282 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO.

2, MARCH/APRIL 2000

Multiple Signal Representation Simulation of

Photonic Devices, Systems, and Networks
Arthur Lowery, Senior Member, IEEE, Olaf Lenzmann, Member, IEEE, Igor Koltchanov, Rudi Moosburger,
Ronald Freund, André Richter, Stefan Georgi, Dirk Breuer, and Harald Hamster

Abstract—Photonic systems design requires simulation over continue an exponential growth in fiber capacity, denser WDM
a wide range of scales; from wavelength-sized resonances in systems will be required as the fiber bandwidth is used up,
lasers and filters, to interactions in global networks. To design which will have to operate with channel spacings reduced to a
these global systems, while considering the effects of the smallest
component, requires sophisticated simulation technology. We have few times the channel bit rate [4]. These systems will require
developed the Photonic Transmission Design Suite, which includes sophisticated modulation techniques, such as phase/ampli-
five different signal representations, so that the details of device tude/polarization modulation, perhaps including duobinary [5]
performance can be efficiently considered within a large network or single-side band modulation [6]. Furthermore, the design of
simulation. Alternatively, a design can be studied using a coarse optical filters for wavelength multiplexers will have to become
signal representation before switching to a detailed representation
for further refinement. We give examples of the application of more sophisticated, because the filters will have to have flat
these representations, and show how the representation of a passbands, good rejection, and low differential group delays
signal is adapted as it propagates through a system to optimize (low dispersion). This design becomes problematic as the
simulation efficiency. channel bandwidths become a significant fraction of channel
Index Terms—Communication systems, data communication, spacing.
design automation, intersymbol interference, optical amplifiers, A push to all-photonic networks, or at least networks with
optical crosstalk, optical fiber communication, optical propagation photonic switching, will require careful consideration of optical
in nonlinear media, semiconductor lasers.
crosstalk and multipath interference [7]. Low levels of crosstalk
can have a significant effect because of the coherent mixing of
I. INTRODUCTION optical fields. Even if the fields are from different transmitters,
or carrying different data, or even from the same transmitter but
T HE DESIGN of photonic systems has reached a stage in
which simulation is no longer a luxury, but a necessity.
This situation has developed over only a few years, because sys-
over a ghost path longer by more than the coherence length of
the laser, coherent mixing will cause large penalties. Thus, all
paths should be considered in a photonic network, and this re-
tems performance has reached a number of limits. Until the last
quires significant computation if all possible phase combina-
decade, optical communications systems were chiefly limited
tions are considered in networks with complex switch topolo-
by loss, dispersion, and transmitter and receiver performance
gies [8].
[1]. However, loss is easy to calculate on the back of an enve-
lope, and dispersion can be estimated by rule of thumb, aided All-photonic networks will require optical amplification to
by experience. It is the advent of optical amplifiers, enabling compensate for losses in switches and multiplexers on top of
high powers and long unregenerated distances that have caused fiber losses. Cascades of amplifiers could cause power transients
significant fiber nonlinearity that necessitated the use of nu- and strong interaction between WDM channels as the channels
merical modeling: to calculate crosstalk caused by four-wave are switched on and off [9]. Transients are caused by the mil-
mixing and the interplay of nonlinearity and dispersion, such as lisecond dynamics of the amplifiers, but they have nanosecond
in near-soliton and soliton systems [2]. In addition, long unre- features, which is a difficult modeling problem because of the
generated systems suffer from polarization mode dispersion as range of time scales. In the steady-state condition, the gain spec-
a system limitation. trum of amplifiers should be flattened to avoid large differences
Furthermore, new problems requiring computer-aided design in signal-to-noise ratio (SNR) between channels [10].
are beginning to come to light [3]. These problems include Fig. 1 summarizes the challenges to modeling a photonic
the design of components for dense wavelegnth-division mul- communications system, from transmitter, through add–drop
tiplexing (WDM) systems, with several tens of channels. To multiplexers, optical cross connects, long-haul links, and,
finally, at the receiver. The design of an optical component can
directly and significantly affect the performance of an optical
Manuscript received July 20, 1999; revised February 7, 2000. This work was system. The system being affected could cost hundreds of
supported in part by the European Union’s ACTS DEMON Project.
A. Lowery is with Virtual Photonics Inc., Carlton 3053, Australia. millions of dollars: the component could cost tens of dollars.
O. Lenzmann, I. Koltchanov, R. Moosburger, R. Freund, A. Richter, S. It would be too expensive to develop every component and
Georgi, and D. Breuer are with Virtual Photonics Inc., D10587 Berlin, optimize it by testing within a whole system. It would also
Germany.
H. Hamster is with Virtual Photonics Inc., San Bruno, CA 94066 USA. take considerable time to optimize component designs by
Publisher Item Identifier S 1077-260X(00)03856-9. developing a series of prototypes. It may be impossible to
1077–260X/00$10.00 © 2000 IEEE
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 283

Fig. 1. Example of the modeling challenges within a photonic communications network.

compare component technologies not yet in mass production scales of abstraction, from component to large network, each
in large systems. However, the telecommunications industry is with the optimum simulation regime.
demanding rapid improvements and lower costs. Third-generation tools require a mixture of signal represen-
Because of the pressures of increased performance, increas- tations, because it is often necessary to consider a component
ingly sophisticated systems, and reduced design cycles, new de- in a system in great detail, while treating the system or network
sign methods must be found [11]. One possibility would be to more abstractly. Furthermore, in frequency space, it may be nec-
tightly specify the performance of each component to ensure essary to treat some WDM channels in great detail while only
the successful operation of the system as a whole. However, this considering the effect of other channels on the channels under
process would lead to overly conservative design, which is not consideration. A further example, it is the separate treatment of
sustainable in a highly competitive industry. An attractive al- signals and noise: the signal channels may occupy far less band-
ternative is to employ computer-aided design and optimization width than the noise from, say, an erbium-doped fiber amplifier
to photonic systems and to replace the hardware prototype with (EDFA), but the noise can saturate other amplifiers or produce
software simulations. This replacement brings with it several electrical noise on detection.
advantages, not forgetting the ease of communicating and doc- The key to developing a third-generation simulator, opposed
umenting software simulations. to a solitary model, is to provide a flexible data interface repre-
This paper discusses the design philosophy that led to the de- sentation between the modules [14]. Each module can represent
velopment of a sophisticated photonic design automation (PDA) a component or subsystem, but the key to a powerful and fu-
product [12], which is based on many tens of years of original ture-proof simulator is the ability for many modules to interact,
research. The importance of having a wide range of signal rep- providing novel solutions, or highlighting potential pitfalls in a
resentations is discussed in Section II. The provision of a range design.
of models from abstract to physical is discussed in Section III.
With this in mind, we have developed a flexible basis for
Examples of systems and network simulation are given in Sec-
treating signals and noise for our simulator photonic transmis-
tion IV.
sion design suite (PTDS). PTDS is based on the Ptolemy simu-
lation engine [15], with a proprietary graphical user interface
II. SIGNAL REPRESENTATIONS FOR INTERCONNECTING MODELS and proprietary signal representations. Furthermore, we have
developed an extensive library of optical and electronic mod-
Photonic simulation is not new: over the years, many re-
ules, covering many levels of abstraction. Ptolemy gives sophis-
searchers, scientists, and engineers have developed numerical
ticated control of the sequencing of modules during a simulation
and semi-analytical models to solve particular problems.
and provides a large library of communications and signal pro-
Groups of engineers have also worked on simulators for sys-
cessing models. Its tcl scripting language [16] allows parame-
tems, for large design projects, such as transoceanic systems.
ters to be specified as functions of higher level parameters or as
What is new, however, is the recent emergence of commercial
random variables, which gives several powerful features as fol-
software for photonic simulation: first-generation commercial
lows.
software focused on specific design problems, such as inte-
grated optics and wave propagation. Second-generation tools • Parameters can be made functions of global variables,
allowed systems or components to be simulated using a single such as a global filter bandwidth.
signal representation or simulation paradigm [13]. Third-gen- • Parameters can include any form of temperature sensi-
eration tools provide flexible platforms for modeling at many tivity.
284 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

Fig. 2. Block and sample modes of simulation, showing unidirectional and bidirectional propagation and the firing sequence of modules.

• Parameters can be swept (using any functional form, from allowing a simulation to be partitioned spectrally into appro-
a central control) to analyze sensitivities. priate signal representations as follows.
• Parameters can be optimized automatically using itera- • Sampled optical field signals, which contain full infor-
tion. mation from which optical and detected waveforms and
Two modes of simulation exist in PTDS: sample mode spectra can be reconstructed. A single frequency band
and block mode. Sample mode is for bidirectional simu- (SFB) can be used to cover all data channels (so that full
lation of closely coupled components, similar to that used interactions are calculated), or these can be represented in-
in Optoelectronic, Photonic and Advanced Laser Simulator dividually using multiple frequency bands (MFB’s), each
(OPALS) [11], but with a complex envelope signal represen- with a center frequency and each covering one or more
tation for phase accuracy over the whole optical bandwidth. channels. MFB’s, thus, can save on memory and compu-
Block Mode passes data as arrays (blocks) of the complex tation when large unused gaps are in the spectrum.
envelope of the optical field, restricting bidirectionality to • Statistical signals carrying average and deviations over
components spaced by more than a block length, such as the time-window of the block. Noise Bins (NB’s) repre-
optical switches separated by fibers, or to within a modules, sent broad noise spectra efficiently as a mean power spec-
such as in filters. The iteration schemes for block mode and tral density within a defined frequency range. NB’s are
sample mode are shown in Fig. 2. In block mode, the simu- effective for the amplified spontaneous emission (ASE)
lation progresses module by module. Usually, the module is in an optical amplifier. Parameterized signals represent
run only once, with one block propagating from transmitter continuous wave (CW) signals or defined pulse shapes
to receiver. However, multiple iteration can be performed, with mean power and jitter characteristics. They are useful
particularly if the system undergoes state changes, such as for signal-to-noise calculations and to represent pumps or
optical switching. The data within the blocks can be con- saturating signals in amplified systems. Noise generated
sidered to be periodic or aperiodic. In the aperiodic case, within the spectral range of SFB or MFB signals can ei-
the models remember their state from run to run, and linear ther be added to these signals or propagated separately as
convolution is performed in all filters. In periodic mode, the NB’s.
data within each block is considered to be independent, and
circular convolution is used in the models. In addition, PTDS passes logical information along a system,
In sample mode, modules communicate bidirectionally which can be used to identify the transmitter in a switched
during iteration to simulate complex interactions and reso- system, the modulation sequence, center frequency, and pulse
nances between the components. Thus, every module must be shape (if applicable). Logical information is used in some
fired to provide up-to-date information to its neighbors. Sample forms of bit error rate (BER) estimation to compare transmitted
mode allows complex devices to be constructed from primitive and received sequences. BER’s are estimated as follows:
components, such as mirrors, delays, gratings, and active re- • fitting distribution functions to received bit sequences, in-
gion. It has been applied to many modeling problems, including cluding noise, after they have been grouped into pattern se-
high-speed, single-mode, Bragg-grating, stabilized and tunable quences to isolate deterministic intersymbol interference
lasers, picosecond pulse sources, clock regenerators, optical from the stochastic noise [18];
filter designs, and many more [17]. • propagating noise and signal separately (using SFB/MFB
Sample mode has a single signal representation, covering all and NB’s) so that the noise statistics are presented de-
simulated optical frequencies and commonly assuming a single terministically to the receiver model [19]. This process
polarization. Block mode has both sampled and statistical sig- neglects the interaction of noise and signal in nonlinear
nals, containing polarization information and center frequency, fibers, but it is deterministic.
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 285

Fig. 3. Tree of simulation modes (sample, block) and signal representations. In

block mode, the spectrum can be covered by four different signal representations
for efficiency.

A. Conversion Between Signal Representations

Fig. 3 showed how the simulated spectrum can be divided into
different block-mode signal representations according to optical
frequency. A simulation can also be divided into different signal
representations along its length, which implies conversion be-
Fig. 4. Changing signal representations along a simulation. (a) Optical
tween representations along the signal path. This conversion amplifier modeling with parameterized signals to represent pumps and (b)
can be done automatically or can be forced, using “nonphys- systems modeling with noise added to an SFB before nonlinearity calculations.
ical” modules. Furthermore, sampling rates can be changed, for
example:
signal representation is controlled by the source modules, and
• to increase the simulation bandwidth, to accommodate it can be changed automatically (for example, when overlap-
four-wave mixing products during a nonlinear optical fiber ping SFB’s are combined in a multiplexer, they become a single
simulation; SFB). Conversion modules are also provided between signal
• to reduce data size when the optical or electrical band- representations, including between block and sample modes.
widths are reduced by filtering. Global parameters can be used to choose signal representation,
As an example of changing the signal representations along allowing coarse first-cut simulations, followed by detailed simu-
a system, Fig. 4(a) shows an optical amplifier schematic, with lations. Also, network simulations tend to use the more abstract
the signal representations annotated. The transmitters produce signal representations, whereas component modeling requires
SFB’s, each with a distinct carrier frequency. When multiplexed the sample mode to represent the interactions between closely
together, the SFB’s become an MFB (a group of SFB’s), al- spaced devices.
though they will combine into a single band if their carriers
overlap or if forced to. The pump laser adds a parameterized
III. MODEL ABSTRACTION
signal (PS), which feeds into a length of doped fiber. This pro-
duces wideband ASE in the form of NB. Noise within the sam- The simulation of photonic networks covers many scales
pled bands can be added to the bands or propagated separately. of problem, from the details of the dynamics of quantum
Fig. 4(b) shows signal propagating through an amplified fiber wells to interaction in fibers within global networks. It is
system. Again, four SFB’s are combined at the WDM coupler therefore impossible to model a complete system on the scale
to form an MFB, and the EDFA puts all noise into NB’s. In of its smallest component; however, it is possible to vary the
order to calculate the full interaction between all of the channels scale of the simulation from component to component. We
and the carriers and the noise, the fiber model first converts the have adopted a range of models for all but the most trivial
MFB’s and NB’s (within the MFB spectral range) to a single- of components. For example, our laser models range from
sampled band (SFB). This conversion allows for full nonlinear CW sources with linewidth, through pulsed laser models, to
interaction between all signal channels and all noise within the single-mode-rate equations, to multisection wide-spectrum,
signal sampled bands. The NB’s outside the signal band will large-signal transmission-line laser models (TLLM’s). Our
continue to propagate along the system. optical amplifier models are described by simple measured
The above examples show the spatial and spectral mapping parameters, such as gain and noise, through frequency- and
of signal representations onto a system simulation. The type of power-dependent external measurements, to full forward
286 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

TABLE I
COMPARISON OF SEMICONDUCTOR LASER MODELS

and backward simulation of an amplifier built from pumps,

doped-fiber, and passive components. Our fiber models range
from simple delays to frequency decomposition methods op-
erating simultaneously on four signal representations, through
split-step Fourier methods, to fast semi-analytical methods for
ultrafast TDM/WDM systems.
A feature of PTDS is that most models select their algorithms
automatically, depending on the signal representations they are
given as inputs. Thus, each model contains a wide-range of
abstractions. Where appropriate, the interactions between dif-
ferent representations will be considered. For example, an op-
tical amplifier model made from individual components will
Fig. 5. Bragg grating, stabilized transmitter schematic, using sample mode to
process statistical representations of pumps and noise, together pass signals bidirectionally between two closely spaced components and block
with multiple signals representing individual WDM channels. mode for the remainder of the simulation.
Examples are given below.
multicavity lasers, such as grating stabilized lasers and tunable
A. Optical Sources lasers. An example of a Bragg-Grating, stabilized laser design
The performance of the optical source can have a profound modeled in sample mode is given in Fig. 5 and is discussed in
impact on the performance of a system. For example, it is detail later.
well known that chirp in directly modulated lasers causes
significant pulse broadening in dispersive fibers [20]. External B. Optical Fibers
modulators can be designed or driven to have zero chirp, or to Although the Kerr nonlinearity in optical fibers is small,
have an optimized chirp. We have laser models from an abstract the use of extremely long fiber links, operated at high powers,
pure-sine wave at one end of the scale, to a full longitudinally means that the effect of the nonlinearity can be large and
inhomogneous model at the other [21]. In between, the models becomes a limiting factor in WDM systems. Nonlinearity leads
assume single-modedness and homegeneity. The range of to self-phase modulation within a channel, giving pulse shaping
models is shown in Table I. and the possibility of soliton systems. In WDM systems, it
Note that dual mode lasers can be formed using two leads to crosstalk between channels and timing jitter caused by
single-mode models to enable the effect of a single side cross-phase modulation. Our fiber models are mostly based on
mode on a system to be assessed. Furthermore, complex and the split-step method, in which the fiber is divided into sections.
novel laser designs can be studied by interconnecting separate Within each section, the effects of dispersion and nonlinearity
sample mode laser models to form multicontact, multisection, are treated separately [22]. The dispersion is treated in the
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 287

frequency domain as a frequency-dependent phase shift, and The range of fiber models, at the time of writing, is summa-
the nonlinearity in the time domain, as a phase shift dependent rized in Table II. It should be noted that the flexible signal rep-
on instantaneous power. The step length is adaptable to give a resentations in PTDS gives the ability to model at many degrees
maximum phase shift per step. Split-step models are provided of abstraction and to include proprietary code using Matlab,
for aperiodic or periodic boundary conditions. Python, or C code. This option is useful for researchers and en-
For generality, all signal representations are converted into gineers working on specialist applications. Note the inclusion of
a single sampled signal, covering the whole wavelength range a bidirectional fiber model, which is simply a time delay. This
(Fiber NLS module). This conversion treats all interactions be- model is useful for constructing photonic circuits, such as filter
tween the WDM channels. However, the independent channels, networks, ring resonators, and mode-locked lasers.
represented as MFB’s can be calculated separately if the effects
of four-wave mixing between the bands are negligible. This cal- C. Optical Amplifiers
culation can be useful for simulating the degradation of the cen-
tral channels in a system because of FWM, without considering Optical amplifiers can be treated with many degrees of ab-
the minimal effect of the channels well away from those under straction, as shown in Table III. The simplest of models assume
consideration. The remaining channels are propagated as PS, flat gain, whereas blackbox [28] models interpolate the gain
so that they can saturate the gain of amplifiers along the link, spectrum from two measured spectra at two saturation powers,
and Raman effects can be quickly estimated using parameter- and a input–output saturation curve. The parameters for our
ized signals and semi-analytical techniques. blackbox model can also be precalculated using a detailed inho-
The NLS Frequency-decomposition module allows control mogenous “Giles” EDFA model [29], perhaps of a multistage,
of the modeling of nonlinear interactions between different fre- multiply pumped amplifier, based on measurements of the gain
quency types. This module is useful for identifying the cause of and absorption cross sections of the fiber. We have also imple-
degradation in a system. Interactions (excluding FWM) between mented a dynamic EDFA model based on [30] for millisecond
PS, MFB’s, and NB’s can be controlled. In the general case, the transients in systems.
contents of NB’s and MFB’s can be converted into an SFB at the Semiconductor optical amplifiers are modeled using rate
beginning of the fiber to give all interactions. Propagating the equations (assuming constant carrier density, implying an
noise independently of the signal to allows fast signal-to-noise exponential power growth) [31], or longitudinally discretized
analysis (though interactions between the noise and the signal models with full dynamics using the TLLM [32]. Most of
are neglected, for example, modulation instability [23]). the amplifier models operate in block mode, except for the
For estimating the effect of polarization dispersion, the TLLM, which is sample mode. It is impractical to formulate
Random Birefringence PMD module propagates two polariza- EDFA models with gain saturation in sample mode, as the
tions represented by coupled, nonlinear Schrödinger equations. average power in a signal would have to be obtained from a
At each step of the split-step algorithm, the polarizations long average of the signal. In block mode, the contents of the
are scattered randomly on a Poincaré sphere, with a uniform block represent the signal over all time, as it is assumed by
distribution of polarizations [24]. This distribution will give the amplifier to be periodic. This signal allows the state of
an increase in pulse spreading, which tends to be proportional saturation to be calculated from the input signal. An example of
to the square-root of the propagation distance. The worst-case using blackbox amplifiers to equalize the signal-to-noise of a
PMD can also be calculated by turning the random scattering WDM signal propagating through a chain of saturated EDFA’s
off. is given later. Here, PS and NB’s are used for efficiency.
Future optical links and networks with speeds of 10 Gb/s
and beyond are likely to be based on return-to-zero coding D. Optical Filters
schemes because of their advantageous interplay between dis-
persion and fiber nonlinearities [25]. Here, two physical ef- The performance of optical filters will become more critical
fects mainly determine the transmission performance. First, as WDM channel spacing becomes denser and the bit rate per
severe pulse-shape deviations in time and amplitude deveop channel is increased. This process will require the evaluation of
from the impact of ASE noise introduced by optical inline filter designs in systems models, as the filter’s impulse response
amplifiers. Neglecting the nonlinear impact of noise onto will dramatically affect intersymbol interference as the ratio of
signal propagation, pulse degradation caused by ASE noise filter bandwidth to data rate is reduced [33].
can be derived analytically for any arbitrary-chirped optical Optical filters can be modeled from using ideal filter forms,
pulse [26]. Second, nonbalanced frequency shifts caused by measured characteristics, or using sample-mode (time-domain)
interchannel pulse collisions in WDM transmission systems models of filter lattices. Bragg gratings are modeled either
result in additional timing jitter. Using the approach of elastic from a frequency-domain transfer-matrix analysis [34] or a
collisions [27], an expression for the timing jitter can be time-domain scattering-matrix analysis based on the TLLM.
found for any arbitrary-chirped optical pulse, provided that These analyses give identical results, but the frequency do-
the main energy of a pulse stays within a bit slot. These main models have more sophisticated design rules to allow
approximations are the basis of efficient semi-analytical es- dispersion compensation or bandwidth to be specified directly.
timation techniques used in PTDS. Compared with split-step Also, the frequency-domain model will operate with periodic
methods, these modules achieve a reduction in computational boundary conditions, allowing long impulse responses to
time of two orders of magnitude. be wrapped-around. This model is useful when modeling
288 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

TABLE II
COMPARISON OF OPTICAL FIBER MODELS

dispersion compensation in which the walk-off of the pulses is experimental results and as part of European-wide
far longer than the modeled sequence. projects, including the COST-240 project on measuring
Most filter modules operate on MFB/SFB signals, samples and modeling advanced photonic telecommunications
signals and NB’s. NB’s offer an efficient way of determining the devices and the ACTS DEMON project.
response of a network by exciting the network with white optical • Cross-checking numerical methods: PTDS contains two
noise (which is deterministic in the NB representation). Alter- dynamic laser models compared in a simulation example,
natively, testing with an impulse in SFB/MFB/sample mode and and several fiber models, all of which have been cross
using a Fourier transform will reveal the spectral response of the checked to prove their ranges of applicability.
network, including its group delay and phase characteristics. • Amplifier and some laser models allow a choice of numer-
ical techniques, with specified accuracy. Other models are
E. Simulation Accuracy based on techniques whose accuracy scales with compu-
tational effort (for example, the TLLM is based on phys-
It is important to be able to build a level of trust in the results ical equivalent circuit analog to the laser, whose inaccu-
of simulations. This trust has been obtained as follows. racies are presented as well-understood “parasitics.” Run-
• Comparing with other numerical models: PTDS has ning at two different sampling rates identifies inaccuracies
been developed from earlier products, such as BroadNeD and their worst-case magnitude.)
(BNeD GmbH), GOLD, and OPALS (Virtual Photonics • Standard regression tests are regularly and automatically
Pty Ltd.), and models at HHI (Germany), the Australian run on the software to detect compilation errors. These
Photonics CRC, and at our partner universities. This tests are based on analytical results, where available.
development has allowed extensive checking against • Comparison with published work: when developing appli-
independently developed numerical models. OPALS, cations examples, PTDS results are compared with exper-
GOLD, and BroadNeD were themselves tested against imental, numerical, and analytical published work.
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 289

TABLE III
COMPARISON OF OPTICAL AMPLIFIER MODELS

• Customer acceptance: Virtual Photonics, Inc. (VPI) has • sample mode for transmitter (laser) design;
over 100 customers, may of whom have compared re- • PS for jitter estimation in long-haul RZ systems;
sults from PTDS with their own numerical models before • combined PS and NB’s for iterative signal-to-noise opti-
making a purchasing decision. mization in an amplified WDM system;
The propagation of errors along a system can be checked by • SFB for dispersion map planning in a TDM system;
monitoring waveforms, spectra, and power along a simulation, • PS and MFB’s for assessing the performance and crosstalk
which is an excellent way to test numerical validity. For ex- in wavelength-converting cross connects;
ample, the optical spectrum shows the results of nonlinear in- • a comparison between split-step (SFB) and frequency-de-
teractions of carriers, and it is easy to see if these fall within composition (MFB) fiber models for modeling short-pulse
the simulated bandwidth (indicating a valid simulation band- interaction caused by cross-phase modulation.
width), and whether they are expected frequencies or are spec-
trally broadened. Each component in a simulation can be made A. Semiconductor Laser Design (Sample Mode)
active or inactive to identify its effect. For long-haul communications, the goals for semiconductor
laser design include the folling:
IV. EXAMPLE APPLICATIONS • high output power;
• single-mode spectrum, with better than 35-dB difference
Hundreds of different designs and proprietary techniques are between the power in the main mode and a side mode;
in photonics, and from our experience, PTDS is helpful in most • low-intensity noise, especially for analog or high-bit-rate
cases to achieve greater understanding of an individual device, systems;
the performance of a device in a system, and the optimization • narrow spectral width under direct modulation (chirp);
of a system overall. • tunability, if possible;
The following applications have been chosen to be illustra- • fast modulation response, if directly modulated, with low
tive of the range of problems that can be solved with PTDS. overshoot;
These examples do not include standard solutions of fiber non- • low threshold current and high efficiency;
linearity, as these are well covered elsewhere; however, they do • temperature insensitivity.
illustrate the power of the signal representations in speeding a Simulation using sophisticated models can be used to de-
design process. The examples are as follows: sign lasers with optimized characteristics to design novel
290 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

Fig. 7. Schematic of a multihop WDM system, in which the input powers are
iteratively optimized to give equal channel SNR’s.

Fig. 6. Unmodulated optical spectrum of the Bragg grating, stabilized laser,

showing the Bragg cavity modes within a dominant supermode, and the laser
chip modes spaced at approximately 80 GHz.

lasers for specialist applications, or to identify the causes

of performance imperfections in real devices. For these pur-
poses, we have enhanced the TLLM [21], so that it can sim-
ulate over the broad spectral ranges required in WDM sys-
tems. The TLLM divides the laser into longitudinal sections,
and then it propagate samples of the optical field between
these sections, modifying the samples to represent stimu-
lated and spontaneous emission, attenuation, reflections, and
phase changes. The output of the module is a series of sam-
ples representing the optical waveform. All resonances of the
cavity (including those from external components) are solved
in the time domain, and the lasing spectrum can be found by
Fourier transformation of the samples. The waveform also
includes the dynamics of the laser, because the electronic
processes are included into the laser model as rate equations.
The TLLM operates in sample mode, so that external
components can interact with the laser by passing samples of
the optical field back to the laser model at each iteration. This
process allows complex lasers (tunable, multisection, multicon-
Fig. 8. SNR through five iterations of optimization, leading to equal signal to
tact, integrated mode-locked) to be built from interconnected noise ratios.
models, and photonic circuits with active elements (wavelength
converters, clock regenerators, limiters, photonic switches) to
be simulated. A wide range of filters, couplers, delays, phase
shifters, and modulators also operate in sample mode, allowing
novel circuits to be designed. B. 10-Gb/s Amplified WDM System Signal-to-Noise
As an example of the application of sample mode to circuit Optimization (NB’s and PS)
design, a semiconductor laser stabilized by a Bragg grating The optimum information-carrying performance of a
is simulated. The schematic is shown in Fig. 5 and com- long-haul saturated amplifier link is obtained when the SNR’s
prises a laser module connected to a time-domain model is equalized over all channels. Fig. 7 shows the schematic of
of a Bragg grating (also based on TLLM techniques). The a 16-channel WDM system with a chain of six amplifiers,
1-cm Bragg grating reflects over a narrow stop-band, se- three sections of dispersion-compensating fiber (DCF), and
lecting one of the weak modes of the imperfectly (2%) an- two spans of single-mode fiber (SMF). The amplifiers have no
tireflection-coated laser chip. The output of the laser is con- gain equalization, so they suffer from a large spectral ripple.
verted to block mode for efficient unidirectional propagation The object of this simulation is to optimize the input spectrum
along the remainder of the system. Unfortunately, because of the chain so that each WDM channel has the same SNR at
of the long length of the compound cavity, several of the the output of the chain, which gives the maximum information
compound cavity’s modes will lase, forming a supermode, capacity for the link, but it is difficult to calculate as the gain
as shown in Fig. 6. Weak resonances of the laser chip, be- spectrum of the amplifiers depends on their input powers.
cause of imperfect antireflection coating, are also present. Thus, a self-consistent solution must be found iteratively.
Many design parameters can be investigated using this sim- This process can be performed using a simple optimization
ulation because of the close relationship between the model loop, automatically included in the simulation using Ptolemy
topology and the real device. scripting language tcl.
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 291

For efficiency, the eye diagrams of each channel are not cal-
culated during the optimization process. Rather, PS are used to
represent the mean power in a WDM channel over a data se-
quence, and the NB’s can be used to represent the noise in and
around each channel. The EDFA models are able to calculate the
saturation of the amplifier using a blackbox model [28], hence,
the amplifier’s gain spectrum from the input signals and noise.
This model uses a simple, single-saturating wavelength mea-
surement of an amplifier’s gain to predict the gain for any set of
input wavelengths and powers. Experimentally, we have shown
excellent (within 0.5 dB) predictions of the gain of fully loaded
WDM spectrum for a commercial amplifier [35], [36].
The output SNR’s of the 16-channels, for the SNR optimiza-
tion, for each iteration step are shown in Fig. 8. These channels
converge in a few iterations. If the gain spectrum of the am-
plifiers were independent of the input power, the convergence
would occur in an iteration step. The converged output spectrum
is shown in Fig. 9. This figure shows a constant SNR (the PS are
equal ratios above the NB’s) for channels. Note that the NB’s
represent the noise within a 39-MHz range, whereas the SNR is
calculated for a 0.1-nm bandwidth receiver, so the optical spec-
trum analyzer (OSA) display’s SNR appears larger than it ac-
tually is. Also, the widths of the NB’s have been automatically Fig. 9. Output spectrum after equalization for SNR. Note the noise bins (bars)
reduced around the ASE peak to maintain amplitude accuracy. are used to represent the ASE noise, whereas parameterized signals (arrows)
This feature is designed to increase efficiency by optimizing the represent the mean channel powers.
number of NB’s covering the spectrum.
Once the SNR has been optimized, it is a simple task to switch
the transmitters to give SFB signals so that the eye diagrams and
bit-error rates (BER’s) of the channels can be assessed. Simi-
larly, multiple sweeps of the system can be performed, for ex-
ample, to assess the performance of the system with one or more
channels disabled.

C. 10-Gb/s Long-Haul System Design (Single Frequency

Band)
The positioning of optical amplifiers in a long-haul system
is a nontrivial problem because of fiber nonlinearities and the Fig. 10. Multihop dispersion-compensated system in which the input powers
interplay between nonlinearities and dispersion. Amplifiers to the dispersion-compensating fiber (DCF) and the single-mode fiber (SMF)
are adjusted for maximum transmission distance in number of spans.
may be placed before sections of dispersive (single-mode,
SMF) fiber, before sections of dispersion-compensating fiber
(DCF), or both. The amplifier power will affect signal-to-noise,
but less obviously, the shaping of the pulses by nonlinearities.
The design is also affected by existing plant, such as installed and the optimum amplifier powers. The numbers of spans
fiber types, position of regenerator stations, and so on. that can be covered are plotted as contours, for initial SMF
Fig. 10 shows a 10-Gb/s single-channel system to be lengths of 10 and 30 km. The 30-km system can operate over
optimized that includes alternate 80-km sections of single 60 spans for a of 6 by increasing the input power to the SMF,
mode (SMF) and DCF to give 99.5% compensation, which compared with the 10-km SMF case, which can only operate
was found to be optimum. The parameters for the fibers are over 48 spans and requires lower amplifier output powers.
given in Table IV. The transmitter is a zero-chirp external Similar results, including experimental conformations using
modulator, and the amplifiers include 1-nm filters. The receiver recirculating loop experiments, have recently been presented
was assumed not to affect performance. The 128-bit sequences in [37].
were simulated. Interestingly, the system has an initial length of
SMF and the ability to set the output powers of the amplifiers. D. Long-Haul WDM Return-to-Zero (RZ) Design—Estimation
The design problem is to find the optimum amplifier output of Timing Jitter (PS)
powers, and the best initial length of SM fiber to give the Accumulated timing jitter due to interchannel pulse collisions
maximum transmission distance (that is, the maximum number and ASE-noise becomes the system limiting factor for RZ prop-
of DCF–SMF spans). As Figs. 11 and 12 show, the initial length agation over long-haul WDM links with bit rates of 10 Gb/s and
of SMF has a profound effect on the performance of the system, beyond. This example illustrates semi-analytical techniques for
292 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

TABLE IV
FIBER PARAMETERS IN LONG-HAUL SIMULATION

Fig. 11. Results of multiple simulations to determine the optimum input

powers to the DCF and SMF. The labeled contours represent the number of Fig. 13. Accumulated timing jitter of RZ propagation over an optically
hops that can be achieved for a particular combination of powers. The chart is amplified WDM link at 10 Gb/s, using two modeling techniques.
for an initial length of SMF of 10 km.

calculating timing jitter. These techniques increase the compu-

tational efficiency by about two orders of magnitude compared
with split-step simulations. Our example is a 10-channel WDM
transmission system, using Gaussian pulses of 16.75 ps width at
10 Gb/s, and a dispersion managed fiber link. The length of the
symmetrical dispersion map is 200 km; the average dispersion
is 0.078 ps/nm-km. The amplifier spacing is set to 50 km, and
each amplifier operates with a noise figure of 6.34 dB.
Fig. 13 shows the simulation results from the semi-analytical
model in split-step models for collision-induced and ASE-in-
duced jitter. The jitter for the split-step methods is estimated
from 100 simulations by averaging the pulse time with respect
to the same statistical propagation properties. Note that the mod-
ules performing the semi-analytical estimation techniques are
operating with PS, and therefore pass data as modules as av-
erage pulse shapes and jitter values. This example shows that
Fig. 12. Results of multiple simulations to determine the optimum input PS are efficient for optimizing long haul links with respect to
powers to the DCF and SMF. The chart is for an initial length of SMF of 30
km. Note the increased transmission distance that can be obtained over a short amplifier spacing and positioning and to the applied dispersion
initial length of SMF. map.
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 293

Fig. 14. Schematic to test the crosstalk performance of a wavelength-interchange, optical cross-connect, with two inputs each carrying four WDM channels.
AWGM’s are used to multiplex and demultiplex the test channels, and an text viewer (scroll icon) can display the signal powers in all possible signal paths for
crosstalk analysis.

Fig. 15. Intenal configuration of the optical cross connect of Fig. 13, showing AWGM’s for demultiplexing the WDM input channels, followed by a 8 2 8 space
switch feeding into eight arbitrary, input-frequency, fixed-output frequency wavelength converters. Two AWGM’s multiplex the outputs to two ports.

E. Crosstalk in WDM Network Design (MFB’s and PS) nonblocking, wavelength converting (using cross-gain,
cross-phase, four-wave mixing, and optoelectronic technolo-
The increase in used bandwidth of optical fibers requires gies). In our example, we investigate the performance of an
a similar increase in the capacity of interconnects. Photonic optical cross connect with two fiber inputs, each carrying four
switching gives the possibility of building large-capacity WDM channels (Fig. 14). The outputs are demultiplexed using
switches. However, photonic switches may not offer the arrayed-waveguide demultiplexers (AWG) [39]. The switch
regeneration that is implicit in electronic switches, although itself (Fig. 15) comprises AWG demultiplexers, an
wavelength converters offer some regeneration because of space-switch (made from distributors and collec-
their nonlinearity. Photonic simulation can be used to assess tors), and eight fixed-output-frequency wavelength converters.
the performance of optical cross connects within systems. Of The eight outputs are remultiplexed using AWG’s to two output
particular interest is optical crosstalk, which can severely limit ports.
the number of optical interconnects in a system [38]. Many Fig. 16 shows the output spectra of the output of the top
different technologies can be compared, including blocking, AWGM, created using MFB (sampled) signals. Ideally, one
294 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

dominant channel per frequency should exist. However, the ef-

fects of the space-switch crosstalk, imperfectly demultiplexing
filters, imperfect AWGM filtering, imperfect wavelength
conversion degrade the channels. Turning off transmitters
or converters, or globally setting filter parameters, crosstalk
amplitudes and phases, can identify these effects individually.
To investigate coherent crosstalk, the simulation can be driven
through a number of phase states using swept parameters, or
random phase parameters. Fig. 17 shows the eye diagram of one
switched and converted channel. This figure has slow-leading
edges because of the transient response of the cross-phase
wavelength converters and large fluctuations because of
crosstalk in the wavelength converters because of imperfect
input filtering.
The simulation can also be globally switched to use parame-
terized signals. The parameterized signal split at every coupler
to form new parameterized signals, which can be monitored on
spectrum analyzers to estimate optical crosstalk, or presented
as a text list of all signals, including frequencies and powers for
further analysis using analytical crosstalk estimates for multi-
Fig. 16. Spectra of all of the outputs of the top AWGM of the optical cross
path propagation. Fig. 18 shows the output at one fiber of an connect simulated using MFB signals. Each channel represents one output of
AWGM demultiplexer. Note the large number of PS (arrows) the AWGM.
caused by the large number of crosstalk paths in the network.
Also, the wavelength converters generate MFB signals. This ex-
ample shows how the performance of a device in a subsystem
can potentially affect a large network.

F. Interaction of Solitons in Nonlinear Dispersive Fibers

(MFB versus SFB)
Solitons at two different wavelengths will walk through each
other as they propagate along a dispersive fiber, because of their
different group velocities. As they pass through each other, they
will modulate each others’ phases, via the nonlinear index of
the fiber, whose slowly varying term depends on the sum of the
powers in both waves. This process will cause frequency shifts
in the pulses.
Soliton interaction can be modeled in two ways in PTDS as
follows:
• by using the split-step Fourier method of Fiber_NLS
acting on the combined fields of the two pulses within an
SFB;
• by using the frequency-decomposition method in
Fig. 17. Waveform of one switched and wavelength-converted channel
FiberNLS_FD acting on individual fields represented in (from top fiber WDM channel 1 to bottom fiber WDM channel 4) showing
MFB’s. This method generally is much more numerically slow-leading edges and large fluctuations caused by crosstalk.
efficient.
Fig. 19 shows the spectrum of a 2-mW 300-ps pulse cal- range of numerical modules representing photonic devices
culated using the two methods, when a 20-mW pulse walks and subsystems. The multiple signal representations allow
through it in a dispersive nonlinear fiber. Both spectra are simulation at the optimum abstraction level for a problem.
dynamically broadened by cross-phase modulation, and the This simulation allows a design to be “roughed-out” using
agreement between the two methods is excellent. The saving abstract signal representations, and then simulated thoroughly
is computation by using the frequency-decomposition method using detailed signal representations. The problem can also be
is a factor of 21. partitioned spectrally, with abstract signal representations for
noise and channels of little interest, or partitioned spatially,
with subsystems being represented in more detail than the
V. CONCLUSION
remainder of the network. We believe that our multirepresenta-
We have developed a flexible framework for photonic de- tion approach offers a future-proof platform for physical layer
vices, systems, and networks simulation, together with a wide photonic device, system, and network simulations.
LOWERY et al.: MULTIPLE SIGNAL REPRESENTATION SIMULATION OF PHOTONIC SYSTEMS 295

Fig. 18. Cross-connect simulation using parameterized signal inputs. The large number of parameterized signals (arrows) is caused by the large number of
crosstalk paths in the network.

[4] T. Ono, “Approaching 1 bit/s/Hz spectral efficiency,” in Tech. Digest

OFC’99, San Diego, CA, Feb. 23–26, 1999, Paper FE4, pp. 410–411.
[5] A. Djupsjöbacka, “Prechirped duobinary modulation,” IEEE Photon.
Technol. Lett., vol. 10, pp. 1159–1161, Aug. 1998.
[6] G. H. Smith, D. Novak, and A. Ahmed, “Technique for optical SSB gen-
eration to overcome dispersion penalties in fiber-radio systems,” Elec-
tron. Lett., vol. 33, pp. 74–75, 1997.
[7] Y. Shen, K. Lu, and W. Gu, “Coherent and incoherent crosstalk in WDM
optical networks,” J. Lightwave Technol., vol. 17, pp. 759–764, May
1999.
[8] C. X. Yu, W.-K. Wang, and S. D. Brorson, “System degradation due
to multipath coherent crosstalk in WDM network nodes,” J. Lightwave
Technol., vol. 16, pp. 1380–1386, Aug. 1998.
[9] J. Nagal, “Dynamic behavior of amplified systems,” in Tech. Dig.
OFC’99, San Diego, CA, Feb. 23–26, 1999, Paper Th03, pp. 319–320.
[10] M. F. Mendez and M. A. Ali, “Simulation of 64 2 2.5 Gbit/s WDM
transmission over 1056 km standard single-mode fiber using gain-flat-
tened silica-based EDFAs,” IEEE Photon. Technol. Lett., vol. 10, pp.
300–302, Feb. 1998.
[11] H. Hamster and J. Lam, “PDA: Challenges for an emerging industry,”
in Lightwave. New York: Penwell, Aug. 1998.
[12] PTDS. Virtual Photonics Inc., Berlin, Germany. [Online]. Available:
www.virtualphotonics.com.
Fig. 19. Spectra calculated using (a) frequency decomposition and (b) [13] A. J. Lowery and P. C. R. Gurney, “Two simulators for photonic com-
split-step methods for a 2-mW pulse walking through a 20-mW pulse. puter-aided design,” Appl. Opt., vol. 37, pp. 6066–6077, Sept. 1998.
[14] A. J. Lowery, “Computer-aided photonics design,” IEEE Spectrum, vol.
34, pp. 26–31, Apr. 1997.
ACKNOWLEDGMENT [15] Ptolemy simulation environment. University of California, Berkeley.
[Online]. Available: www.ptolemy.eecs.berkeley.edu
The authors would like to thank their development team, cus- [16] J. K. Ousterhout, An Introduction to TCL and TK. Redwood City, CA:
tomers, university partners, and Scientific Advisory Board for Addison-Wesley, 1994.
help in steering the development of the multirepresentation sim- [17] A. J. Lowery, P. C. R. Gurney, X.-H. Wang, L. V. T. Nguyen, Y.-C. Chan,
and M. Premaratne, “Time-domain simulation of photonic devices, cir-
ulation concept. cuits and systems,” in Tech. Digest Phys. and Simul. Optoelectron. De-
vices IV, vol. 2693, San Jose, CA, Jan. 29–Feb. 2 1996, pp. 624–635.
REFERENCES [18] C. J. Anderson and J. A. Lyle, “Technique for evaluating system perfor-
mance using Q in numerical simulation exhibiting intersymbol interfer-
[1] J. Lightwave Technol., May 1988, vol. 6, Special Issue on Factors Af- ence,” Electron. Lett., vol. 30, pp. 71–72, 1994.
fecting Data Transmission Quality. [19] G. Jacobsen, K. Bertilsson, and Z. Xiapin, “WDM transmission system
[2] A. E. Willner, “Mining the optical bandwidth for a terabit per second,” performance: Influence of non-Gaussian detected ASE noise and
IEEE Spectrum, vol. 32, pp. 32–41, Apr. 1997. periodic DEMUX characteristic,” J. Lightwave Technol., vol. 16, pp.
[3] IEEE J. Quantum Electron., Nov. 1998, vol. 34, Feature Issue on Funda- 1804–1812, Oct. 1998.
mental Challenges in Ultrahigh-capacity Optical Fiber Communications [20] I. Kaminow and T. Koch, Optical Fiber Telecommunications III. New
Systems, pp. 2053–2103. York: Academic, 1997.
296 IEEE JOURNAL OF SELECTED TOPICS IN QUANTUM ELECTRONICS, VOL. 6, NO. 2, MARCH/APRIL 2000

[21] A. J. Lowery, A. Keating, and C. N. Murtonen, “Modeling the static

and dynamic behavior of quarter-wave-shifted DFB lasers,” IEEE J. Arthur Lowery (M’92–SM’96), photograph and biography not available at the
Quantum Electron., vol. 28, pp. 1874–1883, Sept. 1992. time of publication.
[22] G. P. Agrawal, Nonlinear Fiber Optics. New York: Academic, 1995.
[23] R. Hui, M. O’Sullivan, A. Robinson, and M. Taylor, “Modulation
instability and its impact in multispan optical amplified IMDD sys-
tems: Theory and experiments,” J. Lightwave Technol., vol. 15, pp.
1071–1082, 1997.
[24] D. Marcuse, C. R. Menyuk, and P. K. A. Wai, “Application of Man- Olaf Lenzmann (S’96–A’98), photograph and biography not available at the
akov-PMD equation to studies of signal propagation in optical fibers time of publication.
with randomly varying birefringence,” J. Lightwave Technol., vol. 15,
pp. 1735–1745, Sept. 1997.
[25] C. Caspar, H.-M. Foisel, A. Gladisch, N. Hanik, F. Kuppers, R. Ludwig,
A. Mattheus, W. Pieper, B. Streubel, and H. G. Weber, “RZ versus NRZ
modulation format for dispersion compensated SMF-based 10-Gb/s
transmission with more than 100-km amplfier spacing,” IEEE Photon. Igor Koltchanov, photograph and biography not available at the time of publi-
Technol. Lett., vol. 11, pp. 481–483, Apr. 1999. cation.
[26] V. S. Grigoryan and C. R. Menyuk, “A new linearization approach for
modeling timing and amplitude jitter in dispersion-managed optical fiber
communications,” in Tech. Digest OFC’99, San Diego, CA, Feb. 23–26,
1999, Paper WM34, pp. 295–297.
[27] A. Richter and V. S. Grigoryan, “Efficient approach to estimate colli-
sion-induced timing jitter in dispersion-managed WDM RZ systems,” in Rudi Moosburger, photograph and biography not available at the time of pub-
Techn. Digest OFC’99, San Diego, CA, Feb. 23–26, 1999, Paper WM33, lication.
pp. 292–294.
[28] J. Burgmeier, A. Cords, R. Marz, C. Schaffer, and B. Strummer, “A
black-box model of EDFA’s in WDM systems,” J. Lightwave Technol.,
vol. 16, pp. 1271–1275, July 1998.
[29] C. R. Giles and E. Desurvire, “Modeling erbium doped fiber amplifiers,”
Ronald Freund, photograph and biography not available at the time of publi-
J. Lightwave Technol., vol. 9, pp. 271–283, Feb. 1991.
cation.
[30] A. Bononi and L. A. Rush, “Doped-fiber amplifier dynamics: A system
perspective,” J. Lightwave Technol., vol. 16, pp. 945–956, 1998.
[31] M. J. Adams, H. J. Westlake, M. J. O’Mahoney, and I. D. Henning, “A
comparison of active and passive bistability in semiconductors,” IEEE
J. Quantum Electron., vol. QE-21, pp. 1498–1501, Sept. 1985.
[32] A. J. Lowery, “New inline wideband dynamic semiconductor laser am- André Richter, photograph and biography not available at the time of publica-
plifier model,” Proc. Inst. Elect. Eng., vol. 135, pp. 242–250, June 1988. tion.
[33] M. M. Garcia and D. Uttamchanani, “Influence of the dynamic response
of the Fabry–Perot filter on the performance of an OFDM-DD network,”
J. Lightwave Technol., vol. 15, pp. 1778–1783, Aug. 1997.
[34] J. Lightwave Technol., Aug. 1997, vol. 15, Special Issue on Fiber Grat-
ings, Photosensitivity and Poling.
[35] D. Breuer and K. Petermann, “System modeling of high bit rate TDM- Stefan Georgi, photograph and biography not available at the time of publica-
and WDM-systems,” in Integrated Photonics Research, Santa Barbara, tion.
July 1999.
[36] R. Damle, R. Freund, and D. Breuer, “Outside-in evaluation of com-
mercial WDM systems,” in Proc. Fiber Opt. Eng. Conf. (NFOEC), FL,
1999.
[37] C. Caspar, R. Freund, N. Hanik, L. Molle, and C. Peucheret, “Using
normalized sections for the design of all optical networks,” in Proc. Opt. Dirk Breuer, photograph and biography not available at the time of publication.
Network Design Model. Conf. (ONDM’2000), Athens, February 2000.
[38] E. L. Goldstein and L. Eskildsen, “Scaling limitations in transparent op-
tical networks due to low-level crosstalk,” IEEE Photon. Tech. Lett., vol.
7, pp. 93–95, 1995.
[39] M. K. Smit and C. van Dam, “PHASAR-based WDM-devices: Princi-
ples, design and applications,” J. Selected Topics Quantum Electron., Herald Hamster, photograph and biography not available at the time of publi-
vol. 2, pp. 236–250, June 1996. cation.