>> A Brief Introduction to Optical Amplifiers
Because fiber attenuation limits the reach of a nonamplified fiber span to approximately 200 km for bit rates in the gigabit-per-second range, wide area purely optical networks cannot exist without optical amplifiers.
Optical amplifiers are typically used in three different places in a fiber transmission link.
Power amplifiers serve to boost the power of the signal before it is launched on the line, extending the transmission distance before additional amplification is required.
Line amplifiers are located at strategic points along a long transmission link to restore a signal to its initial power level., thereby compensating for fiber attenuation.
Preamplifiers raises the signal level at the input of an optical receiver, which serves to improve signal detection performance (i.e., the receiver sensitivity).
In each of the three cases, the desired properties are different. For power amplifiers, the important feature is high gain; preamplifiers require a low noise figure, and line amplifiers require both.
Optical amplifiers are also employed at various other points in a network (for example, within an optical switching node to compensate for losses in the switch fabric).
Semiconductor Optical Amplifiers (SOA) were developed in the 1980s but they never had a serious impact on long-distance transmission because of a number of negative features. In the case of fiber amplifiers, especially the EDFA (Erbium Doped Fiber Amplifier) and the RA (Raman Amplifier), however, the situation was quite different.
The first papers on EDFAs appeared in 1987. Within a few years of that time, 9000 km unrepeatered transmission was demonstrated. Shortly thereafter, soliton experiments showed that transmission distances could be extended almost indefinitely. All of these experiments used EDFAs. It is not an exaggeration to say that these devices have revolutionized optical communications.
Although the EDFA played a fundamental role in extending the reach of optical transmission systems it still had some drawbacks, including operation confined to a limited band of the optical spectrum and a nonflat gain profile. In contrast, Raman Amplifiers (RA), which were first demonstrated well before the EDFA and then virtually ignored for three decades, have more recently attracted renewed interest. This stems mainly from their ability to increase both the reach and the aggregate bit rate carried on a fiber; that is, the usable fiber bandwidth.
>> Erbium-Doped Fiber Amplifiers (EDFAs)
The EDFA belongs to a family of rare-earth-doped fiber amplifiers, the class of other possible dopants, including praseodymium (used for amplification in the 1300-nm range), neodymium (originally used for very high-power lasers), ytterbium (which has been used as a codopant with erbium), and thulium (amplifying in the S band). The important place of the EDFA in optical communications is due primarily to the fact that the properties of erbium produce amplification in a fairly wide band (approximately 35 nm) within the 1550 nm low-attenuation window in fibers. Furthermore, the EDFA has many other desirable features.
1. EDFA Module Structures
Three different EDFA structures are shown in the following figure.
In each of the above cases, the amplifier is of the travelling wave type, consisting of a strand of single-mode fiber, typically on the order of tens of meters long, doped with erbium. (The points S in the figure represent fiber splices.)
The EDFA is an optically pumped device, so energy is supplied by an optical source (Laser Diode), which injects power into the doped fiber at a wavelength matched to the characteristics of erbium (980 or 1480 nm). Pumping can be forward, backward, or bidirectional. The pump is typically coupled into the transmission fiber via a wavelength-selective coupler (WSC). Amplifications occurs by transfer of power from the pump wave to the signal wave as it propagates down the doped fiber.
Note that EDFA modules used in the field typically include other components, such as optical isolators to eliminate reflected power, and various devices for signal power monitoring, stabilization, and control.
2. EDFA Three-Energy Level System
Like many other forms of amplifiers of electromagnetic radiation, the EDFA operates via a three-energy level system. The model representing this process is shown in the following figure.
Levels E1, E2, and E3 are the ground, metastable, and pump levels, respectively. The populations (fractional densities) of erbium ions in the three energy levels are denoted N1, N2, and N3, where N1 > N2 > N3 when the system is in thermal equilibrium (no pump or signal present). When pump and signals are present, these populations change as ions move back and forth between levels, accompanied by the emission or absorption of photons at frequencies determined by the energy-level difference.
The wavelengths associated with the dominant transitions are indicated in the above figure. The wavelength λ for each transition is given by the quantum relation λ = hc /ΔE, where h is the Planck’s constant and ΔE is the difference in energy levels. In actuality, the three levels in the simplified diagram are narrow bands, so each transition is actually associated with a band of wavelengths rather than a single line.
Two pump wavelengths are typically used for EDFAs: 980 and 1480 nm. As shown in the above figure, by absorbing energy from a 980 nm pump, Er3+ ions in the ground state are raised to state E3. The rate at which these transitions occur is proportional to N1Pp, where Pp is the pump power. These excited ions decay spontaneously to the metastable state E2, and this transition occurs at a rate much faster than the rate from level E1 to level E3. This means that in equilibrium under the action of the pump, the ion population in the ground state is reduced and accumulates largely in state E2. This process is referred to as population inversion because we now have N2 > N1, the reverse of the situation in thermal equilibrium.
The transition rate from level E2 to level E1 is very slow compared with the other transitions, so that the lifetime τ, in the state E2 (the reciprocal of its transition rate to E1) is very long (approximately 10 ms). Similar pumping action can occur at 1480 nm, in which case the ions are raised directly to the upper edge of the E2 band. reliable semiconductor laser pump sources have been developed for EDFAs at both the 980 and 1480 nm pump wavelengths.
The wavelength band for transitions from state E2 to the ground state is in the 1530 nm range, making it ideal for amplification in the lowest attenuation window of fibers. The dominant transitions from E2 to E1 are radiative, which means that they are of two types: spontaneous emission and stimulated emission.
In the case of spontaneous emission , an ion drops spontaneously to the ground state, resulting in the emission of a photon in the 1530 nm band, and this appears as additive noise. Spontaneous emission noise is an unavoidable by-product of the amplification process, predicted by quantum theory. Its phase, direction, and polarization are independent of the signal.
In the case of stimulated emission, an incident photon in the 1530 nm range stimulates the emission of another photon at the same wavelength in a coherent fashion (with the same direction, phase, and polarization). If the incident photon is from a signal, this produces the desired amplification of the optical field. However, the incident photon could also have originated as a spontaneous emission “upstream” on the fiber, in which case this is called amplified spontaneous emission (ASE), which represents the major source of noise in amplified fiber transmission systems.
3. Gain Profile of EDFA
The fairly large amplification bandwidth of the EDFA is due to the finite width of the energy bands. The width of the energy bands is caused by a number of physical phenomena, including the Stark effect, which splits the main energy levels in to many sublevels. Because the population is not distributed uniformly within the E2 band, the gain is not flat.
A typical plot of gain as a function of wavelength is shown in the following figure.
The uneven gain profile, with a peak at approximately 1530 nm, produces significant problems in a multiwavelength system when many amplifiers are cascaded over a long transmission span. Not only does uneven gain amplify different wavelengths unequally, but it also causes a large accumulation of ASE at the peak of the gain profile, which can eventually saturate the amplifier.
Because amplifier cascading on long links accentuates these effects seriously, gain flattening is an important consideration in EDFAs. Several solutions to this problem are currently in use. One approach is to modify the design of the amplifier itself by using different materials such a fluoride glass. Other approaches use gain equalization via controllable attenuators or inverse filtering.
4. Gain Saturation
A. Small-Signal Gain
The gain of an EDFA is approximately independent of the signal power as long as the pump power is made high enough so that the pumping rate is much larger than the stimulated emission rate. This is called the unsaturated gain or small-signal regime. The small-signal gain under these conditions is an increasing function of pump power.
For a given fiber structure and doping, and a given pump power, these is an optimal fiber length that maximizes gain. For lengths smaller than the optimum, the pump power is not maximally utilized, and for larger lengths, pump power is exhausted somewhere along the fiber, and attenuation takes over. Typical optimal lengths are in the range of tens of meters. Maximum small-signal gains for EDFAs are typically 30 to 40 dB.
B. Gain Saturation
All amplifiers eventually exhibit gain saturation as the signal power is increases. In the saturated case, the signal extracts so much power from the pump as it propagates down the fiber that the stimulated emission rate becomes comparable with the pumping rate. The larger the input signal, and the higher the unsaturated gain, the sooner saturation is reached.
As saturation increases, the gain decreases. The saturation output power Psatout is defined as the output power at which the gain is compressed by 3 dB. The values of Psatout for typical EDFAs are in the hundreds of milliwatts. It should be noted that ASE also contributes to saturation in an EDFA . When input signals are very small, it is the ASE that saturates the amplifier first. This is known as amplifier self-saturation.
Because saturation is a nonlinear effect, it produces a number of complications when multiple signals are being amplified. One problem is that the saturated gain for any one signal depends on the aggregate power of the other signals as well as its own power. Thus signals (as well as accumulated ASE) tend to “steal” power each other. An advantageous effect of saturation is that a small amount of it in each amplifier in a cascade of several amplifiers tends to produce a self-regulating effect.
Several other nonlinear effects are a consequence of this power-stealing phenomenon but on a shorter time scale. The amplifier gain at any instant in time is a function of the excited state population N2, which is depleted momentarily by stimulated emission when a signal is present. One manifestation of this occurs when an intensity-modulated digital signal changes from a 0 to a 1. The resultant fluctuation in N2 causes corresponding gain fluctuations, which are most pronounced in the saturated regime and in the presence of large signals. Another manifestation occurs when beats from two signals spaced closely in optical frequency cause gain fluctuations at the beat (difference) frequency.
The gain fluctuations affect all signals being amplified and thus can potentially produce undesirable cross-talk, with one signal’s intensity fluctuations changing the gain for the others. These effects are significant only when the gain dynamics are such that gain can vary on a time scale as fast as that of the signal fluctuations. A simplified interpretation of gain dynamics in an EDFA is based on the assumption that the maximum speed for gain fluctuations is on the order of the reciprocal of the lifetime in the excited state, which is approximately 10 ms. However, actual gain transients in EDFAs can occur on time scales of hundreds of microseconds, which cannot be predicted using the lifetime alone.
In any case, these numbers indicate that signals fluctuating on time scales more rapid than, say, 100 us will cause no significant cross-talk in EDFAs. This corresponds to a minimum bit rate of approximately 10 Kbps to avoid cross-talk (or a WDM signal separation of approximately 10 KHz to avoid beat frequency effects). The lack of this cross-talk effect for bit rates higher than 10 Kbps is one of the important advantages of the EDFA over the SOA.
5. Noise and Noise Figure
The ASE noise generated in an EDFA can be the limiting performance factor in an optical transmission link. It is therefore important to quantify this effect.
For an amplifier with gain G, the ASE noise power spectral density at the output at optical frequency ν (in each polarization state ) is
where nsp, the spontaneous emission factor, is a function of the state population and approaches its minimum value of 1 with full population inversion. The ASE noise spectrum for an EDFA is roughly the same shape as the gain profile.
The significance of the ASE noise is most clearly expressed in terms of SNRs and the amplifier noise figure Fn. These quantities are defined in terms of electrically detected signals in an ideal system, as shown in the following figure.
The noise figure is defined as
where SNRin is the electrical SNR seen when a signal of power Pin is converted to a photocurrent at the output of an ideal photodetector (PD). The noise in this case is shot noise due to the fact that the ideal detector is counting photons, which arrive randomly at the detector. (The detection process must be an integral part of any noise calculation, reflecting the quantum limits of lightwave transmission.) The numerator in the above equation is given by
where (RPin)2 is the square of the average photocurrent, σ2 = 2qRPinΔf is the shot noise power (the variance of the photocurrent), R=q/hν is the responsivity of an ideal detector, q is the electron charge, and Δf is the bandwidth of the electrical detector.
The quantity SNRout is the electrical SNR seen with the amplifier inserted before the photodetector. To find SNRout, we compute the variance of the photocurrent after amplification with gain G. Because the detector acts as a square-law device, the photocurrent variance contains terms due to shot noise and ASE noise by themselves, as well as signal-spontaneous emission beat noise because of the mixing between the signal and the ASE in the photodetector. It turns out that that latter is the dominant term, provided that G >> 1, and most of the ASE noise is filtered out at the input of the detector. This can be done by making Δf small enough to exclude extraneous noise but include the desired signal. Then we have
Using the above equations, we get
which corresponds to at least a 3-dB SNR degradation in the high-gain case. (In real systems, Fn is typically at least 4 dB.)
6. Amplifier Chains
Over a long transmission link, it is necessary to use several EDFAs interconnected by fiber sections to compensate for fiber attenuation. The gain of each amplifier is normally adjusted so that it compensates for the attenuation on one section of fiber. The question of optimal amplifier spacing then arises. It turns out that this is a fairly complex issue that depends, among other things, on the way in which the amplifiers are pumped, effects of fiber nonlinearities, and practical issues such as amplifier accessibility, cost, and so forth.
We examine a fairly simple model here, in which a fiber of length L is divided into N sections of spacing s = L/N. An amplifier is placed after each section, with a saturated gain that just compensates for the fiber attenuation on one section: G = eαs. The total accumulated noise power spectral density at the end of this chain (taking into account both polarization states) is then
Note from the above equation that for a fixed amplifier spacing the effect of accumulated noise in the cascade grows linearly with the length of the link but decreases as the amplifier spacing decreases (i.e., as the number of amplifiers increases). Thus, the optimal strategy in this case is to place a very large number of low-gain amplifiers very close together, with the limiting case being one long, distributed amplifier. Cost, however, dictates the opposite strategy! In current practice, a compromise is reached, with spacings ranging from 20 to 100 km, typically giving an SNR at the receiver of at least 15 dB. The spacings are based on constraints such as maximum permissible power on a fiber, effects of fiber nonlinearities, and receiver sensitivity.
>> Raman Amplifiers (RA)
1. How Raman Amplifier Works
The discussion of the EDFA provides a useful framework for describing the Raman amplifier: They are both fiber amplifiers, with important similarities as well as differences, so they can often complement each other in applications.
Stimulated Raman scattering (SRS) can cause transmission impairments in fibers, but it can also be used for amplification. When SRS is used for amplification, pump power is introduced into a fiber carrying an optical signal, with the pump operating at a frequency higher than the signal frequency, just as in the EDFA (and other rare-earth-doped fiber amplifiers).
The pump photons interact with the material in the fiber through inelastic collisions, producing scattered photons at lower energy (and frequency) than the pump photons, with the remaining energy imparted to the fiber medium in the form of vibrational waves, called optical phonons. If the frequency of the scattered photon is the same as that of a signal photon propagating in the fiber, it can stimulate the emission of a second signal photon, thereby amplifying the signal, a process identical to that which occurs in the EDFA.
The performance of the RA can be expressed in terms of a Raman gain coefficient (RGC). An illustration of the form of a normalized RGC as a function of frequency shift between the pump and signal appears in the following figure.
As the figure shows, RA gain is polarization dependent. The gain coefficient for copolarized pump and signal waves is an order of magnitude higher than in the orthogonally polarized case. Polarization dependence is mitigated by the averaging effect of the polarization mode dispersion in the fiber medium and can be circumvented by using either polarization diversity pumping or a single depolarized pump.
An important difference between the RA and the EDFA is that the energy levels of Er3+, which determine the gain profile of the EDFA are fixed, thereby fixing the position of the amplification band of the device, as well as the possible pump frequencies. The amplification band for the EDFA is fixed in the vicinity of 1530 nm – the middle of C-band – which is a primary reason for its importance in optical communication but which limits its flexibility in exploiting other transmission bands in optical fibers.
In contrast, for the RA it is only the pump/signal frequency difference (a band centered around 13 THz) that is fixed by the physics of the process, and any pump frequency can be used. Changing the pump frequency automatically shifts the waveband where amplification occurs. Thus the amplification band of an RA can be centered at any desired frequency in the optical fiber transmission window by adjusting the pump frequency appropriately. Furthermore, for a single pump the amplification bandwidth is large (about 6 THz), and this band can be extended by superimposing several pumps at different frequencies. This makes the Ra an excellent tool for widening the usable bandwidth of long-haul WDM transmission systems beyond C- and L-bands into the S- and U-bands and beyond.
2. Raman Amplifier Configurations
The RA can be configured either as a distributed or discrete (lumped) amplifier. A typical distributed RA (DRA) consists of a long transmission fiber into which a counterpropagating (backward) Raman pump is injected. (Backward pumping reduces the effect of pump noise, as explained below.) The distributed amplification results in reducing the perceived loss along the span, which effectively improves the reach of the span and/or increases its capacity.
In a discrete RA, the amplifier consists of a coil of fiber together with pump(s) and ancillary equipment for monitoring, control, and perhaps other purposes such as dispersion compensation, gain flattening, or adding and dropping channels. Isolators are used to keep the pump power from escaping into the line. The fiber medium used in the discrete case is shorter than in the DRA, but it still is typically of the order of kilometers – two orders of magnitude longer than the EDFA. A significant advantage of the discrete RA is that the amplifying fiber can be chosen at will to suit a number of criteria. For example, a dispersion compensating fiber can be used to provide dispersion compensation for the transmission fiber, with the additional benefit of improving the Raman gain coefficient. The primary purpose of discrete RAs is generally to expand the usable bandwidth of a transmission link, whereas the primary purpose of a DRA is to improve the reach of a fiber span.
When several amplified spans are placed in tandem, with lumped line amplifiers placed at the junction points, the result is a hybrid arrangement as shown in the following figure.
The advantage of this arrangement is illustrated by a comparison of signal powers along the line with and without the DRAs (see the following figure).
Without the DRAs the signal level drops linearly along each span. Due to the overall span loss a high signal power must be launched into each span, which tends to produce nonlinear impairments at the beginning of the span. But at the far end of the span, the attenuation drops the signal into the noise level. Clearly the span is too long for discrete amplification alone. However, by adding distributed amplification throughout the span, the signal initially attenuates at almost the same rate as without RA but then increases in power toward the far end of the span as it encounters stronger pump power. The DRA pump signal power is also shown, decreasing from right to left in the figure, which is why distributed gain is highest toward the far end of the span. The net effect of the distributed amplification in the spans is to improve the overall system performance by reducing noise as well as nonlinear effects.
Distributed amplification keeps the signal above the noise level at the far end of each span, so the optical SNR at the input of each line amplifier is improved. Furthermore, distributed amplification makes it possible to launch the signal into each span at a lower power level, thereby reducing nonlinear impairments due to high signal levels.
3. The Good and Bad of Raman Amplifiers
There are a number of additional considerations that work for and against the RA. On the positive side, it operates in ordinary silica fibers, requiring no special materials or dopants. This makes it ideal as a means of adding distributed amplification to existing long transmission links. Furthermore, it has better ASE noise properties than the EDFA. RA acts like an EDFA with full population inversion.
However, there are additional sources of noise in RAs that can be more serious than ASE: in particular, multipath effects caused by reflections and double Rayleigh scattering. Rayleigh scattering causes forward propagating signals (or noise) to be scattered backward, but when a signal encounters this phenomenon twice, the doubly scattered signal propagates in the forward direction, recombining with the original signal after a multipath delay. Discrete double reflections due to imperfections, splices, and connectors in the fibers cause similar multipath effects.
Because the Raman effect is weak, long fibers are required in RAs, which tend to increase the multipath effects. Unintended reflections and Rayleigh scattering are present in all fiber systems, but they are attenuated in a passive fiber. However, when the fiber is pumped the Raman gain magnifies these effects to the point where the multipath interference places a limit on the usable gain in an RA.
Another drawback of the RA is that it has a very fast response to pump fluctuations. This can lead to coupling of pump noise into the amplified signals. These effects can be mitigated by using a counter-propagating pump, in which case the effects of the pump fluctuations are averaged out over the length of the pumped fiber. Otherwise, they require the use of “quiet” pumps; i.e., pumps with very low relative intensity noise.
In deploying Raman amplifiers as discrete amplifiers, there are some other practical concerns due to the high pump powers employed. Connectors should be minimized in favor of splices to reduce reflections and attenuation, and when connectors are required they must be designed to survive the high pump powers. Also, to protect personnel, automatic laser shutdown systems must be employed.
4. The Efficiency of Raman Amplifiers
RAs are normally less efficient than EDFAs in converting pump power to output signal power. However, their efficiency improves, exceeding that of the EDFA, at the large aggregate signal powers that occur in long-haul WDM systems with high channel counts. Furthermore the gain in the fiber medium depends strongly on the type of fiber being used. Because gain is proportional to pump intensity, it increases when a given amount of pump power is confined to a small fiber core. Thus, fibers with smaller cores such as Dispersion Shifted Fibers (DCFs) produce significantly higher Raman gain. This is a particular advantage in discrete Raman Amplifiers, where there is some choice in the type of fiber being used.
>> Semiconductor Optical Amplifiers (SOA)
1. How Semiconductor Optical Amplifier (SOA) Works
The structure of an SOA is similar to that of a semiconductor laser. It consists of an active medium (a p-n junction) in the form of a waveguide, with a structure much like the stripe geometry laser. The mobile carriers (holes and electrons) now play the role of the Er3+ ions in the EDFA.
The energy levels of the electrons in a semiconductor are confined to two bands: the conduction band, containing those electrons acting as mobile carriers, and the valence band, containing the nonmobile electrons. A hole, representing the absence of of an electron in the valence band, also acts as a mobile carrier. Mobile electrons and holes are abundant (i.e., are majority carriers) in n-type and p-type material, respectively.
The two energy bands in semiconductors play a role analogous to band E2 and E1 in the EDFA, but they are much broader than the EDFA bands. A band gap, Eg, separates the lower edge of the conduction band from the upper edge of the valence band so that the energy change involved in moving from one band to the other is at least Eg. Transfer of an electron from the valence band to the conduction band (with the absorption of energy) results in the creation of an electron-hole pair. One way in with this occurs is through the absorption of a photon, as in a photodetector. The reverse phenomenon, electron-hole recombination (with release of energy), occurs either nonradiatively (by transferring energy to the crystal lattice) or radiatively, with the emission of a photon.
The radiative case is of interest to us here for applications in light sources as well as amplifiers. Radiative electron-hole recombination occurs either spontaneously or through stimulated emission involving interaction with an identical photon. These two processes are analogous, respectively, to the spontaneous and stimulated emission processes in an EDFA. By proper choice of the semiconductor materials (e.g., InGaAs or InGaAsP), bandgaps that yield emission and/or absorption wavelengths in the ranges desired for optical communications (e.g., 1300 or 1550 nm) can be produced.
For photon emission to occur by electron-hole recombination at an optical frequency, ν, an electron-hole pair must be present with energy levels separated by an amount ΔE = hν. Furthermore, if the recombination is by stimulated emission, a photon of the same frequency must be present to interact with the electron-hole pair. The conditions for these effects to occur depend on the various carrier concentrations and the photon flux in the active region (the layer around the p-n junction).
In an unbiased p-n junction, a “depletion layer” exists around the junction caused by diffusion of majority carriers across the junction and subsequent recombination on the other side. This creates a net charge on each side of the junction and hence a retarding electric field, preventing further diffusion and draining carriers from the layer around the junction. The depletion layer can be broadened by reverse-biasing the junction, thereby augmenting the retarding field. This is the condition for operation of the p-n junction as a photodetector.
On the other hand, by forward-biasing, the retarding field is reduced, allowing more majority carriers to cross the junction, becoming minority carriers on the other side. This creates a condition favorable to recombination in the active region because once the mobile electrons from the n side cross over to the p side (at which point they become minority carriers), they encounter a large concentration of holes with which to recombine. A similar situation occurs for the mobile holes moving in the opposite direction. This effect, which increases the population of minority carriers in the active region on each side of the junction, is called minority carrier injection.
The current flow through the forward-biased junction acts as an electrical pump, supplying the energy necessary to produce an inversion of the carrier population in the active region. This is analogous to the Er3+ ion population inversion in the EDFA produced by optical pumping. The light-emitting diode (LED) is a simple application of radiative recombination. It is a forward-biased p-n junction producing its radiation by spontaneous emission. This effect is called injection electroluminesence.
Now suppose an optical signal is introduced into a waveguide embedded in a forward-biased p-n junction, which we now want to use as an amplifier. By applying sufficient injection current, conditions can be established in which stimulated emission dominates spontaneous emission and absorption in the guide. At this point, optical gain is produced, and the device becomes a semiconductor amplifier. Because the energy bands are broad in a semiconductor, the SOA amplifiers cover a much wider band than an EDFA.
2. The Good and Bad of Semiconductor Amplifiers (SOA)
Although its broadband gain characteristic is a positive feature, the SOA has a number of negative features.
First, the carrier lifetime in the high-energy state is very short (on the order of nanoseconds). As indicated earlier, this means that signal fluctuations at gigabit-per-second rates cause gain fluctuations at those rates, producing cross-talk effects between simultaneously amplified signals. These effects do not occur in EDFAs until the bit rate drops into the 10 Kbps range.
Second, because of its asymmetrical geometry, the SOA is polarization dependent. The EDFA, with its cylindrical geometry, is not.
Third, the coupling losses between the fibers and the semiconductor chip reduce substantially the usable gain and output power.
Fourth, the noise figure of a typical SOA is slightly higher than that of a typical EDFA due to fiber-chip coupling losses, although advances in packaging technology have improved that.
Because of recent improvements in broadband SOAs, polarization-dependent gain (PDG) and noise figure (rather than gain flatness and saturation-induced cross-talk) are becoming the predominant limiting performance factors. The best commercial SOAs can be specified having PDG as low as 0.5dB over the C-band (30 nm bandwidth). However, CWDM-capable SOAs typically exhibit PDGs of 1 dB or more over a 70 nm band.