Active networks

As optical networks continually strive to increase line rates, capacity, flexibility, and reach, significant changes to both active and passive components are fast becoming necessary and will soon be commercialized for wide-scale deployment. Traditional passive components such as dispersion compensation modules (DCM), gain flattening filters (GFFs), multiplexing/demultiplexing (MUX/deMUX) filters, optical add drop module (OADM) filters, and polarization-mode-dispersion (PMD) maintaining fiber will soon be offered as active modules and sub-systems. Although these changes will surely add to the overall robustness, flexibility, reliability, and performance of tomorrow's optical network, they will also add to the overall complexity of the network as well.

Each active module will operate dynamically and will therefore be controlled by software and be integrally incorporated within the system control loop to ensure proper operation of the codependent active components. For example, tunable lasers will have to be synchronized with their paired tunable OADM filter settings to allow for propagation throughout the optical link. However, tuning these wavelengths also requires subsequent changes to the gain profiles of the optical amplifiers (erbium-doped fiber amplifiers (EDFAs) or Raman amplifiers), thereby resulting in adjustments to the dynamic GFF modules as well. Since all of these active components are inherently related at the optical propagation layer, any changes in the properties of one module will have a system-wide implication. These optical codependences mean that the control system will surely increase in complexity.

Many components in today's deployed 10 Gb/s based optical networks are passive (not electrically powered). Due to the huge influx of capital into the optical networking industry over the past few years, however, many vendors emerged that now offer active versions of these once passive modules, claiming improved performance, reach, and capacity. However, most of these devices are offered in isolation, due to the specific expertise required for their development. Thus, the onus is on the system integrator to incorporate these exotic technologies into a unified control system in wihich performance and flexibility leap forward, but with an increase in system complexity. This increased complexity must be designed into the system to enable simple network deployments and reduce operation expenditures. The level of complexity of today's optical networks requires significant expertise that incurs considerable expenses which existing carriers, in the present environment of fiscal constraint, can ill afford to maintain going forward.

Although some passive devices such as dispersion compensation modules will go active, existing active modules such as lasers will increase performance and flexibility through advanced design. Each promises improvements in network performance and flexibility only if they can be successfully integrated into a control system managed by software both at the local control layer and the network control layer. There must be provisions in such a system to effectively cope with failures in a given module without affecting overall system performance. For instance, if the communications link between the primary controller and its subtending active optical modules is lost due to a fiber break, these active modules must continue to operate in a holdover mode until the communications path has returned. This is analogous to synchronous network elements operating in a local holdover timing mode until the lost system timing source is restored. Telecom companies will not and should not tolerate network availability below the defacto standard of 99.999%, which is often cited as the "five nines" of availability.

The remainder of this article discusses modules that are most likely to become active in the coming generation of commercial optical networks based on current industry trends and requirements. These emerging technologies have been demonstrated in numerous lab experiments and described in generally available technical conference papers.

Next-generation 40 Gb/s networks will implement quadrupled line rates that result in bit unit intervals that are four times narrower than 10 Gb/s-based networks. Consequently, the inverse square relationship between dispersion susceptibility and line rate means that 40 Gb/s networks are actually 16 times more susceptible to pulse spreading and distortion than 10 Gb/s networks. Existing dispersion management strategies implement passive dynamic dispersion slope compensation modules (DSCMs) that may prove inadequate for 40 Gb/s networks thus mandating the introduction of dynamic DSCMs instead. This will enable dynamic dispersion compensation for any index of refraction changes along a given fiber route over time. Fixed dispersion compensation can still be deployed with advanced active dispersion compensation modules managing the changing residual dispersion over time.

In an ideal world, an optical amplifier system would provide the same gain profile across the entire wavelength spectrum to ensure relatively identical signal-to-noise ratios (SNRs) at the receiving demultiplexer site. Deployed wavelengths with the maximum spacing between one another would experience the same gain in this theoretical yet unattainable model. Of course, the practical world deviates significantly from the ideal world, which is essentially the primary need for the field of engineering in general. EDFAs actually exhibit a tilted gain that is wavelength-dependent across the supported spectrum of wavelengths, with local fluctuations referred to as ripple.

Other non-linear factors such as polarization dependent loss and stimulated Raman scattering also contribute to increase the resulting gain disparity between wavelengths. Consequently, the optical SNR will be quite different from one wavelength to another at the individual receivers. Since the overall system performance is limited by the performance of the weakest channel, equalization processes (methods to balance the received optical SNR values) are required to optimize link performance and ensure overall network robustness. The idea is simple: steal from the rich (strongest wavelength) and feed the poor (weakest wavelength).

Passive GFFs used today compensate for the non-linear gain profile of optical amplifiers, resulting in a window known as the design flat gain where input channel powers achieve a marginally flat gain profile. Other more sophisticated mechanisms reside in the amplifier control loop that measure incoming signals and automatically compensate them such that the amplifier operates in the flat gain region even in the event of power transients (sudden loss or gain of wavelengths). Although these devices minimize gain tilt, which provides some improvements in system performance, they do not offer individual wavelength control required to maximize system performance. For smaller systems, a centralized equalization strategy in which passive filters would serve to optimize link performance is sufficient. As the demand for optical reach continues to increase, passive filters are no longer sufficientægreater individual wavelength control is required. In larger systems, a distributed equalization scheme, such as one based on dynamic GFFs, is instead required.

Figure 1. Operation of the dynamic GFF

A dynamic GFF is actually a wavelength-dependent attenuator that can be used to control the amplifier tilt and ripple (see figure 1). The power spectrum of the incoming optical signal, P_in (f), is shown where P is the power of the input signal expressed in dBm and f is the frequency. P_out(f), the desired output power spectrum, is used to calculate the attenuation profile of the dynamic GFF as a function of frequency and can be derived using the relationship:

A(f) = P_in (f) - P_out (f)

Phenomena that could create amplifier tilt and ripple, and consequently affect system performance, include component aging, non-linear effects, and changes in ambient operating conditions. A variety of technologies allow dynamic GFF designs to actively compensate for unwanted energy transfer from one wavelength to another.

MUX/deMUX filters serve two fundamental roles when deployed in optical networks. First, they multiplex (combine) individual wavelengths into one fiber for dense wavelength division multiplexing transmission and demultiplex (separate) these combined signals at the terminal sites back into independent wavelengths. Second, they filter unwanted noise that has accumulated across the optical link as the signal traverses a series of optical amplifiers. An OADM is actually a special case of these optical filters.

Despite meeting these basic requirements, purely passive filters impose limitations in networks that are actually dynamically changing over time. Their fixed wavelength assignment provides little flexibility for wavelength rerouting, which will be discussed later. In addition, the fixed wavelength assignment provides few operational efficiencies, as different modules are required to cover the entire wavelength spectrum.

Figure 2. System components of active filtering

Incorporating some intelligence into the network architecture and utilizing active filters with tunable sources greatly increases the flexibility of an optical system. By using active control mechanisms such as temperature changes to alter the refractive index and by controlling input and output power, an active MUX/DeMUX filter overcomes the drawbacks of today's passive filters. Some systems considerations include (see figure 2).

The range over which these active filters can be tuned. The channel spacing (difference in frequency between two adjacent wavelengths). The maximum number of wavelengths or capacity supported. Monitor and control points to dynamically change the input and output powers.

As optical reach requirements rise, non-linear effects become increasingly important to manage, making dynamic control fundamental. Furthermore, as carriers eventually move away from static networks to reconfigurable networks, active filters will become necessary to react to constant network changes. Active compensation of the filtering response allows tight channel spacing over time and compensates for any system perturbations that would otherwise affect system performance. To achieve these goals, the filters must provide high channel isolation, low loss, low frequency offset, low distortion, and capabilities for temperature readings and control, as well as power readings and control.

There are two primary reasons for considering tunable lasers. The first reason is to replace the inventory of many fixed wavelength transmitters with one transmitter that can tune over a given range of wavelengths. From a strictly operational perspective, this application is attractive. In this application, the tuning speed is of little importance since there is no system dependency to achieve the desired tuned wavelength quickly.

The second reason is dynamic wavelength routing, which allows you to change the path of a wavelength from its ingress point (transmitter) to its egress (receiver) point via a series of passive/active OADMs and/or switches. The latter application requires a faster response time than the former given that the rerouting of wavelengths may be the result of a protection switch. In practice, the total recovery time, including the wavelength tuning, must not exceed 50 ms in order to comply with existing Telcordia standards.

Figure 3. Wavelength routing example

Consider a typical link failure (see figure 3). Under normal operation, links ACDE transport traffic from one terminal to the other terminal. The first cross-connect aggregates traffic from link B with the traffic from link A and transports it on link C. In the event of a fiber cut on link C, traffic on links A and B would be rerouted through link F. However, this poses potential issues since the wavelengths on link B are exactly the same as those on link A. In order to respect the 50 ms switching time, the transmitters are required to re-tune their wavelengths quickly enough that each wavelength is distinct. In other words, wavelength "re-coloring" is required and achievable by using tunable lasers. The colorless MUX/DMUX filters, also dynamic in nature, accept the re-tuned wavelengths, allowing this architecture to be truly robust to system changes.

Receivers operate in their optimal region when the input power is within a certain range that is specified by their detectors. Using passive components, manual external adjustments are made using variable optical attenuators (VOAs) to ensure that the receiver operates in its optimal region. Due to real world ambient conditions, receivers drift from the optimal region as defined by the VOA setting. Furthermore, channel addition and deletion will add another layer of complexity as the total power will be distributed differently, making the original VOA settings sub optimal. Uncontrolled transient events, which result from a fiber pull or a failed transmitter, further increase complexity since these events can typically cause traffic hits on the surviving channels if the control system is not designed to properly manage them.

To overcome these dynamics, receivers must have a tracking and control algorithm built into their designs that allow them to dynamically adapt to varying line condition changes. In the absence of dynamic tracking, costly procedures must be followed to add or delete channels. The dynamic tracking ensures the receiver operates below its overload parameter (the maximum input power that the receiver can accept) and within its optimal sensitivity range (the average optical power required to ensure a certain bit error rate).

This required functionality is actually quite simple. Dynamic receivers possess control circuitry that is located at the optical input and essentially represents the first stage in the signal decoding process. The control circuitry measures the incoming optical power and adds the necessary attenuation to ensure the receiver module remains within its predefined optimal region of operation. In the event the received signal power is too high (overload condition), the signal power is attenuated appropriately. No action is taken if the signal power is within range for optimal performance.

Unfortunately, pre-1994 fiber does not meet today's rigorous optical fiber specifications due to limited manufacturing expertise at the time this burgeoning technology was introduced, as well as the ignorance of the effects of polarization mode dispersion on 10 Gb/s operation. Thus, pre-1994 fiber has became problematic for effective deployment of 10 Gb/s and is not recommended for 40 Gb/s optical links. There is also some poorly installed post-1994 fiber, specifically G.652-based fiber plants, that is also high in PMD. Active PMD compensation experiments have been demonstrated for years but rarely deployed in commercial applications due to other methods of compensating for PMD, such as derated link budgets and robust forward error correction. To better leverage the installed base of existing fiber plants, especially at 40 Gb/s line rates, PMD compensation has become a topic of discussion once again. Different methods of PMD compensation already demonstrated and documented in numerous technical papers.

Implementing these newly active optical components into the network of tomorrow will introduce significant changes into the overall system architecture, both at the hardware and software layers. A robust and redundant communications path between the central controller and each active optical module will have to be developed. Implementing a robust optical control plane architecture will ensure that changing conditions arising from inevitable component aging, constantly changing ambient conditions, and unavoidable manufacturing tolerances can be effectively managed. Efficient software control algorithms must manage the inherent interdependence between these network modules, but at a minimum cost, both from a capital and operational expense point of view.

TCP/IP is an obvious candidate to enable the inter-module communications architecture given its commercial ubiquity coupled with its built-in suite of networking protocols. For instance, each active optical module could appear to the overall control system as an available IP address and would communicate with other IP-enabled optical modules using standard networking protocols. This architecture proposal would then allow equipment vendors to adopt and then adapt existing IP-based routing protocols. Using the wealth of available IP development tools and knowledge could also speed the commercial introduction of these intelligent optical networks as well. Leveraging TCP/IP lowers the cost of implementing a more active optical control layer, due to the abundance of TCP/IP tools, knowledge, and experience available.

The shift from passive optical components to active and dynamic optical components holds great promise in terms of increased network flexibility and robustness. However, adding active module control will also increase cost as a result of the additional electrical hardware, software, and firmware requirements. Simply making an optical component active will not guarantee its success, especially if the cost is higher, and there is no additional and desirable benefit in the mind of the customer. The tight fiscal environment of the telecom industry today has forced service providers to scrutinize all new technologies for a positive return on investment. Thus, for active optical modules to achieve commercial acceptance in the telecom industry, they will first have to offer significant benefits over and above those currently offered in passive optical modules.

Further reading

1. J. Wilson and J. Hawkes, Optoelectronics, an Introduction. 3^rd Ed., Prentice Hall, (1998).

2. Govind P. Agrawal, Fiber-Optic Communication Systems. 2^nd Ed., John Wiley and Sons Inc. (1997).

3. A. Girard et al., EXFO Electro-Optical Engineering Inc., Guide to WDM. 2^nd Ed. (2000).

4. R. Ramaswami and K. Sivarajan; Optical Networks, A Practical Perspective; Morgan Kaufmann, (1998).

5. Y. Chen, C. Visone, et al., "Role of the Dynamic Gain Equaliser as a Network Equaliser," Global Optical Communications 2002.