Introduction

The Need for Speed

Telecommunications for local access has evolved slowly: telephones and televisions, the primary terminal types in today�s access networks, have remained essentially unchanged in function and bandwidth for half a century or more. The limited capacity of these terminals, and the fixed nature of the services they support, has allowed them to smoothly evolve into efficient carriers of their respective services.

The last decade has shown that this situation is no longer stable. Although services such as fax and Internet can be delivered over telephone lines today, it is clear that this is patchwork adaptation. Even high-speed cable modems will not permanently solve the problem of telecommunications for access in future decades, since the rapid advances of personal computer processing power has changed the likely trajectory of telecommunications evolution. Recent experiences with the World-Wide Web suggest that as soon as subscribers view images, they desire video clips. Hyperlinks and low cost memory suggest that many of us will become servers, and video sources are likely to pop up in practically anyone�s backyard. Thus, the demand for a variety of service options and the supply of inexpensive computer processing speed and memory seem likely to equilibrate at a point that is far away from the capabilities of today�s networks: the networks will have to be upgraded far beyond today�s capabilities. Today, the tendency is towards optical communications.

While there are several contenders for the protocol and architecture standards (e.g. ATM), what is clear is the need for faster physical layer technology, to 1 Gbps and beyond. One key difficulty is that the most commonly installed fiber in local area networks does not support this bandwidth over distances of 500 meters due to modal dispersion, which limits the effective bandwidth distance product.

WDM offers an attractive solution to increasing LAN bandwidth without disturbing the existing embedded fiber, which populates most buildings and campuses, and continue to be the cable of choice for the near future. By multiplexing several relatively coarsely spaced wavelengths over a single, installed multimode network, the aggregate bandwidth can be increased by the multiplexing factor.

Multichannel optical systems were relatively unknown in 1980, but much technological progress has been achieved since then. The applications can include a multiplexed high-bandwidth library resource system, simultaneous information sharing, supercomputer data and processor interaction, a myriad of multimedia services, video applications, and many undreamed-of services. As demands for more network bandwidth increase, the need will become apparent for multiuser optical networks, with issues such as functionality, compatibility, and cost determining which systems will eventually be implemented.

Fiber Bandwidth

The driving force motivating the use of multichannel optical systems is the enormous bandwidth available in optical fiber. The high-bandwidth characteristic of the optical fiber implies that a single optical carrier can be baseband modulated at ~25,000 Gbps, occupying 25,000 GHz surrounding 1.55 nano-meter, before transmission losses of the optical fiber would limit transmission. Obviously, this bit rate is impossible for present-day optical devices to achieve, given that heroic lasers, external modulators, switches or detectors have bandwidths < 100 GHz. As such, a single high-speed channel takes advantage of an extremely small portion of the available fiber bandwidth.

Popular Multiplexing Methods

The 100-Gbps channel mentioned in the previous section probably will be a combination of many lower-speed signals, since very few individual applications today utilize this high bandwidth. These lower-speed channels are multiplexed together in time to form a higher-speed channel. This time-division multiplexing (TDM) can be accomplished in the electrical or optical domain, with each lower-speed channel transmitting a bit (or allocation of bits known as a packet) in a given time slot and the waiting its turn to transmit another bit (or packet) after all the other channels have had their opportunity to transmit ( Figure 1 ). TDM is quite popular with today�s electrical networks, and is fairly straightforward to implement in an optical network at < 100-Gbps speeds. This scheme by itself cannot hope to utilize the available bandwidth because it is limited by the speed of the time-multiplexing and -demultiplexing components.

To exploit more of the fiber�s THz bandwidth we seek solutions that complement of replace TDM. One obvious choice is WDM (wavelength division multiplexing), in which several baseband-modulated channels are transmitted along a single fiber but with each channel located at a different wavelength ( Figure 2 ). Each of N different wavelength lasers is operating at the slower Gbps speeds, but the aggregate system is transmitting at N times the individual laser speed, providing a significant capacity enhancement. The WDM channels are separated in wavelength to avoid cross-talk when they are (de)multiplexed by a non-ideal optical fiber. The wavelengths can be individually routed through a network or individually recovered by wavelength-selective components. WDM allows us to use much of the fiber bandwidth, although various device, system, and network issues will limit the utilization of the full fiber bandwidth. Note that each WDM channel may contain a set of even slower time-multiplexed channels.

Another method conceptually related to WDM is subcarrier multiplexing (SCM). Instead of directly modulating a ~terahertz optical carrier wave with ~100s Mbps baseband data, the baseband data are impressed on a ~gigahertz subcarrier wave that is subsequently impressed on the THz optical carrier. Figure 3 illustrates the situation in which each channel is located at a different subcarrier frequency, thereby occupying a different portion of the spectrum surrounding the optical carrier. SCM is similar to commercial radio, in which many stations are placed at different RF (Radio Frequency) such that a radio receiver can tune its filter to the appropriate subcarrier RF. The multiplexing and demultiplexing of the SCM channels is accomplished electronically, not optically. The obvious advantage of cost-conscious users is that several channels can share the same expensive optical components; electrical components are typically less expensive than optical ones. Just as with TDM, SCM is limited in maximum subcarrier frequencies and data rates by the available bandwidth of the electrical and optical components. Therefore, SCM must be used in conjunction with WDM if we want to utilize any significant fraction of the fiber bandwidth, but it can be used effectively for lower-speed, lower-cost multiuser systems.

Additional method is the code-division multiplexing (CDM) ( Figure 4 ). Instead of each channel occupying a given wavelength, frequency or time slot, each channel transmits its bits as a coded channel-specific sequence of pulses. This coded transmission typically is accomplished by transmitting a unique time-dependent series of short pulses. These short pulses are placed within chip times within the larger bit time. All channels, each with a different code, can be transmitted on the same fiber and asynchronously demultiplxed. One effect of coding is that the frequency bandwidth of each channel is broadbanded, or �spread�. If ultra-short (<100 fs) optical pulses can be successfully generated and modulated, then a significant fraction of the fiber bandwidth can be used. Unfortunately, it is difficult for the entire system to operate at these speeds without incurring enormous cost and complexity.

Yet another optical multiplexing scheme is called space-division multiplexing (SDM), in which the channel-routing path is determined by different spatial position (i.e., a different output fiber). A simple example of this is shown in Figure 5, in which the optical output of a fiber is split into N different and parallel optical beam paths. Each of the N output beams is passed through a light-modulating switch and then coupled to a different output fiber. By controlling the transmissivity of each optical modulator, a signal on the input fiber can be routed to any fiber output port. By extending this scenario, N input fiber ports can be fully interconnected with N output fiber ports by an array of N² optical switches. The technology for implementing moderate-speed systems is already commercially available. In contrast to all other methods, however, each channel occupies its own spatial coordinate, and all other channels cannot be transmitted simultaneously on the same fiber. In other words, we are not more fully utilizing the high bandwidth of the fiber, but we are creating a high-bandwidth space-switching matrix, with the result that a high overall switching capacity can be realized.

Wavelength Division Multiplexing

Until the late 1980s, optical fiber communications was mainly confined to transmitting a single optical channel. Because fiber attenuation was involved, this channel required periodic regeneration, which included detection, electronic processing, and optical retransmission. Such regeneration causes a high-speed optoelectronic bottleneck and can handle only a single wavelength. After the new generation amplifiers were developed, it enabled us to accomplish high-speed repeaterless single-channel transmission. We can think of single ~Gbps channel as a single high-speed lane in a highway in which the cars are packets of optical data and the highway is the optical fiber. However, the ~25 THz optical fiber can accommodate much more bandwidth than the traffic from a single lane. To increase the system capacity we can transmit several different independent wavelengths simultaneously down a fiber to fully utilize this enormous fiber bandwidth. Therefore, the intent was to develop a multiple-lane highway, with each lane representing data traveling on a different wavelength. Thus, a WDM system enables the fiber to carry more throughput. By using wavelength-selective devices, independent signal routing also can be accomplished. The highway principle is illustrated in Figure 6.

It is expected that WDM will be one of the methods of choice for future ultra-high bandwidth multichannel systems. Of course, this could be changed as the technology evolves.

Basic Operation

As explained before, WDM enables the utilization of a significant portion of the available fiber bandwidth by allowing many independent signals to be transmitted simultaneously on one fiber, with each signal located at a different wavelength. Routing and detection of these signals can be accomplished independently, with the wavelength determining the communication path by acting as the signature address of the origin, destination or routing. Components are therefore required that are wavelength selective, allowing for the transmission, recovery, or routing of specific wavelengths.

In a simple WDM system ( Figure 7 ), each laser must emit light at a different wavelength, with all the lasers� light multiplexed together onto a single optical fiber. After being transmitted through a high-bandwidth optical fiber, the combined optical signals must be demultiplexed at the receiving end by distributing the total optical power to each output port and then requiring that each receiver selectively recover only one wavelength by using a tunable optical filter. Each laser is modulated at a given speed, and the total aggregate capacity being transmitted along the high-bandwidth fiber is the sum total of the bit rates of the individual lasers. An example of the system capacity enhancement is the situation in which ten 2.5-Gbps signals can be transmitted on one fiber, producing a system capacity of 25 Gbps. This wavelength-parallelism circumvents the problem of typical optoelectronic devices, which do not have bandwidths exceeding a few gigahertz unless they are exotic and expensive. The speed requirements for the individual optoelectronic components are, therefore, relaxed, even though a significant amount of total fiber bandwidth is still being utilized.

The concept of wavelength demultiplexing using an optical filter is illustrated in Figure 8. In the figure, four channels are input to an optical filter that has a nonideal transmission filtering function. The filter transmission peak is centered over the desired channel, in this case, l₃, thereby transmitting that channel and blocking all other channels. Because of the nonideal filter transmission function, some optical energy of the neighboring channels leaks through the filter, causing interchannel, interwavelength cross-talk. This cross-talk has the effect of reducing the selected signal�s contrast ratio and can be minimized by increasing the spectral separation between channels. Although there is no set definition, a nonstandardized convention exists for defining optical WDM as encompassing a system for which the channel spacing is approximately 10 nm.

Topologies and Architectures

Let us consider a simple point-to-point WDM system ( Figure 9(a) ) in which several channels are multiplexed at one node, the combined signals are transmitted across some distance of fiber, and the channels are demultiplexed at a destination node. This facilitates high-bandwidth fiber transmission. Additionally, high-bandwidth routing can be facilitated through a multiuser network ( Figure 9(b) ). The wavelength becomes the signature address for either path through an optical network. Because nodes will want to communicate with each other, either the transmitters or the receivers must be wavelength tunable to facilitate the proper link set-up (in this example, the transmitters were chosen to be tunable).

Two common network topologies can use WDM, namely, the star and the ring networks ( Figure 10 ). Each node in the star has a transmiter and a receiver, with the transmitter connected to one of the central passive star�s inputs and the receiver connected to one of the star�s outputs. WDM networks can also be of the ring variety. Rings are popular because so many electrical networks use this topology and because rings are easy to implement for any network geographical configuration. In this example, each node in the unidirectional ring can transmit on a specific signature wavelength, and each node can recover any other node�s wavelength signal by means of a wavelength-tunable receiver.

In both the star and the ring scenarios, each node has a signature wavelength, and any two nodes can communicate with each other by transmitting on that wavelength. This implies that we require N wavelengths to connect N nodes. The obvious advantage is that data transfer occurs with an uninterrupted optical path between the origin and the destination, known as a single-hop network. The optical data start at the originating node and reach the destination node without stopping at any other intermediate node. A disadvantage of a single-hop WDM network is that the network and all its components must accommodate N wavelengths, which may be difficult (or impossible) to achieve in a large network. Current fabrication technology cannot provide and transmission capability cannot accommodate 1,000 distinct wavelengths for a 1,000-user network.

An alternative to requiring N wavelengths to accommodate N nodes is to have a multihop network, in which two nodes can communicate with each other by sending through a third node, with many such intermediate hops possible. A dual-bus multihop eight-node WDM network is shown in Figure 11 for which each node can transmit on two wavelengths and receive on two other wavelengths. The logical connectivity is also shown. As an example, if node 1 wants to communicate with node 5, it transmits on wavelength l₁ and only a single hop is required. However, if node 1 wants to communicate with node 2, it first must transmit to node 5, which then transmits to node 2, incurring two hops. Any extra hops are deleterious in that they:

1)     Increase the transmit time between two communicating nodes, since a hop typically requires some form of detection and retransmission

2)     Decrease the throughput, since a relaying node can transmit its own data while it is in the process of relaying another node�s data

However, a multihop networks do reduce the required number of wavelengths and the wavelength tunability range of the components.

 

Wavelength Shifting and Wavelength Reuse

In an ideal WDM network, each user would have its own unique signature wavelength. Routing in such a network would be straightforward. This situation may be possible in a small network, but it is unlikely in a large network whose number of users is larger than the number of provided wavelengths. In fact, technologies that can provide and cope with 20 distinct wavelengths are the state of the art. There are some technological limitations in providing a large number of wavelengths, for instance: due to channel-broadening effects and non-ideal optical filtering, channels must have minimum wavelength spacing. Wavelength range, accuracy, and stability are extremely difficult to control.

Therefore, it is quite possible that a given network may have more users than available wavelengths, which will necessitate the reuse of a given set of wavelengths at different points in the network.

Passive Wavelength Routing

In case we have a limited number of available wavelengths, a network can use passive routing of a signal through the network based only on its wavelength. The routing is designed to reuse wavelengths in non-shared links. For example, we can see in Figure 12 that user I can use wavelength l₁ to establish a link with user II, while simultaneously user V can reuse the same wavelength, l₁, to establish a connection with user III. This functionality is accomplished by the proper arrangement of the cross-connects that route an input signal to a wavelength-determined output. A simple example of the operation of a passive WDM cross-connect is shown in Figure 13. The cross-connect is composed of wavelength demultiplexers for the input stage, wavelength multiplexers for the output stage, and fibers interconnecting the two stages. In the example, although there are only two wavelengths, there are four possible non-interfering routing paths based on both wavelength and origin. In general, instead of N wavelengths and N possible connection paths, now there are N wavelengths and N² connections. The same wavelength could be reused by any of the input ports to access a completely different output port and establish an additional connection. This technique increases the capacity of a WDM network.

Active Wavelength Shifting

In contrast to passive routing, which is limited to a static network conditions, active wavelength shifting is dynamically deals with changes of the network condition. It does that by changing the routing depending on the available links and wavelengths. This concept of a network requiring active wavelength shifting is illustrated in Figure 14. In the figure there are two small LANs connected to a larger WAN, and each LAN can transmit on only two available wavelengths (l_a and l_b). Node I wishes to communicate with node II. When node I wishes to transmit, the only wavelength available is l_a. However, when the signal reaches the right LAN, it is revealed that l_a is already being used by the right LAN. Therefore, the only way for the signal to reach node II is to be actively switched onto the available l_b.

Another scenario that would require active wavelength switching is where one set of wavelengths are used exclusively by each LAN, whereas another set of wavelength is used exclusively for communication between LANs. The wavelengths that are used in a LAN can be reused by each LAN since it will not interfere with another LAN. This situation is demonstrated in Figure 15.

Shifting one wavelength to another wavelength complexes the network functionality. One method to perform the active wavelength switching is to employ optoelectronic wavelength shifters. This method necessitates optoelectronic conversions and will cause an eventual optoelectronic speed bottleneck. In order to overcome this problem the final goal is to achieve all-optical active wavelength shifting to retain a high speed data path. All-optical means that all the shifters are purely optical, i.e. not using optoelectronic conversion of the optical data. There are several methods for all-optical wavelength shifting. Each method has its advantages and disadvantages, and it is not clear if any method will eventually be implemented. There is room for more research in this area.

Switching

We know that networks establish communication links based on either circuit or packet switching. For high-speed optical transmission, packet switching holds the promise for more efficient data transfer.

Network packet switching can be accomplished in a straightforward manner by requiring a node to optoelectronically detect and transmit each and every incoming optical data packet. As for the routing, all the switching functions can occur in the electrical domain prior to optical retransmission of the signal. Unfortunately, this approach suffers from an optoelectronic speed bottleneck. Alternatively, much research is focused toward maintaining an all-optical data path and performing the switching functions all optically with only some electronic control of the optical components. However, there are many difficulties with optical switching, for instance:

1)     A redirection of an optical path is not easy since photons do not have as strong interaction with their environment as electrons do.  

2)     Switching has to be extremely fast due to the high speed of the incoming signal.

3)     Switching nodes cannot easily tap a signal and acquire information about the channel.

Contention Resolution

Consider a situation in which two or more input ports request a communications path with the same output port, known as output-port contention. Since we are dealing with a high-speed system, a rapid contention resolution is required, in which one signal is allowed to reach its destination while the other signal is delayed or rerouted in some fashion. In our multiplexing scheme, the issue of contention exists when signals from two different input ports would request routing to the same output port and contain identical wavelengths.

Several approaches exist for resolving contention. One of them is buffering:

The packet is retained locally at the switching node and then it is switched to the appropriate output port when that port is available. The local buffering can be implemented either in electrical or optical form. Electronic buffering is straightforward but requires undesirable optoelectronic conversions and may require very large buffers. On the other hand, optical buffering is difficult because many buffering schemes require updating a priority bit (it is difficult to change a priority bit of an optical data stream), and optical memory is not an advanced art, consisting mostly of using an optical delay line.

Synchronization

A high-speed network transmitting digital signals must have adequate time synchronization to recover the data stream. Time synchronization is especially required with packet switching, asynchronous packet arrival times, and long-distance transmission.

In a WDM network, it is also possible that wavelength synchronization will be required in addition to time synchronization. In such a scenario, a wavelength standard could be broadcast through the network. However, the hope is that the network wavelength stability and accuracy will be robust and will not require its own system overhead and complexity.

Data-Format Conversion

In a large network, it is quite possible that a combination of data formats will be used. This may occur, for instance, if some links may more efficiently use TDM signaling, whereas other links may more effectively use WDM. This explains the need for data-format conversion at network gateways, as illustrated in Figure 16.

Protocols

A standardized network protocol must be used to ensure that data packets are all formatted with recognizable routing information so that the packet can be switched through the network with full global compatibility. There are two standards that show the most promise of full adoption for a global optical network:

1)     SONET - Synchronous Optical Network.

2)     ATM - Asynchronous Transfer Mode.

These two standards can be combined in one network as follows:

Data and header information are bounced into small ATM packets. These packets arrive at a switching node at random times and are grouped together into a large SONET frame ( Figure 17 ), which makes its way in predetermined synchronous time slots through the network. The ATM packets are unloaded by the SONET frame when its direction is switched through the network and it can be placed into a different SONET frame. We can think of the ATM packets as people randomly boarding a time-scheduled SONET train.

Table: Record-setting WDM transmission experiments

Conclusion

Twenty years ago, who could have predicted the success of personal computers, much less the growth of the Internet and the Web? The next 20 years are likely to bring surprises, too. If we recognize this and install robust communications plant, we will be prepared for them.

References

[1]       "Fiber-Optic Communication Systems - Second Edition"
                   Written by Govind P. Agrawal

[2]       "Photonic Networks - Advances in Optical Communications"
                   Written by Giancarlo Prati (Ed.)

[3]       "Optical Fiber Communication Systems"
                   Written by Leonid Kazovsky, Sergio Benedetto & Alan Wilner
                   Permission was granted to use figures from this book