ELSEVIER FIRST PROOFS

Transcription

1 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers 1 OPTICAL AMPLIFIERS A5 S5 P5 P1 Semiconductor Optical Amplifiers M J Connelly, University of Limerick, Limerick, Ireland q 24, Elsevier Ltd. All Rights Reserved. Michael J Connelly, Dept. Electronic and Computer Engineering, University of Limerick, Limerick, Ireland Key words: noise; optical communications; optoelectronics; photonic switching; semiconductor optical amplifier; wavelength converter Introduction There has been rapid growth in the deployment and capacity of optical fiber communication networks over the past twenty-five years. This growth has been made possible by the development of new optoelectronic technologies that can be utilized to exploit the enormous bandwidth of optical fiber. Today, systems are operational which operate at bit rates in excess of 1 Gb/s. Optical technology is the dominant carrier of global information. It is also central to the realization of future networks that will have the capabilities demanded by society. These capabilities include virtually unlimited bandwidth to carry communication services of almost any kind, and full transparency that allows terminal upgrades in capacity and flexible routing of channels. Many of the advances in optical networks have been made possible by the optical amplifier. Optical amplifiers can be divided into two classes: the optical fiber amplifier and semiconductor optical amplifier (). The former dominates conventional system applications such as in-line amplification to compensate for optical link losses. However, due to F5 Figure 1 basic structure. P in advances in optical semiconductor fabrication techniques and device design, the is showing great promise for use in evolving optical communication networks. It can be utilized as a general gain element but also has many functional applications, including optical switching and wavelength conversion. These functions, where there is no conversion of optical signals into the electrical domain, are required in transparent optical networks. We will review basics, technology (materials P15 and structures), signal transmission performance (pattern effects, crosstalk, and ultrashort pulse amplification), and some important functional applications (optical switching and wavelength conversion). Basic Principles Active Optical field R waveguide profile R The is based on similar technology as a laser diode. Optical gain is achieved by electrically pumping a suitable semiconductor material, such that a population inversion occurs between the material conduction and valence bands. An incoming photon can then be amplified when the resulting stimulated emission exceeds losses due to stimulated absorption and internal material losses. s can be designed to operate in either the 13 nm or 155 nm optical communication windows. The principle of operation of an with low residual reflectivities ðr < Þ is shown in Figure 1. An input optical power experiences a single-pass gain G ¼ expðglþ after traveling through the active waveguide of length L. The net gain coefficient g ¼ Gg m 2 a int ; where G is the optical confinement factor (the fraction of the propagating signal power confined to the waveguide), g m the material gain, and a int the optical loss coefficient. g m is a function of the injected carrier (electron) density and wavelength. The single-mode active waveguide can support two orthogonal polarization modes: transverse electric (TE) and transverse magnetic (TM). l l Input spectrum L P = GP out Signal l in l Output spectrum ASE S1 P2

2 2 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers P25 P3 S15 P35 The amplification process also adds broadband noise, amplified spontaneous emission (ASE) to the output signal. The amplifier noise figure (NF) is a measure of the signal-to-noise ratio (SNR) degradation of the signal after amplification. It is defined as the ratio of the SNR of the input to that of the output when the input noise is shot noise limited. Expressions for the output SNR are usually calculated assuming that the is followed by a narrowband optical filter and an ideal photodetector, such that the main source of detector noise is the beat noise between the signal and the ASE falling within the filter bandwidth. In this case the noise figure is given by NF ¼ 2n sp K with the excess noise factor K given by K ¼ ð1 þ RGÞðG 2 1ÞGg m ð1 2 RÞGðGg m 2 a int Þ ½1Š ½2Š where n sp is the effective population inversion parameter. To achieve a low NF, the internal losses must be small and a value of n sp close to one is required. A favorable value of n sp is achieved by operating the at high gain. A low R is required to prevent the from oscillating at high gains. The residual reflectivity is manifest as ripples in the amplifier gain spectrum and ASE spectrum. Structures The key parameters required for a practical are: P4. low reflectivity (,1 24 ) to ensure low gain ripple (,.5 db); P45. low optical loss coefficient to achieve a high gain; P5. high material gain to allow low drive current operation; Electrode p-inp n-inp p-inp. low polarization sensitivity (,.5 db) because the polarization state of the optical signal coming from a link fiber is usually random; P55. high saturation output power ðp sat Þ; defined as the output power at which the gain is reduced by 3 db; and P6. low fiber-to-chip coupling losses (,3 db per facet). P65 A schematic diagram of a commercial chip is P7 shown in Figure 2. The active waveguide consists of a.2 mm thick InGaAsP bulk active layer sandwiched between.1 mm thick InGaAsP separate confinement heterostructure (SCH) layers. The central section of the active waveguide is 6 mm long with a constant width of 1.4 mm. The mode-expanding active waveguide tapers are 15 mm long with a width that changes linearly from 1.4 mm to.4 mm at the tip. The tapers allow optical coupling to an underlying passive waveguide enabling efficient coupling to the input and output optical fibers. The structure provides a high confinement factor because of the refractive index mismatch between the layers in the gain section. The p-n junction formed by the p-inp and n-inp layers acts as a current block, thereby providing good confinement of the injected carriers from the drive current to the active layer. Very lowreflectivity (,1 26 ) is obtained by combining buried windows with antireflection-coated tilted facets (78 tilt angle). Because of the active waveguide asymmetry, the TE P75 confinement factor is larger than the TM confinement factor. g m is isotropic in bulk material. The introduction of tensile strain, due to the lattice mismatch between the active layer and SCH layers, causes the bulk material band structure to change in such a way that the TM material gain is greater than the TE material gain. The introduction of the correct amount of tensile strain compensates for the difference in the TE and TM confinement factors leading to low polarization dependence. Tensile-strained InGaAsP Passive bulk active layer surrounded waveguide by InGaAsP SCH layers (a) Cross-section of the SCH structure. AR coating 15 µm InGaAsP buried active waveguide 1.4 µm Buried window 6 µm 15 µm (b) Top view of the active waveguide. AR coating 15 µm 7 o F1 Figure 2 SCH with tensile-strained bulk active layer.

3 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers 3 T5 Table 1 Typical specifications of a commercial P8 P85 Wavelength of peak gain 155 nm Peak fiber-to-fiber gain 22 db Noise figure 6.5 db Saturation output power 1 dbm Polarization sensitivity.2 db Gain ripple.1 db 3 db bandwidth 4 nm Drive current 2 ma Other structures that can achieve low polarization dependency are based on active waveguides with near square cross-section having almost the same TE and TM confinement factors or the combination of compressively strained quantum wells (higher TE gain) and tensile strained quantum wells (higher TM gain). Typical specifications and characteristics of a commercial are listed in Table 1 and shown in Figure 3. As Figure 3 shows, gain saturation in conventional s is manifest at output signal powers well below P sat. Because the gain recovery time (carrier lifetime) in s is fast (typically.1 1 ns), this can lead to pattern effects in single-channel systems and severe crosstalk in wavelength division multiplexed (WDM) systems. This problem can be greatly reduced by the use of a gain-clamped (GC-). In a GC-, lasing action is produced, at a wavelength remote from the operating wavelength range, by the introduction of wavelength specific feedback. Once lasing begins, the carrier density in the active layer is clamped at a fixed value. Changes in the input signal power lead to opposing changes in the lasing mode power. This has the effect of keeping the carrier density fixed (i.e., clamped), thereby making the signal gain relatively insensitive to changes in the total input power. A common method of providing this feedback is the use of distributed Bragg reflectors (DBRs) as shown in Figure 4. A typical GC- gain versus output power characteristic is shown in Figure 5. Small-signal gain (db) Basic Applications of s in Optical Communication Systems Bias current (ma) S2 The three basic applications of s in optical P9 communication systems are: booster amplifier, in-line amplifier, and preamplifier. The main requirements of s for such applications are listed in Table 2. A Booster amplifier is used to increase the power of P95 a high power signal. Boosting laser power in an optical transmitter can be used to overcome external modulator losses, compensate for splitting and tap losses in optical distribution networks and to increase the power budget of optical links. The most critical requirement of a booster amplifier is a high P sat,to obtain a high output signal power and minimize pattern effects. s with P sat in excess of 1 dbm are now available commercially. The sensitivity of a conventional optical receiver is P1 limited by thermal noise. The sensitivity is the minimum signal power required at the receiver input to achieve a desired bit-error-rate (typically 1 29 ). An optical preamplifier can be used to increase the power level of an optical data signal prior to detection and demodulation, leading to an increase in sensitivity. The performance of a preamplified optical receiver is dependent on the detector signal-to-noise ratio, SNR ¼ i 2 sig = ı2 noise : The signal photocurrent i sig is proportional to the amplified signal power. The mean square noise current ı 2 noise includes circuit noise (thermal noise and detector dark current noise), signal shot noise, spontaneous emission shot noise, signal-spontaneous beat noise, and beat noise between the spectral components of the spontaneous emission. The best improvement in SNR occurs when the is operated in the signal-spontaneous beat noise limit. In this regime, the signal power is sufficiently large such that the dominant receiver noise is the signal-spontaneous beat noise that falls within the signal bandwidth. This usually requires that the spontaneous emission from the be Noise figure (db) Gain (db) db P sat Output power (dbm) F15 Figure 3 Typical commercial small-signal gain and noise figure versus bias current and gain versus output power characteristics.

4 4 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers P15 P11 Active waveguide P in reduced using a narrowband optical filter. The noise figure is a critical parameter in this application and should be as low as possible. In long-haul optical transmission systems, in-line optical amplifiers can be used to compensate for link losses thereby increasing the spacing between optical regenerators. The main advantages of in-line s are transparency to data rate and modulation format (unsaturated operation), bidirectionality, WDM capability, simple mode of operation, low power consumption, and compactness. The latter two advantages are important for remotely located optical components. To avoid noise accumulation in the link, it may be necessary to follow each by a narrowband optical filter, but this may prevent capacity enhancements using WDM. An example of a single channel optical transmission system utilizing booster, in-line, and preamplifier s is shown in Figure 6. The transmitter Gain (db) DBR section Bias current F2 Figure 4 Schematic structure of a GC-. DBR section Output power (dbm) F25 Figure 5 Typical GC- gain versus output power characteristic. T1 Table 2 Requirements of basic applications P out comprized a 139 nm gain-switched laser diode directly modulated with a 1 GHz sinusoid to produce a train of 4 ps wide pulses at a repetition rate of 1 GHz. At 139 nm fiber dispersion is small and the maximum transmission distance is mainly limited by the link losses. The laser output was connected to an external modulator, driven by a pseudo-random bit sequence (PRBS) to produce an optical data stream with an extinction ratio of 13 db. A booster was used to increase the average transmitted power to between and 2 dbm (7 9 dbm peak power). The transmission fiber length was 42 km with twelve in-line s, used to compensate for fiber loss, spaced at 38 km intervals. At the receiver the signal was passed through a 1 nm bandpass filter, to reduce the accumulated spontaneous emission, amplified by an preamplifier and filtered by a 1 nm bandpass filter. The signal was then detected by a p-i-n photodiode followed by an electronic clock and data recovery circuit. The preamplifier and filter increased the receiver sensitivity from 214 dbm to 231 dbm for a BER of The receiver penalty after 42 km was 5 db. In this experiment, the main limitation on transmission distance was the accumulation of spontaneous emission within the optical filter bandwidth. Pattern Effects and Crosstalk S25 When an is operated in the unsaturated regime, P115 the amplifier gain is independent of the number of input signals and signal data rate. Outside this regime, the will cause distortion because at high input powers, the gain saturates and compresses. Dynamic gain saturation occurs on the same time scale as the gain recovery time. This leads to pattern effects where the power of an incoming bit affects the gain experienced by subsequent bits. This is particularly important when the bit rate is of the same order as the inverse of the gain recovery time as shown in Figure 7. In WDM systems cross-gain modulation (XGM) between the channels can lead to severe interchannel crosstalk. A further complication in WDM systems using P12 s is interchannel crosstalk caused by four-wave mixing (FWM). FWM is a coherent nonlinear process Requirement Booster amplifier In-line amplifier Preamplifier High gain Yes Yes Yes High P sat Yes Yes Not critical Low noise figure Not critical Yes Yes Low polarization sensitivity Not critical Yes Yes Optical filter Not necessary Not critical Yes

5 Article Number: OPTC 662 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers 5 Eye diagram at transmission fiber input 38 km single-mode fiber Transmitter 1 Gb/s PRBS Booster 1 GHz 139 nm laser diode In-line External modulator x8 Sublink preamplifier p-i-n photodiode receiver Clock and data recovery Receiver Optical filters PR O of short optical pulses. The transmission distance is limited by the fiber group velocity dispersion, which is proportional to the pulse spectral width. Soliton transmission is also possible where the pulse shape is preserved as it propagates in the fiber. Because an has a very large bandwidth P13 (typically 5 THz) it is capable of amplifying pulses as short as 1 fs. An input pulse can be amplified without significant distortion if the pulse energy is much less than the saturation energy Esat of the. Typical saturation energies are of the order of a few pico Joules. As the pulse energy approaches Esat, considerable spectral broadening and distortion can result. For an input pulse width tp (full width at half maximum) of the order of 1 1 ps spectral broadening is primarily due to self phase modulation (SPM). SPM is caused by gain saturation, which leads to intensity-dependent changes in the active layer refractive index in response to carrier density variations. The degree of spectral broadening and distortion also depends on the input pulse shape. If tp is much less than the carrier lifetime, the output ER FI R between two optical fields within the, resulting in gain modulation at the beat frequency between the fields, and in the process generating new sidebands. In WDM systems with equally spaced wavelengths, the net effect of FWM is the generation of crosstalk terms interfering with the desired signals. Although the crosstalk can be relatively low in power terms, it can produce significant power penalties due to the coherent beat noise phenomena. The level of FWM crosstalk increases as the channel spacing decreases and the channel output power increases. FWM crosstalk is of particular importance in dense WDM systems where the inter channel spacing is less than 1 GHz. Figure 8 shows a typical output spectrum for an 8-channel multiplex with FWMgenerated crosstalk signals. O FS Figure 6 1 Gbit/s transmission experiment utilizing booster, in-line and preamplifier s. (adapted from Kuindersma et al., 1 Gbit/s RZ transmission at 139 nm over 42 km using a chain of multiple quantum-well semiconductor optical amplifier modules at 38 km intervals (European Conference on Optical Communications 1996, with permission).) ST F3 Q1 Ultrashort Pulse Amplification P125 Optical time division multiplexing is an efficient method for increasing the bit rate of optical transmission systems. This requires the time interleaving EL SE VI S3.1 ns (a) F35 (b) Figure 7 Typical eye diagrams of 1 Gb/s nonreturn to zero data (a) before and (b) after amplification by an. ASE noise is present in the output eye diagram.

6 6 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers Power (dbm) pulse power P out ðtþ and phase f out ðtþ are given approximately by with P out ðtþ ¼P in ðtþ exp½hðtþš f out ðtþ ¼f in ðtþ ahðtþ! hðtþ ¼ " G exp U in ðtþ ¼ ðt 21 P in ðtþdt ½3Š ½4Š 2U!# 21 inðtþ ½5Š E sat ½6Š where P in ðtþ and f in ðtþ are the input pulse power and phase, respectively. G and a are the amplifier unsaturated gain and linewidth enhancement factor, respectively. Typical values for a are in the range 2 1 and depend on the amplifier active region material and operating conditions. The output pulse spectrum can be obtained from SðvÞ¼ Wavelength (nm) ð WDM channels FWM crosstalk F4 Figure 8 Typical output spectrum for an 8-channel WDM multiplex with 1 GHz spacing. Channel 6 has been dropped to show the FWM crosstalk signal caused by channels 5 and 7. ½P out ðtþš 1=2 exp½if out ðtþþiðv2v ÞtŠdt 2 ½7Š where n ¼v =2p is the pulse optical frequency. The output pulse chirp (frequency variation) is given by Dn out ðtþ¼2 1 f out ¼Dn 2p t in ðtþþ a h ½8Š 4p t where Dn in ðtþ is the input pulse chirp. Using the above theory the shape, chirp and spectrum of an amplified zero-chirp (transform limited) Gaussian pulse of input power P in ðtþ¼ E! in pffiffi exp 2 t2 t p t 2 is shown in Figure 9. E in is the pulse energy and t p <1:665t : The amplified pulse is asymmetric because the leading edge of the pulse experiences a larger gain than the trailing edge. The amplified pulse spectrum is broader than the input pulse and also develops a multipeak structure. This is due to the SPM-induced frequency chirp imposed on the pulse as it propagates through the amplifier. In this case the chirp is linear and can be compensated for by transmitting the pulse through an optical fiber with anomalous group-velocity dispersion. In practice optical pulses can be far from Gaussian and can also have an initial chirp. In this case the induced chirp and resulting pulse spectrum can be more complex and more difficult to compensate. When the input pulse width is less than,1 ps, the P135 above theory is no longer adequate and other nonlinear effects within the such as carrier heating and spectral hole burning must be taken into account. The resulting pulse power and spectral distortions induced by an on such pulses can be very complex. Normalised power Chirp, Dn out t Normalised power db t/t 2dB 2 db 1 db 1dB 3dB t/t 3 db Input 2 db 1 db Input (n-n )/t Figure 9 Amplified pulse shape, frequency chirp and spectrum for a transform-limited Gaussian input pulse with E in =E sat ¼ :1 and a ¼ 5: n is optical frequency. The parameter is the amplifier unsaturated gain. F45

7 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers 7 S35 P14 P145 P15 Functional Applications s can be used to perform functions that are particularly useful in optically transparent networks. Recent advances in photonic integrated circuit and optoelectronic device packaging technology have made the deployment of functional elements more feasible. Two of the most important applications of s are optical switching and wavelength conversion. An example of a 2 2 switch module used for optical routing is shown in Figure 1. Larger switch matrices can be constructed using this basic element. The module is based on a hybrid structure consisting of four integrated s mounted on a Silicon submount. The -array is aligned to input and output polymer waveguides through V-grooves and alignment indentations. An incoming data packet can be routed to any output port by switching on the appropriate. Switching times of the order of 1 ns are possible. Ultrafast switching (,1 fs) can be achieved by incorporating s in non-linear loop mirror structures, such as the terahertz optical asymmetric demultiplexer (TOAD) shown in Figure 11. Switching is achieved by placing an offset from the center of an optical fiber loop mirror and injecting data into the loop via a 5:5 coupler. The two counterpropagating data pulse streams arrive asynchronously at the. A switching pulse is timed to arrive after one data pulse but just before its replica. Incoming data packets Guard band F5 Figure hybrid switch module. Input fibers Polymer waveguides The switching pulse power is adjusted to impart a phase change of p radians onto the replica, so the data pulse is switched out when the two counterpropagating components interfere on their return to the coupler. The TOAD can also be used to demultiplex high-speed time division multiplexed pulse streams. All-optical wavelength converters will play an important role in broadband optical networks. Their most important function will be to avoid wavelength blocking in optical cross-connects in WDM networks. Wavelength converters increase the flexibility and capacity of a network using a fixed set of wavelengths and can be used to centralize network management. In packet switching networks, tunable wavelength converters can be used to resolve packet contention and reduce optical buffering requirements. Wavelength conversion in s can be achieved using XGM, cross-phase modulation (XPM), or FWM. The design of an wavelength converter based on XPM requires that one or more s be incorporated into an interferometer, an example of which is the Mach Zehnder interferometer (MZI) shown in Figure 12. An input data signal at wavelength l 1 is used to modulate the upper refractive index, which controls the phase shift experienced by a second unmodulated input signal at l 2 in one of the interferometer arms. When the input data signal is low (logic ), the arms are in phase and the l 2 signal appears at the top output. When the input array Output fibers Data pulse 1 Data pulse 2 Pulse to be switched Routed packets Input data pulses (l 2 ) Centre x 5:5 coupler Data and switching pulses out Switching pulses (l 2 ) P155 P16 F55 Figure 11 Ultrahigh speed optical switching using a TOAD.

8 8 OPTICAL AMPLIFIERS / Semiconductor Optical Amplifiers Input data l 1 l 2 Unmodulated input l 2 l 2 Converted outputs F6 Figure 12 Mach Zehnder wavelength converter. S4 P165 pump signal is high (logic 1) it produces an extra p phase shift between the two arms, causing them to be totally out of phase and the l 2 signal to appear at the bottom output. Both outputs contain a copy of the original data signal imposed onto the new wavelength. Up and down conversion is possible. The lower is used to equalize the gain experienced by the l 2 signal in each of the arms. An optical filter is required at the output to remove l 1. Such wavelength converters can operate at very high bit rates (.1 Gb/s). An important advantage of the MZI structure is that it also gives 2R regeneration (re-amplification and reshaping) of the input data signal with wavelength conversion. This is because the interferometer nonlinear response increases the converted signal extinction ratio compared to the input data. All-optical 3R (2R þ retiming) regenerators are of great interest in future optical communication networks to restore degraded transmission signals. -based interferometric structures can be used to achieve 3R (2R þ retiming) regeneration at data rates.8 Gb/s. See also Coherent Lightwave Systems (668). Lasers: Semiconductor Lasers (91). Optical Amplifiers: Basic Concepts (661); Erbrium Doped Fiber Amplifiers For Lightwave Systems (666); Optical Amplifiers for Long Haul Transmission Systems (672); Optical Communication Systems: Optical Time Division Multiplexing (674); Wavelength Division Multiplexing (673); Semiconductor Physics: Band Structure and Optical Properties (617). Further Reading Agrawal GP and Olsson NA (1989) Self phase modulation and spectral broadening of optical pulses in semiconductor laser amplifiers. IEEE Journal of Quantum Electronics 25: Connelly MJ (22) Semiconductor Optical Amplifiers. Boston: Kluwer Academic. Desurvire E (1994) Erbium-Doped Fiber Amplifiers: Principles and Applications. New York: Wiley. Durhuus T, Mikkelsen B, Joergensen C, Lykke Danielsen S and Stubkjaer KE (1996) All-optical wavelength conversion by semiconductor optical amplifiers. Journal of Lightwave Technology 14: Ghafouri-Shiraz H (1995) Fundamentals of Laser Diode Amplifiers. New York: Wiley. Shimada S and Ishio H (1992) Optical Amplifiers and their Applications. New York: Wiley.