IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 15, NO. 10, OCTOBER 2003
1351
EDFA Gain Transients: Experimental Demonstration of a Low Cost Electronic Control A. Bianciotto, A. Carena, Member, IEEE, V. Ferrero, Member, IEEE, and R. Gaudino, Member, IEEE
Abstract—EDFA gain transients are dynamic performance-degrading effects that need to be effectively suppressed in next-generation reconfigurable optical networks. In this paper, we experimentally assess the feasibility of an electronic dynamic gain control based on low-cost optoelectronic components, showing that a linearized electrical circuit and a control bandwidth smaller than 1 MHz can achieve the same performance as the commonly proposed all-optical feedback solutions. Index Terms—Gain transient, optical amplifier, optical communication.
I. INTRODUCTION
A
S shown in several previous works ([1], [2]), erbium-doped fiber amplifiers (EDFAs) show fast gain fluctuations after any abrupt change in input power. A sudden change in the input power leads to gain variations on surviving channels, resulting in possible out of service of some of them. Such abrupt power changes (in the ms range) are expected to become commonplace in next-generation dynamically reconfigurable wavelength-division multiplexing (WDM) networks, and in future optical packet switched networks. Previous studies showed that fast power changes might result in optical signal-to-noise ratio changes and may trigger nonlinear effects in fiber propagation resulting in significant bit error rate (BER) or Q-factor degradation [3], [4]. Moreover, power bursts can also cause damage to the receiver. Several strategies to counteract EDFA gain transients have been proposed [5]–[10], such as the following: 1) electrical feed-forward control with optical-to-electrical-to-optical (OEO) conversion [5], [6] (see Fig. 1), a setup that will be indicated in the following as the electrical control (EC) approach; 2) all-optical feedback control, based on principles of ring lasers [7], [8] (see Fig. 1), indicated in the following as the optical control (OC) approach. Other proposed strategies, such as [9] and [10], try to mix both OC and EC, and can be considered as minor modifications of these two categories. In this paper we focus on EC, and we investigate a solution aiming at reducing its cost by: 1) using a linearized feed-forward control; 2) finding its minimal speed requirements (i.e., its bandwidth).
Manuscript received January 2, 2003; revised June 26, 2003. The authors are with the Dipartimento di Elettronica, Politecnico di Torino, Corso Duca degli Abruzzi, 24-10129 Torino, Italy (e-mail:
[email protected]). Digital Object Identifier 10.1109/LPT.2003.818267
Fig. 1. Experimental setup. Two different EDFA gain locking approaches are shown: EC and OC.
We experimentally assess the performance of the proposed EC system, and we compare it to a standard OC system. We demonstrated that our linearized EC control performs similarly to OC. Moreover, the required EC bandwidth can be smaller than 1 MHz, allowing an implementation based on low-cost electronics. II. EXPERIMENTAL SETUP We considered a scenario of a typical WDM reconfigurable wavelengths, which requires EDFA’s network using high-output power and high gain to compensate for the losses due to fiber and node passive elements, such as microelectromechanical machine. The EDFA gain transient behavior is studied under dynamic add–drop of channels. The experimental setup is shown in Fig. 1. We studied a worst case scenario, channels being dropped, and corresponding to leaving only one surviving channel at the input of the EDFA. In the experiment we used five channels, selected in the ITU 1546.52 nm, grid, spaced 200 GHz apart ( 1.6 nm): 1548.11 nm, 1551.32 nm, 1552.93 nm, 1549.72 nm. Channels , and were and periodically added and dropped using an acoustooptical switch (with a switching time of 120 ns), while the channel at was always on, being used for surviving channel gain measureat the input ments. We indicated the power of channel . The number of used lasers was five of the amplifier as due only to limited laboratory inventory. However, in order to emphasize the magnitude of gain transient effect, we set the , so that the EDFA power of the other channels to 4 , total input power, when all channels were on, was 17
1041-1135/03$17.00 © 2003 IEEE
1352
IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 15, NO. 10, OCTOBER 2003
Fig. 2. Gain fluctuations in the nominal configuration for an add–drop event, at the output of first amplifier.
allowing us to mimic the transient events corresponding to the add–drop of 16 channels over 17. In our experimental setup, shown in the upper part of Fig. 1, only the first EDFA was actually gain controlled but the locking signal was not extracted at its output, so that it also acted on the second optical amplifier. In this way, we simulate a situation in which only the first element of a linear chain of EDFAs is actively gain controlled. We used commercial EDFAs (Ditech STAR-L16, an in-line type EDFA with maximum output power of 16 dBm). Our target was to demonstrate an EDFA-based constant-gain optical module. Consequently, EDFA pump power has been set to achieve a gain equal to 23 dB, on both EDFAs and with all channels on. The reference input power per channel was set to 23.7 dBm, corresponding to 11.4 dBm total EDFA input power. In the following, this condition will be referred as the nominal condition. In the upper part of Fig. 2, we show the gain fluctuation on the surviving channel (at the output of the first amplifier) in this condition when no control is present. The gain variation during an add–drop event was more than 5 dB, and the transient had a time constant of the order of 300 ms. For the EC technique, as shown in the left lower part of Fig. is detected by a 1, the amplifier total input power photodiode, then electronically processed in order to control the output power of a laser that is added to the signal at the amplifier input. This laser acts as an active gain locker, since it allows to dynamically control the EDFA saturation/inversion and wavelength level. The locking laser has power 1535.52 nm, close to the EDFA gain peak. Gain . We locking is obtained by a proper dynamic setting of controlled its power through a simple linear law, imposing , where is the locking laser power in absence of signal at the EDFA input. We observed, by numerical simulation, that the optimal feed-forward is only slightly nonlinear; relation thus, we linearized it. We verified that the linear approximation gives negligible errors, while leading to a great simplification in the practical implementation of the control circuit. The control key issue becomes the proper setting of the and parameters. These values were experimentally set by means of a calibration procedure. We verified by simulation and then by experiment that this calibration can be obtained by simple measurements under steady-state conditions, i.e., when channels are not added/dropped dynamically. Since we assume a linear relation, only two measurements are needed in order to obtain the two parameters. In particular, we developed an
Fig. 3. Gain fluctuations of controlled EDFAs. Continuous line refers to measurement after first EDFA, dotted line to measurement after second one.
experimental routine that requires a first measurement when all channels are on, and a second one when only the single surviving channel is on. The EC circuit in our experiment was implemented using standard operational amplifiers with programmable gain and offset, in order to easily set the two calibration parameters. It should be noticed that the calibration depends on the required nominal gain, set to 23 dB in our experiment. Consequently, it is in principle possible to store (for example. in a lookup table) different calibrations for different nominal gains, thus implementing a flexible, gain-controlled, optical amplification block. For comparison, we also set up the OC system shown in Fig. 1. In the optical feedback loop, we used a commercial optical filter (3-dB bandwidth equal to 80 GHz, centered at 1536.84 nm). Attenuation of the optical loop was set in order to obtain 23-dB gain in the nominal configuration. III. RESULTS In Fig. 3, we compare the performance of the two techniques during an by measuring the maximum gain fluctuations add–drop event for different input power levels. We measured on the surviving channel at , defining it as the maximum (positive or negative) gain variation relative to the nominal gain during an add–drop event, as graphically shown in the takes upper part of Fig. 2. This definition of the gain error into account both transient and steady-state gain error during the full duration of an add–drop event. This measurement has been carried out at the output of first and second EDFA. has been measured under difThe dynamic parameter ferent conditions. In particular, the amplifier input power has been swept using a variable optical attenuator (VOA) from the nominal configuration over a 5-dB range, in order to evaluate the gain control behavior when operating at very different EDFA and paramsaturation levels. It is worth noting that the eters have been optimized only once, for the nominal condition. For both techniques, any power level below the nominal condition shows a maximum gain transient below 0.7 dB after the first EDFA and 1.0 dB at the output of the second optical amplifier. These residual fluctuations are mainly due to nonflatness
BIANCIOTTO et al.: EDFA GAIN TRANSIENTS: EXPERIMENTAL DEMONSTRATION OF A LOW COST ELECTRONIC CONTROL
Fig. 4. Gain fluctuations after the first EDFA as a function of the EC equivalent circuit -3 dB electrical bandwidth.
of the EDFA gain variation over the used wavelength range, resulting in small wavelength dependent saturation effects. In contrast, by turning off the control, we measure gain fluctuations larger than 5 dB, thus showing the effectiveness of the proposed technique (see Fig. 2 for the gain time-domain fluctuation in the nominal configuration). Moreover, these results shows that , then it if the calibration is done only once for a given to . still works properly for any power lower than Above the nominal level, both systems show a threshold-like behavior. For the EC technique, the threshold corresponds to the situation requiring zero power for the locking laser. For the OC technique, the threshold corresponds to the situation in which the EDFA inversion does not allow ring lasing, thus completely destroying the feedback control effect. Above threshold, both systems present also a static gain error (i.e., the gain is smaller than the required nominal 23-dB gain). These results show that the two techniques are very similar, . The EC technique, with minor differences in the gain error as can be seen in Fig. 2, does not show the typical ring laser fast oscillations of the OC technique [7]. Regarding the residual fluctuations (for both techniques), it must be taken into account that we are considering a somehow extreme dynamic scenario, corresponding to 16 channels being dropped out of 17. As a further assessment of the requirement on EC, we forced a bandwidth limitation on the electrical circuit using a programmable single-pole low-pass filter. In Fig. 4, we show the maximum gain fluctuation at the output of the first EDFA, during an add–drop event as a function of the circuit electrical dB bandwidth . Measurements have been taken in the 23.7 dBm. previously described nominal condition: In order to keep the fluctuations below 0.5 dB, the graph shows -dB bandwidth should be kept approximately that the circuit above 500 kHz. For less saturated EDFA, we observed that even slower time constants are acceptable. We, thus, expect that an electronic bandwidth of approximately 1 MHz is acceptable for the EC approach for all reasonable input power conditions. IV. CONCLUSION We have shown that the EC technique can be implemented using a very simple linearized input-output relation, and that
1353
the required electronic bandwidth is smaller than 1 MHz, even for deeply saturated amplifiers. Consequently, our results allow envisioning a low cost electronic implementation, either based on analog circuits, as we have done in our experiment, or on analog-to-digital and digital-to-analog conversion together with simple digital signal processing. In both cases, the cost of the electronic control will be mainly determined by the cost of the directly modulated locking laser, which should typically be a distributed feedback laser (a stable control wavelength is required), with very loose requirements on direct modulation bandwidth. In fact, a 1-MHz modulation bandwidth is usually available even in nominally continuous wave lasers, since it is commonly used for laser dithering. It is to be further noted that the EC technique can also be implemented by directly controlling the power of the EDFA pump laser, by varying the laser current injection. In this case, the EC setup does not require an additional laser and the optical coupler shown in Fig. 1, thus reducing costs and avoiding wasting a fraction of the EDFA bandwidth for the control wavelength. We assessed this approach through simulation (not shown here for space limitation), obtaining the same performance as with an extra locking laser. Finally, we observe that in this paper that we did not consider any issue related to EDFA gain tilt versus frequency and input power. We believe that a fast, wavelength-independent gain control, such as the one proposed in this paper, can be a good match with a slower wavelength-dependent gain equalization controller, such as [11]. REFERENCES [1] E. Desurvire, “Analysis of transient gain saturation and recovery in erbium-doped fiber amplifiers,” IEEE Photon. Technol. Lett., vol. 1, pp. 196–199, Aug. 1989. [2] Y. Sun, A. K. Srivastava, J. L. Zyskind, J. W. Sulhoff, C. Wolf, and R. W. Tkach, “Fast power transients in WDM optical networks with cascaded EDFAs,” Electron. Lett., vol. 33, no. 4, pp. 313–314, Feb. 1997. [3] M. I. Hayee and A. E. Willner, “Transmission penalties due to EDFA gain transients in add-drop multiplexed WDM networks,” IEEE Photon. Technol. Lett., vol. 11, pp. 889–891, July 1999. [4] W. S. Wong, H. Tsai, C. Chen, H. K. Lee, and M. Ho, “Novel timeresolved measurements of bit error-rate and optical-signal-to-noise-ratio degradations due to EDFA gain dynamics in a WDM network,” in Proc. Optical Fiber Communication Conf. 2002, pp. 516–517. [5] E. Desurvire, M. Zirngibl, H. Presby, and D. Di Giovanni, “Dynamic gain compensation in saturated erbium-doped fiber amplifiers,” IEEE Photon. Technol. Lett., vol. 3, pp. 453–455, May 1991. [6] A. K. Srivastava, J. L. Zyskind, Y. Sun, J. Ellson, G. Newsome, R. W. Tkach, A. R. Chraplyvy, J. W. Sulhoff, T. A. Strasser, C. Wolf, and J. R. Pedrazzani, “Fast-link control protection of surviving channels in multiwavelength optical networks,” IEEE Photon. Technol. Lett., vol. 9, pp. 1667–1669, Dec. 1997. [7] M. Karasek, A. Bononi, L. A. Rusch, and M. Menif, “Gain stabilization in gain clamped EDFA cascades fed by WDM burst-mode packet traffic,” J. Lightwave Technol., vol. 18, pp. 308–313, Mar. 2000. [8] M. Zirngibl, “Gain control in erbium-doped fiber amplifiers by an alloptical feedback loop,” Electron. Lett., vol. 27, no. 7, pp. 560–561, Mar. 1991. [9] Y. Liu, “Optical gain control of EDFA’s using time-dependent feedback loop,” Electron. Lett., vol. 35, no. 16, pp. 1371–1373, Aug. 1999. [10] S. Seregeyev, E. Vanin, and G. Jacobsen, “Gain-clamped dynamics in EDFA with combined electronic feed-forward-optical feedback control,” in Proc. Optical Fiber Communication Conf. 2002, Mar., pp. 518–519. [11] M. C. Parker, S. D. Walker, A. Yiptong, and R. J. Mears, “Applications of active arrayed-waveguide gratings in dynamic WDM networking and routing,” J. Lightwave Technol., vol. 18, Dec. 2000.
