Modeling optimizes EDFA design
The number of channels deployed in long-haul dense wavelength-division multiplexing (DWDM) systems is rapidly increasing beyond 100 over the C-band (1528-1563 nm) and L-band (1575-1610 nm). This demand for more bandwidth is driving the demand for more-sophisticated erbium-doped fiber amplifiers (EDFAs), which means amplifiers flattened over a very wide band for maximum bandwidth transmission over the long haul.
In this article we describe how to optimize the gain of gain-flattened amplifiers by adjusting the fiber length and gain-flattening filter shape to meet three design criteria: spectrally flat gain at the target operating point, maximum signal output power, and maximum optical signal-to-noise ratios. These design considerations, along with rules for minimizing multipath interference, are applied to split-band amplifiers designed to simultaneously amplify both C- and L-bands.
The typical optical configuration of a gain-flattened EDFA consists of two gain sections separated by a filter section. Two monitors measure the total input and output light tapped from the signal by fused taper couplers. The pump laser is coupled to the erbium fiber by a fused or dielectric pump coupler (see Fig. 1).
The noise figure of an EDFA improves (decreases) as its population inversion increases, which increases as the pump power increases. Backward pumping, while having a higher noise figure, also has higher output power because the stronger pump power at the output delays the onset of gain saturation. Therefore, forward pumping is preferred to maintain a low noise figure, and backward pumping is preferred to achieve a high output power. Hybrid pumping with 980-nm forward pumping and 1480-nm backward pumping takes advantage of the high population inversion provided by 980-nm pumps and the high power- conversion efficiency of 1480-nm pumps to achieve a low-noise, high-output-power amplifier.
The interstitial gain-flattening filter attenuates the signal gain peak at 1532 nm, and it also increases the population inversion by attenuating the backwards-amplified spontaneous emission (ASE) at the 1532-nm emission peak (see Fig. 2). The middle isolator attenuates reflections from the filter, and it also attenuates the backwards ASE, which saturates the population inversion in the first stage, thereby decreasing the gain and increasing the noise figure.
The filter stage is ideally located between the two gain stages-the filter stage reduces the backwards ASE and so actually increases the net inversion and average gain even though it has an insertion loss of 1-2 dB.1 Locating this loss at the amplifier input would degrade the noise figure by 1-2 dB, and locating this loss at the amplifier output would degrade the output power by 1-2 dB.
An input optical isolator prevents ASE signals from propagating in the backward direction. Otherwise, reflected ASE would reduce the population inversion, thereby reducing the gain and increasing the noise figure. The output isolator prevents light from output reflections from re-entering the EDFA. Together, both isolators prevent Q-switching when the external connectors are disconnected and reduce the multipath interference (MPI). An MPI power penalty occurs whenever a signal has more than one path through the network from its transmitter to receiver. This penalty is increased when the interfering path is amplified; therefore, amplified networks have to be carefully designed to strongly attenuate the secondary paths.
Lower-cost designs use a single 980-nm pump to pump both stages. The remnant pump from the first gain stage is directed around the filter stage by pump couplers because 1550-nm filters and isolators are not transparent at the pump wavelengths. High-isolation pump couplers must be used to direct the pump light around the filter stage to prevent MPI from signal light propagating through the pump path.
For WDM applications, the optimum length of each gain stage is the length that maximizes the gain of the weakest signal, or the length at which the differential gain at the output is equal to zero. The design is optimized by iterating between adjusting the erbium fiber length of the two gain sections to maximize the gain of the weakest signal and adjusting the filter shape to attenuate the stronger channels so that their gain equals the weakest channel`s gain.
For a given set of passive components, fiber type, pump powers, and signal wavelengths, a simulation can determine the gain-flattening filter shape that perfectly flattens the gain spectrum. In practice, batch and temperature variations and inaccuracies in component parameters will give a gain ripple of 1-2 dB over 35 nm.
Note that even if the filter perfectly flattens the EDFA at one operating point, or average inversion, its gain spectrum does not remain flat when the average gain changes. This gain variation, or tilt, must be accounted for in the amplified system design. The EDFA can be optimized for its maximum gain, and applications that require lower gain can pad the link loss or bias the transmitters to equalize the received power or optical signal-to-noise ratio.2 The latter method is usually preferred because it has a higher operating margin.
L-band and split-band EDFAs
L-band amplifiers based on standard erbium-doped silica fiber have recently been commercialized. These amplifiers make use of the fact that the erbium gain peak shifts to longer wavelengths as the inversion decreases.3 Indeed, the EDFA gain spectrum changes with population inversion and fiber length (see Fig. 3). These figures were calculated by adjusting the fiber length to give 20 dB maximum gain.
L-band amplifiers typically have a short, high-inversion gain section pumped by a 980-nm laser to give a good noise figure, followed by a longer, low-inversion gain section that provides enough gain at the longer wavelengths. To accurately design an L-band amplifier, care must be taken to accurately measure the weak cross sections in the gain region, to design the passive components to have their passband in the L-band, and to correct for signal excited-state absorption for wavelengths longer than approximately 1610 nm.
A C-band amplifier has very weak gain in the L-band, and an L-band amplifier absorbs signals in the C-band. Therefore, wideband EDFAs use a band-splitting filter to separate the amplifier into two independent, parallel gain sections, one covering 1530-1560 nm and the other, 1570-1610 nm (see Fig. 4). There can still be a short, shared, initial gain section with high inversion, which provides enough gain to overcome the initial passive component losses.
A dead band of approximately 10 nm between the two gain sections gives enough isolation between the two bands to minimize the MPI penalty. To increase the isolation, extra filters can be inserted after the first band-splitting filter. In this design, the C-band and L-band sections can be independently optimized for their respective bands.
With properly designed filters and fiber lengths, amplifiers based on erbium-doped silica fiber can be designed to provide gain over the C- and L-bands with only 1-2 dB of gain ripple. Nonsilica erbium-doped fibers, such as fluoride and telluride, have also been proposed as alternatives because they do not require as much filtering to achieve a flattened spectrum. These materials widen the erbium spectra because they have a lower phonon energy, which increases the lifetime of the highest metastable levels.
A drawback of the lower phonon energy is that it can preclude 980-nm laser pumping. Concerns about the volume manufacturability and reliability of these new materials have also kept the materials from supplanting the established technology. Erbium/ytterbium codoped silica fiber, however, has been commercialized for several years for applications such as CATV that require much higher output powers (>23 dBm).4
Fiber Raman amplifiers have also been recently commercialized. Although these require much higher pump power than EDFAs, hybrid designs are attractive because the Raman pump wavelength can be chosen to tailor the gain spectrum.5 Raman amplifiers can also be used as distributed amplifiers. This application can provide the same gain as EDFAs but with lower nonlinear penalties. Compared to lumped amplification, distributed amplification is also better suited for supporting soliton transmission.
1. A. Yu et al., IEEE Photon. Technol. Lett. 5, 773 (1993).
2. A. R. Chraplyvy et al., IEEE Photon. Technol. Lett. 3, 920 (1992).
3. J. F Massicott et al., Electron. Lett. 26, 1645 (1990).
4. S. G. Grubb et al., Electron. Lett. 28, 1275 (1992).
5. H. Masuda et al., Electron. Lett. 33, 753 (1997).
Chris Barnard is director of optical system architectures at ONI Systems, 166 Baypointe Pkwy., San Jose, CA, 95134; Jackson Klein is a research scientist and Velko Tzolov is director of scientific research at Optiwave Corp., 16 Concourse Gate, Ste. 100, Nepean, Ontario K2E 7S8, Canada. Contact Velko Tzolov at 613-224-4700 or email@example.com.
FIGURE 1. Typical configuration of a two-stage gain-flattened EDFA has a gain-flattening filter between the gain sections. The signal and ASE spectra at the amplifier output show the flat response (results presented here and in the other figures were calculated with EDFA_Design 2.0 from Optiwave Corp.).
FIGURE 2. Modeling of gain spectra for each stage of the amplifier shows first-stage gain (a), gain-flattening filter transmission (b), second-stage gain (c), and total gain spectrum (d).
FIGURE 3. Gain spectra vary with length of fiber.
FIGURE 4. In a typical configuration of a wideband EDFA, a band-splitting filter separates the amplifier into independent gain sections for C-band (bottom inset) and L-band (top inset).