**New engineering rules focus on the underlying principles of Raman gain generation, enabling the two main levers in Raman amplification—the installed fiber plant and the Raman amplifier module—to enhance current 10-Gbit/s and next-generation 40-Gbit/s networks.**

**James Passalugo**

Early in development, it was clear that the system noise figure is a big limitation for 40-Gbit/s networks. These networks are incapable of transmitting data across spans of more than approximately 40 km before needing signal regeneration. To make them economical, it became imperative that standard 80- to 100-km hut spacing be maintained. As a result, Raman amplification—which provides a dramatic improvement in the noise figure that enables a signal to travel further along a network before regeneration is required—has become an integral part of 40-Gbit/s network transmission.

The physics behind Raman have been described extensively. While the principles have been known for more than 70 years (discovered by Sir Chandrasekhara Raman in 1928), the ability to implement Raman into telecommunication systems is just now becoming possible. This capability is due to the recent development of high power "14xx" nm (1400 to 1510 nm) laser-diode pumps. With powers exceeding 250 mW becoming commonplace and higher powers on the way, the laser sources needed for Raman are now available.

As Raman implementation progresses, system designers require a solid understanding of design parameters for Raman amplifiers. Parameters such as gain, ripple (gain flatness), and noise must be addressed to optimize system performance. In current systems, for example, all gain takes place inside a single "black box" (the erbium-doped fiber amplifier; EDFA). Because of the distributed nature of Raman, gain occurs in the transmission fiber itself. As a result, designers must analyze the interrelated impacts both fiber and module have on key system parameters (see Table 1).

It should be emphasized that Raman amplification is not limited to 40-Gbit/s transmission rates. Enhanced system margin can be applied wherever it is needed: in a single high-loss span; to improve noise accumulation with the added loss of wavelength switching and routing elements; or to extend span/reach distances. In fact, we are now seeing extensive development and initial deployments of ultralong-haul Raman + EDFA systems. Higher spectral efficiencies also can be attained with Raman, because lower launch powers are possible. Consequently, more channels can be launched without causing additional nonlinearities.

**GAIN/ PUMP POWER REQUIREMENTS**

Raman is a nonlinear process that converts higher-energy (lower-wavelength) pump photons into lower-energy (higher-wavelength) signal photons. Corning's normalized nominal large-effective-area-fiber (LEAF) gain spectrum will be used here as a reference (see Fig. 1).

With a clearly defined gain profile, it is possible to determine the pump power required to achieve a given gain level. Through extensive simulations and experimental verification, it has been determined that 500 mW of pump power provides 10 dB of average Raman gain on Corning LEAF fiber.

Once the pump power for a gain level is defined, the pump power required for a secondary gain level in the same fiber can be determined. This relationship is a simple ratio, defined as the ratio of the pump powers (in milliwatts) proportional to the ratio of the gains (in decibels). In other terms,

(P_{Pump} and P_{Pump,o} are in milliwatts, G and Go are in decibels) Where P_{Pump,o} and Go are the pump and gain at the reference level.

Scaling pump powers to go to other fiber types is straightforward, but requires fiber-specific information, including the fiber's effective area at the pump wavelength (A_{eff}, pump) and the ratio of the Raman gain coefficient (gR) to the effective area (A_{eff}). A calculated effective length, L_{eff}, should also be considered. With this information, a fiber figure of merit (FOM) can be determined (see Table 2). The FOM is a convenient way to scale between different fiber types, and is simply the ratio of L_{eff} to gR/A_{eff}.

With the FOM, the required pump power (for a constant gain) in a different fiber type can be determined:

(*Pump* is in milliwatts; Gain is in decibels.)

Using the above relationships, one can also determine the gains (for a constant pump power) when changing fiber types.

**DESIGN RULES AFFECT PERFORMANCE**

To this point, several assumptions have been made to determine the gain for a given pump power (or the required pump power for a desired gain). These assumptions, and some design rules that show the impact of a certain parameter on the resulting performance of the given specification, are as follows.

Fiber attenuation/distributed losses. Nominally, uncabled fiber is assumed to be 0.20 dB/km at 1550 nm. As this is uncabled fiber, one may want to quantify the impact of an additional loss (say, 0.05 dB/km.) This addition could be attributed to cabling or a distributed installation loss, which can be accounted for by scaling the pump power as

Loss_{o} is typically 0.25 dB/km and Loss can be higher or lower than Loss_{o}, where Loss and Loss_{o} are measured at the pump wavelength.

Budgeting of installation losses. Typically, installation losses are the discrete losses between the Raman amplifier and the installed fiber plant. On average, these losses can range from 0.5 dB to 1.5 dB. To account for this, one uses the equation

(P_{pump} and P_{pump,o} are in milliwatts, and Jumper Loss is in decibels.)

The equations discussed above enable an optical module designer to take an initial pump power for a given set of conditions and modify it if the conditions change.

**RIPPLE/GAIN FLATNESS**

Ripple, or gain flatness, is another key system parameter. There are many different dependencies, advantages, and disadvantages to consider.

Understanding the fiber contribution is important since the Raman process depends on the structure of the material. Normalized gain-spectrum performances between different fibers of different effective areas are small (see Fig. 2). Although the changes are small, the variations must be accounted for because of today's stringent requirements on system performance. To understand how to control the gain ripple, one must first determine the sources of variation.

When considering the module contribution to ripple, it is important to recognize four main sources of gain ripple. These sources include the intrinsic Raman gain shape (based on the number of pumps used), pump wavelength variability, the gain-flattening filter (GFF) insertion-loss error function if a GFF is used, and the module-insertion spectral attenuation. In the installed fiber plant, there are three additional sources: gain variations (due to the fiber's Raman efficiency or FOM), fiber-type gain-shape variations, and within-fiber gain-shape variations.

To determine the impact on the system's total accumulated ripple, these contributions should be broken up into two factors—systematic and random. Systematic ripple (sometimes referred to as deterministic ripple) comes from a constant gain-shape error for all modules. As a result, the ripple will grow linearly with the number of spans, as the maximums and minimums line up, adding together. Random (or statistical) ripple has a varying gain shape, which lies around a probability density function. This function generally grows as the square root of the number of spans.

When designing the Raman pump module, the designer is attempting to balance the systematic and random ripple contributions with cost. With a large number of wavelengths, very low ripple values can be achieved. With the added wavelengths comes added cost, as each wavelength requires two orthogonally polarized pumps to maintain depolarization.

Another method for flattening the gain, common in EDFA modules, is a GFF. By adding a dielectric (or fiber Bragg grating) GFF, the module's ripple can be reduced significantly, while keeping pump counts to a minimum. The designer must note that when deciding to use a GFF, several degrees of freedom are being traded off for the lower pump count, such as gain flexibility (as the designer must choose to design around a nominal gain level).

Using a GFF in Raman is truly a designer's choice: the desire to lower the nominal gain ripple at a fixed gain level with a GFF versus adding extra pump wavelengths to achieve the same effect. The use of a GFF in a Raman amplifier is a complex decision. Other aspects, such as footprint and heat-dissipation issues must also be carefully considered.

**NOISE FIGURE**

Another challenge to implementing Raman amplification in system design is the noise figure of a Raman amplifier. Because Raman is a distributed process, the effective noise is less in a distributed Raman amplifier as compared with a discrete EDFA. This lower noise is a result of the amplified spontaneous emission (ASE) being attenuated over a distance, instead of being built up at a single point (and not attenuated).

To compare the performance of a distributed amplifier to a discrete amplifier, an equivalent (lumped) noise figure, FR, for the distributed Raman amplifier can be defined as

Where rASE is the ASE density at the end of the fiber, GR is the on/off Raman gain, h is Planck's constant, and u is the signal's frequency.

As defined here, one could achieve a noise figure less than 3 dB, or even have it be negative. Using this equation, expected Raman noise figures for 100 km of Corning LEAF fiber can be determined (see Fig. 3). It is important to note an effect in the noise performance. As shown, when the Raman gain increases, the noise performance improves.

There is a limit to this noise improvement due to an effect called double Rayleigh backscatter (DRBS). DRBS is the result of Rayleigh scattered light being captured and guided by the fiber. Rescattering of this light back in the original direction allows it to interfere with the signal, giving in-band crosstalk. Considered a noise penalty due to the distributed nature of the scattered light, DRBS starts to dominate at higher gain levels because the scattered light passes through the gain medium twice, while the signal only passes through once. DRBS therefore limits the amount of Raman gain because of the noise penalty it induces.

Armed with this knowledge, it is apparent that the noise figure in Raman-assisted systems is driven by two factors: changing the fiber's pump attenuation and changing the fiber's signal attenuation.

Looking first at pump attenuation, the fiber has an effective length, which is driven mainly by the fiber's pump attenuation. The lower the attenuation, the longer the effective length, and the pump light can penetrate further into the fiber. The accumulated ASE from the gain over a greater distance can be attenuated, resulting in a lower noise figure.

Next, looking at the signal attenuation, it is clear that Raman's benefit lies in the fact that the gain is distributed, enabling the attenuation of ASE over a distance. By increasing the fiber's signal-band (and therefore noise-band) attenuation, one would reduce the amount of ASE even further, thus lowering the noise figure. Higher signal attenuation, however, results in the need for more gain, so there is a limit to this benefit.

The noise figure in Raman amplification is the most challenging parameter to quantify because it is dependent upon multiple interrelated variables. This has been a qualitative analysis, so no equations have been attached. It is clear that optical module and system designers must properly manage pump and signal attenuations. The balance of these attenuations is key to the overall signal-to-noise performance of the system.

**James Passalugo** is a senior market development engineer in Photonic Technologies at Corning Incorporated, One Riverfront Plaza, Corning, NY 14831. He can be reached at passalugjr@corning.com.