JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO.

1, JANUARY 2004

63

Amplifiers for the Masses: EDFA, EDWA, and SOA

Amplets for Metro and Access Applications
Donald R. Zimmerman and Leo H. Spiekman, Member, IEEE
Tutorial Paper

AbstractSmall erbium-doped amplets and semiconductor

optical amplifiers will be used in current and future metro and
enterprise networks in various configurations. Many new system
architectures will be enabled as these low-cost technologies are
used to compensate for transmission and impairment-compensating component losses. This paper discusses the definition, use,
and technologies associated with these new classes of optical amplifiers which, though little, will impact next-generation networks
a great deal.
Index TermsMetropolitan area networks, optical communication, optical fiber amplifiers, optical planar waveguide
components, semiconductor optical amplifiers.

I. INTRODUCTION

INCE the early days of optical amplifier usage in networks,

the variety of applications served and the value provided
has increased dramatically. Early applications used single-pump
erbium-doped fiber amplifiers (EDFAs) in booster configurations to extend the range of 1.5 m transmission links. Shortly
thereafter, coarse wavelength-division multiplexing (CWDM)
was being employed with booster amps to double the capacity
of installed fiber routes. Cascades of EDFA were under investigation for undersea use; first in single wavelength configurations. It soon became apparent that multiple wavelengths could
be supported with cascaded optical amplifier (OA) systems and
an entire industry was born based on dense wavelength-division
multiplexing (DWDM) and EDFA technology.
A. Leveraging DWDM
The success of optically amplified DWDM systems in
long-haul applications was primarily driven by the cost
efficiency of sharing these relatively new and expensive
gain-flattened EDFAs across many wavelengths. Initial
DWDM deployments were used in back-bone networks where
network reconfigurations were few and far between. The economics of DWDM were next applied to metropolitan networks
where reconfigurations were much more frequent. Wavelength
routing and optical protection switching have become commonplace in these networks. This forces additional complexity

Manuscript received June 26, 2003; revised October 7, 2003.

D. R. Zimmerman is with the Light Systems Associates, Farmingdale, NJ
07727 USA (e-mail: [email protected]).
L. H. Spiekman is Vrijkensven 17, 5646HP Eindhoven, The Netherlands
(e-mail: [email protected]).
Digital Object Identifier 10.1109/JLT.2003.822144

onto the OA designer to suppress transient crosstalk between

wavelengths. As OAs are pushed further toward the edges of
the network, cost and complexity become key concerns.
As with many maturing technologies, the bifurcation of the
amplifier space into higher and lower performance solutions has
occurred. Costs for providing basic amplification functionality
have plummeted, driven by improvements in technology and
manufacturing efficiency developed during the golden age of
DWDM systems. The everchanging landscape of technological
development and the economics afforded by each solution force
us to ponder the question: Are there better and more cost-effective ways to deploy the available technologies than we have
done in the past? In the early days, it was all about deploying the
maximum bandwidth from point A to B with the lowest overall
capital cost. Now carriers are much more concerned with first
deployment cost, capacity growth profile, and operational expenditure. Architectural decisions from the past may not be the
best solutions for the future.
B. The Amplet Arrives
New architectures that take advantage of low-cost, modest
performance amplifiers are being developed today for many different applications. These amplifiers have been called amplets
by the early adopters of the technology to distinguish them from
their larger and more complex predecessors. Although it is difficult to give one overarching definition for this class of optical
amplifier, the common thread is reduced performance and lower
cost as compared to a traditional broadband DWDM amplifier.
As an example, subbanded DWDM system application amplets
are being used with only 4 to 8 ITU channels (compared to 32 to
40 channels of typical broadband DWDM OA) therefore, total
power requirements are reduced by as much as 9 to 10 dB. For
an EDFA, this allows for less pump power and fewer gain stages
resulting in a much lower cost product.
Each application dictates the required performance and
operational characteristics of an amplet but a few critical
requirements are shared by all. Amplets are expected to be
low cost. Price is dictated by performance and function but
must be sufficiently low to supplant other architectural choices.
Amplets must be small in size. Functional packing densities of
systems will increase over time to lower both first installation
and operational expenses. Amplets must have low power
consumption. As systems get smaller cooling becomes more
problematic. Amplets must have at least the reliability of their
big brothers in the telco central office environment. As they

0733-8724/04$20.00 2004 IEEE

64

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 1, JANUARY 2004

and boost signal levels for transmission. This application is

ideal for an erbium-doped waveguide amplifier (EDWA) array
where pump sharing and VOA control could be utilized.
B. High-Speed Systems Improvement

Fig. 1. Typical specified performance of three amplet types: EDFA, EDWA,

and SOA.

make their way out to the network edge, they may be required
to withstand the harsher conditions of underground vaults and
customer premises. Lastly, amplets must be easy to use. If an
amplet has a limited operational range or must be controlled
very accurately it may limit the usefulness of the solution.
As can be seen in Fig. 1, the performance of the various
amplet technologies under consideration has improved dramatically to the point where they are all quite comparable.
Throughout the remaining few sections the authors will endeavor to describe various amplet applications, technologies,
and design considerations, to shed light on the benefits and
pitfalls with each available solution.
II. APPLICATIONS OF AMPLETS
Current and future architectural solutions will certainly take
advantage of lower cost, modest performance amplets to provide
better overall network performance with a modularity that scales
with deployed capacity.
A. DWDM Subbanded Line Systems
Single wavelength and subbanded architectures can take
advantage of amplets to provide gain on an incremental growth
basis. For example, a 32-channel DWDM system could be
designed using transient controlled DWDM amplifiers and
32-channel wavelength-division multiplexers (WDMs) at the
terminus points. This is a very cost-effective architecture for
fully loaded systems when no wavelength add/drop occurs.
Next, consider the situation where two nodes require wavelengths to be dropped using fixed WDM and back-to-back
DWDM amplifiers. Finally, consider using up to 16 amplets per
node with subbanded multiplexing as shown in Fig. 2. When
the number of add/drop nodes is high and the channel count is
low, this architecture can be most cost effective. There may be
a small price premium for a fully loaded system but the control
is simpler and it offers the added flexibility of simpler network
reconfigurations. Its first installed cost may be substantially
lower than a DWDM amplifier solution. Fig. 3 shows how in a
fully reconfigurable wavelength add/drop node amplets would
be used at both drop and add ports to overcome device losses

As transmission speed increases to 40 Gb/s and beyond, it

becomes harder to maintain OSNR through the labyrinth of dispersion compensation devices in a system. Use of amplets in
both the transmitter and receiver circuit modules should be considered to increase system margin through OSNR improvement
(see Fig. 4). Wavelength tunable transmitters typically do not
have as much power as their DFB counterparts and high bit-rate
modulators are considerably more lossy than lower bit-rate devices. In addition, dispersion precompensation is required in
many links. An amplet located after modulating and before any
dispersion precompensation allows greater launch power into
either a DWDM mux or a fiber link preserving OSNR. Receiver
sensitivity is enhanced through the use of a preamplifier amplet before a PIN receiver. In either single channel or DWDM
systems, the link performance can be improved. Improvements
of greater than 2 dB have been seen at 10 Gb/s with an EDWA
preamplifier [1].
C. SOAs in Dynamic Channel Count Applications
Since semiconductor optical amplifiers (SOAs) amplifying
WDM channels will nearly always be run as a linear system,
such a system will automatically be suited to operate at a nonconstant total average power. Both changes in the number of
WDM channels, e.g., due to reconfiguration of the network, and
use of bursty data, e.g., in a packet network, do not change the
gain of the amplifiers in such a configuration. This is in contrast with networks using erbium which, operated in heavy saturation, reacts strongly to such slow changes in average power
with its millisecond gain dynamics [2].
Demonstrations of SOA-amplified systems with a varying
number of channels use 8 or 16 10-Gb/s DWDM channels, half
of which are switched on and off at a slow (kHz) rate [3], [4].
Fig. 5 shows received spectra from one of these experiments: 16
channels on; 8 channels on and 8 channels off; and 8 channels
on and 8 channels switched at 100 kHz, respectively; all after
four 40-km spans and four SOAs. It is clear that the amplifier
gains, and therefore the received channel powers, do not vary appreciably with the number of channels. Good eye diagrams and
low error rates were observed in both experiments. Repetition
of one of the experiments with nontransient-controlled EDFAs
[4] shows the clear advantage of running a linear system under
these conditions; see Fig. 6.
D. Coarse WDM Systems
Another application area where SOAs can offer an advantage is in the amplification of coarse WDM (CWDM) data. The
CWDM standard defines a coarse wavelength grid of 20-nm
spaced channels with 13-nm passbands, to allow use of cheap
filter technology and uncooled lasers. Eighteen channels are defined from 1270 to 1610 nm. CWDM can add capacity to simple,
e.g., Gigabit-Ethernet, point-to-point links, and can add OADM
flexibility to more complex datacom networks. Introduction of

ZIMMERMAN AND SPIEKMAN: AMPLIFIERS FOR THE MASSES

65

Fig. 2. Example of incremental growth in 32 channel subbanded DWDM system with fixed add/drop.

Fig. 3.

Use of amplets at drop and add sides of flexible wavelength router.

Fig. 5. Received spectra in a SOA-amplified system with dynamic number
of channels. Top to bottom: all channels on; even channels off; even channels
switched on and off at 100 kHz.

Fig. 6. Eye diagrams of one of the surviving channels of Fig. 5 (20 ps/div).
Left: quasi-linear system using SOAs; right: system with saturated EDFAs.

at the edges was obtained, which allowed extension of the reach

of this CWDM system by 30 km.
Fig. 4. High-speed applications with wavelength agile transmitter and
preamplified PIN receiver.

the WDM multiplexers adds loss, however, which can be recovered using amplets. Varying the composition of the active layer,
the gain peak of a SOA can reach any CWDM channel, and the
wide gain bandwidth of the SOA (80 nm 3-dB width typical)
allows it to amplify a decent number of CWDM channels at a
time. A single SOA has even been shown to amplify up to eight
CWDM channels over a bandwidth of 140 nm [5]. The output
spectrum of the amplifier is shown in Fig. 7. A margin improvement varying from 17 dB in the center of the bandwidth to 5 dB

III. OPTICAL AMPLIFICATION BASICS

Optical gain is the most important property of an amplet.
The two families of amplet discussed in this paper, erbiumdoped devices and semiconductor-based devices, provide optical gain based on different but comparable interactions of light
with matter. In erbium-based devices, light from a pump source
elevates ions of the rare-Earth element erbium to an excited state
(see Fig. 8). Optical signals with wavelengths that fall within
the gain spectrum of the erbium induce stimulated emission
and are thereby amplified. In semiconductor devices, the energy
levels of the erbium ion are replaced with the energy bands of

66

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 1, JANUARY 2004

Fig. 10.

Basic EDFA configuration. A design with counter-propagating pump.

Fig. 7. Output spectrum of a SOA amplifying 7 CWDM and 8 DWDM

channels.
Fig. 11. Typical packaged SOA chip. Lenses are often used to make the two
fiber-chip couplings.

Fig. 8.

Energy level scheme of the Er

ion.

Fig. 12. Typical gain versus output power curve of an optical amplifier. The
3-dB gain compression point is indicated, usually denoted by P .

B. Gain

Fig. 9. Carrier recombination in the active layer of a semiconductor amplifier.

the semiconductor crystal, but other than that the gain mechanism is similar. The semiconductor is brought into an excited
state by pumping it electrically, populating the bands with electrons and holes. An optical signal propagating through the device gives rise to carrier recombination, and the associated stimulated emission amplifies the signal (see Fig. 9).
Note that the device properties that are described in this paragraph apply to traditional optical amplifiers as well as amplets.
After all, the characteristics distinguishing amplets from their
larger cousins are not qualitative but rather quantitative.
A. Device Structure
In order to be amplified efficiently, the signal must propagate
through the amplifier in a well-confined manner. Therefore, amplifiers are usually waveguides with gain. The EDFA is the most
well-known example: a waveguide (the optical fiber) is heavily
doped with erbium ions, which provide gain when optically excited by injection of pump light (Fig. 10). Erbium can also be
implanted into a planar waveguide structure, forming an EDWA.
Similarly, a SOA is formed by enclosing an amplifying active
layer, usually indium gallium arsenide phosphide (InGaAsP) of
an appropriate band gap, between cladding layers of lower refractive index, creating a waveguide structure. Light is usually
coupled into and out of it by means of lenses (see Fig. 11). The
cladding layers of the SOA waveguide are p- and n-doped, respectively, allowing electrical pumping by current injection.

The gain spectrum of the optical amplifier is determined by

the energy levels of the erbium ion, or by the bandgap of the
semiconductor. The gain bandwidth of erbium extends from
about 1525 to 1565 nm, covering a considerable part of the
low-loss window of standard single-mode fiber. The spectral
properties of a SOA are determined by the composition of the
InGaAsP active layer, which can be varied to provide gain from
1200 to 1650 nm. For a given composition, the gain bandwidth
is about 80 nm. The gain spectrum is not the only difference between erbium and semiconductor devices. The lifetime of the
excited state is another distinguishing characteristic. The excited state of erbium has an extremely long lifetime ( 10 ms),
leading to slow gain dynamics. As a result, high-data rate signals do not cause any significant gain modulation even in deeply
saturated amplifiers.
In contrast, the carrier lifetime in a SOA typically is 100 ps,
i.e., of the order of the bit period in a 10-Gb/s modulated signal.
Therefore, amplifying such a signal using a saturated SOA will
normally lead to intersymbol interference (ISI). A third difference is the polarization dependence of the device. An erbiumdoped fiber has circular symmetry, and, therefore, the gain of an
EDFA will exhibit negligible polarization dependence. EDWAs
and SOAs based on asymmetric planar waveguides on the other
hand may exhibit polarization-dependent gain. This is reduced
to acceptable levels by proper waveguide design (EDWA) or by
introducing crystal strain (SOA).
C. Output Power
An optical amplifier driven with lots of input power will saturate, i.e., its gain will drop from its small-signal gain value. The

ZIMMERMAN AND SPIEKMAN: AMPLIFIERS FOR THE MASSES

reason is that the power source of the amplifier, the number of

excited erbium atoms or the number of available electron-hole
pairs, is depleted. The saturation of an optical amplifier is usually referenced to the output power at which the gain has been
compressed by 3 dB, as indicated in Fig. 12. An EDFA can be
operated deeply in saturation (when the input power does not
slowly vary, i.e., when the number of optical channels remains
constant).
A saturated SOA, on the other hand, may give rise to ISI and,
in WDM systems, to interchannel crosstalk due to the fast gain
dynamics. Therefore, operation of the SOA is usually restricted
to the quasi-linear regime, and consequently it is more difficult
to get high output power out of a SOA.
D. Noise Figure
Besides the stimulated emission that creates gain, the gain
medium also produces spontaneous emission, which gives rise
to the amplified spontaneous emission (ASE) spectrum of the
amplifier. This ASE noise limits the optical signal-to-noise ratio
(SNR) of a cascade of amplifiers and is quantified in the ampli,
fiers noise figure (NF). This can be denoted as
is the inversion parameter of
in which
the amplifier (i.e., the degree of population inversion, with
and
the fractional number of erbium atoms or carriers in the
ground and excited states, respectively), and is the input coupling loss. Both well-designed EDFAs and SOAs have inversion
factors close to unity, but the fiber-chip coupling loss of the SOA
puts it at a disadvantage. EDFA noise figures typically are 46
dB, while SOA noise figures are usually 68 dB.
E. Gain Ripple
Different phenomena are denoted by the term gain ripple in
EDFAs and in SOAs. Gain ripple in an EDFA refers to the
shape of the gain spectrum which is determined by the wavelength-dependent emission and absorption coefficients of the erbium-doped fiber, weighed by the fractional populations of the
excited and ground states of the erbium. Gain flattening filters
are sometimes used to reduce this gain ripple. If channel loading
or input levels are changed from their design center, inversion
variation and spectral hole burning will affect the gain flatness
of an EDFA. In-line attenuators are often used in DWDM line
amplifiers to control the inversion and fix the erbium gain, thus,
controlling the spectral tilt [6]. This degree of control is seldom
used in amplet applications due to its added cost and complexity.
The overall gain spectrum of a SOA is determined by the
semiconductor bands, and has a smooth parabolic shape without
the excursions seen in an EDFA gain shape. However, SOAs
are extremely short devices ( 1 mm, compared to many meters for an EDFA), so that reflections at the end facets can give
rise to round-trip resonances that lead to a ripple with a period
of a few tenths of nanometers in the wavelength domain. With
countermeasures like antireflection coatings and angled facets,
the magnitude of this gain ripple can be reduced to 0.1 dB.
IV. ERBIUM AMPLET DESIGN AND TECHNOLOGY
There are two fundamentally different amplet technologies
that utilize erbium-doped glass as the gain medium: one is

67

fiber-based (EDFA) and the other is planar waveguide-based

(EDWA). Although the guiding structures and design approach
are significantly different, one can expect them to have similar
temporal, spectral, and saturation performance.
A. EDFA
Erbium-doped fiber has been in use since the late 1980s as the
gain medium of choice for optical amplifiers [7][9]. In the early
1990s, great improvements in efficiency, spectral performance,
splicability, and numerical modeling were made [10][12] resulting in a robust gain medium that was ideal to exploit for
long-haul and metro-area systems. Recent improvements in the
control of concentration quenching and reduction in cladding diameters to 80 m have yielded fibers that are much better suited
for amplet use. Shorter EDF lengths and tighter bend radii allow
the developer greater flexibility in the design of smaller packages.
A typical EDFA amplet consists of a pump laser, pump WDM
coupler, EDF spool, input and output isolators, and input and
output tap/detectors. Each component takes on new characteristics for use in an optimal EDFA amplet. Small size and low cost
drive the design decisions toward new component choices.
The most costly component for the EDFA amplet is the pump.
Design choices favor 980 nm devices for their improved noise
performance and reduced power consumption. New coolerless
mini-DIL pumps are currently being offered at up to 200 mW
operating power. These devices must operate over large temperature and drive current ranges. It is not uncommon to see the
pump chip gain peak shift by greater than 20 nm as the temperature changes from 0 to 70 degrees C and the output power is
varied from 20 to 200 mW. So that the pump energy remains
centered in the peak erbium absorption region, the pumps are
wavelength locked with a fiber grating, often with polarization
maintaining fiber for improved lock-range over all polarization
states. Although the mini-DIL pumps are intended to be lower
in cost than the larger butterfly packages, their small size and
reduced power consumption are the primary drivers for use in
an EDFA amplet [13], [14].
New advances in erbium-doped fiber have created an opportunity to shrink the package size while still maintaining the
performance of larger single pump amplifiers. Newly available
fibers have peak absorptions greater than 30 dB/m while maintaining pumping efficiency and satisfactory noise performance
[15]. These fibers allow shortening of the fiber by as much as 3x
as compared to standard EDF optimized for DWDM. Vendors
are beginning to offer 80 m cladded versions of these fibers so
that EDF spools can be wound tighter without incurring undue
failure risk.
Optical components are chosen for their small form factor
and ease of use. Many optical components are now being offered with 80 m cladded fiber. When high NA fiber designs are
specified, bend losses are reduced allowing very tight package
designs. The fused fiber components are shorter due to a reduced
taper length that the smaller cladding diameter affords. Photodetectors with integrated taps are also worthy of consideration due
to their dual-use status while incurring a minimal size penalty.
For multiple amplifier array applications, one might even consider hybrid architectures that use passive waveguide devices

68

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 1, JANUARY 2004

coupled with EDF to leverage the higher pump efficiency of the

fiber and the multicomponent cost and size savings of passive
waveguides.
EDFA technology is clearly maturing. In the future, EDFA
amplets will leverage component improvements and design experience in new ways to push the boundaries of size, cost, and
performance.
B. EDWA
The great promise of erbium-doped planar waveguide
technology is the integration of many functions onto an easily
mass-produced photonic IC. Great strides have been made
in this area with recent results yielding amplifiers of very
good performance with high degrees of integration. Two
basic technologies have been used to achieve these recent
results: plasma-enhanced chemical vapor deposition (PECVD)
and metal ion exchange (IE). With PECVD the passive and
erbium-doped waveguides are deposited directly onto the
same silicon substrate in an integrated fashion [16]. With IE
technology, metal ions are imbedded into a glass substrate
to selectively raise the refractive index in the waveguide.
Erbium-doped glass is used for the active waveguides and clear
glass is used for the passive waveguides. The active and passive
sections are then joined together for the final integration [17].
One of the main performance differences between EDFA and
EDWA can be seen in pumping efficiency. The concentration of
erbium in a EDWA is approximately 1020 times higher than
that of an EDFA. Due to the high concentrations of erbium in the
waveguide system, concentration quenching occurs at the higher
pumping levels. Additionally, waveguide losses are much higher
in planar waveguides than in fiber. With large input signals as
much as twice the pump power may be required for an EDWA to
reach output powers on parity to an EDFA. For some amplet applications this could be a concern. But as available pump power
continues to go up and pump failure rates diminish, this should
become less of a concern [18].
A key benefit of waveguide technology is the ability to integrate many functions in a cost effective manner, automating
many of the tasks now currently required to assemble an EDFA.
For applications where multiple amplets are required, array amplifier technology has proven effective at reducing size and expense (see Fig. 13). Pump sharing architectures have been developed to utilize either a one or two high-powered pump(s) and
distribute them to either four or eight individual EDWAs [19].
Mach-Zender VOAs are optionally written into the pump paths
to control each EDWA individually as required. Photodetection
of both input and output signals for control purposes has been
demonstrated using numerous schemes. Stray light management
is a key concern that all waveguide designers must consider to
achieve accurate monitoring.
The integration of all necessary amplifier components is
hampered by the availability of Faraday-effect materials for
integrated isolation. Most suppliers are currently experimenting
with methods to attach bulk isolators to their waveguides with
sufficient performance and stability to eliminate fiber coupled
devices. The focus of this activity is cost and size reduction.
As with the EDFA, the cost of the pump is a major concern.
Maybe even more so since the pumping efficiency is lower in

Fig. 13.

Array EDFA with pump sharing and independent pump control.

Fig. 14. SOA device structure. Mesa, blocking layers, and cladding are often
grown in three separate MOCVD runs.

EDWA. One method under consideration for controlling cost

and reducing size is to couple a pump directly to the waveguide
without an intermediary fiber. The difficulties of achieving a
stable and robust pump package are well known and as such
we might expect this activity to take some time to achieve commercial acceptance.
Further integrations with additional network functions are
just around the corner. As market demand picks up and a drive
toward the next generation platform that is smaller and less
costly commences, designers will have greater tools and much
more flexibility to incorporate amplification into their network
routing components.
V. SOA DESIGN, TECHNOLOGY, AND DEVICE PHYSICS
SOA device design is similar to semiconductor laser design.
The typical SOA is an MOCVD-grown layer structure consisting of an active layer sandwiched between p- and n-doped
cladding layers which allow current injection. Lateral optical
confinement is accomplished by etching a mesa, which is
overgrown with a current blocking structure, which can be
semi-insulating InP or a diode structure in reverse direction
(see Fig. 14).
As aforementioned, a SOA is supposed to deliver gain in a
traveling-wave fashion. Unlike a laser structure, that depends
on facet reflections, in a SOA reflections must be avoided as
much as possible, which usually leads to an implementation
with an angled gain stripe [20] and facet antireflection coatings [21]. An other important difference is that a laser emits
in one (usually TE) polarization, while a SOA should amplify
incoming signals independent of their polarization. This is accomplished by tuning the geometry and composition of the active layer. In particular, the type and amount of crystal strain
has a large influence: Compressive strain leads to TE amplification, while a tensile strained layer mainly amplifies TM-polarized light. Careful tuning of the strain in alternating tensile
and compressive quantum wells [22], or control of the amount

ZIMMERMAN AND SPIEKMAN: AMPLIFIERS FOR THE MASSES

69

of tensile strain in quantum wells [23] or in a bulk active layer

[24], can deliver small ( 0.2 dB) polarization dependence.
A. Output Power and Gain Dynamics
The output power of a SOA is reported in terms of its
,
the power at which the gain is compressed by 3 dB. The highest
values
power SOAs that have been reported to date possess
of 17 dBm [25], [26]. For a single-polarization device, a value
of 20 dBm has been reported [27]. It must be noted that in am,
plification applications, the SOA can not be operated at its
since the fast gain dynamics of the device (carrier lifetime 100
ps) would cause its gain to be modulated by the bit pattern on
the input signal. Likewise, cross-gain modulation (XGM) will
cause crosstalk in amplified WDM signals. When the device is
operated in its (nearly) linear regime (see Fig. 12), the gain modulation is negligible and WDM operation is feasible, as will be
discussed later.

Fig. 15. Transmission of 32 WDM channels modulated at 10 Gb/s across four

40-km spans of standard fiber using SOAs as line amplifiers.

B. Four-Wave Mixing
The phenomenon of four-wave mixing (FWM) occurs in the
SOA as a result of intraband processes such as spectral hole
burning and carrier heating [28]. Compared to FWM in fiber,
the interaction length in a SOA is so short that no walkoff occurs between different wavelength signals, so the strength of the
mixing products is solely determined by the power of the interacting signals
and by the FWM-efficiency,
of the inwhich strongly varies with the frequency spacing
teracting signals. The signals must be copolarized for FWM to
occur.
above and below the
FWM mixing products appear one
interacting signals. In a WDM system, this usually means they
interfere with an other channel. Therefore, the power levels in
a SOA-based WDM system must be controlled to minimize the
occurrence of FWM. Since the output power of the SOA must
be confined to the (quasi) linear regime anyway to avoid XGM,
this poses no additional limitation in WDM operation for current generation devices. However, in future higher power SOAs,
FWM and not XGM may be the limiting phenomenon when designing the system power map.
C. SOA-Based WDM Amplification
Design of systems based on SOAs is different from designing
an erbium-based system, in that SOAs are essentially constant
gain devices, that should not be saturated in order to avoid
XGM, while EDFAs are typically used in constant output
power mode under heavy saturation. Consequently, the SOA
gain has to be matched to the (span or passive component) loss
it is meant to compensate. Between the minimum per-channel
input power required to maintain good optical signal-to-noise
ratio (OSNR) and the maximum total output power limited by
XGM, this leads to moderate span lengths and channel counts.
As an example, a 32-channel (10-Gb/s) system is shown in
Fig. 15. Here, four SOAs with a gain of 13 dB are used to compensate the loss of 40-km spans of standard single mode fiber
plus appropriate amounts of dispersion compensating fiber. The
SOAs are operated at an average output power of 7 dBm,
which puts the peak power about 2 dB below the
of these

Fig. 16. Q-factors measured at the end of the system as shown in

Fig. 15. Varying the launched optical power reveals the limits of SNR and
nonlinearities. The left curve (squares) reflects a quasi-linear system; the right
curve (diamonds) shows the effect of adding a reservoir or ballast channel.

devices, which is 12 dBm. This way, the maximum gain compression remains below 1 dB [29]. The (per-channel) SOA input
devices
power of 21 dBm is sufficient for these
to yield reasonable OSNR after four spans.
Fig. 16 shows Q-factors measured at the receiver versus
launched power. In the optimum, with an OSNR 20 dB, an
for all
average Q-factor of 16.8 dB is observed (
channels). Based on the OSNR alone, a Q of 18 dB would be
expected (left dashed line). The XGM distortion due to gain
compression in the SOAs (right dashed line) deteriorates the Q
for all channels.
with 1.2 dB. Still, the BER is
The method of adding a ballast or reservoir channel has been
suggested to reduce XGM distortion. The always-on reservoir
channel reduces the power swing at the output of the SOA and
therefore partly suppresses the gain modulation. The effect of
this method depends on the system in which it is used. In an early
32 2.5 Gb/s experiment the reservoir channel made a lot of
difference [30]. On the other hand, in the experiment discussed
here, it allows use of larger output powers, but does not improve
the Q-factor (see Fig. 16).
The output powers delivered by the SOAs in this example are
sufficiently moderate to stay out of the regime of fiber nonlinearities. In such an, essentially linear, system, improvement of
of the devices directly leads
either the noise figure or the
to an equal performance improvement in terms of channel count

70

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 22, NO. 1, JANUARY 2004

or span length. Therefore, with recent

and
devices, large improvements over the results as discussed here are to be expected.
VI. SUMMARY
In this paper, several technologies have been discussed to construct optical amplifiers that are suitable for the low-cost, moderate performance application space. These amplifiers must be
small in size and easy to control to allow their use in many places
in the network. The different technologies, EDFA, EDWA, and
SOA, have different properties making them suitable for a variety of applications. Gain, noise figure, and output power of
amplets currently made in these technologies seem to be suitable for single- and multichannel metro and access operation.
The best choice among them is highly application and architecture dependent. Therefore, it pays to be aware of the commonalities and differences between the members of this class of
devices.
REFERENCES
[1] A. J. Bruce, R. Durvasula, and M. Itzler et al., Performance of a high
sensitivity receiver with an EDWA, in OFC 2002.
[2] A. K. Srivastava, Y. Sun, J. L. Zyskind, and J. W. Sulhoff, EDFA transient response to channel loss in WDM transmission system, IEEE
Photon. Technol. Lett., vol. 9, pp. 386388, Mar. 1997.
[3] A. H. Gnauck, L. H. Spiekman, J. M. Wiesenfeld, and L. D. Garrett,
Dynamic add/drop of 8-of-16 10-Gb/s channels in 4 40 km semiconductor-optical-amplifier-based WDM system, in Optical Fiber Communication Conf. OFC 2000, Baltimore, MD, Mar. 510, 2000, paper
PD39, pp. 284286.
[4] J. J. J. Crijns, L. H. Spiekman, G. N. van den Hoven, E. Tangdiongga,
and H. de Waardt, Static and dynamic switching performance of a metro
WDM ring using linear optical amplifiers, IEEE Photon. Technol. Lett.,
vol. 14, pp. 14811483, Oct. 2002.
[5] P. Iannone, K. Reichmann, and L. Spiekman, In-service upgrade of
an amplified 130-km metro CWDM transmission system using a single
LOA with 140-nm bandwidth, in Optical Fiber Communication Conf.
OFC 2003, vol. 2, Atlanta, GA, Mar. 2328, 2003, pp. 548550.
[6] K. Wundke, Advanced amplifier design: physics and systems limitations, in Optical Fiber Communication Conf. 2003, Mar. 2328, 2003,
pp. 373374.
[7] E. Desurvire, J. R. Simpson, and P. C. Becker, High-gain erbium-doped
traveling wave fiber amplifier, Opt. Lett., vol. 12, no. 11, pp. 888890,
1987.
[8] C. R. Giles, E. Desurvire, J. R. Talman, J. R. Simpson, and P. C. Becker,
:  in an erbium-doped
2-Gbit/s signal amplification at 
single-mode fiber amplifier, J. Lightwave Technol., vol. 7, pp. 651656,
Apr. 1989.
[9] E. Desurvire, C. R. Giles, and J. R. Simpson, Gain saturation effects in high-speed, multichannel erbium-doped fiber amplifiers at

:  , J. Lightwave Technol., vol. 7, pp. 20952104, Dec.
1989.
[10] C. R. Giles and E. Desurvire, Modeling erbium-doped fiber amplifiers,
J. Lightwave Technol., vol. 9, p. 271, 1991.
[11] C. R. Giles, C. A. Burrus, D. J. DiGiovanni, N. K. Dutta, and G. Raybon,
Characterization of erbium-doped fibers and application to modeling
980-nm and 1480-nm pumped amplifiers, IEEE Photon. Technol. Lett.,
vol. 3, pp. 363365, Apr. 1991.
[12] E. Desurvire, J. W. Sulhoff, J. L. Zyskind, and J. R. Simpson, Study
of spectral dependence of gain saturation and effect of inhomogeneous
broadening in erbium-doped aluminosilicate fiber amplifiers, IEEE
Photon. Technol. Lett., vol. 2, pp. 653655, Sept. 1990.
[13] S. Mohrdiek, T. Pliska, and C. Harder, Coolerless operation of 980 nm
pump modules, in Optical Fiber Communication Conf. Exhibit, 2001,
vol. 3, Mar. 1722, 2001, pp. WDD77-1WDD77-3.
[14] S. Mohrdiek, M. Achtenhagen, C. Harder, and A. Hardy, Detuning
characteristics of fiber Bragg grating stabilized 980 nm pump lasers,
in Optical Fiber Communication Conf. 2000, vol. 3, Mar. 710, 2000,
pp. 168170.

= 1 53 m

= 1 53 m

[15] S. Tammela, M. Hotoleanu, P. Kiiveri, H. Valkonan, S. Sarkilahti, and

K. Janka, Very short Er-doped silica glass fiber for L-band amplifiers,
in Optical Fiber Communication Conf. 2003, Mar. 2328, 2003, pp.
376377.
[16] K. Shuto, K. Hattori, T. Kitagawa, Y. Ohmori, and M. Horiguchi,
Erbium-doped phosphosilicate glass waveguide amplifier fabricated
by PECVD, Electron. Lett., vol. 29, no. 2, pp. 139141, Jan. 21, 1993.
[17] D. Barbier, P. Bruno, C. Cassagnettes, M. Trouillon, R. L. Hyde, A.
Kevorkian, and J. M. P. Delavaux, Net gain of 27 dB with a 8.6-cm-long
Er/Yb-doped glass-planar-amplifier, in Optical Fiber Communication
Conf. Exhibit, 1998, Feb. 2227, 1998, pp. 4546.
[18] G. Nykolak, P. C. Becker, J. Shmulovich, Y. H. Wong, D. J. DiGiolifetimes in
vanni, and A. J. Bruce, Concentration-dependent
-doped fibers and
-doped planar waveguides, IEEE Photon.
Technol. Lett., vol. 5, pp. 10141016, Sept. 1993.
[19] J. Shmulovich, High gain planar optical amplifiers/arrays, in Optical
Fiber Communication Conf. 2003, Mar. 2328, 2003, pp. 378379.
[20] A. E. Kelly, I. F. Lealman, L. J. Rivers, S. D. Perrin, and M. Silver, Polarization insensitive, 25-dB gain semiconductor laser amplifier without
antireflection coatings, Electron. Lett., vol. 32, no. 19, pp. 18351836.
[21] L. F. Tiemeijer, P. J. A. Thijs, T. van Dongen, J. J. M. Binsma, and E.
J. Jansen, Polarization resolved, complete characterization of 1310-nm
fiber pigtailed multiple-quantum-well optical amplifiers, J. Lightwave
Technol., vol. 14, pp. 15241533, June 1996.
[22] L. F. Tiemeijer, P. J. A. Thijs, T. van Dongen, R. W. M. Slootweg, J. M.
M. van der Heijden, J. J. M. Binsma, and M. P. C. M. Krijn, Polarization insensitive multiple quantum well laser amplifiers for the 1300-nm
window, Appl. Phys. Lett., vol. 62, no. 8, pp. 826828.
[23] M. Joma, H. Horikawa, C. Q. Xu, K. Yamada, Y. Katoh, and T.
Kamijoh, Polarization insensitive semiconductor laser amplifiers with
tensile strained InGaAsP/InGaAsP multiple quantum well structure,
Appl. Phys. Lett., vol. 62, no. 2, pp. 121122.
[24] J. Y. Emery, T. Ducellier, M. Bachmann, P. Doussire, F. Pommereau,
R. Ngo, F. Gaborit, L. Goldstein, G. Laube, and J. Barrau, High performance 1.55  polarization-insensitive semiconductor optical amplifier based on low-tensile-strained bulk GaInAsP, Electron. Lett., vol.
33, no. 12, pp. 10831084.
[25] K. Morito, M. Ekawa, T. Watanabe, T. Fujii, and Y. Kotaki, High saturation output power ( 17 dBm) 1550 nm polarization insensitive semiconductor optical amplifier, in 26th Europ. Conf. Optical CommunicationECOC 2000, Munich, Germany, September 37, 2000, pp. 3941.
[26] A. Borghesani, N. Fensom, A. Scott, G. Crow, L. Johnston, J. King, L.
Rivers, S. Cole, S. Perrin, D. Scrase, G. Bonfrate, A. Ellis, I. Lealman, G.
Crouzel, L. H. K.L. How Kee Chun, A. Lupu, E. Mahe, and P. Maigne,
High saturation power (>16.5 dBm) and low noise figure (<6 dB)
semiconductor optical amplifier for C-band operation, in Optical Fiber
Communication Conf. OFC 2003, vol. 2, Atlanta, GA, March 2328,
2003, pp. 534536.
[27] M. Dagenais, P. Heim, S. Saini, S. Wilson, R. Leavitt, A. Yu, T. Horton,
V. Luciani, D. Stone, and Y. Hu, High power C-band semiconductor
booster optical amplifier, in Optical Fiber Communication Conf. OFC
2003, vol. 1, Atlanta, GA, March 2328, 2003, pp. 8587.
[28] J. M. Wiesenfeld, Gain dynamics and associated nonlinearities in semiconductor optical amplifiers, Int. J. High-Speed Electron. Syst., vol. 7,
no. 1, pp. 179222.
[29] L. H. Spiekman, A. H. Gnauck, J. M. Wiesenfeld, and L. D. Garrett,
DWDM transmission of thirty-two 10-Gbit/s channels through 160-km
link using semiconductor optical amplifiers, Electron. Lett., vol. 36, no.
12, pp. 10461047.
[30] Y. Sun, A. K. Srivastava, S. Banerjee, J. W. Sulhoff, R. Pan, K. Kantor,
R. M. Jopson, and A. R. Chraplyvy, Error-free transmission of 32
2.5 Gbit/s DWDM channels over 125 km using cascaded in-line
semiconductor optical amplifiers, Electron. Lett., vol. 35, no. 21, pp.
18631865.

Er

Er

Donald R. Zimmerman, photograph and biography not available at the time of

publication.

Leo H. Spiekman (M97), photograph and biography not available at the time
of publication.