Tuneable lasers, wavelength selective switches and reconfigurable add-drop multiplexers are all helping to make optical networks more flexible, says E&T.
With Internet traffic growing 40 to 50 per cent a year while the revenue from carrying it stagnates, network operators face a dilemma - how to increase the capacity and flexibility of their networks without spending a lot of money.
What's the problem? Popular conceptions of the Internet imagine data packets flowing like water through whatever infrastructure is available, automatically finding their way around obstacles on their way from source to destination. The reality is that, although data may find its own way, the conduits through which it flows are less like the open sea and more like a Victorian sewer system. Networks of networks have been patched together over time into configurations for which the rationale may have long been forgotten, and that cannot be easily adapted.
Fortunately for networking engineers, increasing capacity is the easy part. Flexibility is more difficult.
Back in the mid-1990s, telecoms engineers invented dense wavelength-division multiplexing (DWDM), which enables a single strand of optical fibre to carry multiple channels of information using multiple wavelengths of light. But this technology comes at a cost - each wavelength of light requires an optical transceiver to convert an electronic stream of digital data into an optical signal and back again.
Fast-forward 15 years and the demand for optical network capacity is rising exponentially. The optical industry has responded by packing ever-increasing amounts of data into a single wavelength (see 'The big squeeze', E&T 20 Feb 2010, p60). Yet the cost of optical transport equipment is still dominated by the cost of those transceivers, necessary because optical signals have to be turned into electronic signals and back again whenever they need to be routed from one fibre to the next.
What if this conversion between the optical and electronic domains, and the transceivers necessary to do it, could be eliminated altogether? In the late 1990s this idea led to the concept of the optical add-drop multiplexer (OADM) - hardware that could route a signal in the optical domain, enabling wavelengths to enter (hence the 'add' part of the name) or leave (hence 'drop') a network node, while others continued on their journey unaffected. If the Internet really were a superhighway, OADMs would be the slip-roads enabling traffic to join or leave at major junctions.
The first OADMs were fixed devices - only pre-determined wavelengths could enter or leave the node. To achieve greater flexibility, networks would have to include optical equipment that could be programmed from a remote operations centre, enabling an operator to change a network's routing on demand. The reconfigurable OADMs (ROADMs) necessary to do this could save operating costs by eliminating the need to send an engineer to a network node to install or configure the optical equipment. It would also mean that network set-ups could be reworked in minutes, not days.
The ultimate vision would be a flexible optical network in which an operator could simply call upon network capacity without worrying about the physics of the underlying optical link. Concerns about the kind of fibre used in each link, the distance between nodes, and the number of nodes a signal had already travelled through would be hidden from the operators by the configuration software.
'What is not yet available is an interface similar to having an app store for the Apple iPhone,' says Stephan Rettenberger, vice president of global marketing for metro optical equipment maker ADVA Optical Networking. 'You would write applications for this flexible resource called the optical network, and applications could then use the network according to their needs.'
The ROADM ahead
Turning the 1990s dream of fully automated, flexible networks into a reality has proven more difficult than expected. There were several roadblocks. First,'the necessary optical components were still in the early stages of development. Second, it would require a completely new approach to the planning and design of optical networks, including considerable work on software control.
British firm Marconi was the first to launch a ROADM, in 1999. The SmartPhotonix PMA-32 supported 32 wavelengths flowing in two directions, usually termed 'east' and 'west', and could be used in a point-to-point or ring network (a common network configuration because it provides redundancy - in the event of a fibre being cut or malfunctioning, the traffic can be sent the other way around the ring). The system used a 'broadcast and select' architecture, which meant splitting the incoming optical signal between the drop and straight-through paths, and then using wavelength blockers provided by Corning to eliminate individual wavelengths on each path (see diagram, above right). On the port where signals were added, banks of tuneable lasers were used to insert new wavelengths.
Marconi's design was ahead of its time, but had limitations. One was its lack of architectural flexibility. The PMA-32 was direction-dependent: traffic added to the node could only travel in one direction. And the device could not be used in complex network configurations, such as the interconnected rings or mesh topologies that carriers were starting to want so they could increase the efficiency of their networks. [A mesh network is more robust in the event of failure because there are more alternative routes across the network to choose from; it also enables more efficient use of capacity, by allowing carriers to route traffic away from hotspots to links that are less heavily used.]
Another issue for first-generation ROADMs was that the add/drop ports on the front faceplate were associated with specific wavelengths. Tuneable lasers overcame this limitation for the add ports, but dropped wave lengths had to be routed to a specific faceplate port. If that port wasn't connected to a router, then an engineer would still need to go to the network node and change a patch cord manually.
Marconi's design was typical of first-generation ROADMs.
'They [ROADMs] were deployed partly for their flexibility and partly because they contained signal conditioning - a gain equaliser inside the device could balance the power among all the channels, which provided a distance extension,' says Rettenberger. But the flexibility they offered was of most use to those setting up or expanding a network: once a wavelength was set, its routing was rarely changed.
The exception to this rule was in research and education, which was an early adopter of first-generation ROADM technology because it wanted to build grid-computing systems. A university might need network capacity to access computing resources in another location for a project; once it was over, that capacity could be released for another research group.
Smaller, faster, cheaper
New ROADMs are being developed that aren't restricted in wavelength or port or direction, and so can support mesh networks. These next-generation ROADMs may help enable the flexible optical networks that were promised more than a decade ago.
Next-generation ROADM equipment is the fastest growing segment of the optical equipment business, according to US market research firm Infonetics. Overall optical equipment revenues have grown at a compound annual growth rate (CAGR) of 8per cent since 2002, but ROADM-based optical equipment revenue increased at a CAGR of 46 per cent between 2005 and 2009. This growth has been enabled, in part, because of advances in two key optical components: the tuneable laser and the wavelength-selective switch (WSS).
Early tuneable lasers were bulky, expensive, and had poorer performance than the fixed-wavelength devices they were meant to replace. Marconi claimed a major advance with the PMA-32 because it fitted a wideband tuneable laser (capable of adjusting to any of 32 wavelengths defined for optical communications at 100GHz) onto a single line card for the first time. Almost exactly a decade later, in September 2009, US optical components maker JDSU started shipping the first module containing both a tuneable laser and detector in an industry-standard XFP package measuring just 23.5mm wide by 67mm long.
Other components manufacturers are now making tuneable XFP modules, including competitors Finisar and Oclaro (formed from the former Bookham Technologies, which acquired the Marconi Optical Components group). Although there are many applications for the tuneable XFP, one of the most important is in the latest generation of ROADMs.
Meanwhile, the first-generation ROADM designs have been superseded by designs that use the WSS, which can route each incoming wavelength independently to one of many output ports (see diagram p57, bottom). WSS modules are the building blocks for ROADMs that can handle any wavelength on any port (and so are known as 'colourless') and can connect signals flowing in any direction on any port to any other port (hence 'directionless'). Three WSS modules are needed to support each connection on a colourless ROADM. This means that a colourless ROADM for use in a node on a ring needs six of the switches, while a ROADM for use in a four-connection node on a mesh network would need 12 (see diagram on facing page). Manufacturers have responded by developing WSSs with port counts ranging from 1-2, through a typical 1-9, to recently introduced 1-23 devices.
Many different technologies are used inside those components including free-space micromechanical silicon mirrors and liquid crystal on silicon, and some vendors offer more than one technology depending on the complexity of the product.
Like tunable lasers, WSS components are becoming physically smaller and cheaper as technology improves.
'There are not likely to be dramatic departures from current [ROADM] architectures,' says Andrew Schmitt, directing analyst for optical at Infonetics. 'We're just likely to see better components, or more integration.' The colourless, directionless functionality of the ROADM can use anything between two and four WSS components per line card, and so there's plenty of opportunity to combine functions and end up with a cheaper and more compact device.
Although the optical components for flexible networks are maturing, the control software still needs work. The main contender to control optical networks is an established protocol called generalised multi-protocol label switching (G-MPLS). It offers many of the key functions necessary, such as the ability for a network node to discover the topology of its nearest neighbours and to deliver real-time inventory reports, which show exactly where equipment and bandwidth are available in the network. The protocol also enables, among other things, service creation, routing optimisation and automatic protection switching.
Through GMPLS, control software can understand the resources that are available in the network. The software can also be programmed with information about any constraints that exist, especially in the optical domain, so that it can route around them. Routes may be selected to promote geographic diversity, or to save money. In all-optical networks, certain links could also be excluded from use in complex routes because the overall signal degradation would be too great.
There's always been a trade-off between flexibility and distance, because the analogue optical signal gradually degrades and will eventually need to be regenerated.
'In the optical domain we've always been discussing how flexible and transparent an optical network can be, especially when talking about large distances,' says ADVA's Rettenberger. 'I think we've come a long way. The signals are a lot more robust, we think there are definitely ways to measure the signal health and ways to insert a regeneration point if necessary, and so I think the robustness and flexibility of these solutions are coming along nicely.'
Equipment maker Alcatel-Lucent, for instance, includes a wavelength tracker in its latest ROADM equipment, which monitors the health of the optical signal and provides that information to the control software. Another option being considered by vendors is to include a few transponders at selected nodes to regenerate an optical signal if it becomes too degraded - although this technology is some way from reaching the market.
Nevertheless, analysts predict that the hardware for all-optical networks is ready for use, and that the dream of flexible optical networks is close to fulfillment.
'Mass adoption hasn't quite occurred yet, but I think we're within a number of months of that kicking off - that could be six months to 18 months,' said Schmitt, because it takes time for carriers to choose vendors and test their equipment before introducing it to their networks. But Schmitt is adamant that flexible optical networks are on their way: 'It's just a matter of time.'