously, but would not allow for much of the intelligence typically associated with modern networking.

Existing cross-connects are largely of the microelectromechanical systems (MEMS) variety, and as such use arrays of tiny mirrors to redirect light. Certainly there is much celebrated development in the nanotechnology realm. But there’s a long way to go before such arrays can be scalable to a useful degree, supporting many channels as we progress from the current OC-48 backplane limitations upward into core OC-192 and 768 speeds in a few years.

Cüneyt Özveren , chairman and CEO of Atoga Systems (Fremont, Calif.), says MEMS, while promising, aren’t here yet for the big time. For example, he notes, 2-D MEMS arrays are now becoming viable, but larger, more scalable 3-D ones are only in their infancy.

Still, Özveren said, the industry must find ways to bring optical switching to a new level. There are some significant benefits, he said, to going optical. "If you want to build large switches you do need to go optical. Electrical switches don’t scale well in terms of cost, power, bandwidth and so forth. If you want to scale to get beyond current size limitations and reach terabit switching speeds, you’ve got to go optical."

And with optical backplanes, Özveren said, designers will need to integrate control-plane architectures, using embedded 1,310-nanometer wavelengths alongside the data fiber’s dense 1,550-nm signals (this is applicable mostly in dense wavelength-division multiplexing–DWDM–scenarios; some designers opt for an external control channel with an Ethernet connection). This will provide a method for switching instructions but will not solve all connectivity problems.

For example, connect-Optical backplane shows some lighting line cards externally with fiber is problematic, Ozveren said, calling the myriad fiber connectors difficult for customers to deal with.

Entire systems could be brought down unintentionally if an operator disconnects a switch, Özveren said. The best option, he said, is an all-optical backplane that allows optical line cards interconnected via switching cards.

Louay Eldada, chief technology officer of Telephotonics (Wilmington, Mass.), said: "Optical switches used today in systems are mostly based on MEMS actuation." System integrators would welcome a move away from components with moving parts to solid-state switches. "The most promising form of solid-state actuation is thermo-optic actuation in materials with a large thermo-optic coefficient such as polymers," Eldada said. "These optical switches operate in a transparent fashion."

Brian Wong, the general manager for communications products at Primarion (Tempe, Ariz.), places an emphasis on economics. "We need to find ways to get cheap semiconductor technology into optical modules, components such as drivers, transceivers and transimpedence amps." Wong suggests a move to VCSELs or other type of semiconductor lasers will go a long way toward making optical backplanes feasible, as will the packaging technology to produce cheap modules to interface the waveguide to the latter.

For an OEM like Altamar Networks (Mountain View, Calif.) various approaches to reaching high optical speeds bear consideration. Marc Schwager, vice president of marketing, said: "Regenerating signals at high speeds is costly, especially with DWDM systems. If you have something like 160 wavelengths going down one fiber and every one needs to be muxed, regenerated and remuxed several times, doing that with laser light sources gets very expensive."

An alternative approach, Schwager said, is to go with an inline erbium-doped fiber amplifier (EDFA), which pumps a certain frequency into a fiber, and brings the overall gain for that fiber up some 15 to 20 decibels. Still, after various distances, artifacts are bound to occur, he said, and you’ll need to go back to electronic processing methods to clean it up.

Unlike Telephotonics’ Eldada, Schwager said many of the problems with optical switching are not in the realm of physics but in applications. In a scenario, for example, where a certain wavelength on a fiber is bound from San Francisco to New York, but must pass through a node in Denver en route, it would be optimal to leave that particular wavelength alone. "In the old days it would have stopped in Denver and been regenerated," Schwager said, but today there are methods to reach its destination without regeneration.

Such an approach involves modulation techniques, as well as amplification (i.e. Raman amps) that are typically proprietary from one vendor to another. If done correctly the signal will not degrade significantly from its source to its destination.

Like others, Schwager agrees that true packet processing in the optical realm is a ways off. Indeed, he notes that a few operations are really best left to the old standby of electronic processing–one of which is wavelength translation. "In a DWDM scenario," he said, "you could fit a large set of wavelengths into a fiber but you might have collisions at the switch if two transmissions occur at the same wavelength. To get around this you need to go to OEO as it’s about 10 to 100 times cheaper to do this kind of switching electrically than optically."