Cross-Gain Compression in Semiconductor Optical Amplifiers

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 3, MARCH 2007

915

Cross-Gain Compression in Semiconductor Optical Amplifiers Giampiero Contestabile, Roberto Proietti, Nicola Calabretta, and Ernesto Ciaramella, Member, IEEE

Abstract—In this paper, we present a novel scheme that exploits cross-gain modulation (XGM) in semiconductor optical amplifiers (SOAs) without overall pattern effects. This technique uses two signals with reversed-intensity modulation and different wavelength to exploit propagation and gain-compression dynamics in gain-saturated SOAs at almost constant overall input power. The resulting cross-gain-compression mechanism between copropagating waves can lead to all-optical waveform reshaping. By using this technique, we experimentally demonstrate enhanced wavelength conversion by XGM and wavelength-preserving noise compression at 10 Gb/s. Index Terms—All-optical regeneration, cross-gain compression (XGC), cross-gain modulation (XGM), semiconductor optical amplifier (SOA), wavelength conversion.

I. I NTRODUCTION

T

HE CROSS-GAIN modulation (XGM) in semiconductor optical amplifiers (SOAs) has been largely investigated in the past for its possible application to all-optical signal processing [1]–[5]. Indeed, XGM in SOAs is a very simple technique that may offer polarization independence, compactness, and low-power-consumption operation. In XGM, a strong modulated-pump signal co- or counterpropagates in an SOA with a weaker continuous-wave (CW) probe. The pump modulates the SOA gain, the gain variations are experienced by the probe light, and this transfers the information encoded in the pump to the probe. However, despite the simple working principle, this technique suffers from serious drawbacks. The inversion in the logic of the transferred data is a first intrinsic weakness, but the major drawbacks are related to the carrier and gain dynamics of the SOA. Namely, pattern-dependent distortions lead to eye closure when the working bit rate is close to the gain-recovery time of the amplifier and the differential saturated gain leads to a variation of the extinction ratio (ER) of the converted signal, depending on the conversion wavelength. An ER degradation always occurs in wavelength upconversions, compromising the possibility of cascading XGM devices. Moreover, in a practical XGM converter, the best output signal is always obtained as a tradeoff between the obtainable ER and the eye distortion due to data pattern. Bandwidth enhancement of wavelength conversion via XGM has

Manuscript received January 25, 2006; revised August 12, 2006. This work was supported in part by the European Commission FP6 Program (I.P. NOBEL). The authors are with Scuola Superiore Sant’Anna, 56124 Pisa, Italy (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JLT.2006.890441

been pursued by reducing the SOA gain-recovery time using higher optical powers, higher amplifier-drive current, a longer amplifier [1], or cascading two amplifiers [6]. However, the use of high optical power and high driving current is undesirable and as the bandwidth is enhanced, pattern effects arise at quite higher bit rate. In a different approach [7], [8], the step-edge part of a fiber-Bragg-grating filter has been used to improve the frequency response of XGM converters by sharpening the transitions between bits but, unfortunately, this simple scheme can only reduce the penalties related to pattern modulations [7]. The use of assist lights [9]–[13] is an alternative approach often used to reduce pattern fluctuations in SOAs, especially when working at ultrahigh bit rates [14]. The assist light has usually been CW or, as recently proposed [15], modulated inverted with respect to the copropagating waveform. In this last case, the assist signal provides a partial temporal equalization of the power in the SOA. A similar concept is exploited to obtain a limiting amplifier for packet/burst mode receivers [16]. Moreover, XGM multistage converters [17]–[19] take advantage from a similar principle for obtaining improved wavelength-converted signals. In this paper, we study the cross effect of copropagating two inverted in-logic signals in a saturated SOA. We demonstrate that, taking care to have an almost constant power in the SOA, a cross-gain-compression (XGC) effect, related to both gain compression and propagation in saturated amplifiers, can reshape, at the same time, both traveling signals. Hence, exploiting this scheme, a single device can be realized for both wavelength conversion and wavelength-preserving signal reshaping. This paper is organized as follows. First, in Section II, we describe the XGC concept and demonstrate distortion-free propagation in a saturated SOA. In Section III, we analyze the performance of a 10-Gb/s wavelength converter based on XGM, in which the performance is enhanced by using the XGC principle. Finally, in Section IV, we report on the noise-compression capability of the XGC for application in wavelength-preserving Reshaping, Reamplification (2R) regeneration. II. XGC W ORKING P RINCIPLE The schematic of the XGC effect is reported in Fig. 1. If, starting from an input signal at λ1 , we are able to obtain a new signal at a different wavelength (λ2 ), that is, inverted in logic, it is possible to exploit the gain-saturation effects of an SOA to improve the quality of the signals, i.e., to obtain all-optical reshaping. The working principle is as follows: The two signals at different wavelengths are coupled, synchronized, and then injected in the SOA. The total power has to be high enough to

0733-8724/$25.00 © 2007 IEEE

916

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 3, MARCH 2007

Fig. 1. Schematic of the XGC effect in a saturated SOA between two signals having opposite logic and different wavelength.

Fig. 2. Scheme of the experimental setup. TL = Tunable Laser. OC = Optical Circulator. ODL = Optical Delay Line. Att. = Optical Attenuator. AWG = Array Waveguide Grating. CDR = Clock and Data Recovery.

strongly saturate the amplifier gain. At the same time, controlling the relative power of the two signals, it is set quite constant to prevent the pattern effects. In this configuration, a logical one (or a logical zero) and a logical zero (or a logical one) always propagate at the same time in the amplifier sharing the saturated SOA gain. During the propagation, saturation and dynamical effects affect the signals shape. The saturated gain acts as power equalizer on the ones amplitude. The overall effect is similar to a high-pass filter with around 1-GHz cutoff frequency [20]. At the same time, in a certain wavelength allocation, as we will see in detail in the following, the zeros experience compressed gain by the copropagating ones. Summarizing, the XGC between opposite symbols can lead to noise compression of both logicalone and logical-zero levels. The simplest way to obtain a copy of the signal with inverted logic at a different wavelength is the use of an XGM wavelength converter. As already discussed, due to bandwidth limitations, the signals resulting from this technique suffer from waveform distortions and poor ER. However, their quality is good enough to be usefully employed in XGC applications in the multigigabit range. Moreover, the XGC process shows reshaping effects on the wavelength-converted signal itself. As the first point, we show that copropagating two inverted in-logic signals is possible to overcome overall gain modulation. We compare the signals amplified by the same SOA when using or not the wavelength-converted assist signal. The experimental setup is reported in Fig. 2. The input signal is generated modulating a CW laser at λ1 = 1550.5 nm by means of a LiNbO3 Mach–Zehnder (MZ) intensity modulator driven by a 231 − 1-long non-return to zero (NRZ)-PRBS sequence at 9.95328 Gb/s. This signal is first directly amplified in SOA2. In the second case, it is split in two parts: One part is sent to the XGM-based wavelength converter. The converter is in the counterpropagating configuration. It uses a local CW tunable laser (TL) and SOA1 to generate at λ2 a copy of the input signal, which is inverted in-logic. The other part of the

Fig. 3. (a) Input and output eye diagrams from SOA2 in saturation in the case of usual single-pass amplification (without assist signal) and assisted amplification. (b) Power penalty as a function of SOA2 input power with and without inverted in-logic assist light.

signal passes an optical delay line (ODL) in the other arm and is synchronized with the wavelength-converted one. The two opposite signals are then coupled together, while optical attenuators are used for setting constant power at the SOA2 input. The two SOAs are polarization-independent (PDG < 1 dB) pigtailed devices with about 28-dB small signal gain and 6-dBm output saturation power at 200-mA driving current. The power of the TL at λ2 = 1548.2 nm; in the wavelength converter is 4 dBm. The total input power at SOA2 is controlled by an optical attenuator. At SOA2 output, the two signals at different wavelengths can be selected by means of a common array waveguide grating (AWG) and detected by an optical receiver with clock and data recovery. Due to the fast dynamics of SOA gain, strong signal distortions arise in usual single-pass amplification when the amplifier gain is saturated. As shown in Fig. 3(a), for an input power of −7 dBm, overshoots in the trailing edge and slow relaxations in the falling edge completely misshape the output eye diagram. On the contrary, when the copropagating assist signal is used, no eye distortion is apparent using the same input power. Those considerations are quantitatively expressed in Fig. 3(b), where we report the bit-error-rate (BER) power penalty at 10−9 for the signal amplified with and without assist light. For very low input power, where the amplified spontaneous emission (ASE) noise dominates, the penalty is similar for the two cases. At increasing input power, where pattern-related distortions arise, using only the assist signal is possible to manage the propagation of signals in the saturated SOA with reduced power penalty.

CONTESTABILE et al.: CROSS-GAIN COMPRESSION IN SEMICONDUCTOR OPTICAL AMPLIFIERS

917

In the following two sections, we will exploit this main remark to obtain, by using a similar setup, both wavelength conversion and wavelength-preserving signal reshaping. The physical mechanisms under the two applications are slightly different, especially regarding the ER improvement. A phenomenological description of the different effects will be introduced in the respective sections. III. W AVELENGTH C ONVERSION BY M EANS OF XGC As discussed in the previous section, XGC at constant power can potentially improve the quality of the interacting signals. Indeed, the XGC scheme can be used to improve XGM-based converters when used in a synchronous double-stage configuration. The first stage realizes the usual XGM conversion, with its limited performance. In the second stage, the XGC leads to signal reshaping and ER improvement. This architecture allows keeping simplicity and robustness of XGM converters at the cost of using double-stage geometry involving two SOAs. Clearly, as usual for XGM converters, the output signal is inverted in-logic. Although a multistage XGM wavelength converter has been proposed before [17]–[19], the results reported in literature are very limited. In [17] and [18], 1- and 2.5-Gb/s operation were obtained, reporting very few details on the signal quality. Only ER improvement at 2.5 Gb/s was analyzed in [19], disregarding the analysis of the eye-diagram distortions and their effect on the BER. Moreover, no focus was given on the critical point to have a quite constant power at the SOA input to reduce pattern fluctuations. We underline that effective signal improvement of the converted signal is obtained by a careful balance of XGM and limiting amplification in SOA2. After the first-stage XGM converter, the signal still shows a certain significant amount of CW signal (that corresponds to a low ER). A certain amount of XGM between this signal and the incoming one with higher ER (interacting with a slight power unbalance) is responsible for improving the output ER. On the other hand, limiting amplification at an almost constant power in the saturated SOA2 strongly reduce the pattern effects generated in SOA1. In this section, we report the characterization of an XGCbased wavelength converter working at 10 Gb/s. Using the setup shown in Fig. 2, starting from an incoming signal at λ1 = 1550.5 nm, we analyze the quality of the converted signal as a function of different detuning conversions. We study the evolution of the ER after the two conversion stages and the effect of the conversion on the BER. As a preliminary remark, we note that, at the output of the first stage, it is possible to obtain different eye-diagram shapes by varying the absolute and relative power of the continuous and modulated signals. For example, increasing the power of the pump, we can increase the output ER at the expense of a stronger eye distortion. In the following results, we chose the signal power levels that give a symmetric shape to the converted signal: The power of the TL is 3 dBm, while the power of the counterpropagating modulated signal is 9 dBm. The total input power at SOA2 is around 3 dBm, while the relative power of the two signals is slightly unbalanced at each wavelength in a way to optimize the converted output-signal eye diagram.

Fig. 4. (Left) Eye diagrams corresponding to a 20-nm wavelength downconversion (from 1550 to 1530 nm) after the first and second stage. (Right) Output ER as a function of the input ER for one and two stages.

Fig. 5. (Left) Eye diagrams corresponding to a 2-nm wavelength downconversion (from 1550 to 1548 nm) after the first and second stage. (Right) Output ER as a function of the input ER for one and two stages.

Fig. 4 corresponds to the case of 20-nm wavelength downconversion. On the left, an example of the evolution of the eye diagram after the first and after the second-stage converter. On the right, we report the output ER as a function of the input one after SOA1 and SOA2. We see the beneficial effect of the second-stage converter that produces clear pattern suppression with a corresponding increase of the ER in the eye diagram. An ER enhancement can be observed for any input ER but is more effective at higher ER values. As discussed before, when having a higher input ER, residual XGM in SOA2 is more efficient. In Fig. 5, the signal is converted close to the input wavelength: It is a 2-nm down-conversion. In this case, pattern effects are less relevant but the eye diagram shows again a better shape after SOA2. Both one and zero levels are compressed. The ER improvement (on the right) shows a similar trend, as in the previous case. The last example is reported in Fig. 6. This case corresponds to around 10-nm wavelength up-conversion. As known in [1], the XGM converters perform worst in wavelength upconversion because of the change in the differential amplifier gain due to carrier depletion. This is reflected in the lower output ER after the first stage, as this corresponds to the major ER improvement in the second compression stage. Eye diagram and ER are improved even if no more than 7.1 dB of output ER was obtained. We underline that at 10 Gb/s, an output ER improvement with respect to the input one, foreseen in [1] and [19] for

918

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 3, MARCH 2007

be usefully exploited for the all-optical simultaneous multiconversion of an input signal to several wavelengths to realize the WDM multicasting function [22]. IV. N OISE C OMPRESSION BY M EANS OF XGC

Fig. 6. (Left) Eye diagrams corresponding to a 10-nm wavelength upconversion (from 1550 to 1560 nm), after the first and second stage. (Right) Output ER as a function of the input ER for one and two stages.

Fig. 7. BER power penalty at 10−9 as a function of the wavelength of the converted signal after one and two stages.

the case of wavelength down-conversions, cannot be obtained without strong eye distortions. Hence, in all cases, the output ER is lower than the input one. After this qualitative study on the wavelength dependence of the converter, we measured the BER power penalty. We see, in Fig. 7, the power penalty at a BER = 10−9 after the first and second stage for wavelength conversions from 1550 nm to the entire C-band (1530–1560 nm). The penalty is reduced from 2.1–3.5 dB after SOA1 to 0.9–1.7 dB after SOA2. As expected, down-conversion performs better than up-conversion, and the power penalty increases from shorter to longer wavelengths. Summarizing, the XGC effect can be effectively used to improve wavelength conversion realized by the XGM. The scheme is quite simple: It is made by two synchronous cascaded XGM processes and, even though the wavelength-converted channels are inverted in-logic, it shows several advantages with respect to other techniques. It shows a limited power penalty (less than 1.7 dB) for a broadband conversion in all the C-band (30 nm). As it is based on two SOAs, which are not in an interferometric structure, it offers intrinsic stability and simplicity. Moreover, the converter does not suffer from any polarization dependence (as long as polarization-independent SOAs are used). The operating bit rate critically depends on the operation speed of the first-stage converter, while in SOA2, XGM is performed at an almost constant power, resulting in very low overall gain modulation. At least 40-Gb/s operation could be expected using an optimized SOA for the first stage [21]. Moreover, a modified version of the same scheme can

Most of the techniques proposed in the literature for alloptical regeneration are based on wavelength converters with a step-like transfer function; hence, the wavelength conversion is an intrinsic characteristic of such regenerators [23]. However, in real networks, as well as in point-to-point WDM links, the wavelength conversion of the regenerated signals can be undesired. On the other hand, regeneration without wavelength conversion is, in general, a harder task to obtain. Following from the considerations on the XGC effect made in Section II, we recently proposed to use this mechanism to have wavelengthpreserving signal reshaping. In [24], we introduced for the first time the XGC scheme showing preliminary results on its reshaping capability. In detail, in that letter, we showed eyediagram opening of noisy signals, including the static transfer function of the XGC process. In the case of a suboptimum receiver, we reported BER improvement of reshaped signals. Here, we analyze in more detail the effect of XGC on the noise distributions of NRZ signals. We report a detailed experimental characterization of the input–output evolution of the eye diagrams. To do this, we study, in the case of 10-Gb/s signals, the BER as a function of the receiver–threshold value at varying-input optical signal-to-noise ratio (OSNR). This kind of analysis permits us to follow the evolution of the probabilitydensity functions of both logical one and zero level. We want to outline here that, while the mechanism responsible for the compression of noise on the one level can be easily understood due to the limiting saturation of the SOA gain, the mechanism responsible for noise compression on the zero level require a deeper insight into the physics of the SOA gain saturation. The situation can be depicted in a simplified picture, as in Fig. 8. If we have a wavelength allocation, in which the assist signal at λ2 is at a longer wavelength in respect of the incoming signal at λ1 , we can consider the differential gain due to semiconductor saturation to be responsible for the noise reduction on the zero level. Indeed, while the logical one at λ1 experiences a saturated gain G(1) , the logical zero experiences a lower saturated gain G(0) because of the SOA gain profile depletion due to the saturating effect induced by the signal at λ2 . It is easy to understand that, working in deep-saturation condition, the gain depletion at λ1 , i.e., G(1) − G(0) is very low and that the strength of this effect is lower than the others responsible for the XGC process. Moreover, it increases at the increasing length of the SOAs and it also depends on the wavelength detuning between the two signals. We want to stress that a model for the XGC process should include fast-saturating effects of the SOA and, at the same time, should consider the wavelength modulation of the SOA gain curve due to the two signals at different wavelength, and this is out of the scope of our paper. The experimental setup for the regenerator characterization is reported in Fig. 9. The input signal is generated using a

CONTESTABILE et al.: CROSS-GAIN COMPRESSION IN SEMICONDUCTOR OPTICAL AMPLIFIERS

Fig. 8. Scheme of the gain-depletion effect due to gain saturation at different wavelength in an SOA.

Fig. 9. Scheme of the experimental setup. TL = Tunable Laser. BPF = Bandpass Filter. WC = Wavelength Converter. PD = Photodetector.

LiNbO3 MZ to modulate a CW lightwave at λ1 = 1548.5 nm (231 − 1 PRBS sequence at 9.95328 Gb/s). The noise loading is realized by adding the ASE from an erbium-doped fiber amplifier. The input signal, with around 12-dBm power, is split into two parts. One is sent to the XGM wavelength converter in copropagating configuration and then filtered out. The converter uses a local CW TL and SOA1 to generate the assist signal at λ2 = 1558 nm: around 10 nm at a longer wavelength with respect to the input signal. In the other arm of the regenerator, the other part of the signal passes through an ODL and is synchronized with the wavelengthconverted one. These two signals are coupled and an optical attenuator controls their relative power before being injected into the XGC stage. The total input power to SOA2 and SOA3 is set quite constant to avoid significant gain modulation. We underline that the overall noise compression depends on the SOA gain compression and on the interaction length between the copropagating signals. For this reason, in this application, we use two consecutive SOAs to enhance the efficiency of the XGC process. It’s easy to understand that a similar result can be expected by using a single longer device, and higher noise-compression factors could be in principle obtained using ultralong SOAs [25].

919

SOA1 is a polarization-independent (PDG < 1 dB) multiquantum-well device with about 30-dB small signal gain and 10-dBm output saturation power at 200-mA driving current, while SOA2 and SOA3 are the same devices used in the setup of Fig. 2. The power of the TL at λ2 in the wavelength converter is 2 dBm. The total input power at SOA2 is around 4 dBm, corresponding to around 25-dB gain compression. At the SOA3 output, the two interacting signals at different wavelengths can be selected by means of a bandpass filter. We expect that both signals experience waveform reshaping. We have seen in the previous section that the converted signal at λ2 experiences ER improvement and pattern-distortion reduction. Here, we focus on the study of the noise-distribution evolution of the signal at λ1 . In Fig. 10, we report the comparison of the eye diagrams and of the BER as a function of the receiver–threshold curves for various input OSNR values ranging from 15 to 30 dB. We see that the eye-diagram opening corresponds to a compression of the probability-density function of the logical levels. The noise compression is more pronounced on the logical-one level at lower OSNR values (despite a slight increment on the zero level for OSNR = 15 dB), while it is effective on both levels at increasing OSNR. Moreover, we note that, as the SOA gain is in constant deep saturation, a fine power balancing of the two interacting signals in the SOA is not critical. Hence, we can, to some extent, control the compression of the two logical levels by simply unbalancing the power of the two signals. For this reason, it is possible, in principle, to adapt the amount of compression on marks or spaces, depending on the kind of noise affecting the eye diagram. As an example, in case of strong crosstalk noise that affects mostly the mark level of a signal, we can selectively act obtaining a higher compression on the marks level. Finally, we also analyzed the case of a noiseless input signal. From theoretical considerations, we expect that a regenerator driven by an ideal input signal can, at best, preserve the overall data quality but cannot improve it. This is confirmed from the BER results found in [24], which can be extrapolated from the diagram of Fig. 3(b). The output signal from XGC shows about 0.3-dB power penalty with respect to the signal in back-to-back. Even if those general considerations are true for the overall BER, we see in Fig. 11 that the XGC effectively compresses the data levels also in the back-to-back case. This effect can be clearly observed looking directly at the eye diagram. To explain this, we have to consider that the input signal is affected by some intersymbol interference due to the finite electrical and optical bandwidth of the components in the transmitter. This leads to a certain spread in the traces of the eye diagram. We see that at the cost of having a slight asymmetry in the output eye diagram, it is also possible to have compression of the symbol distribution, in this boundary case. The results reported here show the potential of the XGC technique for wavelength-preserving 2R regeneration. However, a complete regenerator characterization should involve cascading optical regenerations realized in a transmission-line or in recirculating-loop arrangements. We underline that contrary to all-optical regeneration techniques employing wavelength converters, the XGC scheme

920

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 3, MARCH 2007

Fig. 11. BER versus receiver–threshold curves comparing regenerator input and output signal in case of input back-to-back signal. The corresponding eye diagrams are also reported.

assist signal. When using XGM, as in our setup, we expect that at least 40-Gb/s operation would be possible [6], [21]. V. C ONCLUSION We have studied a simple scheme that can perform at the same time wavelength conversion and wavelength-preserving signal reshaping. The basic physical effect is the XGC in an SOA between two inverted in-logic signals at different wavelengths. Used as the second stage after an XGM converter, it can greatly improve the quality of the wavelength-converted signal, increasing the output ER, and strongly reduce the waveform distortions related to pattern effect. For this application, we obtained less than a 1.7-dB power penalty for a broadband wavelength conversion from 1550 nm to all the C-band at 10 Gb/s. Moreover, we showed the potential of the same scheme to realize a wavelength-preserving 2R regeneration. We found a net compression of the noise statistic of logical-symbol distribution for 10-Gb/s signals. We expect that even higher compression factors could be obtained with longer SOAs. Concluding, we observe that the XGC scheme has a simple architecture, and it may be potentially integrated on a chip. As it is not an interferometric structure, it is very stable, and when using polarization-independent SOAs, its polarization dependence is very low (limited by the SOA-residual PDG). Finally, we expect that at least 40-Gb/s operation could be obtained using the optimized devices. ACKNOWLEDGMENT Fig. 10. BER versus receiver–threshold curves comparing regenerator input and output signals for input OSNR ranging from 15 to 30 dB. Positive voltages correspond to logical ones and negative voltages to logical zeros. The corresponding eye diagrams are also reported.

does not produce an improvement of the output OSNR. We have, instead, a redistribution of the probability-density functions of both marks and spaces with a change in their noise statistics. Obviously, this process cannot correct errors in bits that have already exceeded the threshold level, but it can limit BER accumulation when cascaded in an amplified-transmission link. As it mostly involves saturation effects at constant power in the SOA, the maximum working bit rate of the XGC effect critically depends on the availability of an inverted in-logic

The authors would like to thank G. Gavioli from the University College of London for the useful discussions. R EFERENCES [1] T. Durhuus, B. Mikkelsen, C. Joergensen, S. Lykke Danielsen, and K. E. Stubkjaer, “All-optical wavelength conversion by semiconductor optical amplifiers,” J. Lightw. Technol., vol. 14, no. 6, pp. 942–954, Jun. 1996. [2] D. Marcenac and A. Mecozzi, “Switches and frequency converters based on cross-gain modulation in semiconductor optical amplifiers,” IEEE Photon. Technol. Lett., vol. 9, no. 6, pp. 749–751, Jun. 1997. [3] D. Nesset, T. Kelly, and D. Marcenac, “All-optical wavelength conversion using SOA nonlinearities,” IEEE Commun. Mag., vol. 36, no. 12, pp. 56–61, Dec. 1998. [4] I. White, R. Penty, M. Webster, Y. J. Chai, A. Wonfor, and S. Shahkooh, “Wavelength switching components for future photonic networks,” IEEE Commun. Mag., vol. 40, no. 9, pp. 74–81, Sep. 2002.

CONTESTABILE et al.: CROSS-GAIN COMPRESSION IN SEMICONDUCTOR OPTICAL AMPLIFIERS

[5] H. Lee, H. Yoon, Y. Kim, and J. Jeong, “Theoretical study of frequency chirping and extinction ratio of wavelength-converted optical signals by XGM and XPM using SOA’s,” IEEE J. Sel. Topics Quantum Electron., vol. 35, no. 8, pp. 1213–1219, Aug. 1999. [6] S. L. Danielsen, C. Joergensen, M. Vaa, B. Mikkelsen, K. Stubkjaer, P. Doussiere, F. Pommerau, L. Goldstein, R. Ngo, and M. Goix, “Bit error rate assessment of 40 Gb/s all-optical polarization independent wavelength converter,” Electron. Lett., vol. 32, no. 18, pp. 1688–1690, Aug. 1996. [7] H.-Y. Yu, D. Mahgerefteh, P. S. Cho, and J. Goldhar, “Optimization of the frequency response of a semiconductor optical amplifier wavelength converter using a fiber Bragg grating,” J. Lightw. Technol., vol. 17, no. 2, pp. 308–315, Feb. 1999. [8] X. Zheng, F. Liu, and A. Kloch, “Experimental investigation of the cascadability of a cross-gain modulation wavelength converter,” IEEE Photon. Technol. Lett., vol. 12, no. 3, pp. 272–274, Mar. 2000. [9] M. Yoshino and K. Inoue, “Improvement of saturation output power in a semiconductor laser amplifier through pumping light injection,” IEEE Photon. Technol. Lett., vol. 8, no. 1, pp. 58–59, Jan. 1996. [10] M. A. Dupertuis, J. L. Pleumeekers, T. P. Hessler, P. E. Selbmann, B. Deveaud, B. Dagens, and J. Y. Emery, “Extremely fast high-gain and low-current SOA by optical speed-up at transparency,” IEEE Photon. Technol. Lett., vol. 12, no. 11, pp. 1453–1455, Nov. 2000. [11] R. Inohara, K. Nishimura, M. Tsurusawa, and M. Usami, “Experimental analysis of cross-phase modulation and cross-gain modulation in SOAinjecting CW assist light,” IEEE Photon. Technol. Lett., vol. 15, no. 9, pp. 1192–1194, Sep. 2003. [12] H. Wang, J. Wu, and J. Lin, “Spectral characteristics of optical pulse amplification in SOA under assist light injection,” J. Lightw. Technol., vol. 23, no. 9, pp. 2761–2771, Sep. 2005. [13] H. S. Chung, R. Inohara, K. Nishimura, and M. Usami, “Alloptical multi-wavelength conversion of 10 Gb/s NRZ/RZ signals based on SOA-MZI for WDM multicasting,” Electron. Lett., vol. 41, no. 7, pp. 432–433, Mar. 2005. [14] S. Arahira and Y. Ogawa, “160-Gb/s all-optical encoding experiments by four-wave mixing in a gain-clamped SOA with assist-light injection,” IEEE Photon. Technol. Lett., vol. 16, no. 2, pp. 653–655, Feb. 2004. [15] S. Bischoff, M. L. Nielsen, and J. Mork, “Improving the all-optical response of SOAs using a modulated holding signal,” J. Lightw. Technol., vol. 22, no. 5, pp. 1303–1308, May 2004. [16] R. Sato, T. Ito, Y. Shibata, A. Ohki, and Y. Akatsu, “40-Gb/s burst-mode optical 2R regenerator,” IEEE Photon. Technol. Lett., vol. 17, no. 10, pp. 2194–2196, Oct. 2005. [17] J. C. Simon, L. Lablonde, I. Valiente, L. Billes, and P. Lamouler, “Twostage wavelength converter with improved extinction ratio,” presented at the Optical Fiber Commun. Conf., San Diego, CA, 1995, Paper PDP lS. [18] W. Shieh, S. H. Huang, and A. E. Willner, “A polarization-independent and contrast-ratio-enhancing module for all-optical wavelength shifting using SOAs,” IEEE Photon. Technol. Lett., vol. 8, no. 4, pp. 533–535, Apr. 1996. [19] S. Chelles, F. Devaux, D. Meichenin, D. Sigogne, and A. Carenco, “Extinction ratio of cross-gain modulated multistage wavelength converters: Model and experiments,” IEEE Photon. Technol. Lett., vol. 9, no. 6, pp. 758–760, Jun. 1997. [20] K. Sato and H. Toba, “Reduction of mode partition noise by using semiconductor optical amplifiers,” IEEE J. Sel. Topics Quantum Electron., vol. 7, no. 2, pp. 328–333, Mar./Apr. 2001. [21] G. Contestabile, N. Calabretta, M. Presi, and E. Ciaramella, “Single and multicast wavelength conversion at 40 Gb/s by means of fast nonlinearpolarization-switching in an SOA,” IEEE Photon. Technol. Lett., vol. 17, no. 12, pp. 2652–2654, Dec. 2005. [22] G. Contestabile, N. Calabretta, R. Proietti, and E. Ciaramella, “Double stage cross gain modulation in SOAs: An effective technique for WDM multicasting,” IEEE Photon. Technol. Lett., vol. 18, no. 1, pp. 181–183, Jan. 2006. [23] O. Leclerc, B. Lavigne, E. Balmefrezol, P. Brindel, L. Pierre, D. Rouvillain, and F. Seguineau, “Optical regeneration at 40 Gb/s and beyond,” J. Lightw. Technol., vol. 21, no. 11, pp. 2779–2790, Nov. 2003. [24] G. Contestabile, R. Proietti, N. Calabretta, and E. Ciaramella, “Reshaping capability of cross gain compression in semiconductor amplifiers,” IEEE Photon. Technol. Lett., vol. 17, no. 12, pp. 2523–2525, Dec. 2005. [25] G. Bramann, H.-J. Wunsche, U. Busolt, C. Schmidt, M. Schlak, B. Sartorius, and H.-P. Nolting, “Two-wave competition in ultralong semiconductor optical amplifiers,” IEEE J. Quantum Electron., vol. 41, no. 10, pp. 1260–1267, Oct. 2005.

921

Giampiero Contestabile received the Laurea degree in physics from “La Sapienza” University, Rome, Italy, and the Ph.D. degree in electrical engineering and telecommunications from “Tor Vergata” University, Rome, in 1998 and 2001, respectively. Between 1996 and 2000, he was with the Semiconductor Devices Group of “Fondazione Ugo Bordoni,” Rome. In 2001, he was with Optospeed Italia. Since September 2002, he has been an Assistant Professor with the Scuola Superiore Sant’Anna, University Studies and Doctoral Research, Pisa, Italy. He coauthored more than 40 papers published in international peerreviewed journals and presented in leading international conferences. His main research interests are in advanced wavelength-division-multiplexing systems, optical packet-switched networks, and, in general, applications of semiconductor optical amplifiers.

Roberto Proietti was born in Livorno, Italy, in 1977. He received the Laurea degree in telecommunication engineering from the University of Pisa, Pisa, Italy, in 2004. He is currently working toward the Ph.D. degree at Scuola Superiore Sant’Anna, Pisa, where he works on new technical solutions for advanced wavelength-division-multiplexing systems. Mr. Proietti received scholarship from CNIT National Photonic Networks Laboratory, Pisa, on optical-communication systems, in 2004.

Nicola Calabretta received the B.S. and M.S. degrees in telecommunications engineering from Politecnico di Torino, Turin, Italy, in 1995 and 1999, respectively, and the Ph.D. degree from COBRA Research Institute, the Eindhoven University of Technology, Eindhoven, The Netherlands, in 2004. In 1999, he received scholarship to carry out his Master’s thesis at the Eindhoven University of Technology, where he has investigated a novel cost-effective system for monitoring wavelengthdivision-multiplexing channels. During his Ph.D. studies, he investigated all-optical signal processing in nonlinear medium to achieve functionalities for all-optical packet-switched cross-connect nodes. He is currently working as a Researcher with the Scuola Superiore Sant’Anna University, Pisa, Italy. His fields of interest are all-optical signal processing for optical packet switching, semiconductor optical amplifiers, all-optical wavelength conversion and regeneration, and advanced modulation formats for optical packet switching. He has coauthored more than 35 papers published in international journals and conferences. Dr. Calabretta is the recipient of the KIWI Telecom Awards 2004, Schipool, The Netherlands, for the innovative content reported in his Ph.D. thesis.

Ernesto Ciaramella (M’06) was born in Rome, Italy, in 1967. He received the Laurea degree (cum laude) from “La Sapienza University,” Rome, in 1991. In 1992, he received a scholarship from Alcatel. In 1992–1994, he was a Researcher with the “Fondazione Ugo Bordoni,” working on nonlinear optical effects. In 1994–1998, he was with CSELT, Turin, where he was first concerned with linear and nonlinear propagation effects in optical fibers and then with numerical modeling of high-capacity optical systems. During this period, he contributed to the CSELT-Telecom Italia working group on wavelength-division-multiplexing systems. In 1998–2000, he was Scientific Researcher with the “Fondazione Ugo Bordoni,” working on optical-transmission systems and network architectures. In 2001–2002, he was Research Manager at CNIT National Photonic Networks Laboratory, Pisa. Since 2002, he has been an Associate Professor with the Scuola Superiore Sant’Anna, Pisa. His research activity covers various issues in optical communications (components, systems, and networks). He participated in various European research projects, published approximately 90 papers, and is author/coauthor of five international patents.