Long-haul DWDM optical communication systems require broadband in-line amplification along the optical transmission path. The rapid growth of these systems also dictates that the in-line amplifiers should be transparent to the bit rate and to the communication protocols used. In the 1550-nm transmission window, erbium-doped fiber amplifiers (EDFAs) are the in-line amplifiers of choice because they meet these criteria. Without EDFAs, the evolution of DWDM systems would certainly not have progressed so rapidly. Conversely, EDFA technology would not have found as much interest without the driving force of DWDM systems.
The amplifiers have undesired impacts on signals as the transmission path lengthens. Cumulative signal-distortion phenomena, such as chromatic and polarization-mode dispersion or nonlinear effects, become a concern. The nonuniform EDFA gain spectrum must also be equalized if the transmission bandwidth is to be preserved after passing through multiple concatenated amplifiers. Finally, if DWDM systems are to be fully exploited, channel adding and dropping functions should be implemented along the optical transmission link. For this reason, the gain per channel of EDFAs must be stabilized to avoid gain fluctuation with variations of the total incident input power.
A low-cost data-conditioning EDFA that would address its own undesired effects would thus be very helpful. A judicious combination of fiber Bragg gratings (FBGs) and fiber amplifier sections is a promising candidate to perform this task.
Drawbacks to current components
The use of highly efficient power-conversion EDFAs is a good method for limiting the undesirable effects because it limits the number of fiber amplifiers to be used along the transmission link. Low noise figure, high small-signal gain, and high saturation output power are all characteristics that are required of these amplifiers, and so they must incorporate highly efficient power conversion erbium-doped fiber.
This requirement means that the background loss of the erbium-doped fiber must be decreased and, in the case of 980-nm fiber pumping, a rare-earth dopant confinement must be used within the fiber core. Ultimately, it means that the fiber manufacturer should have a proven quality-controlled erbium-doped fiber fabrication process (see Fig. 1). The doped-fiber fabrication recipe can also be adapted to enhance the DWDM gain bandwidth.
Polarization-mode dispersion (PMD) cannot be compensated for in this way. In fact, PMD-a stochastic phenomenon-is very difficult to compensate for unless a feedback-controlled polarization controller is used in front of the receiver. Such a compensating scheme has questionable efficiency. A better solution is to limit the impact of PMD on the DWDM system by using devices such as low-PMD EDFAs and low-PMD fiber made by introducing many polarization-coupling points along its path. Most of the modern fiber installed does not suffer excessively from PMD.
Chromatic dispersion can be compensated for because it is a predictable phenomenon in a given fiber and creates a cumulative optical pulsewidth broadening along the fiber. Dispersion leads to inter-symbol interference, which increases the bit error rate. Dispersion can be managed by specially designed fiber that produces zero dispersion at a specific wavelength. However, if the zero-dispersion wavelength falls within the DWDM transmission window, the nonlinear phenomenon of four-wave mixing can occur. This phenomenon can also degrade the quality of the transmission because new frequencies can interfere with DWDM channels.
The fiber can be optimized to mitigate the impacts of nonlinear effects and dispersion by placing the zero-dispersion wavelength slightly outside the DWDM transmission window. Unfortunately, these transmission windows are continuously expanding and most of the fiber already installed is standard single-mode fiber with dispersion values of approximately 17 ps/nm/km at 1550 nm, with a zero-dispersion wavelength of approximately 1310 nm.
Currently, the primary method of dispersion compensation is to insert dispersion-compensating fiber (DCF), which offers opposite dispersion compared to the transmission fiber and is inserted periodically along the communication link. This fiber is very lossy compared to standard fiber and has a low nonlinear-effect threshold.
Fiber Bragg grating alternative
A better solution is chirped fiber Bragg gratings (FBGs), which offer dispersion values higher than 1700 ps/nm-sufficient to compensate more than 100 km of standard fiber. Using a circulator in combination with such dispersion-compensating FBGs, losses are not a major concern; neither are nonlinear effects. However, unlike dispersion-compensating fiber, dispersion-compensating FBGs are not yet considered a proven technology by DWDM system manufacturers.
Conservatism explains part of this situation. Also, the results obtained with dispersion-compensating FBGs developed so far have not been that convincing. Once the FBG industry really attacks this challenge at its starting point (the fabrication process), things may quickly change. To date, the DWDM system manufacturers are mainly questioning the ripples in the delay curves of FBGs. However, it is known that if a perfectly apodized FBG can be made, no ripples would be present on the delay curve. Barring perfection, the ripples amplitude can be limited to a very small value (~10 ps). Delay-curve ripples should also be put in a proper perspective: ripples create distortion, but a certain level of distortion is acceptable as long as it does not create power penalties at the receiver.
Equalization of the EDFA gain spectrum must be implemented to preserve the complete DWDM transmission window while going through multiple concatenated in-line fiber amplifiers. There are many gain-flattening filter technologies available on the market, including acousto-optic, long- and short-period gratings, all-fiber Mach-Zehnder, and thin-film dielectric filters. The technology that will prove best should be low-cost, flexible (easy adaptation to the spectral response), made to fit as precisely as possible with the needed spectral response, and not dependent on a separate, unique technology (not dependent on a special fiber, for example). Spectrally designed short-period gratings (gain-flattening FBGs) meet all these requirements as long as they are combined either with an isolator or a circulator.1 Because EDFAs are associated with isolators, this technology could prove an optimal choice.
As for the add/drop DWDM channel functions, a mid-stage access EDFA offers this possibility. To avoid gain spectrum distortion or a variation of the obtained-gain-per-channel value when adding or dropping channels, a gain-locking scheme must be implemented. Two gain-locking possibilities can be proposed: gain clamping and residual-pump-power gain locking. Gain clamping is an automatic gain control scheme-a laser channel, slightly outside the DWDM transmission window, is created by inserting spectrally overlapping FBGs at each end of the amplification section. This laser channel uses the excess gain available when DWDM channels are dropped.2 For this method to be implemented efficiently, the reflectivity of the FBGs must be adjusted very precisely.
The second approach, which is also the most common approach, uses a comparison between the pump residual power at the output of the EDFA and the pump power launched into the amplifier fiber.3 By keeping the ratio between these two values constant using a feedback loop, one can obtain a constant gain spectrum and gain-per-channel value. It has been experimentally verified that a greater than 10-dB gain-locking dynamic range can be obtained on a two-stage EDFA (with mid-stage access) using a gain-locking feedback loop on the second stage only.
We believe that spectrally designed FBGs can be combined with a dispersion-compensating, gain-equalizing, gain-locking, mid-stage access EDFA to improve performance. First, an optical circulator is inserted at mid-stage of a two-stage EDFA. The circulator leads to a cascade of FBGs, each of which is associated with a DWDM channel. By adjusting the relative reflectivity and the delay curves associated with the FBGs, gain equalization and dispersion compensation are obtained. Moreover, the circulator prevents counterpropagating amplified spontaneous emission (ASE) from reaching the first stage of the EDFA, thus leaving more photons available for the useful signals. If a channel needs to be dropped at mid-stage, the FBG associated with the channel would either be detuned or simply not inserted in the cascade (see Fig. 2).
Of course, such a cascade of gratings can be implemented only if the DWDM channel grid is well established and each channel source is frequency-locked. If that is not the case, it is still possible to obtain a similar result with broadband FBGs that would cover multiple DWDM channels simultaneously. There is a tremendous advantage to using the one-FBG-per-channel approach: the ASE between the channels that is not reflected by the FBGs does not reach the second stage of the amplifier. This approach leaves more photons available for the useful signals and improves the overall performance of the data-conditioning EDFA.
A one-FBG-per-few-DWDM channels approach might be a good compromise to filter out some ASE and also have channel wavelength flexibility. This last approach can also be a good way of optimizing the required volume to package a cascade of gratings. Another interesting way of minimizing packaging size is to insert a few gratings one over the other in a fiber section. This would require the use of photosensitive fibers (see Fig. 3).
Extending the DWDM transmission window would certainly allow the transmission of more channels simultaneously. The L-band window has been recently opened (approximately 1560 to 1610 nm), and the S-band (shorter than 1530 nm) may be opened soon. Most L-band EDFA configurations use a combination of 980- and 1480-nm pumps along with a large quantity of erbium-doped fibers. Such configurations are costly. The data-conditioning EDFA described here could be readily modified to provide an alternative with little additional cost (see Fig. 4).
Fiber-Bragg-grating technology is well suited to improve the performance of EDFAs. This all-fiber, compact, low-cost technology with excellent spectral filtering potential usually suffers from acting as a reflective filter. Combined with an EDFA, where isolators are common, this disadvantage vanishes.
The fabrication flexibility of FBGs is ideal for EDFAs where the gain spectrum varies from one configuration or manufacturer to the other. This fabrication flexibility can be an advantage if one wants to implement unequal DWDM channel spacing, contrary to other technologies where there would be a high price to pay. FBGs can even be fabricated on-line within an EDFA to insert data-conditioning functions.
1. M. Rochette et al., IEEE Photon. Tech. Lett.11(5), 536 (May 1999).
2. M. Zirngibl, Electron. Lett.. 27(7), 560 (March 28, 1991).
3. J. C. van der Plaats, F. W. Willems, and A. M. J. Koonen, "Dynamic pump-loss controlled gain-locking system for erbium-doped fiber amplifiers in multi-wavelength networks," ECOC `97, Edinburgh, Scotland, (Sept. 1997).
Jocelyn Lauzon is director of the Photonics and Guided Wave Optics Sector, Martin Guy heads the optical communications division, and Martin Rochette is a Ph.D student at INO, 369 Franquet, Ste-Foy, Québec, Canada, G1P 4N8. For more information, contact Martin Guy at 418-657-7006 or firstname.lastname@example.org.
FIGURE 1. Compputer-controlled draw tower ensures quality of erbium-doped fiber used in power-conversion EDFAs needed in the transmission links of long-haul DWDM systems.
FIGURE 2. A circulator, inserted at mid-stage of a two-stage EDFA leads to a cascade of FBGs, each associated with a DWDM channel. Gain equalization and dispersion compensation can be obtained by adjusting the relative reflectivity and the delay curves of the FBGs.
FIGURE 3. The performance of a FBG-based data-conditioning EDFA varies if discrete or continuous FBGs are implemented at mid-stage of a two-stage EDFA. Performance is excellent for both, with a clear advantage for the discrete FBGs cascade approach.
FIGURE 4. The L-band EDFA section (top) is pumped by the signal and ASE residuals not reflected by the FBG cascade. The reflectivity of the FBGs in the cascade can potentially be adjusted in order to vary the pumping power available for the L-band stage.