Optical Ampliﬁers for Access and Passive Optical Networks: A Tutorial

: For many years, passive optical networks (PONs) have received a considerable amount of attention regarding their potential for providing broadband connectivity, especially in remote areas, to enable better life conditions for all citizens. However, it is essential to augment PONs with new features to provide high-quality connectivity without any transmission errors. For these reasons, PONs should exploit technologies for multigigabit transmission speeds and distances of tens of kilometers, which are costly features previously reserved for long-haul backbone networks only. An outline of possible optical ampliﬁcation methods (2R) and electro/optical methods (3R) is provided with respect to speciﬁc conditions of deployment of PONs. We suggest that PONs can withstand such new requirements and utilize new backbone optical technologies without major ﬂaws, such as the associated high cost of optical ampliﬁers. This article provides a detailed principle explanation of 3R methods (reampliﬁcation, reshaping, and retiming) to reach the extension of passive optical networks. The second part of the article focuses on optical ampliﬁers, their advantages and disadvantages, deployment, and principles. We suggest that PONs can satisfy such new requirements and utilize new backbone optical technologies without major ﬂaws, such as the associated high cost.


Introduction
Passive optical network (PON) technologies find their major deployment in access networks [1][2][3][4][5][6][7] owing to their low requirements on optical distribution networks (ODNs), such as single and shared optical fibers between customers and the central office (CO).
This technique uses point-to-multipoint (P2MP) shared infrastructure, but it should be noted that a shared fiber means some limitations on the customer's side, such as shared bandwidth, and upstream transmission must be secured with another control mechanism [8][9][10][11][12][13]. Passive optical networks are able to transmit signals from the optical line terminal (OLT) to optical network unit(s) (ONUs) up to 20 km, but in some cases, this distance limitation has to be broken or extended due to extensions of signal transmission in rural areas, remote offices, remote cities, etc. For these purposes, standardization organizations, such as the International Telecommunication Union (ITU) or Institute of Electrical and Electronics Engineers (IEEE), proposed PONs with longer reach [14][15][16][17][18][19]. Furthermore, the extended reach networks require optical amplifiers to extend the distance between the OLT and ONUs [20][21][22][23][24][25][26][27][28][29][30][31][32]. In the following sections, the methods for reach extensions are discussed.
Optical fiber amplifiers were invented back in 1964, three years after the first fiber laser was developed by Elias Snitzer and his colleagues. Both the first laser and amplifier used neodymium as three categories: 1R, 2R, and 3R. While the current research interest is full optical amplifiers, we discuss all three categories due to the potential usage of 3R amplifiers in xPONs [43][44][45][46][47][48][49].
The main signal degradation in fiber optic systems arises from amplified spontaneous emission (ASE) due to optical amplifiers, pulse spreading due to group velocity dispersion (GVD), which can be corrected by passive dispersion compensation schemes, and polarization mode dispersion (PMD). Nonlinear distortions are attributed to Kerr nonlinearity, such as cross-phase modulation, which can be responsible for time jitter in wavelength division multiplexing (WDM), or Raman amplification, which can induce channel average power discrepancies [50].
The 1R category represents the simplest amplifier of an optical signal. Only the input signal is amplified and transferred to the output. Note that an input signal is not recovered (the shape, position, and phase are exactly the same as those of the input signal). However, 1R amplifiers are simple, which presents some advantages. For example, a processed signal does not depend on the modulation format, transmission speed, or other parameters of a signal. The basic principle of 1R amplifiers is shown in Figure 1. The input signal is degraded, but the output signal is only amplified because 1R amplifiers do not consider the shape and timing of the input signal; they only consider amplification. All known optical amplifiers can be placed in the 1R category.  The second category of R amplifiers works more complexly with an input signal because they are based on the 1R category and add reshaping of an input signal. The shape of a carried signal is degraded with increasing distance from the transmitter side. We consider optical networks and take into account the attenuation of optical fibers. We cannot eliminate the attenuation of optical fibers because we are not able to produce clean silica fibers without admixtures and impurities (additional details about optical fiber manufacturing are provided in [51,52]). The standard attenuation values of the fibers are 0.35 and 0.22 dB/km for 1310 and 1550 nm, respectively. Other important factors are dispersion (additional details about dispersions are provided in [53][54][55]). In general, dispersion causes a carried signal to become deformed in the fiber and spread in the time domain, which produces a range restriction by decreasing the signal-to-noise ratio (SNR), transmission speeds, and improperly logical 0 or 1 decision in the receiver. A 2R amplifier is referred to as a regenerator. A regenerator has an optical signal at an input port, which is converted into an electrical signal; decisions are subsequently made. A decision entails recognition of logical 0 and 1 of the input signal. The signal is subsequently transferred to a transmit circuit. The transmitting circuit converts the signal to the optical domain and transfers to a fiber path. Note that the output signal has recovered its shape and has a higher power level (was amplified), but timing recovery does not occur (the positions of the signal samples are unchanged); refer to Figure 1.
The 3R amplifier adds time synchronization to the basic principle of 2R. The 3R amplifier converts the input signal from the optical domain to the electrical domain, amplifies it and reshapes it. A clock rate is recovered and reconstructed before sending a time position (for example, by a comparator). This output signal is equivalent to the original signal that was transferred to the fiber. Figure 2 shows the principle of 3R amplifiers, and Figure 3 shows the block scheme of a reach extended passive optical network (RE-PON). Regeneration of 3R can occur in two ways: Inline 3R regeneration and in-node regeneration. Inline 3R regeneration is usually implemented when the physical distance between the end points exceeds the maximal power budget of the optical network. In-node regeneration can occur in the optical cross-connect nodes, where some OEO regenerators are usually deployed [56].
Note that OEO 3R regenerators are dependent on the signal waveform (modulation formats). If the waveform is changed, the 3R regenerator must be adapted to it. A second significant limitation of 3R regeneration is the bit rate. The maximal bit rate for OEO 3R regenerators is approximately 40 Gb/s. Both problems are solved in all-optical 3R regenerators.
The standard for a GPON optical reach extension was ratified by ITU-T G.984.6 in 2008. This standard includes the architecture and interface parameters for GPON systems with extended reach using a physical layer mid-span extension between the OLT and the ONU that uses an active device in the remote node. The GPON reach extender enables operation over a maximum of 60 km of fiber with a maximum split ratio of 1:128 [48]. Two ways to amplify a signal are presented in ITU-T G.984.6. The first method is based on optical amplification of the optical signal: Bidirectionally. This principle is based on 1R regeneration. This kind of amplifier can be based on an EDFA, Raman amplifier or semiconductor optical amplifier (SOA).
The second approach is to use an OEO regenerator, as shown in Figure 2. The regenerator consists of a couple of branches for each way using diplexers. In both branches, the receiver and the transmitter are dimensioned for the wavelength band, which explains why the optical signal must be converted to an electrical signal. The electrical signal is recovered and converted to the optical domain. The important function of this part is to recover the clock signal. This step is resumed by the receiver downstream-continual mode-but upstream, the burst mode is used. ITU-T G.984.6 also considers the combination of both systems, e.g., the OEO regenerator downstream and the SOA amplifier upstream. All-optical 2R is also possible; however, it is not transparent to modulation of the input signal [57]. Full optical 3R regeneration is not considered in standardized PONs but is suggested for future networks [58].
Full optical 3R regeneration with a real function of retiming requires clock recovery, which can be achieved either electronically or all-optically. The main difference between both types of retiming is that electronic functions are narrowband compared with broadband optical clock recovery [59]. Full optical 3R regeneration can be realized in two different ways:

1.
Data-driven 3R regenerator-nonlinear optical gate. This scheme mainly consists of an optical amplifier, that is, a clock recovery block providing an unjittered short pulse clock stream, which is then modulated by a data-driven nonlinear optical gate block [50].

2.
Synchronous modulation 3R regenerator-this technique is particularly efficient with pure soliton pulses. It consists of combining the effects of a localized "clock-driven" synchronous modulation of data, filtering, and line fiber nonlinearity, which results in both timing jitter reduction and amplitude stabilization (see Figure 4). The high-dispersion fiber first converts the amplified pulse into a pure soliton. The filter blocks the unwanted ASE but also has an important role in stabilizing the amplitude in the regeneration span. Data are then synchronously and sinusoidally modulated through an intensity or phase modulator, driven by the recovered clock [50].

Optical Amplifiers in Telecommunications Networks
Optical amplifiers are an essential part of any optical transmission system and are not limited to long-haul systems, such as submarine systems. There are excellent books that address optical amplifiers, for example, [60][61][62], which is used as the basic reference for the following paragraphs. For 1 Gb/s and 10 Gb/s transceivers, the maximum fiber distance is usually 80 km; some transceivers can reach 120 km. While 80 km distances can be overcome without any correction control, for longer distances, forward error correction (FEC) mechanisms must be implemented. This situation changed with the emergence of coherent systems in 2008 [63]. In 2019, coherent systems with maximum transmission rates of 200 Gb/s are very common, and those with rates of 400 Gb/s are also available; however, these systems are expensive, with the optical reach limited to a few hundred kilometers. A new generation of silicon electronics-digital signal processing (DSP)-will be able to increase this rate to 600 Gb/s, with the potential to extend the optical reach to 400 km compared with the current situation.
The first concepts of optical amplifiers were introduced in the early 1960s, and the first optical amplifier was invented in 1964 by Professor E. Snitzer, who used neodymium and worked in a 1060 nm spectral window. Professor Snitzer also demonstrated the first erbium glass laser. Other experiments with neodymium followed in 1970, but it was too early for real deployment. These principles were also applied for the first single-mode fibers in early 1980 at Bell Laboratories. Erbium was used for amplification at the University of Southampton and AT&T Bell Laboratories in 1985. The key advantage was the capability of erbium to work at 1550 nm-the most important part of the spectrum in silica fibers [60].
Optical amplifiers are referred to as all-optical (OOO) compared with OEO regenerators. We note that optical amplifiers are referred to as "regenerators" in the submarine world, which may be confusing for readers from the terrestrial telco world.
The main advantage of optical amplifiers is that one device is able to amplify many optical signals at once. This feature is in sharp contrast with OEO regenerators, where one regenerator can be used for only one signal and expensive multiplexing and demultiplexing techniques are necessary.
Optical amplifiers amplify optical signals by stimulated emission; this mechanism is the same mechanism used for lasers. Optical amplifiers are sometimes described as lasers without feedback. An optical amplifier is pumped (fed with energy) optically or electrically to achieve population inversion of the dopant elements. Population inversion indicates that some parts of the system-photons in the case of optical amplifiers-are in higher energy or excited states than would be possible without pumping. These excited states are unstable and revert to normal states with population relaxation times that are approximately in the range 1 ns to 1 ms (other limits are possible and are discussed in more focused resources on optical amplifiers) [60]. Figure 5 shows different configurations of optical amplifiers used in practical applications. A configuration with a booster only is typically used for shorter distances of 150 km (see Figure 5a). A configuration with a preamplifier only is used when we want to avoid high optical powers produced by boosters; in this configuration, it is often necessary to use an optical filter to suppress noise (see Figure 5b). When distances are longer, for example, 250 km, it is necessary to use a configuration with both a booster and preamplifier (see Figure 5c). For longer cascaded optical spans, it is necessary to deploy inline amplifiers (see Figure 5d). Optical filters may be necessary for all configurations with preamplifiers to reduce noise; usually, this is not needed for booster and/or inline amplifiers. The last configuration utilizes Raman pumping, and in this configuration, it is possible to achieve a distance of ≈350 km, but Raman pumping must use high optical powers (up to 1 W) due to the weak Raman effect in silica glass, which may necessitate serious eye safety measures. It must be noted that the provided distances are proximate only and are very dependent on real transmission equipment (the most important parameter is the receiver sensitivity).
General parameters of optical amplifiers [64]: • gain-ratio of output and input power, • gain waveform-should be flat in an ideal case, • saturation power-capability to absorb high input power, • saturation gain-energetic efficiency of the optical amplifier, • insertion loss and insertion loss of the switch-off amplifier, • bandwidth, • noise figure-signal-to-noise ratio, • temperature stability.

Erbium-Doped Fiber Amplifiers
The real revolution with optical amplification started in late 1980 when amplifiers based on rare earth elements became commercially available. The most significant research was performed by D. Payne and E. Desurvire. A detailed description of EDFAs is available in [65], and a detailed theory of EDFAs is provided in [66]. These fiber-doped amplifiers were investigated in the 1960s; however, techniques such as fabrication were not sufficiently mature. Many rare earth elements can be used as dopants in fibers, for example, neodymium, holmium, thulium or ytterbium; these amplifiers can operate in the wavelength range from 500 nm to 3500 nm. However, some combinations of rare earth elements and fibers (fiber is host medium only) can be produced at reasonable prices, and some nonsilica fibers are not easily produced and maintained.
Statements about explosive and exponential growth in data traffic and worldwide fiber networks are almost cliché. Twenty years ago, these networks carried telephone traffic and cable television signals; however, the real explosion started with the World Wide Web (WWW). At that time, deployment of optical amplifiers in local networks was expensive; this situation has changed in the last few years.
The first EDFA was demonstrated in 1989, and the initial users of these new devices were submarine (or undersea) systems because all-optical amplifiers could replace expensive and unreliable electronic regenerators. Trans-Atlantic transmission (TAT) systems are usually cited as the first long-haul systems to fully utilize the strength of EDFAs in 1996. Other systems followed (US and Japan); note that the amplifier spacing range is 30 km to 80 km.
Terrestrial communication systems followed their aquatic counterparts for the same reason-to replace electronic regenerators. It is interesting to note that the first transmission systems supported only a single-channel configuration, and even in early 1990, top commercial transport systems could transport a maximum of 16 channels on a single fiber (with speeds of 2.5 Gb/s, with the latest step to 10 Gb/s), with predictions to support a maximum of 100 channels in the future [60].
The most important of all rare earth elements for telecommunication fiber networks is erbium because it can amplify signals in the most important frequency spectrum in silica fiber: The third window or the conventional C-band near 1550 nm. EDFAs started a new era of optical communication. For example, the usual spacing of EDFAs is 80 km but may be longer; in some links where the total distance is shorter, the spacing can exceed 200 km. This fact is in sharp contrast to the spacing of OEO regenerators, where the spacing was typically 10 km, and as previously mentioned, regenerators can be used for one signal only. EDFAs can amplify a maximum of 100 signals in the C-band, which covers 1530-1565 nm. Almost all optical dense wavelength division multiplexing (DWDM) transmission systems operate in the C-band. However, if the capacity is not sufficient, EDFAs can be customized to amplify signals in the long L-band, which covers 1565-1625 nm [60].
EDFAs must be pumped to achieve gain by population inversion (refer to Figure 6). Figure 6a shows energy levels for erbium atoms. Electrons are pumped from the low energy level to the high energy level, with a relatively short lifetime of 1 microsecond. On the metastable level, with a lifetime of 10 milliseconds, electrons "wait" for incoming photons and amplify them via a radiative transition process. The levels are described by the well-known Russell-Saunders notation, and a detailed description is beyond the scope of this text. Figure 6b shows a more detailed description of the energy levels with splitting due to both spin-orbit coupling and fine splitting owing to the structure of the host silica glass. From this figure, we can deduce the mechanism of amplification in different spectral areas; in this case, for the erbium amplifier, the spectral area is from 1530 nm to 1565 nm. Different pumping schemes are possible, and the most efficient pumping wavelengths are 980 nm and 1480 nm. Two different configurations can be realized. The first configuration has a pump and signal propagating in the opposite direction; this configuration is referred to as backward pumping. The forward pumping configuration is the second configuration, in which the pump and signals propagate in the same direction. Both schemes are frequently used; a combination of backward and forward pumping is employed when more uniform gain is required. As with any real device, optical amplifiers have some limiting factors in practical deployment. The most important factor is amplifier noise, which is usually expressed as a noise figure (NF). The cause of this behavior is ASE. ASE is the random return from an excited state to a normal energy state. ASE can be used to produce a broadband light source, which is undesired in optical amplifiers. The ideal theoretical NF for EDFAs is 3 dB; typical NFs vary from 4 dB to 8 dB. A total of 980 nm pumps can provide a better NF than 1480 nm pumps.
Amplifier noise is a very limiting factor in long-haul transmission because not only data signals are amplified [52].
The described EDFAs are referred to as lumped amplifiers, in contrast to the distributed Raman amplification techniques that are described in this section. However, even EDFAs can be used as distributed-gain amplifiers when the fiber is doped with erbium. These distributed EDFAs were investigated but were never massively deployed in reality.
Another rare earth element used for amplification is praseodymium. Praseodymium-doped fluoride fiber amplifiers (PDFFAs), which have sometimes been referred to as PDFAs to make the name more visually similar to EDFAs, can be used to amplify signals in the original O-band, which covers 1260-1360 nm. Compared with EDFAs, these amplifiers for the O-band are different in one important aspect. Pr (and Nd) operate on the four-level principle, which implies slightly worse parameters, such as output powers or noise figures. In contrast to a three-level system, the population inversion in four-level systems is permanently positive. However, this issue is beyond the scope of this paper. When pumping does not occur, for example, after pump failure, transmitted signals do not suffer gain or achieve attenuation. This effect is in contrast to the three-level Er system, which becomes a strong absorber, and in reality, no signal is transmitted [52].
People may ask why signals should be amplified in the original lossy area of 1310 nm when all long-haul systems use C-and L-bands. The answer is chromatic dispersion (CD) and higher speeds. Even for 10 gigabit Ethernet (GE), 1310 nm transceivers were available and substantially less expensive than their 1550 nm counterparts. Czech Education and Scientific Network (CESNET) performed few experiments with PDFFAs in the 2000s [67], especially when extending the all-optical reach of 10 GE server adapters or network interface controllers (NICs). Note that NICs with 1550 nm transceivers were not available on the market partly due to the high prices of 1550 nm transceivers. From the original distances of approximately 10 km, we were able to reach more than 100 km with PDFFAs and almost 200 km with PDFFAs augmented with Raman amplification. The only drawback of PDFFA is fluoride fibers, which are difficult to manufacture (fluorine is very hazardous, the fluoride composite glasses are hygroscopic and the mechanical properties are not as relaxed as those of the silica glasses used for EDFAs); therefore, few vendors can manufacture them. PDFFAs are noisier than EDFAs as well. However, problems with chromatic dispersion and higher speeds occur with pluggable transceivers for 100 Gb/s and 200 Gb/s. The price difference between the shorter reach of 1310 nm and the longer reach of 1550 nm transceivers is significant, and PDFFAs can offer economically profitable solutions. Thulium fiber amplifiers are used for PONs for signals in the 1490 nm spectral window. Ytterbium is frequently used as a codopant in EDFAs to achieve higher optical output powers [52].
Fiber optic amplifiers operate based on the principle of stimulated emission. The principle is similar to that of lasers. An EDFA amplifier consists of a laser pump diode (laser source of optical radiation) and special erbium (Er)-doped fiber. Due to the radiation added from the pump to the Er fiber, the gain is achieved in the range of C-band wavelengths. A simple schematic is shown in Figure 7. The principle of working is referred to as "3-layer" [64].

•
Optical radiation from the laser pump is coupled to an Er + fiber with a length of a few meters (10-100 m). • Due to this process, the atoms of erbium (Er 3+ ions) are excited. • Absorbed energy allows migration to higher energetic layer E3. • Ions in this so-called metastable state remain only for a short time (a few milliseconds). • Then, the atoms migrate to the conductivity layer-E2 (nonradiative transition). • After the state of "population inversion" is achieved, the highest proportion of Er ions is in an excited state, and the energy is released via the transmitted signal. • The excited ions return to the basic energy layer E1 in the valence band. This is accompanied by the stimulated emission of radiation with the same wavelength and phase as the transmitted signal. • This is how to temporally store the energy achieved by the laser pump.
The transmitted signal is amplified in the C-band in the area of 1550 nm. Note that a useful signal and noise are amplified in the amplified band. While the use of 980 nm and 1480 nm is possible, only 980 nm pumps are currently used due to a higher degree of population inversion.
With the exception of the C-band (1530-1565 nm), we can use EDFAs for amplification in the L-band (1570-1625 nm). A difference is primarily observed in the Er fiber length-for the C-band, the Er fiber must be longer.
The gain of EDFA amplifiers is approximately 30-50 dB depending on the Er fiber length and the power of the pump laser. The higher the quantity of ions is, the higher the energy level and the more frequent the stimulated emission. These phenomena increase the gain of the optical amplifier.
Amplification is the result of the population inversion state of doped ions due to the pump laser. If the power of the optical signal increases or the power of the pump decreases, the inversion state is reduced and the power is decreased. This phenomenon is known as "saturation". EDFA amplifiers are used below the saturation threshold. Spontaneous emission and ASE are reduced, which is referred to as "gain compression" [64].
EDFAs are the most usable, and their advantages are described as follows: • full optical system, • high gain, 30-50 dB, • low noise Figure (4

Semiconductor Optical Amplifiers-SOAs
SOAs are another possible solution for data transfer in optical communications. An excellent review is provided in [68]. Note that SOAs were explored in the 1960s, when semiconductor lasers were invented. While the principle of the laser dates to 1958, a solid-state ruby laser was demonstrated in 1960, and the semiconductor laser was subsequently considered. Early SOAs used GaAs/AlGaAs, but more complex InGaAsP/InP materials, which operate in the 1300 nm to 1600 nm wavelength window, were subsequently introduced for use in optical transmission systems.
SOAs are important devices in many optoelectronic systems, such as optical recording or high-speed printing. In the reality of telecommunication networks, SOAs were deployed in the 1980s, but they exhibited some drawbacks, such as a rather high noise figure and polarization sensitivity, as well as serious problems when amplifying more than one signal due to effects such as cross-phase modulation.
On the other hand, SOAs can be manufactured in specific ways and are able to function in nearly every optical band, covering an almost empty spectral window 1460-1530 nm, the so-called the S-band (no silica host glass rare element-based amplifiers can operate in this band); additionally, these amplifiers can be integrated on chips. For this reason, SOAs are reused for high-speed 100 GE transceivers, where four SOAs are integrated within transceivers and each SOA amplifies only one 25 Gb/s optical signal. SOAs can be used as all-optical wavelength converters and even all-optical switches [59].
SOAs have a structure similar to that of Fabry-Pérot lasers (see Figure 8). However, this Fabry-Pérot configuration is practically improper for data transmission applications because the available bandwidth is very small (less than 10 GHz). To make SOAs suitable for the data world, conversion to traveling wave (TW) devices must be accomplished, which can be performed by the suppression of reflections from the end facets of an SOA with antireflection coatings. The reflectivity must be very small (less than 0.1%) to achieve the desirable behavior. For this reason, other techniques for suppressing reflections were invented, for example, angled-facet or tilted-stripe structures [59].  SOAs are small and electrically pumped (in contrast to EDFAs/PDFFAs or Raman amplifiers) and can be easily integrated with other semiconductor elements and devices, such as lasers and modulators. Undesirable properties, such as a high noise figure, low output power and polarization sensitivity, restrain SOAs from massive deployment as amplifiers, even when many techniques, such as series, parallel or double-pass configurations, have been introduced and studied.
Other novel areas exist where SOAs can find potential use; examples include wavelength conversion, optical demultiplexing of very-high-speed (100 Gb/s) signals to low-speed (10 Gb/s) tributary signals or optical clock recovery units. However, commercial equipment based on these principles is not available.
The gain of semiconductor amplifiers is not generated in the fiber optic material but is generated in the structure of a semiconductor amplifier. Pumping is not optically performed but must support electrical energy (electrical field). Typical materials used for SOA amplifiers are GaAs, AlGaAs, InGaAs, InGaAsP, InAlGa As and InP. These materials have excellent quantum efficiency, which provides a maximum number of generated photons. The principle of SOA operation is similar to that of photon emission in lasers [60]: • Stimulated absorption. • Media excitation. Excitation of a semiconductor medium in the P-N transition is the result of energy pumping and depends on stimulated absorption. Absorbed energy is transferred to an electron in the valence band, which is excited to a higher energetic layer in the valence band. The energy of an incident photon must be sufficient to overcome the forbidden band of the semiconductor. • Population inversion. In a pro-polarized P-N transition, it is possible to achieve population inversion by molecular excitation to a higher energetic layer. The state of population inversion means that the quantity of electrons in the valence band is higher than the quantity of electrons in lower-energy bands. SOAs are manufactured as a chip situated in a standard housing with the capability of temperature control, which allows wavelength stability and the possibility of achieving maximal gain. A high concentration of carriers in an active area causes an increase in the refractive index, which is higher than that in the coating. This region serves as a lightweight circuit for newly generated photons [60].
Advantages of SOAs: • Saturation of an SOA is achieved by a strong input signal, which depletes free carriers in an active area. The gain decreases with increasing input power. Saturated power is achieved by a 3 dB increase in the maximal value (see Figure 9). The influence of carrier depletion can be partially limited by so-called holding beam injection (optical copumping) [69].

Raman Amplifiers
Another principle used for optical amplification is based on stimulated Raman inelastic scattering (SRS). This process differs from stimulated emission, as exhibited by EDFAs and SOAs, where incident photons stimulate emission of another photon with the same energy (i.e., frequency). In SRS, incident photons create another photon with lower energy (i.e., with lower frequency), and the remaining energy is absorbed in the fiberglass as molecular vibrations (optical phonons). Materials absorb energy, which is subsequently emitted. If the energy of emitted photons is lower than the energy of absorbed photons, the effect is referred to as Stokes Raman scattering. If the energy of emitted photons is higher than the energy of absorbed photons, materials lose energy, and the effect is referred to as anti-Stokes Raman scattering. This scattering process is spontaneous, i.e., in random time intervals, and occurs when signal photons (sometimes referred to as Stokes photons) are injected into materials with pump photons, as known to occur in EDFAs.
While Raman amplification in optical communications was demonstrated in the early 1970s, the Raman effect was predicted in the 1920s and published in 1928 [70]. The first demonstration of Raman amplification in optical fibers was performed in the early 1970s, and many research papers indicated the potential of the Raman effect and amplifiers in fiber optic networks. However, as with coherent systems, Raman amplifiers were overtaken by EDFAs. In the 2000s, Raman amplification started to emerge in real transmission systems, especially with long-haul and ultralong-haul systems, but with improved devices. The Raman effect in silica fiber is weak, and much higher pump powers are required than those with EDFAs. Polarization dependency is also a problem, but it can be solved with the use of two orthogonally polarized pump sources, and the gain profile is not spectrally constant (refer to Figure 10). The problem with gain is true for every amplifier, and solutions to mitigate this effect are known [71].
The principle of the Raman amplifier is based on the interaction between photons that spread in the optical environment and this environment (material). The result of the interaction is a frequency shift. Raman amplifiers produce stimulated Raman scattering (SRS) in the material of the optical fiber.
Due to optical pumping at specific wavelengths, interaction between photons and phonons of the material is possible, where the energy of molecules is added to the energy of photons (refer to Figure 11). Due to this change, a new mode with a 100 nm wavelength shift is created. The wavelength is shifted to longer wavelengths. Therefore, if we need to amplify optical signals in the 1550 nm band, 1450 nm pumping sources must be used. Raman scattering is an elastic scattering mechanism that does not require a population inversion. The maximal gain is approximately 30 dB [64]. As previously mentioned, the amplified band is given by the wavelength of the pumping diode. Due to this capability, Raman amplifiers can function in an extensive range of wavelengths.  Amplification in Raman amplifiers is very different from that in EDFAs, PDFFAs and SOAs: The transmission fiber is used as the media for amplification, and therefore, Raman amplifiers are distributed. Other optical amplifiers may be considered to be "lumped". DRA uses backward pumping. When the pump is situated at the end of an optical link, the gain contributes to all-optical links, and the power loss is continually compensated. DRA amplifiers have a low noise figure, high gain and low nonlinear distortion [71].
It should be noted that lumped Raman amplifiers were also introduced in telecommunication transmission systems but in a slightly different manner. Raman pumping was combined with a dispersion compensating fiber (DCF). The diameters of DCFs are smaller than those of standard single-mode fibers, so the interaction in DCFs is stronger, and Raman amplification is more efficient. DCFs are "lumped" because they are periodically inserted into the transmission line. Thus, adding Raman pumps to previously lumped DCFs can create lumped Raman amplifiers. With the deployment of coherent transmission systems, DCFs are removed because compensation of chromatic dispersion is not necessary [59].
The Raman effect is broadband, but the drawbacks are the polarization dependency (where a common solution is to use pump depolarization) and the low gain coefficient in silica glass (see Figure 12). For this reason, high optical powers must be applied. In CESNET, experiments were performed in which the launched powers often exceeded 500 mW. These powerful lasers introduce serious safety eye hazards, even when automatic laser shutdown (ALS) is implemented (in some cases of fiber cuts or angled physical contact (APC) connectors, ALS may encounter difficulties in detecting fiber failure). For this reason, Raman amplifiers are rarely deployed in common optical systems. They are used in specific cases when very long fiber spans must be lightweight-for example, submarine links between the mainland and islands or similar specific conditions. The present DWDM system transporting coherent signals over long distances needs to deploy Raman amplifiers together with EDFAs to cope with lossy segments with high attenuation. This hybrid solution helps to keep the overall optical signal-to-noise ratio (OSNR) acceptable. It is interesting to note that in some of the literature [60], because of the rather weak interaction in silica glass, the maximum required pump power to achieve a gain of 30 dB is calculated to be 5 W. Experiments performed in CESNET showed that even pump powers less than 1 W caused very strong distributed Rayleigh scattering (DRS) and could not be used. Some vendors of transmission equipment do use pump powers below 500 mW for Raman amplification, which is another peculiarity on the other end of the power scales; both SRS and stimulated Brillouin scattering (SBS) are nonlinear effects, so some threshold optical powers are required to "kick-start" the mechanism.
We tested and verified the Raman amplification for pump powers of 10, 50, 100, 150, 200, 250 and 300 mW of optical power. The central wavelength of the pump was ≈1455 nm, and the signal wavelength was 1552.064 nm, which is adequate for a 97 nm wavelength shift (refer to Table 1). The continuous wave signal from the laser diode was coupled to a fiber spool with a length of 50 km. If we consider a fiber attenuation of approximately 4% per kilometer (corresponding to 0.18 dB/km, which is the normal attenuation coefficient for 1550 nm in a standard single-mode optical fiber), then after 50 km, the signal is attenuated by approximately 9 dB (≈87% loss of power!). In the scheme depicted in Figure 13 we can see two important things. First, the amplification is more effective if the pump signal propagates in the opposite direction from that of the data signal. Second, the Raman gain coefficient is highly polarization dependent; therefore, two pump diodes (Pump 1 and Pump 2) are used to generate depolarized light. It is also very important that the fiber length must be long enough to generate Raman scattering [72]. As shown in Figure 14, the saturation power of the amplified signal is quasilinear. The difference between the saturation power for the 10 mW pump power and the saturation power for the 300 mW pump power is approximately 2.5 dB.  Disadvantages of Raman amplifiers: • high pump power requirements, • lower efficiency for a specific wavelength then EDFAs (for the same pump power), • sophisticated gain control is needed.

Brillouin Amplification
The last amplification technique discussed is based on SBS. SBS is similar to SRS but with notable exceptions: SBS occurs only in the backward direction, the scattered light is shifted by approximately 9-11 GHz (compared with 13 THz or 100 nm for SRS) and the gain spectrum is 100 MHz (for SRS, it is 30 THz). For SBS, energy absorbed by the fiberglass has the form of acoustic phonons (in contrast to an optical phonon in the case of SRS).
Brillouin scattering is a "photon-phonon" interaction that occurs when annihilation of a pump photon simultaneously creates a Stokes photon and a phonon. The created phonon is the vibrational mode of atoms, which is also referred to as a propagation density wave or an acoustic phonon/wave. In a silica-based optical fiber, the Brillouin Stokes wave propagates dominantly backward, although very partially forward. The frequency (9-11 GHz) of a Stokes photon at a wavelength of 1550 nm differs considerably from that of Raman scattering (is smaller by three orders of magnitude and is dominantly downshifted due to the Doppler shift associated with the forward movement of created acoustic phonons [73]. Depending on the frequency offset, the interference of the counterpropagating pump light with the signal light causes a moving density grating. The density grating coherently scatters pump photons into the signal beam, which is amplified. The characteristics of the SBR amplifier are the narrowband spectrum of approximately tens of MHz (based on an optical gain medium). Brillouin amplifiers have a built-in narrowband optical filter, which enables amplification of specific signals. In contrast to broadband amplifiers (EDFAs, SOAs, and Raman amplifiers), Brillouin amplifiers enable a maximum signal stage of 50 dB or higher [74].
Brillouin amplifiers are not suitable for standard data communication due to their very narrow gain spectrum; however, new applications use different optical signals. We can provide examples of two new applications that have been extensively investigated: Accurate time transfer (that is, atomic clock comparison) and ultrastable frequency transfer. These signals are very slow (hundreds of MHz for accurate time, continuous wave (CW) for stable frequency), and therefore, their spectra are very narrow, which renders them suitable for Brillouin amplification, especially for ultrastable frequency transfer using very narrow laser sources. In this case, SBS can be used for very powerful amplification [75].
Advantages of Brillouin amplifiers: • high gain and saturation power for narrowband signals, • wavelength conversion, • can enable amplification of a very small input signal (a few nanowatts) by more than 50 dB in a single gain step.
Disadvantages of Brillouin amplifiers: • limited range of use, • nonlinear phenomena.

Amplifiers for PONs
Long-reach optical access is a promising technology for future access networks. This technology can enable broadband access for a large number of customers in access/metro areas while decreasing capital and operational expenditures for the network operator.
Almost all the described optical amplifiers can also be used in passive optical networks (see Table 2), with the notable exception of Brillouin amplifiers (which are not suitable due to a very small gain spectrum, as previously described). Prospects for PONs with reaches of 100 km and 10 Gb/s speeds are being investigated, but these devices are not commercially available. A typical PON can reach 20 km with a maximum split ratio of 64. For example, the GPON standard established optical budgets of 28 dB with 2.488 Gb/s for downstream transmission and 1.244 Gb/s for upstream transmission. This standard is the current standard of PONs. In long-haul systems, optical amplifiers are extensively employed to extend the reach of systems to hundreds or thousands of kilometers. The cost of optical amplifiers is sufficiently low; thus, we can consider their use in PONs. The cost of amplifiers can be shared among numerous customers. The GPON protocol can support a logical reach of 60 km and a split ratio of 1:128. Optical amplifiers may be used to extend the reach. The transparency of optical amplifiers indicates their use for GPONs and gigabit Ethernet PONs (GEPONs). Optical amplifiers are a main technology for next-generation access (NGA) PONs.
Several benefits of extended-reach PONs exist. First, customers are located as far from the CO as they can be connected. Second, where customers are sparsely distributed over a large area, optical amplifiers can be used to ensure good utilization of the shared PON. Third, depending on an end-to-end network design, extending the reach of a PON can enable node consolidation, which entails reducing the number of PON head-end locations that must be managed by the operator [76].
In metro and long-haul networks, EDFAs are extensively employed because they provide a high gain, output power and noise figure from 1530-1565 nm. Existing PON standards apply EDFAs for analog video broadcast (overlay PON). An alternative to fiber amplifiers is the SOA. While SOAs do not provide gain and noise figures that are comparable to those of EDFAs, their advantage is that they can operate at any wavelength. The gain dynamics of SOAs are also substantially faster than those of EDFAs, so SOAs can be used for bursts upstream [76].
While the Raman amplifier can be theoretically used downstream, we have to consider the high price and necessity of high-power, dangerous pumps.
FEC is another important technology for extending the capability of PONs. While FEC is specified in GPONs and GEPONs, an enhanced version of FEC could be used in future PONs.
Early proof-of-concept experiments have been performed using an optical amplifier at an intermediate powered location in combination with the FEC, which can envisage a 10 Gb/s PON with a split ratio of 1024 [77].
Other PON systems use the C-band, for example, coarse WDM (CWDM) wavelengths of 1530 nm and 1550 nm between the OLT and ONUs). If EDFAs are used as power boosters and preamplifiers, the maximum budget increase is reported to be 34 dB [31]. EDFAs are used for the area of 1550 nm, where video overlay signals are transmitted.
In [78], SOAs with Raman amplification are demonstrated for maximum speeds of 2.5 Gb/s. Raman pumping at 1270 nm is used, with a maximum pumping power of 1 W. The results for extending the reach for rural areas are promising, but 1 W is Class IV, and serious eye safety hazards must be carefully considered.
British Telecom has demonstrated its long-reach PON. The system used EDFAs and SOAs. With the appropriate optical technologies, 10 Gb/s transmission was achieved in the downstream and upstream channels across 100 km to 1024 customers using a low-cost optical transceiver in the ONU situated in the customer premises [79].
The ACTS-PLANET realized the SuperPON in 2000. The implemented system supports a total of 2048 ONUs and achieves a span of 100 km. The 100 km fiber span consists of a maximum feeder length of 90 km and an add and drop section of 10 km.
EDFAs and SOAs were also used [79]. The Photonic System Group of University College Cork in Ireland has demonstrated the wavelength and time-division multiplexing long-reach PON (WDM-TDM LR-PON). The network supports multiple wavelengths, and each wavelength pair can support a PON segment with a long distance (100 km) and a large split ratio (1:256). The LR-PON contains 17 PON segments, each of which supports symmetric 10 Gb/s upstream and downstream channels over a 100 km distance. The system can serve a large number of end-users: 17 × 256 = 4352 users [80].
The authors in [81] cooperated with British Telecom, Alcatel and Siemens, who introduced the second-stage prototype of a photonic integrated extended metro and access network (PIEMAN) sponsored by Information Society Technologies (IST). PIEMAN consists of a 100 km transmission range with 32 DWDM channels, each of which operates at symmetric 10 Gb/s and 32 PON segments. The split ratio for each PON segment is 1:512; thus, the maximum number of supported users is 32 × 512 = 16384 end-users.
Other long-reach topologies considered by researchers include ring-spur topologies for the long-reach PON. Each PON segment and OLT are connected by a fiber ring, and each PON segment can exploit a traditional fiber to the x (FTTx) network with a topology that consists of several "spurs" served from the "ring". The ring can cover a maximum metro area of 100 km. The natural advantage of the ring topology is two-way transmission and failure protection [79]. An example of this topology was demonstrated by ETRI, a Korean government-funded research institute, which has developed a hybrid LR-PON named WE-PON (WDM-E-PON). In the WE-PON, 16 wavelengths are transmitted on the ring and can be added and dropped to local PON segments via the remote node (RN) on the ring. The RN can include an optical add-drop multiplexer (OADM) and an optical amplifier. The split ratio of the PON segments is 1:32, and the system can support 512 end-users [82].
Another demonstration of ring-based technology, which is called scalable advanced ring dense access network architecture (SARDANA), also implements "ring-and-spur" technology. In this system, 32 wavelengths are transmitted on the ring, with a split ratio of 1:32 for each wavelength. More than 1000 end-users are supported. The ONU units are based on a reconfigurable semiconductor optical amplifier (RSOA) [83]. The comparison of LR-PON projects is depicted in Table 3. Many tests of different optical amplifiers in the PONs were conducted. In general, we suggest that use of the Brillouin amplifier is not feasible in this area of optoelectronics due to specific properties [93][94][95][96][97][98][99][100][101]. EDFAs can be employed for analog radio frequency (RF) overlay video services or WDM-PONs, where the C-band or L-band is used [102][103][104][105][106][107]. Other types of fiber amplifiers can be used for PONs: A thulium-based amplifier downstream and a praseodymium-based amplifier upstream [108][109][110][111][112][113][114]. Raman amplifiers can be used for PONs; however, if we take into account the cost and hazardous optical power, it is not the best solution for downstream transmission [115][116][117][118][119]. SOAs can be used as one of the most suitable candidates for future next-generation long-reach PONs. The low cost, sufficient gain and small size positions SOAs for future development [21,[120][121][122][123].

Conclusions
In this paper we focus on reach extension in passive optical networks whereas applications in access and passive optical networks are being considered. Achieving longer distances without amplifiers or repeaters is not possible so the article explains both the basic principles of repetition and amplification, as well as the optical fiber amplifiers themselves. History, the general principles of operation and the basic configurations are explained for all types of amplifiers.
While many standards for high-speed PONs exist and additional standards are being prepared, there are also new trends that have been barely documented. However, the lack of standards should not hinder the creation of new approaches, for example, deployment of optical amplifiers in PONs, mainly EDFAs, Ramans and SOAs. An evaluation measurement to verify the dependence of the power level of the Raman amplifier on the saturation power was performed. Measurements have shown that even a relatively small powers of the pump diodes of the Raman amplifier (≈300 mW) can amplify the transmitted signal.
In addition to explaining the basics of amplification and measuring the amplification itself with a Raman amplifier, the article provides a comprehensive overview of the current state of research in the use of optical fiber amplifiers in PON networks. Both simple solutions that would be easily implementable in practice and complex solutions with signal regeneration are presented.
New trends of open networking promoted by hyperdata center companies should be considered for new trends in PON deployment to avoid any undesirable vendor dependencies and lock-ins. Open networking can ensure that technologies are replaced or migrated to new equipment as needed, especially when deploying out-of-box optical equipment, whether 2R or 3R, in PON ecosystems. These new open trends are yet not standardized in many cases but should not be disregarded because they are emerging in many parts of the world, especially in North America and Asia.
Additionally, we believe that, with optical amplification, the support of new applications, such as accurate time transfer or distributed fiber sensing, could be important for PON end-users. This new class of applications may not appear to be appropriate for a PON environment at first, but future user requirements and new open approaches are to be utilized here. Acknowledgments: Tomas would like to give thanks to Ales Buksa for his support at the University in memorial. Ales taught and inspired him regarding many things in his personal life. Acknowledgment is also given to CESNET for technical support and the equipment used for the measurement.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript: