This document discusses recent advances in optical fiber amplifiers. Key points include: 1. Erbium-doped fiber amplifiers (EDFAs) have increased in bandwidth from 35nm to over 80nm, enabling higher capacity WDM systems. 2. Raman fiber amplifiers provide even wider bandwidth amplification. 3. Control schemes have been developed to mitigate fast power transients in amplifier chains caused by network reconfigurations. 4. Advances include wider bandwidth EDFAs combining C-band and L-band amplification, automatic gain control, and techniques to protect signal channels during network changes.

1. 1Advance in Optical Fiber Amplifier
Chun Jiang, Qingji Zeng, Hua Liu, Xiaodong Tang, Xudong Yang
Center for Broadband Optical Networking Technology, College of Electronics & Information, Shanghai
Jiaotong University, Shanghai 200030, P.R.China
ABSTRACT
Advances in optical fiber amplifier for optical network are reviewed in this paper. Considerable progress has
been made in optical amplifier technology in recent years. The bandwidth of amplifiers has increased several
times and flat gain amplifiers with more than 80 nm of bandwidth have been demonstrated. With the advent of
Raman fiber amplifiers, more wider bandwidth is obtained. Progress has also been made in the understanding of
amplifier gain dynamics. Several control schemes have been successfully demonstrated to mitigate the signal
impairments due to fast power transients in a chain of amplifiers and will be implemented in optical network
design. Terrestrial optical systems have been increasing in transmission capacity. In this review, we focus on the
recent progress in some important aspects of several optical fiber amplifier technology.
Keywords: Erbium doped fiber amplifier, Raman fiber amplifier, Optical network
1.INTRODUCTION
In the past decades years, tremendous progress has been made in the development of optical amplifier
components and technology, including erbium-doped fiber amplifier, Raman fiber amplifier and waveguide
amplifier, semiconductor pump lasers, passive components, and splicing and assembly technology. In the
research area, Optical amplifier with a bandwidth of 80 nm was achieved for the first time. In the meantime, an
enormous effort has been under way to incorporate optical amplifier into commercial optical communication
systems. After intensive laboratory research and development, optical amplifiers technology offer an
unprecedented cost-effective means to meet the ever-increasing demand for transport capacity, networking
functionality, and operational flexibility. In this paper, we focus on the recent progress in some important aspects of
several optical fiber amplifier technology.
2.ERBIUM DOPED FIBER AMPLIFIER
2.1 Optical Network Demand
An standard EDFA is equipped with the features of gain, noise figure, output power, dynamic range and
reliability. In practice, however, different network functions require only some of these features. Amplifier
features can generally be divided into static parameters and dynamic parameters. To obtain good static
1: Correspondence Author: Emai1:cjiangonhine.sh.cn; Telephone:0086-021-62932166;Fax:0086-021-62820892.
In Rare-Earth-Doped Materials and Devices IV, Shibin Jiang, Editor,
318 Proceedings of SPIE Vol. 3942 (2000) • 0277-786X/O0/$1 5.00

2. parameters, EDFAs with two or more stages are generally used. Zyskind et al.' discuss the basics of two-stage
amplifier and some of the related design issues. In this subsection, we concentrate on the recent progress in
several important aspects of EDFAs technology. In the next subsections, we represent wideband amplifier for
high capacity applications and conclude with a discussion ofthe current status ofthe EDFA.
2.2 Wideband EDFA
The recent exponential growth in data communications places urgent demands on high capacity communication
networks. To increase total capacity, research and development teams can works on one or more of the following
parameters: High speed, which is currently limited by high speed electronics, fiber dispersion, and nonlinear
effects; Channel spacing, which is limited by filtering technology and non-linear effects. Amplifier bandwidth,
which has paid much attention in recent years. For WDM application, uniform gain is desired for all signals
channels. Generally, the gain is flat somewhere between 1540nm and 1 560nm for an inversion level of 40% to
60%.Actually, it is this generic gain that was used in initial WDM systems. Since the ASE power around the
1530nm region can be high enough to cause saturation, an ASE filter can be added in the middle stage to block
the ASE in this band 2 This type of optical amplifier has been successfully used in CATV and early WDM
optical networks3. To fully utilize the gain band between 1530nm and 1565nm, gain equalization filters(GEFs)
can be used to flatten the gain spectrum. Several technology have been studied to fabricate GEFs, including thin
film filters, long period gratings ", short-period gratings5,silica wavelength structure6, fused fiber, and acoustic
filters7.A GEF is inserted between two erbium doped fibers in order to form a wideband optical amplifier.
Figure 1 Design of wideband optical amplifiers with a C-band and L-band structure
319
ocE GEF 3( U.fflp't
DE
Fow
C
•kJtcfl
E ft;r . :::.. EF
ca -D ekrrtit
r(F— gIa.1 ili jCFsU - L.LiJkAI
FDf - TIbpr
I .)CF
GEt fite-
L fl — u ip tIfl.4zui Ir*js-
MU) - Mkj1i
1rJM-Wmirgfti1cInn ri iifl1i

3. Depending on the design, a bandwidth of about 35nm to 4Onm can be obtained in this conventional wavelength
region ( the C—band) 8,9• Additionally there is variable attenuation in the middle stage whose function will be
discussed in the optical amplifier for practical WDM networking systems" section. this kind of amplifier with
35nm of flat bandwidth was used in the long distance transmission of 32nm and 64 channels at 10 Gb/s 8,10
2.3. C-band and L-band
Since the gain drops sharply on both sides of the C-band at 40 to 60 % inversion, it is impractical to further
increase the bandwidth with a GEF. However, a flat gain region between 1565nm and 1615nm (the L-band) can
be obtained at a much lower inversion level (20 to 40%)h112. By combining the C-band and the L-band, a much
wider bandwidth can be realized, with the principle shown in figure 1 .Since the initial demonstration of
principle12"3"4 much progress has been made in the understanding and design of ultrawideband optical amplifiers
with a split -band structure.
2.4. Automatic gain control
EDFAs are employed in present-day multi-wavelength optical networks to compensate for the loss of fiber spans
and network elements. In these applications, the amplifiers are normally operated in a saturated mode. In the
event of either a network reconfigration or a failure, the number of WDM signals traversing the amplifiers would
change and the power of surviving channels would increase or decrease due to a cross-saturation effect in the
amplifiers. Dropping channels can give rise to surviving channel errors, since the power of these channels may
surpass the threshold for nonlinear effects such as Brilluin scattering. Adding channels can cause errors by
depressing the power of surviving channels below the receiver sensitivity. To overcome such enor bursts in
surviving channels in the network, the signal power transients must be controlled. The response speed required
for surviving channel protection is governed by the EDFA transient response, the size of the WDM system, and
the power margins build into the transmission system.
2.4.1 dynamic gain control
Because of the saturation effect, the speed of gain dynamics in a single EDFA is generally much faster than the
spontaneous lifetime of about lOms.The time constant of gain recovery in single-stage amplifier was reported to
be between 110 and 340 s15 The time constant of gain dynamics is a function of the saturation caused by the
pump power and the signal power. With the development of high-power EDFAs for WDM communication
systems, the saturation factor becomes higher and the transient time constant shorter. In a recent report,
characteristic transient times have been reported to be tens ms in a two-stage EDFA.16. Figure 2 shows the
transient behavior of surviving channel power for the cases of one, four, and seven dropped channels in an eight-
channel system. In the case of seven dropped channels, the transient time constant is nearly 52 i- s. As shown in
the figure, the transient becomes faster as the number of dropped channels decreases. The time constant
decreases to 29 i- s when only one out of eight channels is dropped. The rate equations 17 for the photons and
the populations of the upper (I3/2) and lower (I5/2) states can be used to derive the following approximate
formula for the power transient behavior18:
P(t)
c)
(1)
where P(O) and P( ') are the optical powers at time t =0 and t = , respectively. The characteristic time T
the effective decay time of the upper level averaged over the fiber length. It is used as a fitting parameter to
320

4. obtain the best fit with the experimental data. The model has been used to calculate the fractional power
excursions in decibels of the surviving channels for the cases of one, four, and seven dropped channels. The
times required to limit the power excursion to 1 dB are 18 i s and 8 1-' s when four and seven channels are
dropped, respectively. As EDFA technology advances further to support larger numbers of WDMchannels in
optical networks, the transient times will fall below 10 is. Dynamic gain control of EDFAs with faster
response times will be necessary to control the signal power transients.
Tim!
Figure 2 the transient behavior of surviving channel power for the cases of one, four, and seven dropped
channels in an eight-channel system
—I
0 2C IIJ &O ] }]
inc
— Apr L
Al pIfcr
Amir4—
— A'flpi4r*1C
—AflpI1;r*s4
i 1IO 1G) 1*10
Figure 3 The effect of dropped channels on surviving powers in an amplifier chain"
321
1
2:0-
io
-I-
• 1 rhIrub1. i;i
— I Chrm4, ,ixki
• 2 Chrn*l;ti
i ri uiwh. in ii Is
S 7 CF. th I!tb
7 Charuid 'nod.
0 50 5') 2 sc
4—
4cIbme

5. 2.4.2 Power transients
In recent studies, the phenomenon of fast power transients in cascaded EDFA was reported.19'20. The effect of
dropped channels on surviving powers in an amplifier chain is illustrated in Figure 3 .When 4 out of 8 WDM
channels are suddenly lost, the output power of each EDFA in the chain drops by 3 dB, and the power in each
surviving channel then increases toward double the original channel power to conserve the saturated amplifier
output power. Although the gain dynamics of an individual EDFA are unchanged, the increase in channel power
at the end of the system becomes faster for longer amplifier chains. The fast power transients result from the
effects of the collective behavior in chains of amplifiers. The output of the first EDFA attenuated by the fiber
span loss acts as the input to the second EDFA. Since both the output of the first EDFA and the gain of the
second EDFA increase with time, the output power of the second amplifier increases at a faster rate. This
cascading effect results in faster and faster transients as the number of amplifiers increases in the chain. To
prevent performance penalties in a large-scale WDM optical network, surviving channel power excursions must
be limited to certain values depending on the system margin. For the Multi-channel Optical Networking
(MONET), the power swing should be within 0.5 dB when channels are added and within 2 dB when channels
are dropped., In a chain consisting of 10 amplifiers, the response times required to limit the power excursions to
0.5 dB and 2 dB would be 0.85 i,i s and 3. 5 i s, respectively. The response times are inversely proportional to
the number of EDFAs in the transmission system.21 .The time response of EDFAs can be divided into three
regions: the initial perturbation region, the inter- mediate oscillation region, and the final steady-state region. In
the initial perturbation region, the gain of the EDFA increases linearly with time, and the system gain and output
power increase at a rate proportional to the number of EDFAs. Assuming that the amplifiers work under same
conditions, the rate of change of gain at each EDFA is the same and is proportional to the total lost signal power.
The slope increases linearly with the number of cascaded EDFAs. These experimental results have been
confirmed by modeling and numerical simulation from a dynamic model.22. From the results of both
experimental measurements and numerical simulation on a system with N EDFAs, the time to reach the peak is
found to be inversely proportional to N, and the slope to the peak is found to be proportional to N 23 These
properties in the perturbation and oscillation regions can be used to predict power excursions in large optical
networks.
2.5 Channel protection
Channels in optical networks can suffer from error bursts caused by signal power transients resulting from a line
failure or a network reconfiguration. Such error bursts in surviving channels represent a service impairment,
which is absent in electronically switched networks and unacceptable to service providers. The speed of power
transients results from channel loading and there are the speed required to protect against such enor bursts
proportional to the number of amplifiers in the network and can be extremely last for large networks. Several
schemes to protect against fast power transients in amplified networks--pump control, link control, and laser
control-have been demonstrated in recent years.
2.5.1 Pump control
The gain of an EDFA can be controlled by adjusting the pump current. Early reported work addressed pump
control on time scales of the spontaneous lifetime in EDFAs24. One of the studies demonstrated low-frequency
feed-forward compensation with low-frequency control. 16.After the discovery of fast power transients, pump
control on short time scales 25 was demonstrated to limit the power excursion of surviving channels. In the
322

6. absence of gain control, the change in surviving channel signal power exceeds 6 dB. When the pump control on
both stages is active, the power excursion is less than 0.5 dB for both drop and add conditions. The control
circuit acts to correct the pump power within to 8 i s, which effectively limits the surviving channel power
excursion.
2.5.2 Link control
The pump control scheme described above requires protection at every amplifier in the network. Another
technique uses a control channel in the transmission band to control the gain of amplifiers. Earlier work
demonstrated gain compensation in an EDFA at low frequencies using an idle communication network element.
The control channel is stripped at the next network element to prevent improper loading of downstream links.
The power of the control channel is adjusted to hold constant the total power of the signal channels and the
control channel at the input of the first amplifier. This maintains constant loading of all EDFAs in the link. The
experimental demonstration of link controlled surviving channel protection is set up with several signal channels
and 1 control channel. A fast feedback circuit with a 4 ii s response time is used to adjust the line control
channel's power to maintain constant total power. The signal channels and the control channels are transmitted
through amplified spans of fiber, and the bit error rate (BER) performance of one of the signal channels is
monitored. When most out of several signal channels are added/dropped, the surviving channel suffers from a
power penalty over 2 dB and a severe BER floor. With fast link control in operation, power excursions are
mitigated, BER penalties are reduced to a few tenths of a dB, and error floors disappear.
'4(ç ;4:.2LJ:L..I
•i' -' ..
*,
•/
I_cu — Lir* crrtrI unit
•fl(C — ::1I1;3 .1 i 14 1I1HE
PD -P:d o
'- Ji]. 1divc:o.zpkr
Figure 4 Schematic diagram of link control for surviving channel protection in optical networks
2.5.3 Laser control
Automatic gain control by Laser has been extensively studied since it was experimentally demonstrated26. A new
scheme for link control based on laser gain control has recently been reported 271n this work, a compensating
323

7. signal in the first amplifier is generated using an all-optical feedback laser loop; the signal then propagates down
the link. Stabilization is reached within a few tens ms, and output power excursion after 6 EDFAs is reduced
from more than an order of magnitude to a few tenths of a dB. For laser gain control, the speed is limited by
laser relaxation oscillations28 which are generally on the order of tens of ms or slower. Inhomogeneous
broadening of EDFAs and the resulting spectral hole burning can cause gain variations at the signal wavelength,
which limit the extent of control from this technique. The same is true for link control schemes.
2.6.EDFA for WDMSystems
The advent of practical EDFAs has revolutionized almost all aspects of optical communication systems. This
section focuses on the practical issues that must be addressed in order to design and engineer the optically
amplified communication systems.
2.6.1 SNR
In an optically amplified system, the channel power into the receiver is usually well above the receiver
sensitivity. The optical signal is optically degraded by the accumulated ASE noise from the optical amplifiers in
the chain. At the photodetector, ASE noise is converted to electrical noise primarily through signal-ASE beating,
which leads to BER flooring. System performance therefore places a stringent requirement on the optical signal-
to-noise ratio (OSNR) of each of the optical channels. OSNR is thus the most important design parameter for an
optically amplified system. Other optical parameters in system design consideration are channel power
divergence and maximum channel power relative to the threshold levels of optical nonlinearities--for example,
self-phase modulation, cross-phase modulation, and four-wave mixing.29
2.6.2 Noise Figure and Output Power
EDFAs are conventionally classified as power amplifiers, in-line amplifiers, and preamplifiers, state-of-the-art
WDM systems require all three types of amplifiers to have low noise figure, high output power, and uniform
gain spectrum. But here we do not distinguish these three types of amplifiers. The nominal OSNR for a 1 .55
mWDM system with N optical transmission spans can be given by the following :
OSNR = 58 +P0- lOloglO(Nh)-L-NF-lO 1og10(N+l) (2)
where OSNR is normalized to 0.1 nm bandwidth, P01 is the optical amplifier output power in dBm, Nh is the
number of WDM channels, L, is the fiber span loss in dB, and NF is the amplifier noise figure in dB. For
simplicity, we assume that both optical gain and noise figure are uniform for all channels.Equation (2) shows
how various system parameters contribute to the OSNR. This equation indicates that we can make tradeoffs
between the number of channels and the number of spans in designing a system. Note that the tradeoffs may not
be as straightforward in a practical system because of the mutual dependence of some of the parameters. Other
system requirements impose additional constraints, for instance, optical nonlinearities place an upper limit on
channel power, which depends on the number of spans, the fiber type, and the data rate. This simple formula
highlights the importance of two key amplifier parameters--noise figure and output power. While it provides
valuable guidelines for amplifier and system design, it is always necessary to simulate the OSNR evolution in a
chain of amplifiers when designing a practical WDM system. The amplifier simulation is usually based on an
accurate mathematical model of amplifier performance. Amplifier modeling is also a critical part of the end-to-
end system transmission performance simulation that incorporates various linear and nonlinear transmission
penalties.
324

8. 2.6.3 Gain Flatness
Gain flatness is another key parameter for WDM system design. The worst WDM channel that experiences the
lowest amplifier gain will have an OSNR value lower than the nominal value given in Equation (2). This deficit,
which can be viewed as a type of penalty resulting from amplifier gain non-uniformity, is a complicated function
of the individual amplifier gain shapes and the correlation of the shapes of the amplifiers in the chain. It is
assumed the same gain shape for all amplifiers in the chain and calculate the OSNR penalty due to gain non-
uniformity, while the absolute penalty may have a value in practical cases, the experimental result show that gain
flatness is a parameter that can have a significant impact on the bottom-line OSNR, the penalty is especially
severe for a long amplifier chain, as in the case of long haul applications, the impact of gain non-uniformity,
however, is not limited to the OSNR penalty; it also causes power divergence of WDM channels in a long chain,
while the weak channels see an OSNR penalty that limits the system performance, the strong channels continue
to grow in power that may reach the nonlinear threshold, also limiting system performance. Additionally, large
power divergence increases the total crosstalk from other WDMchannels at the optical DMUX output. State-of-
the-art optical amplifiers usually incorporate a gain equalization filter to flatten the gain spectrum. To minimize
the residual gain nonuniformity requires careful design, modeling ,and engineering of the amplifier components
in particular, the gain equalization filters.
2.6.4 Amplifier Control
An optical amplifier may not always operate at the gain value at which its performance, especially gain flatness,
is optimized. Many factors contribute to this non-optimal operating condition, including the fact that the span
loss can be adjusted at system installation and maintained in the system's lifetime only to a finite range with
respect to the value demanded by the amplifiers for optimal performance. As a result, amplifier gain will be
tilted, and such tilt can have significant impact on system performance in ways similar to gain non-uniformity. If
not corrected, gain tilt can result in an OSNR penalty and increased power divergence. Control of optical
amplifier tilt is often necessary to extend the operational range of the amplifiers and compensate for loss tilt in
the system due to, for example, fiber loss variation in the signal band. Control of amplifier gain tilt can be
achieved by varying an internal optical aUenuator°. Implementation of such a tilt control function requires a
feedback signal that is derived from, for example, measured amplifier gain or channel power spectrum, and an
algorithm that coordinates the measurement and adjustment functions. By changing the loss of the attenuator, the
average inversion level of the erbium doped fiber can be adjusted, which affects the gain tilt in the EDFA gain
spectrum. Another important amplifier control function is amplifier power adjustment. In a WDM system, there
is a need to adjust the total amplifier output as a function of the number of equipped channels. The total output
power must be adjusted so that while the per-channel power is high enough to ensure sufficient OSNR at the end
of the chain, it is low enough not to exceed the nonlinear threshold. Additionally, per-channel power must be
maintained within the receiver dynamic range as the system channel loading is changed. Such power adjustment
has been conventionally achieved by a means of a combination of channel monitoring and software-based pump
power adjustment.
The recent advances in WDM optical networking have required a power control fast enough to minimize
channel power excursion when a large number off channels are changed due to, for example, catastrophic partial
system failure. Various techniques have been demonstrated to stabilize amplifier gain, thereby achieving the goal
of maintaining per-channel power. In addition to amplifier dynamics control, practical implementation in a
system also requires a receiver design that can accommodate power change on a very short time scale.
325

9. 2.6.5 Performance Monitoring and Fault Location
To maintain an optically amplified WDM system, it is essential to be able to monitor system parameters and
locate and isolate faults quickly. To achieve this goal, the system must be continually monitored, and the
gathered information must be transmitted in a timely manner to the endpoints. A telemetry channel in a
traditional regenerated system uses the overhead bits on one signal channel to transmit the maintenance and fault
location data to the end terminal and provide communication to remote repeater sites. In an optically amplified
system, there is no access to these bits; Instead, other methods have been developed to transport telemetry
information. The use of a separate optical channel just for telemetry has been widely adopted in practical
optically amplified systems. Telemetry is added and dropped at each repeater amplifier site to serve the function
of collecting, processing, and transporting maintenance information. It is usually operated at 15 10 urn with a
capacity 1 .5 Ivfb/s- a slow data rate for optical networking technology.
2. RAMAN FIBERAMPLIFIER
Recent advances in high power diode lasers and Bragg gratings enable the use of stimulated Raman scattering to
frequency-shift a pump wavelength to amplify optical signals at a long wavelength31, in a 1 .3 1-'m intracavity
cascaded Raman amplifier, Braggings were made at both ends of a fiber loop, to selectively confine the pump
and stokes line within the fiber loop. This configuration permits an efficient transfer over several stokes shifts
from the 1 .064 ii m pump wavelength to 1 .3 i' m for signals amplification. Most fibers employed in Raman
amplifier are standard germanium doped fibers having a high delta and small core diameter for power
confinement. Since a relatively long (0.5-2km) fiber loop is used, a reasonably low attenuation loss is required
between the pump wavelength (around 1 ii m) and the signals wavelength (1 .3 U m). Laser at different
wavelengths can be fabricated by employing configurations such as that illustrated in figure 4. This is composed
of high-NA germanosilicate fiber. Typically 0. 1 to 1.0 km length, nested within cascaded Bragg gratings and
using a high-power pump sources. The gratings selectively confine and re-circulate the pump and Raman
frequencies up to the (n-1)-th order while permitting lasing from the n-th order2. Such lasmg can be contrived at
almost any wavelength using different pump and accessing different order Raman peaks.
Figure 5 Schematic diagram of a Raman fiber laser
Amplification of signals in the frequency ranges of the n-th order Raman peak can also be achieved using
gratings that confine Raman frequencies up to the (n-1)-th order. Gain is obtained through n-th order stimulated
326
A1 A2 A
A0
A A2 A

10. Raman emission as the signal propagates through the same fiber. Using this method, amplification of 13 lOnm
signals was demonstrated in germanosilicate fibers using Nd3 cladding pumped laser at 1064nm and cascaded
gratings for 11 1 ,1 1 5 and 1240nm Gain up to 25dB has been achieved with 350mW pump power4.
Ultrahigh output power of 8.5 W at 14 2nm and having 4 % slope efficiency has been demonstrated in a
cascade Raman fiber laser. Broadband operation is feasible since Raman peaks are relatively broad and the pump
wavelength of cladding-pumped fiber lasers can be varied by as much as lOOnm.It has been observed that the
frequency shift (l320cm1) of the first-order Raman scattering peak in P-doped silica is about three times larger
than that for Ge-doped and un-doped silica Recently, It has been demonstrated that strong Bragg gratings can
be written using ArF excimer (193nm) in P-doped fibers sensitized by deuterium loading 338.Thus ,P-doped
fibers can permit the use of a fewer number of Stokes transitions between the pump wavelength and 1.3 itm for
signal amplification. Since the Raman gain coefficient at 1320 cm1 in P-doped difficulties in incorporatting a
high concentration of P-dopant, the delta is usually less than 1 .5-2%in fibers doped exclusively with phosphorus.
A significant increase in attenuation loss is also observed in fibers containing a high phosphorus dopant
concentration. Thus, further improvement in the fiber processing will be required to fully utilized the advantages
of P-doped fibers for Raman amplification applications.
3.CONCLUSIONS
The availability of high-performance optical amplifiers and other advanced optical technologies, as well as the
market demand of more bandwidth at lower costs, have made optical networking an attractive solution for
advanced networks. Optical network uses the WDM wavelengths not only to transport large capacity but also to
route and switch different channels. Compared to point-to-point systems, optical networking applications need
higher optical amplifier requirements such as gain flatness, wide bandwidth, and dynamic gain control.
Flatness affects system performance in many ways. Flat gain amplifiers are essential for achieving the
system OSNR margin for routed channels and minimizing power divergence to allow practical implementation
of networking on the optical layer. Wide bandwidth can either enable large channel spacing as a countermeasure
of the filter bandwidth narrowing effect or allow more optical channels for more flexibile routing of traffic.
Dynamic gain control is critical to maintaining system performance under varied channel loading conditions
caused by either a network reconfiguration or a partial failure. In addition to the traditional optical amplifier
attributes--output power and noise figure—future amplifiers are not only expected to deliver more (wide signal
band) bandwidth and higher-quality (flat gain spectrum) bandwidth, but managed bandwidth with well-
controlled gain shape and amplifier dynamics.
Considerable progress has been made in optical amplifier technology in recent years. The bandwidth of
amplifiers has increased several times and flat gain amplifiers with 84 urn ofbandwidth have been demonstrated,
made possible by addition of the L-band branch. With the advent of Raman fiber amplifiers, more wider
bandwidth is obtained. Progress has also been made in the understanding of amplifier gain dynamics. Several
control schemes have been successfully demonstrated to mitigate the signal impairments due to fast power
transients in a chain of amplifiers and will be implemented in optical network design. Terrestrial optical systems
have been increasing in transmission capacity. To meet the enormous capacity demand, the presently available
hundreds of Gb/s capacity system with more than 50 channels on a single optical fiber will soon be followed by
systems having terabit capacity.
327

11. ACKNOWLEDGEMENTS
Present work is funded by China Government "863-3 1 ".
REFERENCES
1. J.L. Zyskind, C.R. Giles, J.R. Simpson, "Erbium-Doped Fiber Amplifiers and the Next Generation of
Lightwave Systems," AT&TTech. J., Vol. 1 ,No. 1 , Jan-Feb. 1 992, pp. 53-62.
2. R.G. Smart, J.L. Zyskind, and D.J. DiGiovanni, "Two-Stage Erbium-Doped Fiber Amplifiers Suitable for Use
in Long Haul Soliron Systems," Elect. Left, Vol. 30, No. 1, Jan. 1994, pp. 50-52.
3. A.R. Chraplyvy, J.M. Delavaux, R.M. Derosier,G.A. "1420-km Transmission of Sixteen 2.5-Gb/s Channels
Using Silica Fiber-Based EDFA Repeaters," IEEE Photon. Tech. Left., Vol. 6, No. 11, Nov. 1994, pp. 13 1- 3.
4.A.M. Vengsarkar, J.R. Pedrazzani, J.B. Judkins,et al. "Long-Period Fiber-Grating-Based Gain Equalizers," Opt.
Left., Vol. 21, No. 5, Mar. 1996, pp. 336-338.
5. R. Kashyap, R. Wyatt, and P.F. McKee, "Wavelength-Flattened Saturated Erbium Amplifier Using Multiple
Side-Tap Bragg Gratings," Elect. Left, Vol. 29, No. 1 1 ,May 1 993, pp. 1025-1026.
6. Y.P. Li, C.H. Henry, E.J. Laskowski, et al."Waveguide EDFA Gain Equalization Filter," Elect. Left., Vol. 31,
No. 23, Nov. 1995, pp. 2005-2006.
7. S.H. Yun, B.W. Lee, H.K. Kim, "Dynamic Erbium-Doped Fiber Amplifier with Automatic Gain Flattening,"
Opt. Fiber Comntun. Confi (OFC '99), San Diego, Calif., Feb. 21-26, 1999, postdeadline paper PD28.
8. Y. Sun, J.B. Judkins, A.K. Srivastava, "Transmission of 32 WDM 10-Gb/s Channels over 640 km Using
Broadband Gain-Flattened Erbium-Doped Silica Fiber Amplifiers," IEEE Photon.Tech. Lett, Vol. 9, No. 12, Dec.
199 , pp. 1652-1654.
9. P.F. Wysocki, J.B. Judkins, R.P. Espindola, "Broad-band Erbium-Doped Fiber Amplifier Flattened Beyond 40
nm Using Long-Period Grating Filter," IEEEPhoton. Tech. Left., Vol. 9, No. 10, Oct. 199 ,pp. 1343-45.
10. A.K. Snvastava, Y. Sun, J. L Zyskind, "Error-Free Transmission of 64 WDM 10-Gb/s Channels over 520 km
of TrueWave Fiber," Proc. European Confi on Optical Commun. (ECOC '98), Madrid, Sept. 20-24, 1998, pp.
265-266.
11. Y. Sun, A.K. Srivastava, J.H. Zhou, J.W. Suihoif, "Optical amplifier for WDM optical network",Bell-
Lab.Technique Journal, Janunary-March 1999, P18 -203.
12. Y. Sun, J.W. Suihoff, "Ultra-Wideband Erbium-Doped Silica Fiber Amplifier with 80 nm of Bandwidth,"
Proc. Optical Amplifiers and their Applications, Victoria, Canada, July 199 , postdeadline paper P1-2.
13. M. Yamada, H. Ono, T. Kanamori, "Broadband and Gain-Flattened Amplifier Composed of a 1.55 1-'m
Band and a 1 .58 it m Band Er3-Doped Fiber Amplifier in a Parallel Configuration," Elect. Left., Vol. 33, No. 8,
Apr. 1 99 , pp. 1 0- 11.
14. Y. Sun, J.W. Suihoif, A.K. Srivastava, "An 80-nm Ultra-Wideband EDFA Low Noise Figure and High
Output Power," Proc. European Conference on Optical Commun. (ECOC '9 ), Edinburgh, U.K., Sept. 22, 199
postdeadline paper TH3C,pp. 69- 2.
15. C.R. Giles, E. Desurvire, and J.R. Simpson, "Transient Gain and Crosstalk in Erbium-Doped Fiber
Amplifiers," Opt. Left., Vol. 14, No. 16, 1989, pp. 880-882.
16. A.K. Srivastava, Y. Sun, J.L. Zyskind, "EDFA Transient Response to Channel Loss in WDM Transmission
Systems," IEEE.Photon. Tech. Left., Vol. 9, No. 3, Mar. 199 , pp. 386-388.
328

12. 17. E. Desurvire, "Analysis of Transient Gain Saturation and Recovery in Erbium-Doped Fiber Amplifiers,'
IEEE Photon. Technol. Lett., Vol. 1, No. 8, Aug. 1989, pp. 196-199.
18. Y. Sun, J.L. Zyskind, AK. Srivastava, "Analytical Formula for the Transient Response of Erbium-Doped
Fiber Amplifiers," Applied Optics, Vol. 38, No. 9, Mar. 1999, pp. 1682-1685.
19. J.L. Zyskind, Y. Sun, "Fast Power Transients in Optically Amplified Multiwavelength Optical Networks,"
Opt. Fiber Commun. Confi (OFC '96), San Jose, Calif., Feb. 25-Mar. 1, 1996, postdeadline paper PD31.
20. Y. Sun, A.K. Srivastava, J.L. Zyskind, "Fast Power Transients in WDM Optical Networks with Cascaded
EDFAs," Elect. Left., Vol. 33, No. 4, Feb. 199 ,pp. 313-314.
21. Y. Sun, A.K. Srivastava, "Multiwavelength Optical Networking Consortium (MONET), "Payable Milestone
No. 2, Quarterly Report, 1995.
22. Y. Sun, J.L. Zyskind, A.A.M. Saleh, "Model for Gain Dynamics in Erbium-Doped Fiber- Amplifiers," Elect.
Left., Vol. 32, No. 16, Aug. 1,1996, pp. 1490-1491.
23. Y. Sun and A.K. Srivastava, "Dynamic Effects in Optically Ampifled Networks," Proc. Optical Amplifiers
and their Applications, Victoria, Canada, July 199 , pp. 33 3-353.
24. E. Desurvire, Erbium-Doped Fiber Amplifiers Principles and Applications, New York, 1994.
25. A.K. Srivastava, Y. Sun, J.L. Zyskind, "Fast Gain Control in Erbium-Doped Fiber Amplifiers," Proc. Optical
Amplifiers and their Applications, Monterey, Calif., July 11-13, 1996, postdeadline paper PDP4.
26. M. Zirngibl, "Gain Control in Erbium-Doped Fiber Amplifiers by an All-Optical Feedback Loop," Elect.
Lett., Vol. 2 , No. , 1991, pp. 560-561.
27. J. Jackel and D. Richards, "All-Optical Stabilization of Cascaded Multichannel Erbium-Doped Fiber
Amplifiers with Changing Numbers of Channels," Opt. Fiber Commun. Confi (OFC '9 )Tech. Digest, Dallas,
Texas, Feb. 16-2 1 , 199 , paper TuP4.
28. E. Desurvire, M. Zirngibl, H.M. Presby, "Dynamic Gain Compensation in Saturated Erbium-Doped Fiber
Amplifiers," IEEE Photon. Tech. Lett., Vol. 3, No. 5, May 1991, pp. 453'—455.
29. G. Luo, J.L. Zyskind, Y. Sun, "Relaxation Oscillations and Spectral Hole Burning in Laser Automatic Gain
Control of EDFAs," Opt. Fiber Conimun. Confi (OFC '9 ) Tech. Digest, Dallas, Texas, Feb. 16-21, 199 ,paper
30. 5. Kinoshita, Y. Sugaya, H. Onaka, "Low Noise and Wide Dynamic Range Erbium-Doped Fiber Amplifiers
with Automatic Level Control for WDM Transmission Systems," Proc.Optical Amplifiers and their Applications,
Monterey, Calif., July 1 1-1 3, 1 996, pp. 211-214.
31 .S.G.Grubb,T.Erdogan,V.Mizarhi,T.Strasser, "optical amplifier and their application", paper PD3,Technique
Digest Senes,Vol.l4,Optical Society ofAmerican, Washington DC(1994).
32.E.M.Dinov,D.G.Fursa,A.A.Abramov, "Raman optical fiber amplifier of signals at the wavelength of
1 .3um",Quantum Electronics,24[9] 49- 51(1994).
33.K.Suzuki,K.Noguchi and N.Uesugi, "Selective stimulated Raman Scattering in Highly P2O5-doped Silica
single mode fibers", Optics letters ll[l0] 656-658(1986).
34.T.Kitagawa,K.O.Hill,D.C.Johnson, Conference on Optical Fiber Communication, paper PD-l . Technique
Digest Series, Optical Society of American, Washington DC(l994).
35.B.Malo,J.Alber,F.BiloDeau, "photosensitivity in phosphorus-doped silica glass and optical waveguides".
Applied Phys.Lett.65(4) 394-396.
36.T.A.Strasser,A.E.White,M.F.Yan, "Strong Bragg Phase Grattings in Phosphorus-doped Fiber induced by ArF
Excimer Radiation", OFC,Technique Digest Series, Optical Society of American, Washington DC(1995).
329