VINCENT MORIN and EDOUARD TAUFFLIEB
The advent of wavelength-division multiplexing (WDM) as the best technology to upgrade the capacity of existing fiberoptic networks has pushed research on high-performance optical amplifiers. Erbium-doped fiber amplifiers (EDFAs)—which previously revolutionized the design and performance of optical fiber networks—are compatible with WDM systems that transmit up to eight channels. But the most recent commercially available dense-WDM systems are designed for 16, 32, or 40 channels, and systems with more than 100 channels are in the offing.
Standard EDFAs are highly efficient thanks to their high gain, high output power, low noise, wide bandwidth, and fiber compatibility; however, some of their characteristics are not sufficient for DWDM applications. Still-wider bandwidth and higher output power, as well as flatter gain, are now required for this new generation of EDFAs.
The key advantage of EDFAs compared to other amplification techniques is their transparency to both wavelength and bit rate. However, transmission at very high bit rates (greater than or equal to 10 Gbit/s) is quickly limited by both chromatic and polarization dispersion; consequently, most WDM systems transmit at 2.5 Gbit/s.
The only way to increase the aggregate bit rate while retaining the bit rate per channel is to increase the number of channels. Because the different channels are usually separated by a constant channel spacing (typically 100 GHz, corresponding to 0.8 nm, although narrower channel spacings are under development), the overall bandwidth is 25 nm for a 32-channel WDM system.The key difficulty in using a conventional EDFA in such systems is that the gain spectrum on an EDFA is not constant over the bandwidth (see Fig. 1). The nonuniformity of gain in an EDFA—which has little impact on single-channel transmission—becomes a strong limitation with multiple channels. The accumulation of gain discrepancies results in a level discrepancy and a signal-to-noise ratio (SNR) discrepancy, with a significant bit error rate (BER) penalty for the worst channels.
Another limitation to the easy integration of EDFAs in WDM transmission lines is the gain variation with total input power: in those systems, optical power per channel is usually constant at the input of an EDFA, but the overall number of channels propagating through the amplifier may vary. In that case, adding or dropping some channels will have a strong impact on the gain available for the other channels.
The last parameter that must be carefully studied when designing these amplifiers is output power. The most powerful amplifiers currently marketed have output power in excess of +20 dBm.
The typical high-gain bandwidth of a standard EDFA is limited to the 1530-1560-nm window, thus limiting the number of available channels to 40. In this window, the gain spectrum can be represented by a two-bump curve, with a high (10 dB) and narrow (<10 nm) bump around 1530 nm and a wider (20 nm), shorter (3 to 5 dB) one around 1550 nm. The particular gain spectrum depends on several parameters, such as pump wavelength (980 or 1480 nm), fiber length, and composition, but the general shape and the solution used for gain-flattening are the same. This solution provides EDFAs with flat gain over a 15-20-nm bandwidth (1540-1560 nm), which is enough for WDM systems with up to 16 channels.
Another approach to obtain the necessary gain flatness is to use doped fibers with completely different host material, such as fluoride-based fibers. This solution is efficient but complex, and such fibers are much harder to handle and to splice than silica-based fibers. However, the gain spectra obtained with those approaches are not wide enough or flat enough for advanced DWDM systems.
If the WDM system is designed for more than 16 channels, then the gain spectra have to be flattened by external methods. The overall principle is to use gain-flattening filters. Basically, these filters are designed to approximate the inverse profile of the gain spectrum. Recently, many solutions have been successfully demonstrated with acousto-optic tunable filters (AOTFs), thin-film layers, Mach-Zehnder fiber filters, and fiber Bragg gratings—in particular, long-period fiber gratings (LPGs). Each solution has its advantages and drawbacks, and the two most promising candidates appear be AOTFs and LPGs.
Photonetics (Marly-le-Roi, France) has chosen the LPG approach in developing its gain-flattened amplifier. These gratings are passive components (thus stable and reliable) of an all-fiber construction with insignificant reflection, very low excess loss, and low polarization dependence, and they can be designed to match the entire 30-nm flat-gain bandwidth requirement. Moreover, they are potentially low-cost components and are easy to manufacture in large quantities. However, their temperature dependence (up to 100 pm/C, depending on grating construction) most often requires a thermal control.In practice, the filter required for the 30-nm flat-gain window between 1530 and 1560 nm is synthesized by cascading several LPGs (see Fig. 2).
A gain-equalizing filter is designed with the exact inverse spectrum of the gain spectrum of the amplifier for a particular input power. But the gain spectrum of EDFAs also depends on the input power. The question becomes crucial in advanced DWDM systems, in which adding or dropping channels should not affect the amplification of all other channels.
For instance, in a DWDM network using 32 channels with equal input powers, the difference in total input power when 1 or 32 channels are transmitted is 15 dB, which means that the gain of the EDFA must be the same at any wavelength and for any input power over a minimum 15-dB input power range. In that case, the multichannel system would be fully reconfigurable without having to precisely set the input conditions for each EDFA.
This gain-locking function can be carried out with optical or electrical control. Photonetics has chosen a simple, all-optical gain-control system in which fiber Bragg gratings are placed at both ends of the doped fiber. These two mirrors create a cavity that generates a laser oscillation at the reflection wavelength of the Bragg gratings. This wavelength was selected to avoid any interference between signal transmission and laser oscillation.
In this amplifier, lasing is produced at 1515 nm. The stimulated emission locks the average population inversion of erbium ions. Because erbium has a homogeneous broadening, the saturation (fixed by the lasing effect) locks the gain at any other wavelength. As a consequence, the gain does not depend on the signal input power or the pump power. All channels are amplified by the same gain, independently of total input power, number of channels, or channel allocation.
The overall configuration of the fixed-and-flat gain EDFA is a double-forward pumping scheme, with two pumps emitting 120 mW at 980 nm (see figure at top of this page). The gain-flattening filter is placed within the amplifying medium, but neither at the beginning of the fiber or at the end. This is a compromise between noise figure, which is strongly degraded if the filter is placed in front of the amplifying medium, and output power, which is limited if the filter is placed at the end of the doped fiber.
The best place for the gain equalizer is roughly at one third of the doped fiber, so that the signal is first preamplified to create a sufficient signal level, then equalized and finally power-amplified by the amplifier stage. Furthermore, gain locking provides efficient gain flattening, irrespective of signal input power.
WDM performanceAn amplifier of this design can obtain 22-dB gain in the gain-lock regime (see Fig. 3). The gain flatness is ۪.7 dB over a 30-nm bandwidth (1530-1560 nm) and for an input signal power up to -6 dBm. In this case, the maximum output power in the gain-lock regime is around 16 dBm. Moreover, the gain dependence with signal power is 0.7 dB in the gain-lock regime.
In the same condition, the noise figure has a 5.5-dB average value, which is comparable to 980-nm pumped amplifiers without gain equalizers. The degradation of the noise figure around 1530 nm is less than 0.5 dB.
A gain-locked, gain-flattened EDFA is tolerant of signal variation (numbers of channels, signal allocation, total input power), which simplifies system design, especially for networks using channel adds and drops and for straight-line applications in which amplifiers can be cascaded with no need for pre-equalization.
For future DWDM applications, next-generation amplifiers will have to transmit even more channels over an enlarged bandwidth or with a smaller channel spacing. In both cases, the key parameters that still must be enhanced are output power, bandwidth, and gain flatness. Furthermore, these amplifiers will need to incorporate gain-control systems to obtain true transparency to network reconfigurations such as channel adds and drops.
On gain-locked EDFAs:
R. Lebref et al., "Theoretical Study of the Gain Equalization of a Stabilized Gain EDFA for WDM Application," J. Lightwave Technol. 15, 766 (1997).
On gain-flattened EDFAs:
Y. Sun et al., "Transmission of 32-WDM 10 Gb/s Channels Over 640 km Using Broadband, Gain-Flattened Erbium-Doped Silica Fiber Amplifiers," IEEE Phot. Technol. Lett. 9, 1652 (1997).
Vincent Morin, formerly a research engineer at Photonetics SA, is with Ciena Ltd.; Edouard Taufflieb is a technical manager at Photonetics SA, 52, Avenue de l`Europe - B.P. 39, Marly-le-Roi, France.